A general maximum likelihood framework for modulation classification
Denys Boiteau, Christophe Le Martret · 2002
This paper deals with modulation classification. First, a state of the art is given which is separated into two classes: the pattern recognition approach and the maximum likelihood (ML) approach. Then we propose a new classifier called the general maximum likelihood classifier (GMLC) based on an approximation of the likelihood function. We derive equations of this classifier in the case of linear modulation and apply them to the MPSK/M'PSK problem. We show that the new tests are a generalisation of the previous ones using the ML approach, and don't need any restriction on the baseband pulse. Moreover the GMLC provides a theoretical foundation for many empirical classification systems including those systems that exploit the cyclostationary property of modulated signals.