Subspace methods for identifying neural activity from electromagnetic measurements of the brain

Use of subspace-based signal processing techniques allows the underlying neural sources to be localized from measurements of their external weak electromagnetic fields. The authors review the localization problem and construct a mathematical model whose form is analogous to the diversely polarized source form of the direction-of-arrival problem. A new source localization technique for this application is presented, based on Schmidt´s (1979) MUSIC algorithm, and it is contrasted with existing nonlinear least squares approaches. The locations of the neural sources are analogous to the angles of arrival, the moments of the neural sources are analogous to the polarization, and the magnitudes of the neural sources are analogous to the signal strength