In this paper we investigate automatic modulation classification (AMC) using homogeneous feature-vectors based on cyclic cumulants (CCs) of fourth-, sixth- and eight-orders, respectively, for QAM, PSK and ASK signals within a pattern recognition framework. Analysis of CCs of the baseband signal at the receiver is performed and used for feature selection. The cycle spectrum of the baseband signal at the receiver is derived as a function of excess bandwidth for a raised cosine pulse shape and a necessary and sufficient condition on the oversampling factor is obtained. Theoretical arguments regarding the discrimination capability of the examined feature-vectors are verified through extensive simulations.