Variable Embedding Based on L–statistic for Electrocardiographic Signal Analysis

In this paper, a variable embedding approach for reconstructing attractors of dynamical systems is proposed, using the L–statistic based on noise amplification. Particularly, the variable manifold is obtained from a time-series using delay coordinates and an embedding vector, the last one, is constructed based on a L–statistic matrix which contains the local reconstruction quality of whole attractor. The embedding vector contains the optimal embedding dimension for each point in the manifold. This approach were performed on electrocardiography databases, we obtain the first four statistical moments for the embedding dimension vectors and apply statistical tests to distinguish between normal and pathological signals. Results shown significant differences that lead to new classification strategies, infer about functional states, and establish a new path for processing signals with high embedding dimensions, i.e., high computer complexity.