Efficiency of the Nonlinear Schur-Type Estimation Algorithms for Higher-Order Stochastic Processes

In real-life we are usually faced with non-Gaussian signals (e.g. speech signal, electrocardiograms). Majority of the signals estimation methods developed so far are based on the linear approach, appropriate for Gaussian signals. They are neither adequate nor optimal for most, practically important, real-life non-Gaussian signals. The nonlinear approach can significantly improve estimation results, however it essentially increases the processing complexity. As the nonlinear processing of non-Gaussian signals is a key to enhance the existing as well as to introduce new technologies, in this paper we propose optimization of the nonlinear Schur-type orthogonal parametrization for higher-order and non-Gaussian stochastic processes efficient estimation. We show an advantage of employment of even small number of the nonlinear Schur coefficients over the use of a large collection of the linear Schur coefficients, resulting in very fast convergence of the nonlinear estimates at a comparable complexity.