Comparative Analysis of Blind Source Separation Methods for Biological Signal Processing

Blind Source Separation (BSS) is a signal processing technique that separates individual source signals from their mixtures without prior knowledge of the source characteristics and mixing process. Analyzing biological signals often contaminated by noise and other biological artifacts is crucial. This research compares the performance of seven of the most common BSS algorithms under the influence of sources’ noise intensities and the number of components extracted from the mixed signals. These algorithms are evaluated based on the Signal-to-Interference Ratio (SIR), Mean Square Error (MSE), and Normalized MSE (NMSE) criteria on two simulated and biological ABio5 data sets. Also, proper statistical tests are utilized to explore significant differences $\lt$0. 0 0 1) superior performance than the ERICA algorithm in estimating the sinusoidal and random noise components. The JADE, SOBI, and MULTICOMBI algorithms exhibited strong robustness in estimating primary sources under varying noise intensities. The AMUSE and JADE algorithms outperformed the Erica algorithm significantly (P -Values $\lt$ 0.001)in estimating all biological components, especially the electrocardiogram (ECG) signal. Also, the JADE algorithm performed consistently regardless of whether the number of estimated components from mixed signals is smaller, equal to, or greater than the number of primary sources. It is hoped this research can help to choose the appropriate method for the BSS problem, especially in biological applications, according to the components’ natures, sources’ noise intensities, and the number of estimated components.