Nonlinear complexity reduction: Sparsity of the generalized schur coefficient matrices and frobenius norm criterion

Most of real-life signals are nonGaussian, therefore the linear parametrization/modeling methods, suitable for Gaussian signals are neither adequate nor optimal. Using a nonlinear approach, we can significantly improve filtering results. However, the price that has to be paid is the complexity of the nonlinear treatment. In this paper we consider the nonlinear Schurtype orthogonal transformations of higher-order nonGaussian stochastic processes. The nonlinear filtering (parametrization, modeling) complexity increases tremendously and time necessary for computation makes such a method inefficient or even useless from practical point of view. Therefore, we propose and examine a method based on sparsity of the Schur coefficients matrices that reduces the computational effort.