System interpretation of causality measures in frequency domain used in eeg analysis

Abstract

This paper suggests new measures for the evaluation of an absolute, ratio and relative causal relation in frequency domain in the context of multivariate autoregressive models.The idea is explained on four synoptic artificial data experiments. In each example, a model without causal connection is presented and then, a causal connection is added. The influence of this modification is analysed and interpreted in the scope of LTI digital filters. Using the proposed measures, a comparison of results of state-of-the-art frequency domain methods Generalized Partial Directed Coherence (GPDC) and Direct Directed Transfer Function (dDTF) is performed to evaluate their behavior. The concept is demonstrated on real EEG data of awake resting state of human brain.