Fourth-order CONFAC decomposition approach for blind identification of underdetermined mixtures

We have recently proposed a second-order method for the blind identification of underdetermined mixtures that relies on the constrained factor (CONFAC) decomposition. It consists in storing successive second-order derivatives of the cumulant generating function (CGF) of the observations computed at different points of the observation space in a third-order tensor following a CONFAC model. In this work, we extend this approach to the case of third-order derivatives by resorting to a fourth-order CONFAC decomposition. We show how different third-order derivative types can be combined into a single fourth-order CONFAC tensor model with the goal of increasing the diversity of the observations, so that higher underdeterminacy levels can be handled. Computer simulation results illustrate the performance of a CONFAC-based blind identification algorithm compared to some competing methods.