Detecting Nonlinear Properties of Snoring Sounds for Sleep Apnea Diagnosis

This paper investigates nonlinear properties of snoring sounds by a surrogate analysis which is generally used to verify the existence of nonlinearity in a time series. The ultimate goal of this study is to extract useful information from the nonlinear properties of snores so as to diagnose obstructive sleep apnea. For such purpose, many researchers have examined snoring sounds by linear frequency analysis such as Fourier Transform or Linear Predictive Coding, but the nonlinear properties of snores have not yet been clarified and the existence of nonlinearity has not been proved so far. The author adopts correlation integral to evaluate the geometrical nonlinear structure of snore attractors quantitatively. As a result of experiments, nonlinear properties are found in some kinds of waveform. But a complex waveform, in which no prominent peaks are found in the amplitude spectrum, does not have a nonlinear property.