Partial Information Rate Decomposition in Physiological Networks

Abstract

In the field of information theory, the framework of partial information decomposition (PID) has been extensively applied to networks of two source variables sharing information with a target. A dynamic version of the PID, extended to the case of three sources and providing spectral estimates of the information shared among the involved processes, is still missing. We fill this gap by introducing a coarse-grained partial information rate decomposition (PIRD) for random processes in the time and frequency domains, applied to the network of cardiovascular, respiratory and cerebrovascular oscillations studied in patients prone to postural syncope during a protocol of postural stress.