Signal and noise separation in medical diagnostic system based on independent component analysis

Robustly separating sound components in human body based on 2-ch ICA and EM algorithm with dirichlet distribution

“…However, the existing methods [2,3] do not work well if biosignals do not contain strong components of HS or BS because the EM algorithm is a method to separate BS, BSS and HS by use of expectations of each component (BS, BSS and HS in biosignals). That is, separation results lack robustness in this case.…”

“…Therefore, we propose band pass filter, 2-ch leA and EM algorithm with Dirichlet distribution to solve these problems. Herewith, we propose methods to separate BS, BSS and HS from biosignals more robustly than the existing method [2,3]. This paper is organized as follows.…”

“…In this experiment, we assume that the number of microphone sensors M and sources N are three. Existing methods using independent component analysis (leA) [3] and expectation-maximization (EM) algorithm used in time-frequency (T-F) masking [2] have been proposed to analyze biosignals. However, there is a problem of robustness for separate BS and HS.…”

Separation of cardiorespiratory sounds using time-frequency masking and sparsity

“…The method in [7] presents a sophisticated technique for heart sound recovery in the presence of possible signal conditions of saturation, reverberation and noise. The conventional ICA is used in the frequency domain for source separation.…”

“…The problem becomes more challenging in scenarios when other interfering sounds like snore, speech or traffic noise come into the scene. When the number of sources is greater than the number of observation mixtures, the BSS problem is degenerate and cannot be addressed through matrix inversion unmixing techniques (such as [1]- [7]). Most of the existing techniques developed for cardiac sound recovery (discussed in Section II) separate only the heart and lung sounds this, however, is not sufficient in certain scenarios, e.g.…”