Automatic Modulation Classification: Principles, Algorithms and Applications

7Blind Modulation Classification

7.1 Introduction

From Chapters 3–6, we listed modulation classifiers of different types. These classifiers mostly require the prior knowledge of channel state information (Wei and Mendel, 2000; Wang and Wang, 2010). There are few classifiers which have the certain ability to treat one or two channel parameters as unknown (Panagiotou, Anastasopoulos and Polydoros, 2000). Many classifiers may appear to be able to recognize the modulation type without the need for CSI (Azzouz and Nandi, 1996; Spooner, 1996; Swami and Sadler, 2000). In fact, the classification accuracy is often far inferior if the CSI is not utilized for the preparation of reference values or decision thresholds. Dobre et al. reviewed some of the semi-blind classifiers and suggested the necessity of a blind modulation classifier (Dobre, Abdi and Bar-Ness, 2005).

The classification of modulation types in a channel with unknown CSI is normally divided into two steps. In the first step, channel estimation is performed. The estimation can either acquire all of the needed channel parameters or partial CSI. When the entire CSI is estimated, any classifier that we have mentioned in the previous chapters can be employed to complete the second step. If the CSI is partially estimated, a classifier which requires the prior knowledge of all channel parameters will not be able to complete the classification. Instead, a semi-blind classification that can complement the partial channel estimation ...