Extraction of Signal Waveform Feature Based on Bispectrum

Abstract

This paper presents a new method of feature extraction for signal waveform. Selection of a feature extraction method is probably the single most important factor in achieving high recognition performance. Over the past four decades, considerable work has been done in the area of power spectrum estimation. However, a problem with this method is that it is phase blind. Situations arise in science and engineering whereby signal analysts are required to look beyond second-order statistics and analyze a signal's Higher-Order Statistics (HOS). In this paper, bispectrum is used to extract the feature of signal. Feature of the signal can be extracted by selecting the eigenvector whose corresponding eigenvalue's module is the largest as the template of recognition. The experiment being made by our research group suggests that recognition accuracy rate of bispectrum-method can be no less than 90 percent in additive white Gaussian noise channel when SNR (Signal to noise rate) is no less than 8dB .