Robust nonlinear signal separation using regularised maximum likelihood neural network (2003)

Author(s): Woo WL, Dlay SS

    Abstract: The fundamental problem in independent component analysis (ICA) is to find a set of statistically independent components from the output of a mixing system. Almost all of the existing algorithms are based on the ideal situation where the mixture is a linear. However, in some practical situations, the signals are nonlinearly mixed and thus the problem results in ill-posed solution. A robust nonlinear technique is presented for instantaneous signal separation of nonlinear mixtures based on regularised maximum likelihood estimation combined with multiple-layer neural network. The motivation for such criterion is to incorporate a priori information such as smoothness constraints into the statement of the ill-posed problem so that convergence to undesirable minima can be avoided by the neural network. (8 References).

      • Date: 2003
      • Journal: WSEAS Transactions on Systems
      • Volume: 2
      • Issue: 3
      • Pages: 675-680
      • Publisher: World Scientific and Engineering Academy and Society
      • Publication type: Article
      • Bibliographic status: Published

        Professor Satnam Dlay
        Professor of Signal Processing Analysis

        Dr Wai Lok Woo
        Director of Singapore Operations