Laboratory Experiment

pubs.acs.org/jchemeduc

Fourier Transforms Simplified: Computing an Infrared Spectrum

from an Interferogram
Quentin S. Hanley*
School of Science and Technology, Nottingham Trent University, Clifton Lane, Nottingham NG11 8NS, United Kingdom
*
S Supporting Information

ABSTRACT: Fourier transforms are used widely in chemistry and allied sciences. Examples
include infrared, nuclear magnetic resonance, and mass spectroscopies. A thorough understanding
of Fourier methods assists the understanding of microscopy, X-ray diffraction, and diffraction
gratings. The theory of Fourier transforms has been presented in this Journal, but detailed practical
exercises are limited. A lab-based experiment is described that begins with the acquisition of an
interferogram, provides background and instructions for computing and interpreting transformed
data using widely available software, applies the principles of Connes’ advantage and sampling
theory to calibrate the spectrum, and finishes with an infrared spectrum. Related treatment of free
induction decay data from an NMR spectrometer is provided in Supporting Information.

KEYWORDS: Upper-Division Undergraduate, Analytical Chemistry, Laboratory Instruction, Computer-Based Learning,

Hands-On Learning/Manipulatives, Fourier Transform Techniques

F ourier transforms enter the science of chemistry through

the structural and characterization tools of the working
chemist: NMR, IR, X-ray crystallography, and mass spectrometry.
converts the intensities from mirror position to frequency. In the
related techniques of FT-NMR and FT-MS, the transform
converts time dependent waveforms to frequencies. In the first
They also arise in a host of other techniques and related disciplines. case (NMR), the frequencies are interpreted after conversion to
A number of educationally oriented articles have appeared in this chemical shift. In the second case (MS), frequency is related to
Journal on the topic of Fourier transforms and FTIR going back mass based on knowledge of the magnetic field strength. Common
over 40 years.1−37 These include extensive discussions of the to FTIR, FT-NMR, and FT-MS is a data set, g, consisting of
instrumentation and simulations supported by specialized spread-
N evenly spaced discrete data samples that are transformed over
sheets. Practical Fourier methods supported by computational
the data index, n, to obtain a set of Fourier coefficients, G,
exercises are rarely taught, and although there are a number of
excellent texts on Fourier methods (cf. 38, 39), these tend to be corresponding to specific frequencies, f:
highly mathematical making understanding difficult for the N−1
beginning student. Fourier methods are discussed in introductory Gf = ∑ gn e2πifn / N
textbooks in a conceptual way40 with limited computations. This n=0 (1)
gap may be bridged with available software and requires a level of
sophistication similar to that required for linear regression. This discrete Fourier transform (DFT) (eq 1) when unoptimized
This exercise provides students an opportunity to collect and requires a number of operations proportional to N2. Fast Fourier
process an interferogram from a standard FTIR. The steps consist transforms (FFTs) require fewer operationsproportional to N
of loading the data into a spreadsheet, applying the Fourier log2 Nresulting in large computational savings.39,41,42
transform, computing the complex magnitude, presenting the data Historically, FFT algorithms were discovered independently
as a transmission spectrum, and finally, applying the Nyquist limit multiple times over a period of approximately 150 years with Carl
and Connes’ advantage to calibrate the frequency axis. Friedrich Gauss (1805) currently credited with the first description

■ FOURIER TRANSFORMS
DFTs and FFTs
in a work published posthumously in 1866.43,44 The modern use
of FFTs began with the work of James Cooley and John Tukey45
with one early application being crystallography.46
A modern FTIR instrument collects IR intensity data at different
position settings of a moving mirror. The Fourier transform Published: January 26, 2012
© 2012 American Chemical Society and
Division of Chemical Education, Inc. 391 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396
Journal of Chemical Education Laboratory Experiment

The most commonly applied FFTs require the number of transform will be a table of complex numbers corresponding to
data points to be a power of 2. This requirement is a property different frequencies.48 For example,
of a particular class of algorithm and is neither a requirement of
all “fast” Fourier transforms nor a requirement of the work h(x) = 1 + sin(x) + 0.33 cos(3x) + 0.33 sin(3x)
described by Cooley and Tukey. More extensive discussions of + 0.20 sin(5x) + 0.15 sin(7x)
specific algorithms for performing the transform may be found
elsewhere;38,39 however, it is worth noting that FFTs are can be expressed as a table of Fourier coefficients (Table 1).
available for all data lengths (N) including prime numbers and Although Table 1 contains only integer frequencies, Fourier
multiple dimensions (such as might be useful for 2D-NMR).47
Complex Number Representation of Fourier Coefficients Table 1. Fourier Coefficients and Equations
The Fourier coefficients, Gf, produced by the FT operation, scale a Frequency ( f) Fourier Coefficient (Gf)a
set of sine and cosine terms at different, equally spaced, 0 1 + 0i
frequencies, which recreate the original data set when summed. 1 0 + 1i
By convention, the Fourier coefficients are usually presented as 2 0 + 0i
complex numbers with the real part representing the cosine and 3 0.33 + 0.33i
the imaginary part the sine. This convention arises from Euler’s 4 0+0i
relationship (eq 2) that provides a link between the complex 5 0 + 0.20i
exponential (eq 1) and the more familiar sine and cosine functions. 6 0 + 0i
7 0 + 0.15i
e ix = cos x + i sin x (2)
a
Fourier coefficients corresponding to the equation, h(x) = 1 + sin(x) +
Complex numbers, C, have the form 0.33 cos(3x) + 0.33 sin(3x) + 0.20 sin(5x) + 0.15 sin(7x).
C = A + Bi (3)
coefficients obtained from laboratory data can represent a wide
where A and B are the real and imaginary parts of the complex range of real number values. Discrete Fourier transformation of
number and i is the square root of −1. Complex numbers may be a data set always produces an integer number of frequencies at
represented as a magnitude, |C|, and a phase angle, ϕ. evenly spaced intervals. Calibrating and interpreting the frequencies
requires an understanding of the data collection procedure of the
⎡ B⎤
|C | = A2 + B2 and ϕ = tan−1⎢ ⎥ specific instrument producing the data.
⎣ A⎦ (4)
The Nyquist Limit and Aliasing
|C| is also referred to as the modulus or the absolute value. The The maximum frequency from a Fourier transformed data set is
relationship between |C| and ϕ and A + Bi may be understood determined by the interval spacing between the points. The
graphically (Figure 1). For this exercise, the important parameter Nyquist limit defines the maximum frequency and coincides
with the maximum frequency visible in the Fourier transformed
data:
f
fNyquist = measurement
2 (5)

For example, an interferogram sampled every 500 nm has a

Nyquist limit of 10,000 cm−1 (defined by 1/(2 × 500 nm)).
Wavenumbers above that limit “fold” into lower wavenumbers
via a process referred to as “aliasing”. Calibration of a spectrum
in wavenumber may be done through consideration of the
Nyquist limit.
Zero Filling
A set of N real data points produces N/2 frequencies. Under
these circumstances, resolution is lost due to closely spaced
Figure 1. Illustration of the complex plane. The plot shows the frequencies appearing as a single frequency. This can be
location of 2 + 6i in complex coordinates. The magnitude of the mitigated by padding the set of N points with N zeroes to
complex number |C| is given by (22 + 62)1/2 and ϕ by tan−1(6/2). The produce a data set with 2N points. Visually, improvement can
magnitude is proportional to intensity in an FTIR spectrum. be seen up to 3N zeroes. Zero filling may improve peak widths
in portions of a spectrum and provide better representation of
in the complex number is the magnitude because it is proportional bands lying midway between two frequency points in the
to the intensity of light at the frequency represented by a complex nonzero-filled spectrum. Additional details on zero-filling may
number. be found elsewhere,38 including specific discussion of IR.49
Reconstructing a Waveform from Fourier Coefficients
The real and imaginary parts of a complex number produced by
a FT represent the cosine and sine terms at a particular
■ FTIR
The Interferometer and the Laser
frequency. The complex number C = 5 + 2i represents h(x) = 5 FTIR instruments for routine use contain 4 key components:
cos(x) + 2 sin(x), or, if preferred, h(x) = |C| cos(x − φ). For a an interferometer, IR source, detector (Figure 2), and a HeNe
given data set, the coefficients produced by the Fourier laser. Light entering the interferometer is directed to a beam
392 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396
Journal of Chemical Education Laboratory Experiment

Figure 2. Block diagram of a Michelson-type interferometer. The

locations O, S, and M refer to the positions of the optical beam
splitter, the stationary mirror, and the moving mirror, respectively.
When length(OS) = length(OM) (dashed line), the path lengths are
equal and all wavelengths undergo constructive interference. Displace-
ment of M by λ/4 changes the δ by λ/2. Note that x = length(OM) −
length(OS), and δ = 2[length(OM) − length(OS)].

splitter where it is sent along two paths. The first path goes to a
fixed mirror. The second goes to a moving mirror. It then
returns to the beam splitter where it is recombined. When the
two paths are of equal length, constructive interference occurs
for all wavelengths. This condition is sometime referred to as Figure 3. Simulated interferogram (A) and the magnitude of its Fourier
the “center burst”. Moving the mirror a distance, x, causes the transform (B). The simulation consists of 9 terms where I0 values = [1,
path length to change by 2x. The path length change is referred 2, 3, 4, 5, 6, 7, 8, 9] and 2π/λ = [2, 4, 6, 8, 10, 12, 14, 16, 18].
to as the retardation, δ. Constructive interference for a specific The 9 terms (eq 11) were computed at 256 evenly spaced intervals
wavelength, λ, will occur when the retardation is an integer from −π to π.
multiple, n, of the wavelength:
δ = 2x = n λ where n = 1, 2, 3, 4,... (6)
Destructive interference occurs when the retardation is an
integer multiple of the wavelength plus a half:
⎛ 1⎞
δ = λ ⎜n + ⎟ where n = 1, 2, 3, 4,...
⎝ 2⎠ (7)
As the mirror moves, the intensity, I, measured for a particular
wavelength will vary:
I ⎡ ⎛ 2πδ ⎞⎤
I(δ) = 0 ⎢1 + cos⎜
⎝ λ ⎠⎥⎦
⎟
2⎣ (8)
A single wavelength directed through the interferometer will
produces a series of “fringes” (eq 8) as δ changes. If
some number, K, of wavelengths, λ, are present, a sum Figure 4. Raw interferogram. The full data set has been truncated to
over the intensities, Is, will produce a simulated interferogram show detail near the center burst (δ = 0).
(Figure 3):
wavelength intervals. Fellgett51 also noted the possibility of
K
Ik ,0 ⎡ ⎛ 2πδ ⎞⎤ using interference methods and a single detector to obtain this
IT (δ) = ∑ ⎢1 + cos⎜ ⎟⎥ advantage. Later analyses52−55 indicated that Fourier methods have
2 ⎢⎣ ⎝ λk ⎠⎥⎦ (9)
k=1 a S/N advantage ∼(N/8)1/2 under conditions where the detector’s
All frequencies up to the sampling limit are represented in noise limits the measurement. This latter point explains why FT
measurements of intensity as δ changes. The measurements methods are commonly employed in the IR portion of the
allow the whole IR spectrum to be collected simultaneously spectrum where detectors are comparatively noisy and rare in the
with the individual wavelengths being detected as different UV−vis portion where detectors are nearly perfect.
frequencies within the measured interferogram (Figure 4). The Raw Spectrum
The ability to measure all frequencies together is referred to as
Fellgett’s or the multiplex advantage.12 While comparing spectro- The interferogram (Figure 4) can be imported into a spreadsheet
graphs and spectrometers,50 Fellgett noted that a spectrograph and the Fourier transform computed. This produces a table of
operating for the same time should have a signal-to-noise (S/N) complex numbers that can be converted into magnitudes (eq 4) to
improvement depending on N1/2, where N is the number of produce a raw uncalibrated spectrum (Figure 5). A number of
393 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396
Journal of Chemical Education Laboratory Experiment

the measurements. The maximum frequency in wavenumber is

defined by one-half the frequency of acquisition (double the
wavelength of the monochromatic reference), as governed by
f Nyquist. The Nyquist limit for data collected at 632.8 nm
(15,803 cm−1) intervals in an FTIR instrument is 1.265 μm
(7901 cm−1) and each point along the wavenumber axis is
an equally spaced interval of (2f Nyquist/N). For example, if
N = 512, the spacing of points on the wavenumber axis will be
30.87 cm−1.
Presenting the Finished Spectrum
Presentation of the results is complete (Figure 6) after
conversion of magnitude data to percent transmittance (%T)
from the reference (I0) and sample (I) data.
Figure 5. Raw magnitude spectrum obtained from the source
background. The symmetry about the Nyquist limit is characteristic I
of fast Fourier transformation of a real data set. This symmetry is %T = 100%
sometimes referred to as “half complex” because only half the numbers I0 (12)
need to be stored.42 The shaded regions indicate portions of the
spectrum where low source intensity results in noisy transmittance
values. To calibrate the axis, use the relationship (2fNyquist/N) (see text
for details) with N = 4096. Note: the distribution from an ATR cell is
very different.

features are worth noting.1 The broad distribution is produced by

the source. The IR source produces an approximation to “black
body” radiation. Planck’s radiation law describes the spectral
energy density, Ub, of a black body source at a particular frequency,
ν, and temperature, T:56
8πhν 3 1
U b(ν) = 3 hν
c −1
ekT (10)
where c, h, and k are the speed of light, Planck’s constant, and
Boltzmann’s constant, respectively. Given a temperature, the Figure 6. Student data from a sample of 3.0 mil polystyrene. The
maximum can be estimated using Wien’s displacement law:2,57 spectrum has been restricted to 500−4000 cm−1 to avoid noise from
the shaded regions of Figure 5.
̃ = 1.97(cm−1 K −1)T
vmax
The spectrum was collected without a nitrogen purge resulting
(11)
■ EXPERIMENT
Procedure
in very strong absorbance by atmospheric water and carbon
dioxide.3 The measured intensity is low in portions of the Students are given instructions to set up the FTIR instrument to
spectrum (Figure 5, shaded regions), particularly below 400 cm−1 collect an interferogram and save the resulting data in an ASCII
and above 6800 cm−1. This is from non-ideal source behavior, format that can be imported into a standard spreadsheet. They
optical materials (particularly below 450 cm−1), and intentional transfer the data to a memory device for later work at home. They
filtering at high frequencies to avoid aliasing. are encouraged to bring samples of interest; however, samples are
The Nyquist Limit, Connes’ Advantage, and Wavenumber available for them in the lab. For this exercise, they collect a
Calibration background data set with no sample present, a polystyrene
reference material, and a sample of their choice.
A typical commercial FTIR has a HeNe laser emitting 632.8 nm
They are instructed to process the data through to completed
light. This light is visible to the eye and often mistaken for the
FTIR spectrum to obtain uncalibrated FTIR spectra of the
IR source. The laser light travels through the interferometer
samples. They then convert raw frequency to wavenumber to
along a slightly different optical path than the IR source and is
produce a calibrated spectrum using their understanding of the
detected separately. The laser produces an interferometric
Nyquist limit, verify that the calibration is correct against the
signal used to trigger data acquisition at evenly spaced intervals.
known spectrum of polystyrene, and assign characteristic bands
Often, but not universally, the FTIR triggers acquisition of an
arising from functional groups within the spectrum of the
intensity measurement once per interference fringe. This
background and sample.
provides an inherent calibration to the resulting spectrum.
Using a monochromatic reference interferogram to calibrate a Data Analysis
Fourier transformed spectrum was first described by Janine Data analysis was done with the assistance of the “Analysis
Connes 58 and has come to be known as Connes’ Toolpak” add-in available with MS Excel.
advantage.5,12,59
For a data set consisting of N measurements of the Equipment and Materials
interferogram, Connes’ advantage can be realized by under- The data were collected on a Thermo-Nicolet Nexus 670 FTIR
standing the relationship between the maximum frequency in instrument or a PerkinElmer Spectrum 100 equipped with an
the raw Fourier transformed data and the distance spacing of ATR cell.
394 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396
Journal of Chemical Education

■
Laboratory Experiment

HAZARDS (7) Blunt, J. W. J. Chem. Educ. 1983, 60, 97.

(8) Rabenstein, D. L. J. Chem. Educ. 1984, 61, 909.
There are no hazards.

■
(9) Macomber, R. S. J. Chem. Educ. 1985, 62, 213.
(10) Perkins, W. D. J. Chem. Educ. 1986, 63, A5.
RESULTS AND DISCUSSION (11) Perkins, W. D. J. Chem. Educ. 1987, 64, A296.
Students have universally been able to complete the Fourier (12) Perkins, W. D. J. Chem. Educ. 1987, 64, A269.
transform of the interferogram. Results (Figure 6) show good (13) Glasser, L. J. Chem. Educ. 1987, 64, A306.
correspondence with known compounds and good wave- (14) Glasser, L. J. Chem. Educ. 1987, 64, A260.
number calibration. No systematic discrepancies have been (15) Glasser, L. J. Chem. Educ. 1987, 64, A228.
noted. (16) Chesick, J. P. J. Chem. Educ. 1989, 66, 413.

■
(17) Chesick, J. P. J. Chem. Educ. 1989, 66, 283.
(18) Chesick, J. P. J. Chem. Educ. 1989, 66, 128.
CONCLUSION (19) King, R. W.; Williams, K. R. J. Chem. Educ. 1989, 66, A243.
Fourier processing is easily done with readily available software. (20) King, R. W.; Williams, K. R. J. Chem. Educ. 1989, 66, A213.
Good FTIR instrumentation is available nearly universally in (21) King, R. W.; Williams, K. R. J. Chem. Educ. 1990, 67, A100.
chemistry teaching laboratories. Over a decade of use, all (22) Williams, K. R.; King, R. W. J. Chem. Educ. 1990, 67, A125.
students have been able to do the Fourier transformation and (23) Williams, K. R.; King, R. W. J. Chem. Educ. 1990, 67, A93.
convert the complex numbers to intensity values. This includes (24) Estler, R. C. J. Chem. Educ. 1991, 68, A220.
those with the weakest mathematical backgrounds. A minority (25) Foley, J. A. J. Chem. Educ. 1991, 68, 889.
(26) Bell, H. M. J. Chem. Educ. 1993, 70, 996.
of students have had difficulty calibrating the wavenumber axis (27) Fuson, M. M. J. Chem. Educ. 1994, 71, 126.
using the Nyquist limit. Students struggle most with basic (28) Graff, D. K. J. Chem. Educ. 1995, 72, 304.
concepts and skills, such as confusion about I, I0, T, %T, and (29) Grunwald, E.; Herzog, J.; Steel, C. J. Chem. Educ. 1995, 72, 210.
absorbance, assigning bands to functional groups in spectra, and (30) Bettis, C.; Lyons, E. J.; Brooks, D. W. J. Chem. Educ. 1996,
testing the reasonableness of their results by comparison with 73, 839.
known values. The experiment reinforces these concepts while (31) Doscotch, M. A.; Evans, J. F.; Munson, E. J. J. Chem. Educ. 1998,
introducing practical Fourier methods, sampling criteria, the 75, 1008.
use of complex numbers to represent sine and cosine terms, (32) Sesi, N. N.; Borer, M. W.; Starn, T. K.; Hieftje, G. M. J. Chem.
and Connes’ advantage. The exercise provides an accessible Educ. 1998, 75, 788.
introduction to practical Fourier transforms in chemistry. The (33) Clegg, W. J. Chem. Educ. 2004, 81, 908.
data processing portion may be used on its own to demystify an (34) Besalu, E. J. Chem. Educ. 2006, 83, 1795.
(35) Zielinski, T. J. J. Chem. Educ. 2008, 85, 1708.
intimidating concept or as a supplement to the wide range of (36) Overway, K. J. Chem. Educ. 2008, 85, 1151.
existing highly mathematical treatments. Similar processing can (37) Hoffmann, M. M. J. Chem. Educ. 2009, 86, 399.
be applied to the raw NMR free induction decay (FID) and a (38) Marshall, A.; Verdun, F. Fourier Transforms in NMR, Optical, and
sample data set is included in the Supporting Information.

■
Mass Spectrometry: A User’s Handbook; Elsevier: Amsterdam , New York,
1990.
ASSOCIATED CONTENT (39) Press, W. H.; Teukolsky, S. A.; Vetterling, W. T.; Flannery, B. P.
*
S Supporting Information Numerical Recipes in C: The Art of Scientific Computing, 2nd ed.;
Cambridge University Press: Cambridge, 1992.
Instructions for students; instructor’s notes; discussion (40) Harris, D. Quantitative Chemical Analysis, 7th ed.; W.H.
questions and outline answers are available to assist those Freeman and Co.: New York, 2007.
wishing to adopt this as a laboratory exercise; a practice data set (41) Embree, P. M.; Kimble, B. C Language Algorithms for Digital
consisting of reference and sample interferograms is available in Signal Processing; Prentice Hall: Englewood Clifs, NJ, 1991.
Excel format; for those wishing to apply Fourier processing in (42) Galassi, M.; Davies, J.; Theiler, J.; Gough, B.; Jungman, G.;
another context, an NMR data set and analysis guidance notes Alken, P.; Booth, M.; Rossi, F. GNU Scientific Library Reference
have been provided. This material is available via the Internet at Manual, 3rd (v1.12) ed.; Network Theory Limited: Boston, MA, 2009.
http://pubs.acs.org. (43) Heideman, M. T.; Johnson, D. H.; Burrus, C. S. Archive for

■ AUTHOR INFORMATION
Corresponding Author
History of Exact Sciences 1985, 34, 265.
(44) Briggs, W. L.; Henson, V. E. The DFT: An Owner’s Manual for
the Discrete Fourier Transform; Society for Industrial Mathematics:
Philadelphia, PA, 1987.
*E-mail: quentin.hanley@ntu.ac.uk.

■
(45) Cooley, J.; Tukey, J. Math. Comput. 1965, 19, 297.
(46) Rockmore, D. N. Comput. Sci. Eng. 2000, 2, 60.
ACKNOWLEDGMENTS (47) Frigo, M.; Johnson, S. G. Proc. IEEE 2005, 3, 216.
The author thanks the many students at Nottingham Trent (48) Sometimes, the output of a real FFT consists of paired numbers
University and the University of the West Indies Cave Hill not explicitly including i. For details, see refs 39, 42.
(49) Herres, W.; J., G. Comput. Appl. Lab. 1984, 5, 352.
Campus who did versions of this exercise between 2002 and the (50) Fellgett considered spectrographs to be instruments measuring
present. He thanks Dr. Jon Tepper and CELS for encouragement the elements of a spectrum simultaneously and spectrometers to be
and Dr. Chris Garner for providing the NMR FID data set.

■
instruments that measure the spectrum an element at a time.
(51) Fellgett, P. J. Phys. Radium 1958, 19, 157.
REFERENCES (52) Tai, M. H.; Harwit, M. Appl. Opt. 1976, 15, 2664.
(1) Low, M. J. D. J. Chem. Educ. 1970, 47, A415. (53) Treffers, R. R. Appl. Opt. 1977, 16, 3103.
(2) Low, M. J. D. J. Chem. Educ. 1970, 47, A255. (54) Harwit, M.; Sloane, N. J. A. Hadamard Transform Optics, 1st ed.;
(3) Low, M. J. D. J. Chem. Educ. 1970, 47, A163. Academic Press: New York, 1979.
(4) Pearson, J. E. J. Chem. Educ. 1973, 50, 243. (55) Carli, B.; Natale, V. Appl. Opt. 1979, 18, 3954.
(5) Marshall, A. G.; Comisarow, M. B. J. Chem. Educ. 1975, 52, 638. (56) Ingle, J. D.; Crouch, S. R. Spectrochemical Analysis; Prentice Hall:
(6) Strong, F. C. J. Chem. Educ. 1979, 56, 681. Englewood Cliffs, NJ, 1988.

395 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396

Journal of Chemical Education Laboratory Experiment

(57) The wavelength dependent versions of both Planck’s radiation

law and Wien’s displacement law are more commonly presented in
textbooks and monographs. It is worth noting that the maximum in
equally spaced frequency intervals is not equivalent to the wavelength
dependent version.
(58) Connes, J. Rev. Opt. Theor. Exp. 1961, 40, 41.
(59) Connes’ original work used glass optics and was calibrated with
a red Cd lamp.

396 dx.doi.org/10.1021/ed2002253 | J. Chem. Educ. 2012, 89, 391−396