Automatic Digital Modulation Recognition Based on Locality Preserved Projection

In this paper, we investigate the modulation recognition method based on locality preserved projection (LPP) in AWGN channels. Feature extraction is the precondition of signal modulation recognition. Based on analyzing the characteristic of signal in time and frequency domain, seven feature parameters with fine classification information are selected. In order to wipe off the relativity among different features, and keep the important identity for classification simultaneously, we need to search for a best feature subspace in which different modulation can be apart very well. LPP builds a graph incorporating neighborhood information of the data set to preserve the local structure, it is likely that a nearest neighbor search in the subspace will yield similar results to that in the original feature space. Combined with 1-NN Nearest-neighbor pattern classifier, our method achieves better performance compared with the method based on PCA which is widely used.