Time-frequency analysis of short segments of biomedical data

Investigates methods to analyse the frequency content of EEG data where there are a limited number of samples. The aim is to determine how many frequencies are present at a particular time and to track changes in any of the frequencies. A real-time analysis is not sought; instead, we identify those frequencies present at the onset of an epileptic seizure that aid in the formulation of a real-time analysis. The standard high-order Yule-Walker (HOYW) method is one of the methods used to analyse the EEG data. In addition, two novel off-line techniques are tested: a recursive modelling technique, where polynomial modelling of the data is combined with zero-crossing analysis to determine the dominant frequency components, and a phase compensation technique, where the underlying signal is modelled as a series of cosines with reference frequencies. The derivative of the phase is computed and is used to correct the values of the various reference frequencies. We compare the performances of these two novel techniques with the established HOYW method in the analysis of short segments of non-pathological data.