부산대학교 도서관

Automatic signal recognition algorithms turned out to be useful in many areas including intelligent radio, electronic warfare or surveillance systems. There are various new signals and the emitter recognition is a key problem in rapidly changing electromagnetic environment.Using continuous wavelet transform (CWT) turns out to be effective way to extract signal/modulation features for classification purposes. In this paper the recognition possibilities of selected types of radar signals, like linear and stepped frequency modulated signals (LFM, SFM), phase coded waveforms (PCW) with Barker code and rectangular pulses (Rec), are analysed. Two kinds of algorithms are considered. In the first one higher order statistics (HOS) of continuous wavelet transform (CWT) coefficients are proposed as signal features. Principal component analysis (PCA) is considered to reduce number of features and feed-forward neural network is proposed as classifier. In the second one CWT coefficients are treated as an image and the classification process is carried out using a convolutional neural network (CNN). For evaluating the performance of the simulated (in Matlab environment) classification models a confusion matrix is used.