On the blind classification of parametric quadrature amplitude modulations

Abstract

Automatic modulation classification (MC) for parametric quadrature amplitude modulation (QAM) formats is investigated. The generic θ-QAM is considered, which includes popular constellations, such as square QAM (SQAM) and triangular QAM (TQAM). Both feature-based and optimal likelihood-based classifiers are tested, where in the first we use high order cyclic cumulants (CCs) of the received signals as features for MC, and in the latter we exploit several likelihood functions. Simulations demonstrate that, unlike likelihood-based MC schemes, feature-based MC schemes are not generally suitable to discriminate between different θ-QAMs, unless the true values of θ are perfectly chosen.