Principal component analysis of cyclic spectrum features in automatic modulation recognition

Automatic modulation recognition (AMR) of communication signals is a critical and challenging task in cognitive radio systems. In this work, classifications of four digital modulation types, including BPSK, QPSK, GMSK and 2FSK, are investigated. From the received radio signal, a set of cyclic spectrum features are first calculated, and a principal component analysis (PCA) is applied to extract the most discriminant feature vector for classification. A novel max-multiple layer perceptron (MaxMLP) neural network is introduced for classification of modulation feature vectors through supervised learning. In the experiments, real radio signals with different modulation types were generated from an Agilent vector signal generator, and sampled by an Agilent digital signal analyzer. The proposed AMR method is tested at various channel SNR levels. Experimental results indicate that the performance of this method is highly competitive, and the computational cost is relatively low.