一种用ICA去除瞬态诱发耳声发射伪迹的新方法

A NEW METHOD FOR ELIMINATING ARTIFACT IN TRANSIENT EVOKED OTOACOUSTIC EMISSIONS USING ICA

如何去除伪迹是瞬态诱发耳声发射检测中一个关键的问题。本研究提出了一种用ICA去除伪迹的新方法。首先用四组线性增长的刺激声在耳道内录音 ,得到的波形是瞬态诱发耳声发射和伪迹的混叠。因为伪迹和瞬态诱发耳声发射是统计独立的 ,而且伪迹随刺激声的变化线性增长 ,而瞬态诱发耳声发射随刺激声的变化非线性增长 ,逐渐趋于饱和 ,所以它们在混叠信号中具有不同的混叠系数。用ICA算法可以将各独立分量及混叠矩阵估计出来 ,伪迹是其中的一个独立分量。然后将伪迹的波形置零后再进行一次混叠 ,便达到了去除伪迹的目的。最后通过与传统的DNLR方法比较 。

How to eliminate the stimulus artifact from the TEOAEs is a key problem in TEOAEs measurement. In this paper, a new method to eliminate stimulus artifact by independent component analysis(ICA) is proposed. First, four linear increasing stimulating sounds are used to obtain the mixed waveforms of TEOAEs and stimulus artifacts. Because stimulus artifact and TEOAEs are independent statistically, and stimulus artifacts are line-increasing with stimulus while TEOAEs are nonlinear increasing with the trend to saturation gradually, their mix coefficients in mixed signals are different. The independent components and mix matrix are estimated using ICA algorithm by taking stimulus artifact as one of the independent components. Thas the artifact are subtracted by remixing the other independent components. Finally, compared with conventional DNLR algorithm, it is proved that the method we proposed is more effective.