Generalized Partial Directed Coherence

This paper re-examines the definition of partial directed coherence (PDC) which was recently introduced as a linear frequency-domain quantifier of the multivariate relationship between simultaneously observed time series for application in functional connectivity inference in neuroscience. The present reappraisal aims at improving PDC's performance under scenarios that involve severely unbalanced predictive modelling errors (innovations noise). The present modification turns out to be more robust in estimating imprecisions associated with finite time series samples.