Baseline offset

Changes in the differential reflectance spectra AR of the sensing film upon exposure to different vapors at various concentrations are presented in Fig. 4.9. These spectra illustrate several important findings. For polar vapors such as water and ACN (see Fig. 4.9a, b respectively), the differential reflectance spectra have a stable baseline and consistent well-behaved changes in the reflectivity as a function of analyte concentration. The response of the colloidal crystal film to nonpolar vapors such as DCM and toluene (see Fig. 4.9c, d respectively) is quite different compared with the response to polar vapors. There are pronounced analyte concentration-dependent baseline offsets that are likely due to... 

However, the influence of working electrode potential on background current (electrolysis current of mobile phase constituents) i.e. on baseline offset, stability and noise should also be taken into account. 

Electrolysis of mobile phase constituents will cause a continuous detector response (background current) resulting in a chromatographic baseline level that differs from the electrical detector zero-response level. The difference, baseline- offset, is an important analysis parameter, because baseline fluctuations (noise, drift) due to fluctuations in electrolysis conditions (potential, mobile phase flow rate, temperature) are proportional to baseline offset. See Figure 2-5 for an example of the influence of flow pulsation at different baseline offset... 

Obviously the optimum potential for detection of analyte X in this mobile phase is a compromise a higher potential will increase the peak height, but will also increase the baseline offset and thus the baseline noise and drift. Lower potentials will decrease the peak height, but also decrease the noise and drift. 

However, an investigation into the errors caused by baseline offset and noise in the least-squares estimation of exponential lifetimes has already been carried out by... 

While baseline offsets are usually not a problem in a laboratory cuvette, they can be a significant issue in on-line measurements. There are several sources of such offsets. 

This filtering preprocessing method can be used whenever the variables are expressed as a continuous physical property. One example is dispersive or Fourier-Transform spectral data, where the spectral variables refer to a continuous series of wavelength or wavenumber values. In these cases, derivatives can serve a dual purpose (I) they can remove baseline offset variations between samples, and (2) they can improve the resolution of overlapped spectral features. 

In spectroscopy applications, a hrst derivative effectively removes baseline offset variations in the spectral prohles. Second-derivative pretreatment resnlts in the removal of both baseline offset differences between spectra and differences in baseline slopes between spectra. Its historical effectiveness in NIR diffuse reflectance applications snggests that baseline slope changes are common in these applications, although there is no theoretical basis for snch variations. 

However, the benefit of EMSC over MSC is the ability to explicitly use prior knowledge of the spectroscopy and chemistry of the problem to both (a) enable better estimates of the baseline offset and multiplicative effects in the spectra, and (b) enable filtering/removal of spectral effects that are known or suspected to be irrelevant to the problem. In practice, the challenge in using EMSC effectively in PAT applications is often in the determination of spectral profiles to use in H, P and S. For example, S and P can be populated with measured, estimated, or even library spectra corresponding to relevant pure components and irrelevant... 

In addition to spectra of the reference minerals listed in Table II, the least-squares components in each iteration included 3 "spectra" representing 1) moisture in KBr blank (obtained by subtraction of 2 KBr blank spectra), 2) a constant baseline offset (1 abs from 4000 to 400 cm" ), and 3) a sloping linear baseline (line from 1 abs at 4000 cm" to 0 abs at 400 cm" ). The final mineral component concentrations were normalized to 100%, disregarding the contributions of the three artificial components. The normalized least-squares results for each sample were combined with the ash elemental composition of each reference mineral to calculate the elemental composition of the ASTM oxidized ash corresponding to each LTA. This was done by multiplying the concentration of each reference mineral in a sample by the concentration of each elemental oxide in the reference mineral, then summing over each oxide. 

Figure 3.12. Data with a baseline offset before (a) and after (b) baseline correction using the expfidt modeling approach.

Figure 5.27 displays the spectra of the pure components along with the validation samples. No individual spectrum appears anomalous, but again high absorbances are observed along with a slight baseline offset. 

FIGURE 5.39. Spectral residuals for ICLS Example 1, not accounting for baseline offset. 

To estimate the model (St), the 550 response variables in the region from 1100 to 2198 nm of the pure component spectra and a simple baseline offset are supplied to the computer. The resulting output from the validation samples includes the statistical prediction errors, estimated concentrations, concentration residuals, and spectral residuals. 

The calibration spectra displayed in Figure 5.33 reveal a random baseline offset for these data. No other anomalous behavior is observed. The concentration design has seven independent standards (see Figure 3.32), which are sufficient for estimating the three pure spectra. 

In this what if, tlie pure component spectra for Example 1 are estimated without first removing the baseline. The following results are observed (1) The resulting estimated pure spectra shown in Figure 5.37 have an offset. (2) The plot of the pure spectrum of component B with the 2 sd uncertainty band is shown in Figure 5.38. The 2-sd band is quite large and nearly constant across all variables. The width of the band relative to the features in the spectra qualitatively indicates a problem wdth the estimated pure. (3) The calibration residuals shown in Figure 5.39 look reasonably random except for the presence of a baseline offset. 

FIGURE 5.37. Estimated pure spectra, not accounting for random baseline offset. Solid, component A dashed, component B and dashed-dotted, component C. 

All wavelengths larger than 2200 nm are removed and the baseline offset is removed by zeroing each spectrum at 1100 nm. 

In spectroscopy applications, a first derivative effectively removes baseline offset variations in the spectral profiles. As a result, first derivatives can be very effective in many spectroscopy applications, where spectral baseline offset shifts between samples are rather common. 

---