Modified Neighborhood Determination in Nonlinear State-Space Projective Filtering

Nonlinear state-space projection method (NSSP) is a very useful tool for filtering signals contaminated with broadband noise. The filtering process takes place in a reconstructed embedded space. Time-delay technique is used for reconstructing the embedded space. After that, one signal sample corresponds to a point in a multidimensional space. For each point in the embedded space we seek for its vicinity. This process is time consuming. In this paper we propose an application of clustering method for creation of the signal subspaces. The proposed method is applied to process different types of signals: Lorenz chaotic signal and a real ECG signal. The proposed method appears more effective than NSSP method.