基于小波变换和维纳滤波的半导体器件1/f噪声滤波

Reduction of 1/f noise in semiconductor devices based on wavelet transform and Wiener filter

针对半导体器件中普遍存在的1/f噪声提出了一种结合了提升小波变换和维纳滤波器的处理方法.首先利用重新加权迭代最小二乘法拟合1/f噪声的功率谱曲线得到噪声参数的估计,从而选择恰当的小波.其次,对包含了1/f噪声的信号进行提升小波变换.考虑到小波变换对1/f噪声的白化作用,利用维纳滤波器对每一层小波系数进行处理.设计了最优全通滤波器以校正维纳滤波器的相频特性,使得小波系数经滤波后相位不变.最后利用提升小波逆变换获得被1/f噪声淹没的信号.利用实验检验了提出方法的有效性,实验数据采自用于微创外科手术机器人的力传感器.结果表明提出的方法能够有效抑制1/f噪声,并使传感器的分辨力提高了25%.

In order to reduce universal 1/f noise in semiconductor devices,a method with combining lifting wavelet transform and Wiener filter is presented.Firstly,the iteratively reweighted least square method is introduced to fit the power spectrum of 1/f noise and estimate its parameter,and then an appropriate wavelet can be selected.Secondly,the signal with 1/f noise is decomposed by lifting wavelet transform.Considering the fact that the wavelet transform whitens 1/f noise,Wiener filter is used to treat the wavelet coefficient of each layer.Allpass filter is optimized to adjust the phase frequency response of Wiener filter,and the phase of filtered wavelet coefficient is not changed.Finally,the useful signal embedded in 1/f noise is retrieved by the inverse lifting wavelet transform.Experimental study demonstrates the proposed procedure and verifies its effectiveness,and the experimental data are acquired from a force sensor developed forminimally invasive surgery robot.The results show that the method works very well inminimizing 1/f noise,and so the resolution of the sensor increases 25%.