Выбор меры различия значений образцового и отфильтрованного биомедицинских сигналов

Abstract

The article covers selection of a error measure for the results of filtering distorted quantized periodic signals when searching for optimum values of low band digital filter parameter represented by a "simple moving average" in a non-recursive form. The following error measures were selected for comparison – Mean Squared Error, Root Mean Squared Error, Mean Absolute Error. To perform the experiments, a program library was created for generation of periodic discrete quantized signals, and for their distortion using fluctuating interference (noise) with normal distribution, and for further filtering of the distorted signals. A series of experiments was carried out for the generation and filtering of distorted periodic signals with different noise levels. As a result, error measures were found for filtered signals with different values of the digital filter parameter. Ultimately, a measure was selected based on comparison results, representing greater number of local optimum values of the digital filter parameter at high fluctuating noise levels.