Max62-1020081167

26.10.2023 Views

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/23390631How to Deal with Some Spurious Fringes in Fourier Transform InfraredSpectrometersArticle in Applied Spectroscopy · November 2008DOI: 10.1366/000370208786049132 · Source: PubMedCITATIONS5READS1002 authors:Jean-Joseph MaxUniversité du Québec à Trois-Rivières58 PUBLICATIONS 2,521 CITATIONSSEE PROFILECamille ChapadosUniversité du Québec à Trois-Rivières98 PUBLICATIONS 2,849 CITATIONSSEE PROFILEAll content following this page was uploaded by Jean-Joseph Max on 25 August 2015.The user has requested enhancement of the downloaded file.

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/23390631

How to Deal with Some Spurious Fringes in Fourier Transform Infrared

Spectrometers

Article in Applied Spectroscopy · November 2008

DOI: 10.1366/000370208786049132 · Source: PubMed

CITATIONS

5

READS

100

2 authors:

Jean-Joseph Max

Université du Québec à Trois-Rivières

58 PUBLICATIONS 2,521 CITATIONS

SEE PROFILE

Camille Chapados

Université du Québec à Trois-Rivières

98 PUBLICATIONS 2,849 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jean-Joseph Max on 25 August 2015.

The user has requested enhancement of the downloaded file.


spectroscopic techniques

How to Deal with Some Spurious Fringes in Fourier Transform

Infrared Spectrometers

JEAN-JOSEPH MAX AND CAMILLE CHAPADOS*

ITF Labs, 400 Montpellier, Montréal, QC, Canada, H4N 2G7 (J.-J.M); and Département de chimie–biologie, Université du Québec à Trois-

Rivières, Trois-Rivières, QC, Canada G9A 5H7 (C.C.)

Faulty fringes coming from an infrared spectrometer may creep into a

spectrum. Because these come from one faulty interferogram out of many

used to obtain the spectrum, these may pass unnoticed. However, they

cause some problems in the data treatment of factor analysis and other

spectral analysis. We present a method for detecting the faulty fringes and

give a simple method to eliminate them at the interferogram accumulation

level.

Index Headings: Faulty interferograms; Anomalous fringes; Correcting

procedure; Infrared spectroscopy; IR spectroscopy.

Received 2 March 2008; accepted 8 July 2008.

* Author to whom correspondence should be sent. E-mail: Camille.

Chapados@uqtr.ca.

INTRODUCTION

Fringes are often observed in infrared (IR) spectra. These

result from multiple reflections at the surface of polymer-like

thin films and IR cells with neatly polished IR windows

separated by micrometer spacers. 1–5 These fringes are sample

dependent. In Fourier transform infrared (FT-IR) spectrometry

other fringes are encountered that are instrument dependent. To

decrease the noise level in the spectra, 20, 50, 100, 1000, or

even more interferograms are accumulated, averaged, and

Fourier transformed to obtain the spectrum. There is a small

drift from one interferogram to another that the spectrometer

tracking device replaces at the center burst. In most cases the

displacement is minimal and the tracking device does its job

correctly. However, when the deviation is too great, the

tracking device does not operate properly and a deviant

interferogram is obtained that is averaged with the others. This

will introduce fringes that are added to the species spectrum. If

one deviant spectrum out of 100 or more is obtained, the fringe

pattern may pass unnoticed. However, when subtracting one

spectrum from another in order to look at the variation process

or for more sophisticated data treatment such as factor

analysis 6–12 and band-enhancement techniques, 13 these errant

fringes render the operation impossible.

When the above situation occurs in limited interferogram

accumulations such as twenty, one easily sees the spectrometer

malfunction; that set is discarded and another one is taken.

However, in a lengthy experiment where several interferogram

accumulations are made over a long period of time, one is

frustrated to find after data treatment that an errant interferogram

has crept into the accumulated spectra. One way to deal

with this matter is to redo the experiment, hoping that the

spectrometer will behave properly. Here we present an

experimental situation that illustrates the problem and the

method that we developed to replace the errant interferogram in

the proper track without discarding an otherwise good

experiment.

THEORETICAL CONSIDERATIONS

Most FT-IR spectrometers use a Michelson interferometer to

obtain interferograms. These are Fourier transformed (FT) by a

fast Fourier transform algorithm (FFT) to obtain the IR

intensity spectrum. Taking first the background power

spectrum and, second, the spectrum of the sample, their ratio

gives the transmission spectrum that is transformed into

absorbance units for quantitative analysis. FTs have linear

properties, i.e., the FT of a sum is equal to the sum of the

individual FTs. 14 This property is used in FT-IR spectroscopy

to increase the signal-to-noise ratio. Several interferograms

obtained by individual scans of the interferometer are averaged

so that the noise level decreases with the square root of the

number of scans.

In most commercial FT-IR spectrometers, the exact location

of the mobile mirror in the Michelson interferometer is not

measured. Rather, a laser beam is superimposed on the infrared

(IR) beam to produce an interference pattern in the interferometer.

These laser interferences are used to trigger the IR

interferograms: at each laser interference maximum (or

minimum), a measure of the intensity of light is obtained,

giving one data point of the interferogram. Due to the sampling

interval (k laser ), the maximum frequency obtained through FT of

the interferogram is given by ˜m FFT max ¼ (1/100k laser ). According to

Shannon’s theorem, the maximum frequency recovered for the

IR spectrum is half the latter value: ˜m max ¼ (1/200k laser ). 14

In the default mode, the Nicolet 510P spectrometer shows in

real time the center burst region of the interferograms after

centering. When we look at these we see from time to time that

one scan is far out of the center. More recent models (such as

the Nicolet Impact 420) do not display, in the default mode,

such interferograms. Therefore, the operator cannot see this

type of error, which occurs occasionally.

Because the exact location of the mobile mirror is not

measured, one needs to be sure that the interferograms to be

Volume 62, Number 10, 2008 APPLIED SPECTROSCOPY 1167

0003-7028/08/6210-1167$2.00/0

Ó 2008 Society for Applied Spectroscopy


averaged are properly centered. This is achieved by locating the

interferogram center bursts situated at the interferogram

maximum intensity peaks. These are centered at the same

position (position ‘‘0’’ on the path difference) after which they

are averaged. Every FT-IR spectrometer has its own modus

operandi. However, because of instrument peculiarities, it

occasionally happens that some of the interferograms are outside

the adjustment region, which is around 6 10 samples (this value

is only indicative because it is a manufacturer’s property). The

out-of-bound interferograms cannot be properly centered. In

such cases, the resulting averaged interferograms will contain an

erroneous intergerogram that will generate in the spectrum a

wave similar to a sine. The FT of one single spike is a true sine

wave, the frequency of which depends on its location in the input

function. 14

EXPERIMENTAL AND DATA TREATMENT

Infrared Measurements. The IR measurements are obtained

with a model 510P Nicolet FT-IR spectrometer that has

a DTGS (deuterium triglycine sulfate) detector. Two KBr

windows isolate the measurement chamber from the outside.

The samples are contained in a closed Circle cell (SpectraTech,

Inc.) equipped with a ZnSe crystal rod (8 cm long) in an

attenuated total reflection (ATR) configuration where the beam

is incident at an angle of 458 with the rod’s axis and makes 11

internal reflections of which nearly 6.4 are in contact with the

liquid sample. The cell effective number of reflections was

calibrated with the pure water spectrum at 25 8C given by

Bertie and Lan. 15

The spectral range of this system is 6000 to 650 cm 1 .

Compared to the rest of the range, the intensity of the source is

weaker in the 6000 to 4800 cm 1 region, which will generate

higher noise levels. The spectra are taken under a nitrogen flow

to ensure low CO 2 and water vapor in the spectrometer. Each

spectrum represents an accumulation of 140 scans at 2 cm 1

resolution (0.965 cm 1 sampling interval). Model 510P being a

single-beam spectrometer, a background reference is taken

with the empty cell carefully dried before measuring the sample

spectra.

The IR measurements consisted of obtaining the ATR

background and sample interferograms. These are transformed,

respectively, into spectral intensities R 0 and R using the usual

FFT algorithm, with the Happ–Genzel apodization function

and coefficient (0.46). The ratio of R/R 0 is the intensity I for the

spectral range being studied. Thereafter, the 5547 data points (I

(˜m) vs. ˜m (in cm 1 )) of each spectrum are transferred to a

spreadsheet program for numerical analysis. The intensities I

are transformed into absorbance units, log (1/I) (abbreviated in

some cases as AU). The numerical data of each interferogram

accumulation are transferred to the spreadsheet program for

data treatment.

Chemicals and Solutions. Deionized freshly distilled water

was used. The cell is filled with this water and tightly closed. A

heating device ensures a slow, orderly increase of the sample

temperature. This is measured with a copper–constantan

thermocouple imbedded in the cell. In a series of sample

measurements covering the temperature range 29.1 to 91.1 8C,

we found that out of the 492 spectra taken, five were obtained

at the constant temperature of (90.66 6 0.06) 8C. All the

spectra looked alike. However, closer examination showed

some differences where none were expected. The bad spectra

caused false data treatment, which alerted us to the matter and

prompted us to find a remedy.

Correction of the Improper Interferograms. The

Correction Equation. Each deviant interferogram that produces

an oscillation in the absorbance spectrum can be corrected. Let

Int(d) be the measured power at path length difference d,withd 2

[ a, þ b], where a and b belong to the set of natural numbers.

Int(d) is an accumulation of n scans that are averaged:

IntðdÞ ¼ 1 n

X n

i¼1

Int i ðdÞ

The perturbation of concern is due to one or more scans not

properly aligned to the center burst of the remaining

interferograms. Hence the perturbed interferogram, Int*(d),

can be written as follows:

" #

Int ðdÞ ¼ 1 X n p

Int i ðdÞþ Xp Int m ðd þ d m Þ ð2Þ

n

1

1

where Int i (d) represents the correctly co-added scans, Int m (d) is

the non-properly aligned interferograms, and p is the number of

these interferograms. Equation 2 is transformed into:

n 3 Int ðdÞ ¼ Xn

1

p

Int i ðdÞþ Xp 1

Int m ðd þ d m Þ

Therefore, the properly centered interferograms can be

obtained by the following:

X n

1

p

Int i ðdÞ ¼n 3 Int ðdÞ

X p

1

Int m ðd þ d m Þ

Finally, the corrected averaged interferogram is obtained by

Int corr ðdÞ ¼ 1

n

¼ 1

n

X n

p

Int i ðdÞ

p

" 1

#

p n 3 Int ðdÞ

X p

1

Int m ðd þ d m Þ

Since we do not have the individual interferograms, we use the

following approximation as a first-order perturbation:

Int m ðd þ d m Þ ’ Int ðd þ d m Þ

Equation 6 represents a first-order approximation because

Int m (d þ d m ) contains no perturbation, while Int*(d þ d m )

contains 1/n perturbation (one or more shifted scans out of n

scans). Hence, one obtains the equation used for correcting the

faulty interferograms:

" #

Int corr ðdÞ ’

n

n p Int 1 X p

ðdÞ Int ðd þ d m Þ ð7Þ

n

1

Determination of Parameters p and d m . In Eq. 7, the

parameters p and d m must be determined. We begin with d m .

Each shifted scan co-added to the proper ones will produce a

small sine wave after FFT. The period (in cm 1 ) of this sine

wave is directly determined by the shift, d m , and the

wavelength that determines the interferogram sampling step,

ð1Þ

ð3Þ

ð4Þ

ð5Þ

ð6Þ

1168 Volume 62, Number 10, 2008


FIG. 1. ATR IR spectra of liquid water at 90.66 6 0.06 8C. (a) Five

consecutive spectra (separated by 0.3 AU) obtained from 140 co-added scans.

(b) Difference spectra (separated by 0.05 AU) obtained from the spectra in (a):

spectrum n minus spectrum 4 or 2: (2–4); (1–2); (3–2); (3–4); and (5–4) (see

text). Intensities are in ATR absorbance units (AU).

k laser , as follows:

Period ¼ ð1=100k laserÞ 1

¼

dm 100k 3 dm

The constant k laser is the laser wavelength (here it is 632.8

nm). Since an approximation of the period can be obtained

from the perturbed spectrum, an iterative procedure starting

from the estimated value of d m and monitored by the resultant

interferograms can be operated to obtain the exact value of d m .

The amplitude of each shifted interferogram is a multiple of the

inverse of the total number of co-added scans: 1/n.

RESULTS AND DISCUSSION

Experimental Spectra. Starting Spectra. To illustrate the

usefulness of the method, we present five spectra recorded

successively at a temperature that varied less than 60.06 8C.

Figure 1a shows the five original spectra without any

correction, some of which contain sine waves. These are not

evident on casual observation. However, they can be detected

by careful evaluation in the 5650 to 3650 cm 1 region, where

no sample band is present. We describe a better way to observe

them in the next section.

The Difference Spectra. Figure 1b illustrates the difference

between spectra where the numbers refer to the original spectra

of Fig. 1a. The difference spectra enhance the perturbing sine

ð8Þ

FIG. 2. (a) Interferograms of the water spectra in Fig. 1a (separated by 1000

amplitude units). (b) Differences between the interferograms (separated by 10

amplitude units): (2–4); (1–2); (3–2); (3–4); (5–4) (see text).

waves in three of the five spectra in Fig. 1a. The top spectrum

in Fig. 1b (spectrum 2 minus spectrum 4) shows a null

spectrum (only noise remains). This indicates that no

oscillating pattern is present in spectra (2) and (4) and that

these are of good quality. However, the other difference spectra

in Fig. 1b show three different oscillating patterns: in Fig. 1b

(trace 1–2), a large period (around 895 cm 1 ) is present; the

three others (traces 3–2, 3–4, and 5–4) contain shorter periods

(around 370 cm 1 ), two of which have twice the amplitude of

the bottom difference spectrum (trace 5–4).

At this stage, we can expect that the use of difference

spectra, abundantly used in factor analysis (FA), will present

difficulties in data treatment. If perturbed spectra are obtained,

one can either discard the faulty spectra, which will result in a

gap in the series, or take another series of measurements. The

latter is time consuming, and success is not assured.

The Experimental Interferograms. The oscillations observed

in the difference spectra (Fig. 1b) come from faulty

interferograms generated by the instrument.

Original Interferograms. Figure 2a shows the five interferograms

related to the five spectra of Fig. 1a. For clarity, only

the center burst region is shown ( 100 , d , 100). All five

interferograms look alike, with minute differences not evident

at first sight. These are revealed in the next section.

The Difference Interferograms. The difference interferograms

are shown in Fig. 2b with an amplitude scale expansion

of 120. The top interferogram difference (2–4) shows very low

amplitude in the center burst region (around 1/780 ¼ 0.13% of

APPLIED SPECTROSCOPY 1169


FIG. 3. (a) Difference interferograms and (b) spectra obtained after correction

of the perturbed interferograms: (1), (3), and (5) in Fig. 2a (see text). The

differences and presentation are the same as in Figs. 2b and 1b, respectively.

the maximum intensity). Both averaged interferograms (accumulation

of 140 scans) are almost identical without any side

burst. The small intensity difference comes from low

temperature 16,17 and instrument differences between the two

spectra. These are normal and do not generate a sine pattern in

the spectra. From this we conclude that interferograms (2) and

(4) are good and that the corresponding spectra are devoid of

spurious fringes.

Using the good difference interferogram (2–4) as the

measuring stick, we find (Fig. 2b) that the interferogram

differences (1–2, 3–2, 3–4, and 5–4) are faulty because side

bursts are observed. This indicates that interferograms (1, 3,

and 5) and related spectra are faulty. Interferogram (1) is

perturbed in the following way: the difference interferogram

(1–2) displays two patterns similar to the center burst of

interferogram (1), one positive exactly at the center burst

position and one negative at a location 17. From Eq. 8, we

find that position 17 corresponds to an oscillation period of

930 cm 1 [1/(100 3 632.8 nm 3 17)]. This value is close to the

estimated value from Fig. 1b (895 cm 1 ). The positive

perturbation at the center burst location is due to a missing

scan that could not be properly centered. Recall that the normal

interferogram center burst is negative (Fig. 2a). The amplitude

of 5.5 is related to interferogram (2) center burst intensity value

of 780 to give around 0.705%. One over this value gives

around 142, which corresponds almost to the number of scans

(140). This indicates that the evaluation rationale is correct.

The faulty interferograms (1, 3, and 5) show perturbations

situated on one side or the other of the center burst position, as

illustrated in the interferogram differences (Fig. 2b). For that of

the (1–2) difference, the counterpart is located at position 17.

This perturbation is similar to the original interferogram (Fig.

1a, trace 1), but shifted on the path length axis from the center

burst position. It has no counterpart on the positive side of the

path length axis. This clearly indicates that it represents one

shifted scan that had been added to the other properly centered

scans.

Interferogram (3) is perturbed because the difference

interferograms (3–2 and 3–4) (Fig. 3b) display two patterns

similar to the center burst of the original interferogram, one

positive at the exact center burst position and one negative at a

location (þ43). The position þ43 corresponds to an oscillation

period of 368 cm 1 , obtained from Eq. 8 [1/(100 3 632.8 nm 3

43)]. This value is the same as the fringe difference of 370

cm 1 observed in Fig. 1b (traces 3–2 and 3–4). This indicates

that the interferogram perturbation of spectrum (3) is different

than that of spectrum (1). The perturbation amplitude is almost

twice that in interferogram (1). This means that two shifted

scans were co-added. Since the counterpart to the center burst

perturbation is located at one single place (the þ43 position)

with twice the amplitude of the (1–2) difference interferogram,

it means that, in this case, two faulty scans are present but

shifted exactly at the same position. Finally, interferogram (5)

is perturbed similarly but with half the amplitude of the

perturbation of interferogram (3).

The Corrected Interferograms. With the above information

on the perturbed interferograms, one can now correct the

perturbation by applying Eq. 7 with the determined parameters

d m and p. For interferograms (1, 3, and 5), they are ( 17, 1;

þ43, 2; and þ43, 1), respectively. Using the same presentation

as in Fig. 2b, Fig. 3a shows the interferogram differences

obtained after correction of the three perturbed interferograms.

The four bottom difference interferograms are similar to the top

one, which was obtained with the original two good

interferograms. As before, the small intensity differences come

from the slight temperature and instrumental differences

between the spectra. Figure 3a shows clearly that the applied

corrections from Eq. 7 worked adequately.

Absorbance Spectra Obtained with the Corrected

Interferograms. Given the corrected interferograms, we can

now recalculate the corresponding spectra. To illustrate

convincingly the corrections made, we obtained the same

difference spectra as those illustrated in Fig. 1b, which showed

faulty spectra. These are displayed in Fig. 3b. The five ‘‘null’’

spectra show only noise whose intensities are stronger in

regions of low light intensities where strong absorption bands

are present (m OH , 3600–3100 cm 1 ; d HOH , 1680–1600 cm 1 ;

and m L , , 850 cm 1 ) and in the region where the IR beam is

weak (.4050 cm 1 ). These indicate eloquently that the

corrected spectra are without spurious fringes (Fig. 3b).

CONCLUSION

In the long series of spectra obtained in FT-IR spectrometers

coming from series of interferograms, a faulty interferogram

may creep in among the good ones. The FT of these will give

the IR spectrum, but the faulty one will generate a sine wave

pattern that is added to the spectrum. Because the faulty

The pattern is not exactly a sine wave. Attempts to subtract a sine curve

from the absorbance spectrum were fruitless.

1170 Volume 62, Number 10, 2008


interferogram is only one out of the many (usually more than

20 and often more than 100) taken for the spectrum, the

spurious fringe may pass unnoticed because it is weak. For

qualitative identification of a sample, this minimally affects the

bands and usually does not influence their assignment.

However, for quantitative measurements, the spurious fringes

perturb the data treatment, which cannot be adequately

completed. By taking the difference between two spectra, we

show how to detect the spurious fringes. From these

differences, the correcting parameters to be introduced into

the correcting equation are determined. Applying the correction

to the three faulty interferograms corrected these adequately,

which yielded the spectra after FT. The difference spectra of

these showed no spurious fringes and illustrate eloquently the

effectiveness of the corrective procedure.

ACKNOWLEDGMENTS

This work was supported in part by NSERC of Canada and in part by ITF

Labs.

1. T. Hirschfeld and A. W. Mantz, Appl. Spectrosc. 30, 552 (1976).

2. D. J. Moffat and D. G. Cameron, Appl. Spectrosc. 37, 566 (1983).

3. H. M. Heise, Appl. Spectrosc. 41, 88 (1987).

4. P. J. Farrington, D. J. T. Hill, J. H. O’Donnell, and P. J. Pomery, Appl.

Spectrosc. 44, 901 (1990).

5. M. F. Faggin and M. A. Hines, Rev. Sci. Instrum. 75, 4547 (2004).

6. J.-J. Max and C. Chapados, J. Phys. Chem. A 105, 10681 (2001).

7. J.-J. Max and C. Chapados, J. Chem. Phys. 116, 4626 (2002).

8. J.-J. Max, S. Daneault, and C. Chapados, Can. J. Chem. 80, 113 (2002).

9. J.-J. Max and C. Chapados, J. Chem. Phys. 119, 5632 (2003).

10. J.-J. Max and C. Chapados, J. Chem. Phys. 120, 6625 (2004).

11. J.-J. Max and C. Chapados, J. Chem. Phys. 122, 014504 (2005).

12. J.-J. Max and C. Chapados, J. Phys. Chem. 111, 2679 (2007).

13. C. Chapados, M. Trudel, and J. Miletic, Chemom. Intell. Lab. Syst. 22, 209

(1994).

14. J. Max, in Méthodes et techniques de traitement du signal et applications

aux mesures physiques (Masson, Paris, 1985), Tome 1, 4 e ed.

15. J. E. Bertie and Z. Lan, Appl. Spectrosc. 50, 1047 (1996).

16. J.-J. Max, V. Gessinger, C. van Driessche, P. Larouche, and C. Chapados,

J. Chem. Phys. 126, 184507 (2007).

17. J.-J. Max and C. Chapados, J. Chem. Phys. 126, 154511 (2007).

APPLIED SPECTROSCOPY 1171

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!