Application of advanced signal processing techniques in analysis of heart sound

The performance of different power spectral estimation techniques including classical FFT based techniques and so called modem parametric and nonparametric methods were investigated for their sensitivity to two common practical problems in heart sound recording analysis, i.e. the noise level in the recorded signal and the uncertainty in identifying the exact start and end points of major heart sounds, S₁, S ₂, and murmurs. The results of the comparison described show that no single method has superior performance with respect to all features of interest. However the EV and MUSIC eigen analysis-based techniques show a better performance in locating the dominant frequency peaks in both cases with the addition of noise and truncation in signal duration. The most dominant frequency peak DFP₁ can be evaluated by the Welch periodogram method more accurately than by other methods. However the results indicate that the Welch periodogram method does not show an acceptable performance when the signal is truncated. The results presented here indicate that in a practical situation with high level of noise and uncertainty in signal boundary, it is possible to extract the desired features and draw a conclusion about the optimum method to be used in a given situation