A new coherence estimating method: The magnitude squared coherence of smoothing minimum variance distortionless response

Abstract

The magnitude squared coherence (MSC) is an important method to calculate the connectivity between neural signals. It provides a better spectral resolution than the Welch's method and is often used in analyzing electroencephalograph (EEG) synchronization activity. The minimum variance distortionless response (MVDR) is a spectral estimation method based on matched filterbank theory. The Cheriet-Belouchrani (CB) kernel is provided for measuring the energy of a signal in time-frequency distribution, which has significant interference mitigation and preserves high resolution measure values. By combining MVDR spectra and CB kernel, a new magnitude squared coherence estimating method is proposed in the paper by smoothing the MVDR with the CB kernel (SMVDR). The simulation results show that SMVDR MSC approach has better performances than the MVDR MSC method.