线性预处理改善体音听诊自适应降噪性能的分析  

Improving adaptive noise reduction performance of body sound auscultation throughlinear preprocessing

作　　者：莫鸿强[1,2] 田翔 李彬[1,2] 田军章[4] MO Hongqiang;TIAN Xiang;LI Bin;TIAN Junzhang(School of Automation Science and Engineering,South China University of Technology,Guangzhou 510641,P.R.China;Guangdong Engineering Research Center of Cloud-Edge-End Collaboration Technology for Smart City,Guangzhou 510641,P.R.China;School of Electronics and Information Engineering,South China University of Technology,Guangzhou 510641,P.R.China;Guangdong Second Provincial General Hospital,Guangzhou 510317,P.R.China)

机构地区：[1]华南理工大学自动化科学与工程学院,广州510641 [2]智慧城市云-边-端协同技术广东省工程研究中心,广州510641 [3]华南理工大学电子与信息学院,广州510641 [4]广东省第二人民医院,广州510317

出　　处：《生物医学工程学杂志》2024年第5期969-976,共8页Journal of Biomedical Engineering

基　　金：2021年广东省科技专项资金“大专项+任务清单”(210719145863737);广东省科技计划资助项目(2017B020210002);华南理工大学中央高校基本科研业务费项目(D2182650)。

摘　　要：当前体音听诊多采用基于最小均方误差准则的双通道自适应滤波算法实现环境音降噪;对于包含脉冲分量的非高斯信号,这类方法容易出现滤波器权值失调。与常用的变步长方法不同,本文引入线性预处理来克服这种现象。针对双通道归一化最小均方自适应降噪算法,分析线性预处理对于改善自适应降噪效果所起的作用,并探讨预处理环节的设计原则。分析结果表明:滤波器的稳态均方权值偏差正比于体音的方差而反比于副通道环境噪音的方差,当线性预处理参数设置得当时,可抑制体音信号中的尖峰,大幅减小体音的方差和功率谱密度,而且不明显地减小甚至可能增大副通道环境噪音的方差及其功率谱密度,如此即可减小权值失调从而显著提升环境音降噪效果。最后以心音的环境音降噪为例说明了如何设计预处理环节,并解释了它对于环境音降噪所起的作用。本文研究结果可为体音听诊的自适应降噪算法设计提供理论依据。Adaptive filtering methods based on least-mean-square(LMS)error criterion have been commonly used in auscultation to reduce ambient noise.For non-Gaussian signals containing pulse components,such methods are prone to weights misalignment.Unlike the commonly used variable step-size methods,this paper introduced linear preprocessing to address this issue.The role of linear preprocessing in improving the denoising performance of the normalized least-mean-square(NLMS)adaptive filtering algorithm was analyzed.It was shown that,the steady-state mean square weight deviation of the NLMS adaptive filter was proportional to the variance of the body sounds and inversely proportional to the variance of the ambient noise signals in the secondary channel.Preprocessing with properly set parameters could suppress the spikes of body sounds,and decrease the variance and the power spectral density of the body sounds,without significantly reducing or even with increasing the variance and the power spectral density of the ambient noise signals in the secondary channel.As a result,the preprocessing could reduce weights misalignment,and correspondingly,significantly improve the performance of ambient-noise reduction.Finally,a case of heart-sound auscultation was given to demonstrate how to design the preprocessing and how the preprocessing improved the ambientnoise reduction performance.The results can guide the design of adaptive denoising algorithms for body sound auscultation.

关 键 词：听诊 环境音降噪 归一化最小均方自适应滤波 线性预处理 

分 类 号：R318[医药卫生—生物医学工程] TN713[医药卫生—基础医学]