MEG/EEG Spatiotemporal Noise Covariance Estimation in the Least Squares Sense
A new spatiotemporal noise covariance model were recenlty proposed to consist of the multi-pair Kronecker products of spatial covariance matrices of rank 1 and their corresponding full temporal covariance matrices. Optimized estimators for parameters within this model were derived through the maximum likelihood method and their inversions were accomplished through simple closed formulas, thereby allowing the method to be easily incorporated into parametric source localization methods. However, these maximum likelihood estimators were derived under the assumption that collected spatiotemporal noise samples are Gaussian distributed, which is generally not true for such non-averaged (or single trial) MEG/EEG signals. In this work, an unbiased estimators of spatiotemporal noise covariance in the least squares sense is proposed with using the same multi-pair Kronecker product model without assuming a Gaussian distribution for noise in the data. We found that the least squares covariance estimator for orthogonal spatial components is the same (only differing by constant factors) as the maximum likelihood estimator. However, for independent spatial components the least squares estimator is different from the maximum likelihood estimator.
Minkyu Ahn Sung C. Jun
Gwangju Institute of Science and Technology, Gwangju, South Korea
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-6
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）