Modulation classification based on bispectrum and sparse representation in cognitive radio

Spectrum awareness is a prominent characteristic of cognitive radio technologies. Realizing such awareness in cognitive radio requires a capability to recognize the incoming signal's modulation type. In this paper, a novel approach to classify digital modulated signals is proposed for cognitive radio. This method combines high-order spectra with sparse representation. We cast the modulation classification problem as finding a sparse representation of the test bispectrum features w.r.t. the training set. The sparse representation can be accurately obtained by solving L1-minimization. Unlike conventional modulation recognition method, if sparsity in the recognition problem is properly harnessed, high-dimensional data with highly distinctive features can be applied in the signal identification. The classification results for the modulation types 2-FSK, 4-FSK, QPSK and 16-QAM, obtained from computer simulations, show the proposed feature extraction and classification method has high classification correct ratio in strong noise condition.