Nonstationarity detection

• Main Author: Kiremire, Bunty Byarugaba
• Format: Others
• Language: en
• Published: 2009
• Online Access: http://hdl.handle.net/10539/5993

This dissertation is an investigation into nonstationarity detection. Current methods are looked into and a novel method is proposed. An index was developed during the course of the study called the Stationarity Index. The method of nonstationarity detection proposed is based on this index and it gives a measure that varies directly proportionally to changes in the dynamic behavior of the signal. This allows not only for the detection of nonstationarity but also serves as a means of determining relatively how much it is changing. The Stationarity Index maps changes in the dynamics to a range of 0 − 100 in the similar manner to the mapping of measures to percentages. This allows for the comparison of the variation of dynamics between separate signals. This ability sets the test apart from current tests. Most current tests simply determine if nonstationarity is present or not, and do not allow for the kind of measure that allows relative comparisons between signals. The index is then used to successfully distinguish between stationary and nonstationary chaotic signals. It is then used in the analysis of electrocardiogram (ECG) and electroencephalogram (EEG) profiles. It is able to clearly pick out the changes in the dynamics that are the result of the onset of partial seizures. It shows potential for use in the identification of certain dynamic traits in ECG and EEG profiles. Its ability to discriminate between stationarity and nonstationarity also shows potential for use in the segmentation of nonstationary biomedical signals into partially (quasi) stationary segments.