The Application Research Of Higher-Order Spectra To The Analysis Of Pulse Signals

The unique diagnostic methods and particularly curative effects of traditionalChinese medicine are obvious to all. Along with the development of sensors andcomputer technology, people bend themselves to the objectivity of the Chinese medicine,and hope to apply the modern technologies and instruments to push the modernizationof the traditional Chinese medicine, which is the main motive in this paper.Higher-Order Spectra is an advancing research aspect in signal processing field,which is the main tool for analyzing non-Gaussian signals. Bispectrum is the mostcommon method of Higher-Order Spectra mathematical analysis, and it includes twochief approaches that can be used to estimate higher-order spectra; namely theconventional and the parametric approach. In this paper, the indirect algorithm ofconventional approach is used to analyze the bispectrum estimation of pulse signals for15 heroin addicts and 15 healthy persons to classify the pulse signals. According to thecontour map of bispectrum estimation, it is found that the average phase phof heroinaddicts on a specified time-frequency region is generally lower than that of healthypersons. Using the average phase phas characteristic parameter, a critical parameter isdetermined that is used to classify heroin addicts and healthy persons. Thus, all of the15 healthy persons are identified. Only one heroin addict B13 is misjudged. Inparametric Bispectrum estimation, the model ARMA( p,q ) is created by means ofidentifying the exponent number of the date segments of pulse signals, in whichp = 2, q =1. Then the residual time series algorithm is used to estimate the bispectrum ofpulse signals. Using AR parameter spectra approach, average phase approach and sliceapproach to analyze the AR parameter and bispectrum, it is found that these approachesall can give the well classified identify to healthy persons and heroin addicts. Last, thepulse signals are intercepted 30 data segment between 47 and 76, which is used toidentify the exponent number. With the same algorithm, the model ARMA( p,q ) iscreated in which p = 3,q =1and the bispectrum of pulse signals is estimated. Using ARparameter spectra approach to analyze the AR parameter, it is found that three healthypersons Z01, Z10 and Z15 are misjudged.It is shown that at the aspect of extracting characteristic information of pulsesignals, the conventional Bispectrum estimation and parametric Bispectrum estimationall have the higher discrimination. But there can not suggest that this approach is betterthan that one, because the model ARMA( p,q ) is not the best one which is match therandom process of pulse signals. In addition, it is shown that the conclusion is not thesame when the part data segment is used but not the whole data segment. This paper represents the basic conceptions and theories of Bispectrum estimationin detail, and discusses the physical meaning of the Bispectrum. While using theindirect algorithm and parametric model algorithm to analyze pulse signals, this paperalso derives, verifies and uses them.