Sleep and cardiac signal processing using improved multivariate partial compensated transfer entropy based on non-uniform embedding

Abstract

Based on the symbolic transfer entropy, this paper proposes an improved method to detect the nonlinear interaction of MV time series, namely the effective symbolic phase multivariate partial compensated transfer entropy (EMPcTE). It is based on embedding sequences through sequential non-uniform processes and correcting the bias of transient effects by compensated conditional entropy, eliminating influencing factors in multivariable systems, and estimating information flow from source variables to target variables. The method was validated on short-term implementations of linear stochastic and nonlinear deterministic processes and then evaluated on the Sleeping Heart Health dataset to explore the impact of interactions between electrophysiological signals on cardiovascular-related diseases caused by sleep-disordered breathing. The experimental results show that diseases with high prevalence such as angina pectoris may be related to the close information transfer between EMG and ECG.