Department of Electrical and Computer Engineering , Marquette University , Milwaukee , Wisconsin 53201-1881 , United States.
Department of Electrical Engineering, Center for Integrated Systems , Stanford University , Stanford , California 94305-4075 , United States.
ACS Sens. 2019 Jun 28;4(6):1682-1690. doi: 10.1021/acssensors.9b00564. Epub 2019 Jun 3.
Most chemical sensors are only partially selective to any specific target analyte(s), making identification and quantification of analyte mixtures challenging, a problem often addressed using arrays of partially selective sensors. This work presents and experimentally verifies a signal-processing technique based on estimation theory for online identification and quantification of multiple analytes using only the response data collected from a single polymer-coated sensor device. The demonstrated technique, based on multiple stages of exponentially weighted recursive least-squares estimation (EW-RLSE), first determines which of the analytes included in the sensor response model are absent from the mixture being analyzed; these are then eliminated from the model prior to executing the final stage of EW-RLSE, in which the sample's constituent analytes are more accurately quantified. The overall method is based on a sensor response model with specific parameters describing each coating-analyte pair and requires no initial assumptions regarding the concentrations of the analytes in a given sample. The technique was tested using the measured responses of polymer-coated shear-horizontal surface acoustic wave devices to multi-analyte mixtures of benzene, toluene, ethylbenzene, xylenes, and 1,2,4-trimethylbenzene in water. The results demonstrate how this method accurately identifies and quantifies the analytes present in a sample using the measured response of just a single sensor device. This effective, simple, lower-cost alternative to sensor arrays needs no arduous training protocol, just measurement of the response characteristics of each individual target analyte and the likely interferents and/or classes thereof.
大多数化学传感器对任何特定目标分析物的选择性都只是部分的，这使得分析物混合物的识别和定量变得具有挑战性，通常使用部分选择性传感器阵列来解决这个问题。本工作提出并实验验证了一种基于估计理论的信号处理技术，该技术仅使用从单个聚合物涂层传感器设备收集的响应数据，即可在线识别和定量多种分析物。所演示的技术基于多个阶段的指数加权递归最小二乘估计（EW-RLSE），首先确定传感器响应模型中包含的哪些分析物不存在于正在分析的混合物中；然后在执行 EW-RLSE 的最后阶段之前，将这些分析物从模型中消除，在该阶段，样品中的组成分析物被更准确地定量。该方法基于具有特定参数的传感器响应模型，这些参数描述了每个涂层-分析物对，并且不需要对给定样品中分析物的浓度进行初始假设。该技术使用聚合物涂层切向水平表面声波器件对苯、甲苯、乙苯、二甲苯和 1,2,4-三甲苯在水中的多分析物混合物的测量响应进行了测试。结果表明，该方法如何仅使用单个传感器设备的测量响应，准确地识别和定量样品中的分析物。与传感器阵列相比，这种有效、简单、成本更低的替代方法不需要艰苦的训练协议，只需测量每个目标分析物以及可能的干扰物及其类别个体的响应特性即可。
