抄録
Ambient noise such as machine noise and human voice often disturbs lung sound measurement. Frequency domain filtering is usually ineffective because of the spectral overlap of lung sound and ambient noise. Ambient noise is transmitted to the microphone measuring lung sound through the chest wall around it. Cancelling of the noise component may be possible by identifying this transfer function. The function, however, may vary with subject and measuring site, so that it should be modified dynamically. We applied adaptive filtering technique to solve this problem. An off-line adaptive noise canceller having a transversal filter was implemented in a workstation. Filter coefficients were controlled by LMS algorithm. Wide-band random noise and human voice were used as ambient noise. Noise corrupted lung sound and ambient noise were simultaneously recorded as the primary and reference input of the noise canceller. The performance of the canceller was assessed by changing canceller parameters and ambient noise level. Noise reduction was calculated as the ratio of the power of the canceller output over that of the canceller primary input for the data in the case when the subject held his breath. The result showed that the ambient noise was reduced by about 30dB using a 256-tap filter with the convergence time of several seconds. Clear lung sound can be heard by D/A conversion of the canceller output. The results show that this method is very effective as a preprocessing tool for the lung sound analysis, and that it is promising to realize an electronic stethoscope with high ambient noise immunity by real-time digital signal processing.
