Automatic Modulation Classification: Principles, Algorithms and Applications

2Signal Models for Modulation Classification

2.1 Introduction

Signal models are the starting point of every meaningful modulation classification strategy. Algorithms such as likelihood-based (Huang and Polydoros, 1995; Wei and Mendel, 2000; Shi and Karasawa 2012), distribution test-based (Wang and Wang, 2010; Urriza et al., 2011; Zhu, Aslam and Nandi, 2014) and feature-based classifiers (Azzouz and Nandi, 1996; Spooner, 1996; Swami and Sadler, 2000) all require an established signal model to derive the corresponding rules for classification decision making. While some unsupervised machine learning algorithms could function without a reference signal model, the optimization of such algorithms still relies on the knowledge of a known signal model. Meanwhile, as the validation of modulation classifiers are often realized by computer-aided simulation, accurate signal modelling provides meaningful scenarios for evaluating the performance of various modulation classifiers.

The objective of this chapter is to establish some unified signal models for the development of all modulation classifiers from Chapters 3 to 7, and to provide a level ground for the validation of each modulation classifier in Chapter 8. Through the process, the accuracy of the models will be the first priority as it provides credible evidence to aid the design of specific modulation classification strategies for real world applications in Chapters 9 and 10. That, however, is with a fine balance of simplicity in the ...