Hierarchical Classification of Analog and Digital Modulation Schemes Using Higher-Order Statistics and Support Vector Machines

Yalcinkaya B., Coruk R. B., KARA A., Tora H.

Wireless Personal Communications, vol.136, no.2, pp.827-847, 2024 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 136 Issue: 2
  • Publication Date: 2024
  • Doi Number: 10.1007/s11277-024-11285-y
  • Journal Name: Wireless Personal Communications
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, PASCAL, Applied Science & Technology Source, Communication Abstracts, Compendex, INSPEC
  • Page Numbers: pp.827-847
  • Keywords: Analog modulations, Digital modulations, Feature extraction, Hierarchical modulation classification, Machine learning algorithms, Support vector machine
  • Gazi University Affiliated: Yes


Automatic modulation classification (AMC) algorithms are crucial for various military and commercial applications. There have been numerous AMC algorithms reported in the literature, most of which focus on synthetic signals with a limited number of modulation types having distinctive constellations. The efficient classification of high-order modulation schemes under real propagation effects using models with low complexity still remains difficult. In this paper, employing quadratic SVM, a feature-based hierarchical classification method is proposed to accurately classify especially higher-order modulation schemes and its performance is investigated using over the air (OTA) collected data. Statistical features, higher-order moments, and higher-order cumulants are utilized as features. Then, the performances of some well-known classifiers are evaluated, and the classifier presenting the best performance is employed in the proposed hierarchical classification model. An OTA dataset containing 17 analog and digital modulation schemes is used to assess the performance of the proposed classification model. With the proposed hierarchical classification algorithm, a significant improvement has been achieved, especially in higher-order modulation schemes. The overall accuracy with the proposed hierarchical structure is 96% after 5 dB signal-to-noise ratio value, approximately a 10% increase is achieved compared to the traditional classification algorithm.