Higher-Order Statistics Analysis For Human Pulse Signals

Pulse-feeling is the most characteristic diagnostic methods in traditional Chinese medicine. Along with the development of science and technology, people hope to apply sensors and modern signal processing technology to human pulse diagnosis, in order to carrying on an investigation in the objectivity of the Chinese medicine Pulse-feeling ,which is the main research aspect in this paper. In recent years, higher-order statistics (HOS) have received increasing interest. HOS are the higher-order statistical description of a signal. HOS preserve the magnitude information as well as the phase information. The two main motivations behind the use of HOS are extracting information due to deviations from Gaussianity and detecting nonlinear properties in signals. HOS can be used to suppress additive colored noise in theory. Particular examples of higher-order spectra are bispectrum . Considering the characteristic differences between the pulse signals of heroin addicts and healthy persons, we successfully use direct class of conventional bispectrum estimator and two non-Gaussian AR models to identify heroin addicts from the pulse signals of 15 heroin addicts and 15 healthy persons. Direct class of conventional bispectrum estimators is proposed for calculating the bispectrum of pulse signals. characteristic parameters of bispectrum of the pulse signals are extracted by using horizontal slices of bispectrum. The phases of the sum of bispectrum in the horizontal slices are used to classify heroin addicts and healthy persons. Thus, all of the 15 heroin addicts are identified. Only two healthy persons are misjudged. The first non-Gaussian AR model is identified by using TOR method in analyzing the pulse signals. Non-Gaussian AR model is proposed for calculating the normalized bispectrum of pulse signals. In the same way, the phases of the sum of normalized bispectrum in the horizontal slices are used to classify heroin addicts and healthy persons. Thus, all of the 15 heroin addicts and 14 healthy persons are identified. Only one healthy person is misjudged. The second non-Gaussian AR model is a specified ARMA model( q =0) which is identified by using identification methods of AR part in ARMA model in analyzing the pulse signals. Characteristic parameters of normalized bispectrum magnitude of the pulse signals are obtained by using horizontal slices of normalized bispectrum magnitude. It is found that the peaks of normalized magnitude bispectrum of the pulse signals of heroin addicts are higher than healthy persons , so a critical parameter is determined that is used to classify heroin addicts and healthy persons. Thus, only 1 heroin addict and 1 healthy person are misjudged. Characteristic parameters of principal value of normalized bispectrum phase of the pulse signals are obtained by using horizontal slices of principal value of normalized bispectrum phase. It is found that principal value of bispectrum phase of heroin addicts on a specified frequency region is different from healthy persons, so a critical parameter is determined that is used to classify heroin addicts and healthy persons. Thus, 2 heroin addict and 1 healthy person are misjudged. It is shown that at the aspect of extracting characteristic information of pulse signals, the parametric method of bispectrum estimators which is based on TOR method has superiority over direct class of conventional bispectrum estimators. Besides the three methods of bispectrum estimators mentioned above, HOS are deduced and proved in this paper. Meanwhile, the extracting technology for characteristic parameters of bispectrum is also researched .