A Direct Estimation of Noise Covariance Matrix from Noisy Multichannel Data for Improving Brain Source Localization

Abstract

Abstract. In this paper, we present the new computational method for estimating a noise covariance matrix directly from noisy multichannel data. Our proposal outperforms the conventional time shift difference (TSD) method with an increase of sharpness (focalization) for estimated brain source locations, which are calculated by using the noise covariance incorporated multiple signal classification (MUSIC) algorithm proposed by Sekihara et al. 1997. This conclusion is validated through numerical simulations of a realistic magnetoencephalography (MEG) system.