Karolina Carlsson defends her bachelor thesis

Estimation of ERP:s using the Prony method and Time
Frequency analysis with the aim to find a correlation
between semantic spaces and brain activity


Abstract

In order to obtain the characteristics of an EEG signal, different
statistical data processing is needed. To minimize the noise,
researchers have used various procedures of smoothing and
averaging. In this study two alternative methods is discussed, the
Prony method and Time Frequency analysis.

Much of the cognitive research performed with EEG uses event-related
potentials (ERPs). This study has shown that ERPs can be modelled by a
sum of damped sinusoids (Prony) and that this method can be used to
filter the unimportant details and find the true character of the ERP
signal. Time Frequency analysis with the spectrogram has shown that a
non-parametric method can be used to give information in both the time
and frequency domain.

Finally, a simple correlation method and a linear regression technique
have been applied on the datamaterial to map the ERPs to semantic
representations of words. The results indicate that it is possible to
find a correlation between brain activity and semantic spaces.