On application of input data partitioning to Bayesian weighted averaging of biomedical signals

Averaging in the time domain may be used for noise attenuation in case of biomedical signals with a quasi‐cyclical character. Traditional arithmetic averaging technique assumes the constancy of the noise power cycle‐wise, however, most types of noise are not stationary and the variability of noise power is observed. It constitutes a motivation for using methods of weighted averaging, in particular Bayesian weighted averaging. This paper presents the computational study of Bayesian weighted averaging with traditional (sharp) and fuzzy partition of the input data in the presence of non‐stationary noise. There is presented the known empirical Bayesian weighted averaging method (EBWA), with the parameter p describing the probabilistic model, and its modification NBWA which eliminates the parameter. Both methods can be extended by partitioning of the input data. The performance of presented methods is experimentally evaluated for an analytical signal as well as a real ECG signal and compared with traditional arithmetic averaging method. However, the methods can be applied to any signal with a quasi‐cyclical character. The aim of the paper is to show the influence of the type of partition as well as the number of parts on the quality of the averaged signal.