204
C. W. Huck
For example, Consumer Physics designed a pocket-size spectrometer (SCiO), with
dimensions of 67.7 × 40.2 × 18.8 mm and weight of 35 g, available at a very low
price. This was accomplished through incorporation of a LED (IrED) light source and
a silicon detector in the form of a 4 × 3 photodiode array with optical filters over the
individual pixels. However, the device operates over a narrow Vis/SW-NIR spectral
region (740–1070 nm; 13,514–9346 cm
−1 ) with a rather poor spectral resolution of
ca. 28 nm because of just 12 resolution elements and sub-par S/N of the measured
spectra.
8.3 Application and In-depth Evaluation of Performance
Characteristics of Portable NIR Spectrometers
Variety of the technology solutions and miniaturization itself has a meaningful impact
on the operating parameters and performance of handheld NIR spectrometers. The
key characteristics such as the working spectral region, spectral resolution, sensitivity, and S/N, of such spectrometers differ from those available on benchtop instrumentation. These issues influence the applicability and analytical performance of
miniaturized NIR spectrometers. It is now an active field of research from several
scientific group to perform systematic evaluation studies of different handheld NIR
spectrometers in variety of analytical applications.
Various approaches can be helpful in examining the analytical worthiness of
miniaturized spectrometers. The most straightforward and definitive evaluation of the
analytical accuracy is provided by the statistical errors of multivariate analysis, e.g.,
quantitative models constructed for prediction of the chemical contents or classification models for qualitative discrimination between samples. Correlation coefficients
for regressions (e.g., by means of partial least squares), either for cross-validations
or test-set validations, and root mean square errors (of cross-validation, calibration,
estimation) deliver numerical values indicating the worthiness of a given spectrometer. However, these values are only valid for the given, particular application and
are neither easily interpretable nor transferrable to other scenarios. As such, these
are not sufficient, if one aims for comprehensive evaluation of the instrumentation or
prediction how it should behave in other more or less similar scenarios. Therefore,
some other approaches can be helpful in obtaining a more general overview of the
concerning problem. Comparative measurements of the same sample sets on highperforming benchtop NIR spectrometers, with underlying reference analysis based on
gold standard methods of analytical chemistry (e.g., chromatography coupled to mass
spectrometry) are indispensable for establishing the performance level in best-case
scenarios. Differences between the devices in the wavelength-dependent sensitivity
levels can be easily visualized and assessed by performing 2D hetero-correlation
analysis, in which spectra measured on different NIR spectrometers (e.g., miniaturized vs. benchtop) can be directly correlated. This approach should be repeated for
different samples, as well as experimental conditions, to outline the performance
C. W. Huck
For example, Consumer Physics designed a pocket-size spectrometer (SCiO), with
dimensions of 67.7 × 40.2 × 18.8 mm and weight of 35 g, available at a very low
price. This was accomplished through incorporation of a LED (IrED) light source and
a silicon detector in the form of a 4 × 3 photodiode array with optical filters over the
individual pixels. However, the device operates over a narrow Vis/SW-NIR spectral
region (740–1070 nm; 13,514–9346 cm
−1 ) with a rather poor spectral resolution of
ca. 28 nm because of just 12 resolution elements and sub-par S/N of the measured
spectra.
8.3 Application and In-depth Evaluation of Performance
Characteristics of Portable NIR Spectrometers
Variety of the technology solutions and miniaturization itself has a meaningful impact
on the operating parameters and performance of handheld NIR spectrometers. The
key characteristics such as the working spectral region, spectral resolution, sensitivity, and S/N, of such spectrometers differ from those available on benchtop instrumentation. These issues influence the applicability and analytical performance of
miniaturized NIR spectrometers. It is now an active field of research from several
scientific group to perform systematic evaluation studies of different handheld NIR
spectrometers in variety of analytical applications.
Various approaches can be helpful in examining the analytical worthiness of
miniaturized spectrometers. The most straightforward and definitive evaluation of the
analytical accuracy is provided by the statistical errors of multivariate analysis, e.g.,
quantitative models constructed for prediction of the chemical contents or classification models for qualitative discrimination between samples. Correlation coefficients
for regressions (e.g., by means of partial least squares), either for cross-validations
or test-set validations, and root mean square errors (of cross-validation, calibration,
estimation) deliver numerical values indicating the worthiness of a given spectrometer. However, these values are only valid for the given, particular application and
are neither easily interpretable nor transferrable to other scenarios. As such, these
are not sufficient, if one aims for comprehensive evaluation of the instrumentation or
prediction how it should behave in other more or less similar scenarios. Therefore,
some other approaches can be helpful in obtaining a more general overview of the
concerning problem. Comparative measurements of the same sample sets on highperforming benchtop NIR spectrometers, with underlying reference analysis based on
gold standard methods of analytical chemistry (e.g., chromatography coupled to mass
spectrometry) are indispensable for establishing the performance level in best-case
scenarios. Differences between the devices in the wavelength-dependent sensitivity
levels can be easily visualized and assessed by performing 2D hetero-correlation
analysis, in which spectra measured on different NIR spectrometers (e.g., miniaturized vs. benchtop) can be directly correlated. This approach should be repeated for
different samples, as well as experimental conditions, to outline the performance