ABSTRACT

This introduction presents basic concepts for the analysis of single-trial electroenchephalogram (EEG) data. This is done in a way that is accessible to readers whose main experience rather lies in other fields. While attempting to be as illustrative as possible, our exposition contains the mathematically precise formalism of the methods discussed. Readers not familiar with basic mathematical notions such as vectors and matrices should nevertheless not be discouraged, as they hopefully benefit from reading this tutorial even when skipping the technical parts. For the classification of event-related potentials, spatiotemporal features and classification with regularized linear discriminant analysis is discussed. Modulations of oscillatory brain activity are detected with common spatial pattern analysis and log(arithmized) band-power features. All of those methods are well comprehensible 344and easy to implement and show nevertheless competitive performance. The first two sections motivate the transition from univariate features, such as signal amplitude in a certain channel and at a certain latency toward multivariate features. Readers familiar with the benefits of multivariate analyses are encouraged to step very quickly over those two sections and to start with Section 18.3.