New Methodology for Differentiating Noisy Signals of Different Colours Using Geometric Features

This work focuses on the differentiation of coloured noise signals such as pink noise, red noise, blue noise, violet noise, white Gaussian noise and uniform white noise. For this purpose, an algorithm based on the sweeping technique has been used, defining two input values: a embedding dimension and a discretisation factor. Classification is performed by extracting geometric features such as amplitude, and two new parameters: Zenit angle and a shape factor. The process creates three-component vectors in a systematic way, reducing redundant information. As a new result, the model generates a reduced feature space containing unique vectors that can be visualised and used for further processing. The cardinal of this space, i.e. the number of vectors it contains, is the main parameter used to perform the classification. The proposed algorithm showed a significant differentiation of the processed signals using this parameter.