Abstract

Cardiorespiratory signal separation is a critical task in biomedical signal processing, with significant implications for improving the accuracy of cardiovascular and pulmonary disease diagnoses. The presence of complex mixed signals and noise complicates the separation of cardiac and respiratory sounds from a single-channel recording. Traditional separation methods struggle due to the overlapping frequency ranges of these physiological signals. In this work, we propose a multi-stage Singular Spectrum Analysis (MSSA)-based framework for the separation of cardiac and respiratory sounds. The approach utilizes a two-stage SSA process where the first stage isolates the cardiac components through periodic structure analysis, while the second stage extracts respiratory components based on relative variance and cross-correlation. Key features of this method include a systematic selection of RCs that ensure high-quality signal separation, preserving physiological integrity. The framework was validated on real-world cardiorespiratory recordings obtained via electronic digital stethoscopes, demonstrating robustness across varying recording conditions.