A Robust and Practical Method to Separate Periodic Signals from MEG Data using Second Order Statistics

The analyses of recordings of magnetoencephalography (MEG) and other imaging techniques may require the separation of periodic signals from the observed signals. Blind source separation (BSS) is widely used for the separation of specific signals these days. Though several algorithms based on the BSS scheme for the separation of periodic signals have been proposed, they usually assume the system to be well-posed, satisfactory results often cannot be obtained for practical recordings. In this study, we show that a method based on the joint approximate diagonalization of correlation matrices with several time delays (JADCM) is robust and good results can be obtained by choosing the time delays carefully, especially in practical illposed situations such as signal separation from MEG recordings. The performance of the proposed method is compared with that of Periodic BSS and JADCM using the conventional parameter set.