Automatic Modulation Classification by Using Information Theory Based Features


Kutlu A., Yılmaz D.

V. International Applied Statistics Congress (UYIK 2024), İstanbul, Turkey, 21 - 23 May 2024, pp.408-418

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.408-418
  • Gazi University Affiliated: Yes

Abstract

Automatic Modulation Classification (AMC) is to determine the modulation mode of the signal by processing and analyzing the received signal when the prior information of the signal is insufficient. Machine Learning (ML) algorithms have been increasingly used in the field of AMC in recent years. In this study, the performances of traditional classification methods were investigated with the new features that are based on information theory in this field, by using the two types modulated signals obtained from digital and analog modulation schemes in the levels of 0 dB, 10 dB and 20 dB Signal to Noise Ratio (SNR). For this purpose, the RadiOML2018 dataset which includes 24 types of synthetically produced digital and analog modulation was used. In this dataset, there are signal samples recorded from more than 26 SNRs for each modulation type. Unlike the features used in the literature in this field, the new features based on information theory approaches such as entropies, dimensions and recurrence measures were extracted. To examine the performances of different traditional classifying algorithms for AMC classification, Decision Trees (DTs), Support Vector Machines (SVMs), K-Nearest Neighbors (KNNs), Ensemble Classifiers (ECs), Neural Networks (NNs) and Kernel methods were used with calculated features. The results showed that the highest accuracies for 0 dB and 10 dB SNR ratios were reached with the EC as 92.9% and 96%, respectively. For 20 dB SNR, the highest accuracy is found as 99.3% with DT classifier. When these results are compared with the results in other AMC studies reviewed in the literature, it was seen that the used features give useful information and effective for higher classifying accuracies.