Geometric pattern model of 1D signals based on self-structures for the construction of the feature space for a classifier

Abstract

The model presented in this paper is based on the formulation of novel parameters to distinguish different types of discrete signals. We compute the parameters using the scanning method with a given embedding dimension, which is widely used in Information Theory applications. In this work, we propose three simple parameters, namely: amplitude, Zenith angle and factor of shape. With these parameters, we obtain a 3-tuple for each segment of a signal, and for each signal in a given group. Each signal generates a number of 3-tuples, which are then reduced to single 3-tuple by removing repeated information. We apply the proposal to the analysis of synthetic, random and chaotic signals.

We have proved that the present methodology is efficient in differentiating the signals studied.