You are on page 1of 11

364 Chapter 8

8.1 Fourier Transform Spectrometers


There are two basic FTS types discussed in this chapter: temporal FTS
instruments, which are described in this section as well as Section 8.2,
and spatial FTS instruments, described in Section 8.3. The most widely
used FTS approach is based on a Michelson interferometer (Pedrotti and
Pedrotti, 1987), as depicted in Fig. 8.1, and is called a temporal FTS for
reasons that will become apparent. In a Michelson interferometer, incident
light from a source is split into two paths by a beamsplitter, reflected off
mirrors in both paths, and recombined onto a detector. One path includes
a moving or controllable mirror, for which the optical path difference
(OPD) between the interfering wavefronts can be precisely controlled.
Assume that total transmission through one arm of an interferometer is
⌧1 and the other is ⌧2 . Theoretically, transmission is constrained such that
⌧1 + ⌧2 < 1. Ideally, ⌧1 = ⌧2 = 12 . To cancel the effect of any dispersion in
the beamsplitter material, a transparent substrate of an identical material
with the same thickness but maximum transmission is located in one arm
of the interferometer. This element is called a compensator.
At a certain position of the moving mirror, the OPD between interfering
paths becomes identically zero. This position is called the zero path
difference (ZPD) position and is indicated by the dotted line in Fig. 8.1.
Considering the source as an incident, monochromatic plane wave with
irradiance E0 and wavelength , irradiance at the detector for an arbitrary
OPD is given by
⌧p p 2
E( ) = ⌧1 E0 cos(!t kz) + ⌧2 E0 cos{!t k(z + )}
" p !#
⌧1 + ⌧2 2 ⌧1 ⌧2 2p
= E0 1 + cos , (8.1)
2 ⌧1 + ⌧2

Figure 8.1 Michelson interferometer setup.


Fourier Transform Spectrometer Design and Analysis 365

where k = 2p/ and ! = 2p⌫. In the ideal case, Eq. (8.1) simplifies to
" !#
1 2p
E( ) = E0 1 + cos . (8.2)
2

If d is the displacement of the mirror from the ZPD position and ✓ is the
incident angle of the plane wave relative to the optical axis, then the OPD
can be shown to be

= 2d/ cos ✓. (8.3)

In a nonimaging FTS, the incident angle is assumed to be zero. The


analysis that follows is clearer if wavenumber = 1/ is used instead
of the wavelength. With this substitution, Eq. (8.2) becomes

1
E( ) = E0 [1 + cos(2p )]. (8.4)
2

The behavior of the Michelson interferometer with monochromatic


light at two different wavelengths is depicted in Fig. 8.2. As the OPD
varies, a sinusoidal modulation occurs where the period of oscillation is
related to the wavelength of the light because of the spectral dependence
of the cosine argument in Eq. (8.4). At the ZPD position, light of
all wavelengths constructively interferes so that sinusoidal patterns for
differing wavelengths are all phased up at this corresponding mirror
position. If transmission of the two interferometer arms is not exactly one-
half, the same modulation occurs, but the overall transmission of the bias
component (the mean value of the interference pattern in Fig. 8.2) is

⌧1 + ⌧2
⌧b = , (8.5)
2

and the modulation depth (the ratio of the sinusoidal amplitude to the mean
value) is reduced to
p
2 ⌧1 ⌧2
m= . (8.6)
⌧1 + ⌧2

The transmission and modulation depth are assumed for now to equal 1
for notational simplicity but are factored in when radiometric sensitivity is
examined later in this chapter.
366 Chapter 8

Figure 8.2 Sinusoidal modulation of a Michelson interferometer for light at two


different wavelengths.

8.1.1 Interferograms
Now consider the response of the Michelson interferometer to a
polychromatic source with spectral irradiance E0 ( ). Detector irradiance
as a function of OPD is then given by the integral
Z 1
1
E( ) = E0 ( ) [1 + cos(2p )] d , (8.7)
0 2

which is referred to as an interferogram. As an example, consider the


irradiance spectra given in Fig. 8.3. These spectra are representative of
a 320-K blackbody radiator viewed through 100 ppm m of methane gas
at 280 K, captured by an ideal f /3 imaging system with a cold filter that
truncates the spectrum to the 3- to 12-µm spectral band. Note the strong
methane feature in the 7.5-µm region. This example is chosen because it
includes both an extended spectrum and some fine spectral features with
which to explore FTS operation. Two versions of the irradiance spectrum
are displayed: E ( ) in µW/cm2 µm units in Fig. 8.3(a), and E ( ) in
µW/cm2 cm 1 units in Fig. 8.3(b). The spectra are related by

E ( )= E ( ). (8.8)

Because it is more natural to use wavenumber units when dealing with


an FTS, the produced spectrum is in the form of E ( ) and not E ( ).
Fourier Transform Spectrometer Design and Analysis 367

100 12

Irradiance (µW/m2 cm-1)


Irradiance (µW/cm2 µm)
90
80 10
70
8
60
50 6
40
30 4
20
2
10
0 0
3 10 11 12 1000 1500 2000 2500 3000 3500
Wavelength (microns) Wavenumber (cm–1)
(a) (b)
Figure 8.3 Example methane spectra in both (a) wavelength and (b) wavenumber
units.

The interferogram corresponding to the spectrum example is depicted in


Fig. 8.4 for a maximum OPD of both 0.125 [Fig. 8.4(a)] and 0.02 cm
[Fig. 8.4(b)]. The interferogram in Fig. 8.4(b) is just the central part of the
interferogram in Fig. 8.4(a). In both cases, the mirror moves symmetrically
about the ZPD position, leading to both positive and negative OPDs and
making what is known as a two-sided interferogram. Since all wavelengths
interfere constructively at the ZPD position, there is a large central peak
to the distribution. As the magnitude of an OPD increases, sinusoidal
patterns for constituent wavelengths dephase, and the modulation depth
of the pattern about the bias level decreases. Note that this interferogram
is produced by varying the mirror displacement in time, hence the
designation as a temporal FTS.

0.12 0.12
Irradiance (mW/m2)
Irradiance (mW/m2)

0.1 0.1

0.08 0.08

0.06 0.06

0.04 0.04

0.02 0.02

0 0
-0.02 -0.015 -0.01 -0.005 0 0.005 0.01 0.015 0.02
-0.1 -0.05 0 0.05 0.1
OPD (cm) OPD (cm)
(a) (b)

Figure 8.4
maximum OPD.
368 Chapter 8

8.1.2 Spectrum reconstruction


Recognizing Eq. (8.7) as a discrete cosine transform, one would expect
to be able to reconstruct a spectrum from the interferogram by a Fourier
transform. Consider the Fourier transform of the interferogram, given by
Z 1
S( ) = E( ) ei2p d . (8.9)
1

Inserting Eq. (8.7) into Eq. (8.9) and separating the two terms leads to
Z 1 "Z 1 #
1
S( ) = E( ) d ei2p d 0 0
2 1
Z0 Z
1 1 1
+ E( 0 ) cos(2p 0
) ei2p d 0 d . (8.10)
2 1 0

The first term represents the Fourier transform of the bias term in Eq. (8.7)
and results in a very large peak-at-zero frequency proportional to the total
irradiance over the spectral range of the instrument. This term can easily
be removed by subtracting the mean or bias level from the interferogram
prior to performing the Fourier transform. Removal is not necessary
from a theoretical standpoint but is necessary from a practical, numerical
perspective to ensure that side lobes of the very large zero-frequency peak
do not interfere with the reconstructed spectrum. If the interferogram bias
level is removed prior to computing the Fourier transform, the first term in
Eq. (8.10) is zero, and the cosine in the second term can be expressed as
the sum of two exponentials, leading to the two terms
Z 1 Z 1
1 0
S( ) = E0 ( 0 ) ei2p( + ) d 0 d
4 1 0
Z Z
1 1 1 0
+ E0 ( 0 ) ei2p( )
d 0d . (8.11)
4 1 0

Integration over results in Dirac delta functions (x), such that


Z 1 Z 1
1 0 0 0 1
S( ) = E0 ( ) ( + )d + E0 ( 0 ) ( 0
) d 0.
4 0 4 0
(8.12)

This leads to the final result

1 1
S( ) = E0 ( ) + E0 ( ). (8.13)
4 4
Fourier Transform Spectrometer Design and Analysis 369

The Fourier transform of an interferogram after bias removal leads to


two terms: the irradiance spectrum and a mirror image of the irradiance
spectrum, as is depicted in Fig. 8.5 for the irradiance spectrum example.
The spectrum of the source, therefore, can be produced by capturing
an interferogram with a Michelson interferometer and computing the
Fourier transform over the spectral range of the source. Since the result
in Eq. (8.9) is real-valued due to the symmetry of the interferogram, the
same result can be achieved by using the cosine transform:
Z 1
S( ) = E( ) cos(2p )d . (8.14)
1

8.1.3 Spectral resolution


In practice, movement of the mirror is restricted to a particular physical
range, so that the OPD limits of integration in Eq. (8.7) or Eq. (8.9) are
finite. Consider on-axis radiation (✓ = 0), and let the total displacement of
the mirror from the ZPD in either direction be given by distance d. In this
case, Eq. (8.5) becomes
Z 2d
S( ) = E( ) ei2p d , (8.15)
2d

Figure 8.5
spectrum.
370 Chapter 8

based on the relationship between and d in Eq. (8.3). By recognizing


that Eq. (8.15) is the product of Eq. (8.7) and a rect function of width 2d,
the Fourier transform represented in the ideal case by Eq. (8.13) becomes
( )
1 1 sin(4p d)
S( ) = E0 ( ) + E0 ( ) ⇤ . (8.16)
4 4 4p d

That is, the ideal spectrum is convolved with the sinc function of a width
that is inversely proportional to total mirror displacement. It is common to
define spectral resolution of an FTS as the peak-to-null width of this
sinc function, such that

1
= . (8.17)
4d

Using the definition of resolving power given previously for dispersive


spectrometers,

4d
R= = = . (8.18)

Resolving power is directly proportional to the total OPD 4d from end to


end over the double-sided interferogram. The effect of mirror displacement
on spectral resolution is illustrated in Fig. 8.6, where an example spectrum
is reconstructed based on truncated interferograms with maximum OPDs
of 0.125, 0.0625, and 0.03125 cm, corresponding to decreasing spectral
resolution of 4, 8, and 16 cm 1 . In this way, it is mechanical parameter
d (as opposed to material characteristics) that completely defines spectral
resolution.

8.1.4 Spectral range


Equation (8.15) uses a continuous Fourier transform to describe the
transformation of the interferogram to the spectrum. In reality, the
interferogram is sampled, and a discrete form of Eq. (8.11) is used:

N
1 X
S( ) = E( j ) ei2p j
, (8.19)
N j=1

where N is the number of interferogram samples and j are their


corresponding OPDs. The expression in Eq. (8.19) is called the discrete
Fourier transform (DFT) and is usually processed in a computationally
efficient manner called a fast Fourier transform (FFT), where samples are
uniformly spaced and N is a power of 2 (Oppenheim and Schafer, 1975).
Fourier Transform Spectrometer Design and Analysis 371

Maximum OPD: 0.125 cm Maximum OPD: 0.0625 cm


1 1
0.9 0.9
0.8 0.8
Arbitrary Units

Arbitrary Units
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
800 1000 1200 1400 1600 1800 2000 800 1000 1200 1400 1600 1800 2000
Wavenumber (cm–1) Wavenumber (cm–1)
(a) (b)
Maximum OPD: 0.03125 cm
1
0.9
0.8
Arbitrary Units

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
800 1000 1200 1400 1600 1800 2000
Wavenumber (cm–1)
(c)
Figure 8.6 Reconstructed spectra at different spectral resolutions: (a) 4-cm 1
spectral resolution, (b) 8-cm 1 spectral resolution, and (c) 16-cm 1 spectral
resolution.

Consider a uniformly distributed set of samples with a differential mirror


displacement between samples of . To achieve Nyquist sampling of the
irradiance spectrum, the spectrum must be bandlimited to a maximum
wavenumber of

1
max = , (8.20)
2(2 )

or a minimum wavelength of

min =4 . (8.21)

This defines the spectral range of an FTS. Maximum wavelength max is


independent of interferogram sampling but is practically limited by optics
transmission and detector response.
Fourier Transform Spectrometer Design and Analysis 373

8.2 Imaging Temporal Fourier Transform Spectrometers


An FTS can be extended to accommodate imaging by incorporating an
FPA and imaging optics, as depicted in Fig. 8.8. In this case, a remote
scene is imaged by fore-optics onto an intermediate image plane, and
this intermediate image is relayed onto an FPA through a Michelson
interferometer. Ideally, the Michelson interferometer exists in collimated
space, where the wavefront associated with a specific image location
is planar. This planarity is not necessary for the spectrometer to work
but simplifies the analysis and results in optimal performance. The
spectrometer operates by capturing a set of frames of the FPA as one
mirror is precisely moved across a certain range. The trace of a single FPA
374 Chapter 8

Figure 8.8 Imaging FTS based on a Michelson interferometer.

pixel over the set of frames represents an interferogram associated with


a particular location. Combining the interferograms for all detector pixels
will produce an interferogram dataset organized as depicted in Fig. 8.9.
By Fourier transforming all of the signals in a frame-to-frame direction,
an irradiance spectrum can be formed for each FPA pixel, thus producing
a hyperspectral image.
In an imaging FTS, the frame sequence can be expressed as
Z 1
1
E(x, y, d) = E0 (x, y, ) [1 + cos{4p d/ cos ✓(x, y)}] d , (8.27)
0 2

where (x, y) represents the image-plane spatial coordinates of a particular


FPA pixel, and ✓(x, y) represents the inverse of the lens-mapping function.
For example, ✓(x, y) can be given by
0p 1
1
BBB x2 + y2 CCC
✓(x, y) = tan BB@ CCA , (8.28)
f

Figure 8.9 Frame sequence from a temporal imaging FTS.


Fourier Transform Spectrometer Design and Analysis 375

where f is the focal length of the final reimaging lens. Because of


the angular dependence of Eq. (8.27), the spatial pattern of a single
frame with a uniform input radiance over the FOV represents a bulls-
eye pattern, as depicted in Fig. 8.10, which is the resulting spatial
distribution for a spatially uniform, monochromatic source at different
mirror displacements. The specific results depicted are for a 100-mm
focal length for spectrometer lenses, 50-mm full-image field, and 10-
µm wavelength. Since detector elements integrate spatially over this
pattern, the interferogram modulation depth decreases with increasing
mirror displacement as the spatial frequency content of the spatial pattern
increases. This essentially causes an apodization effect, which degrades
the spectral resolution of the interferometer with field angle.

You might also like