Blind Separation of Heart-Lung Sound Mixed Signals in Reverberation and Noise Environments

To address the difficulty of using a stethoscope to acquire pure heart and lung sound signals with reverberation and noise in clinical environments, we propose a blind separation algorithm based on the time difference of arrival estimation and iterative gradual optimization technology to explore the separation problem of mixed signals of cardiopulmonary sounds in complex environments with reverberations and noises, assisting clinical doctors in intelligent diagnosis. Firstly, a new model of time-frequency domain cardiopulmonary sound mixed signal is constructed, and the estimation of the mixed matrix is completed using the time difference of arrival estimation. Then, the model parameters are updated using iterative gradual optimization methods to reconstruct the iterative gradual optimization cardiopulmonary sound signals. Simulation experimental results show that the proposed algorithm achieves the separation of cardiopulmonary sound signals under linear and convolutive mixed models. Meanwhile, it has better separation performance and robustness than several popular blind separation algorithms.