用盲源分离技术创建独立的虚拟传感器

在振动与声测量中，由于结构对振动的传播作用以及声传播过程中散射与混响效应的存在，传感器(如加速度计或麦克风)所测得的信号往往是多个源的混合。盲源分离作为一种强有力的冗余取消工具，可以正确恢复独立源信号的波形。不过在具体实施中，所有的盲源分离算法都依赖于一个基本假设，即传感观测信号数必须大于或等于系统中的独立源数，而实际机器中独立源的数目往往未知。为此首先提出一种基于奇异值分解的聚类不相关源数估计新方法，估计一个系统中独立源数的上界，并籍此获得足够维数的传感观测信号，保证盲源分离方法在实际应用中的正确实施，从而共同构建一个能获取无法直接观测的独立源波形的虚拟传感观测系统。实验结果表明了该系统潜在的实用意义。

Creating Independent Virtual Sensors with Blind Source Separation

In vibratory and acoustic measurement,the observations by sensors (for example accelerator or microphone) are often mixtures of several independent sources.As a power tool to reduce redundancy,blind source separation(BSS) can restore the waveform of independent sources correctly.However,all algorithms of BSS rely heavily on such a basic hypothesis that there must be more sensors than sources,which,in a real-world system.are often unknown.A new method to determinate the number of incoherent sources based on SVD and clustering is proposed firstly,by which the upper limit of the number of independent sources embedded in a system can be estimated.By the use of our method,enough observations can be acquired,and BSS can be implemented correctly in practice.Thus,a virtual sensor system capable of capturing independent sources,which is inaccessible directly,is constructed.