CORRELATIONAL EFFICIENCY OF FUNCTIONAL NOISE REALIZATIONSCHARACTERISTIC OF RESPIRATORY SYSTEM DISORDERS

Abstract

The study investigates the correlation properties of acoustic signals generated by the functioning of the human
respiratory system. It is shown that the analysis of noise realizations, particularly through the expansion of informative
features, can improve the effectiveness of non-invasive diagnosis of respiratory disorders.
The problems of respiratory sound segmentation are outlined, considering the variability of breathing cycles, the
presence of external noise and technical artifacts, as well as individual anatomical differences. It is determined that
traditional amplitude-based signal processing methods are insufficiently effective in such cases, which necessitates the
use of more informative, particularly probabilistic, characteristics. Consequently, the application of sliding statistical
estimations—local mean, variance, and information entropy—is considered, allowing for the localization of respiratory
movements within time windows. Based on these estimations, a correlation analysis was implemented to detect signal
periodicity, which improved the accuracy of identifying individual respiratory phases (breathing movements).

During the research, an autocorrelation analysis of respiratory noise amplitudes represented by various statistical
estimations was conducted. It was found that variance and information entropy estimations provide the most effective
segmentation of respiratory movements; however, the entropic representation is characterized by lower fluctuation of
correlation function apertures and does not require additional scaling, which simplifies hardware implementation.
Experimental studies have shown that the use of probabilistic estimations—specifically variance and information
entropy—enables more accurate identification of specific types of functional respiratory disorders. For each type of
disorder, reference signals were formed, exhibiting a high degree of cross-correlation with experimental realizations.
Moreover, a comparison of cross-correlation results showed that the representation of signals using sliding
information entropy estimations provides approximately twice the resolution between homogeneous and heterogeneous
disorders compared to the variance-based approach.
The presented results may have practical significance in the development of digital diagnostic systems based on
microcontrollers and computer-based signal processing tools.