Modulation Classification Based on Spectral Correlation and SVM

Abstract

This paper addresses the problem of automatic modulation recognition of digital signals. A classification method based on spectral correlation and Support Vector Machine (SVM) is developed. The spectral correlation theory is introduced and several characteristic parameters which can be used for modulation analysis are extracted. The parameters are used as the input feature vectors to SVM. SVM maps the vectors into a high dimensional feature space, so the problem of non-separable classification in low dimension is resolved and the decision threshold become unnecessary. The experiment results show that the algorithm is robust with high accuracy even at low SNR.