CONFAC Decomposition Approach to Blind Identification of Underdetermined Mixtures Based on Generating Function Derivatives

摘要: This work proposes a new tensor-based approach to solve the problem of blind identification underdetermined mixtures complex-valued sources exploiting cumulant generating function (CGF) observations. We show that collection second-order derivatives CGF observations can be stored in third-order tensor following constrained factor (CONFAC) decomposition with known structure. In order increase diversity, we combine three derivative types into an extended CONFAC decomposition. A detailed uniqueness study this is provided, from which easy-to-check sufficient conditions ensuring essential mixing matrix are obtained. From algorithmic viewpoint, develop CONFAC-based enhanced line search (CONFAC-ELS) method used alternating least squares estimation procedure for accelerated convergence, and also analyze numerical complexities two algorithms (namely, CONFAC-ALS CONFAC-ELS) comparison Levenberg-Marquardt (LM)-based algorithm recently derived same problem. Simulation results compare proposed some higher-order methods. Our corroborate advantages over competing LM-based terms performance computational complexity.