A Cyclostationary Based Signal Classification Using 2D PCA

In this paper, we propose an advanced automatic modulation classification (AMC) method for cognitive radio (CR). Conventional AMC algorithms employ some pattern recognition algorithms such as hidden markov model (HMM) and support vector machine (SVM) to recognize the signal modulations through the characters of spectral correlation, e.g., a-profile, f-profile, average value, and etc. However, these methods are one dimensional approaches and might not extract the whole characteristics of modulations completely. In this paper, we exploit a two dimensional property of cyclostationarity: spectral correlation function (SCF). Compared with those of one dimensional spectral correlation, the SCF exhibit more classification information. Moreover, we employ two dimensional principal component analysis (PCA) which minimize the size of original data not losing own features so that we can have better performance than choice of few characteristics.