Analysis of Distribution Test-based and Feature-based Approaches toward Automatic Modulation Classification

Abstract

Automatic modulation classification (AMC) is a signal processing approach to determine a received signal’s modulation scheme without any a priori information about that signal’s configuration. Three main approaches have been proposed for this task, namely Likelihood-based, Feature-based and Distribution Test-based AMC. Each of these approaches has different levels of computational complexity affecting their ability to conduct AMC in real-time and achieving different levels of modulation classification accuracy. Thus, there is a trade-off between computational complexity and classification accuracy in this domain. This paper investigates the real-world scenarios of AMC applications with the emphasis on the two most feasible low-complexity approaches: Feature-based (FB) and Distribution Test-based (DT) AMC. This study has specifically evaluated these approaches with focus on their classification accuracy and their computational complexity. For comprehensive evaluation purposes the study encompassed signals utilizing 8-ASK, 8-FSK, 32-PSK and 32-QAM modulation schemes, as representatives of the most prominent modulation schemes in current wireless systems. Our analysis shows that DT classifiers exhibit on average a 17.5% lower computational complexity than the investigated FB classifier, with about on average 12% higher classification accuracy in SNRs in the range of 0 to 10 dB. On the other hand, the DT approach classifiers show not only on average 31% higher computational complexity than the FB classifier, but it also has an approximately 10% lower classification accuracy for SNRs in the range of -10 to 0 dB.