Keywords: conjugate gradient optimisation, kernel density estimation, blind source separation, overdetermined BSS, OBSS, Informax principle, singular value decomposition, SVD
Informax based overdetermined blind source separation method with conjugate gradient optimisation algorithm and kernel density estimation
In this paper, we mainly use a combination of the kernel density estimation and the conjugate gradient optimisation to solve the overdetermined blind source separation (OBSS) problem. The Informax principle is used as the basis for introducing the contrast function of the OBSS problem with the singular value decomposition (SVD) technology. For practical optimisation of the contrast function, the conjugate gradient optimising algorithm is employed to derive the learning rules for training the separating matrix. Within the training equations, the term of score function is estimated directly by the kernel density estimation method. Experimental results confirm that the proposed OBSS approach works very fast and reliably.