Automatic Modulation Classification: Principles, Algorithms and Applications

6Machine Learning for Modulation Classification

6.1 Introduction

In Chapter 5 we list a collection of signal features for modulation classification. Some of the classification decision making is based on multi-stage decision trees where each stage utilizes a different feature. However, the need for designing the decision tree and optimization of multiple decision thresholds is not most convenient.

To overcome these problems, various machine learning techniques have been employed to accomplish two major tasks in feature-based modulation classification. First, the machine learning techniques can provide a classification decision making process that is much easier to implement. Second, the machine learning techniques can reduce the dimension of the feature set. This is achieved by feature selection and feature generation, which enables the consideration of a more versatile feature set while maintaining the computational efficiency of the classifier.

In this chapter we first introduce two machine learning-based classifiers, namely k-nearest neighbour classifier and support vector machine classifier, for modulation classification in combination with the features listed in Chapter 5. Next, the issue of feature space dimension reduction is explored through different algorithms including linear regression, artificial neural network, genetic algorithm and genetic programming.

6.2 K-Nearest Neighbour Classifier

The k-nearest neighbour (KNN) classifier is a non-parametric algorithm which ...