RAMAN SPECTROSCOPY / Near-Infrared 105
Near-Infrared
C Ross, Brock University, St Catharines, ON, Canada
K T Carron, University of Wyoming, Laramie, WY, USA
& 2005, Elsevier Ltd. All Rights Reserved.
Introduction
When radiation passes through a transparent medi-
um, the species present scatter a fraction of the beam
in all directions. C.V. Raman discovered that incident
light is inelastically scattered from a sample and
shifted in frequency by the energy of its characteristic
molecular vibrations, a phenomenon known as Ra-
man scattering. Today, laser photons over a wide
range of frequencies from the near-ultraviolet (near-
UV) to the near-infrared (NIR) region are used in
Raman scattering studies. By choosing wavelengths
which excite appropriate electronic transitions, res-
onance Raman studies of selected components of a
sample or parts of a molecule can be performed.
Over the past few years, the range of excitation
wavelengths has been extended to the NIR region in
which background fluorescence is reduced and pho-
toinduced degradation of the sample is diminished.
The NIR region of the spectrum extends from the
upper wavelength end of the visible region at
B770–2500 nm (13 000–4000 cm
1). For many years,
NIR excitation of Raman spectra was not conside-
red viable. However, now high-intensity NIR diode
lasers are easily available, making this region attrac-
tive for compact, low cost Raman instrumentation.
Furthermore, the development of low noise, high
quantum efficiency, multichannel detectors (charge
coupled device (CCD) arrays) combined with high-
throughput single-stage spectrographs used in com-
bination with holographic laser rejection filters, have
led to high-sensitivity Raman spectrometers.
Classical, spontaneous Raman scattering is a pow-
erful analytical tool that allows for the investigation
of the qualitative and quantitative composition of
biological, pharmaceutical, and environmental sam-
ples. The following discussion of NIR-Raman spec-
troscopy will begin with a general review of Raman
spectroscopy, followed by a description of NIR-Ra-
man, with further discussion about instrumentation
and applications of the NIR-Raman technique.
General Aspects of Raman
Spectroscopy
When a beam of monochromatic light is incident on
a sample, some of the light is transmitted, some is
absorbed, and some is scattered. Most of the scat-
tered light has the same wavelength as the inci-
dent light, an effect known as Rayleigh scattering.
However, a small fraction of the scattered light
(B1 out of 108 photons) is shifted in wavelength,
resulting in inelastic scattering or Raman scattering.
In the Raman scattering, the differences in energy
between the incident and the scattered photons are
quantized, corresponding to the energy differences
within the vibrational or the rotational levels of the
molecules.
In Raman spectroscopy, the wavelength shift is
defined as the difference in wavenumber, Dn, (cm
1)
between the observed radiation and that of the
source. Two types of Raman scattering are present,
Stokes and anti-Stokes lines. Stokes lines are found at
wavenumbers that are lower than the Rayleigh peak
while anti-Stokes peaks occur at wavenumbers grea-
ter than the wavenumber of the laser excitation
source. Anti-Stokes lines are generally considerably
less intense than the corresponding Stokes lines. Flu-
orescence may seriously interfere with the observat-
ion of Stokes shifts but not with anti-Stokes, thus
with fluorescing samples, anti-Stokes signals may
sometimes be more useful despite their lower in-
tensities. With highly fluorescing samples, it is more
common to use a different excitation source, cir-
cumventing the production of fluorescence. This
strategy frequently involves moving from the use of
a visible laser source as an excitation source to one
lasing in the NIR region.
Figure 1 depicts an energy-level diagram showing
the sources of Raman and Rayleigh scattering. When
a photon having energy n0 collides with a molecule,
two different events can take place. If the scattered
photon has the same energy as the exciting photon,
the Rayleigh scattering can be observed. The middle
arrow depicts the changes that produce Rayleigh
scattering. No energy is lost in Rayleigh scattering,
thus the collisions between the photons and the mol-
ecule are termed elastic.
The energy changes that produce Stokes and anti-
Stokes emission are depicted on the left and right,
respectively. The two differ from the Rayleigh radi-
ation by frequencies corresponding to 7Dn, the
energy difference between the first vibrational level
and the ground state. In Stokes scattering, the mol-
ecule loses energy
Dn, reaching the vibrational
level n ¼ 1 and not returning to its vibrational state
n ¼ 0. In this case, the scattered photon has energy
n0
n1, where n1 is the frequency of the vibration. In
anti-Stokes scattering, the molecule originally in the
106 RAMAN SPECTROSCOPY / Near-Infrared
Optical field incident
on a molecule
∆


0(cm−1)
0− ∆

0
Stokes Rayleigh
Figure 1 Energy level schematic of Raman spectroscopy.
n ¼ 1 state, is first promoted to a virtual excited state,
then successively returns to the ground vibrational
state, n ¼ 0. The energy of the scattered photon is
n0 þ n1 and the spectral line is shifted to a lower
wavelength and a higher frequency than the incident
radiation. However, since the majority of the mole-
cules are in the ground vibrational state at room
temperature, the Stokes lines is stronger compared to
the anti-Stokes lines.
NIR-Raman Spectroscopy
The Difference in Excitation
Many books on Raman spectroscopy were written
before NIR lasers became a viable excitation source.
Traditionally, the relative magnitude of Raman spec-
troscopy is explained using the n4 factor, which in-
dicates that the Raman intensity for normal Raman
scattering varies with the fourth power of the
observed frequency, an event dependent on the ex-
citation laser frequency. However, since Raman is
too weak to be seen by the naked eye without the aid
of filters, it is difficult to find good analogies for tea-
ching. The n4 process provides the explanation for
several common phenomena, the blue sky and red
sunset. Rayleigh proved that elastic scattering of light
(Rayleigh scattering) is responsible for the blue sky.
When the sky is observed during the day, the vision is
overwhelmed by light scattered by atmospheric mol-
ecules. Since high frequencies scatter more and blue
light is the highest frequency of the visible spectrum,
the sky appears blue. At sunset just the opposite oc-
curs and all the high frequencies are scattered away
to the west. Thus, only the low frequencies of red and
Electronic excited states
Virtual excited states

=2

= 1 Vibrational levels
∆


=0

0+∆

Anti-Stokes
orange are visible. Raman may be considered in this
context as the inelastic equivalent of Rayleigh scat-
tering and follows the same rule.
This theory predicts that Raman spectroscopy
should be optimally performed in the near-UV not
the NIR region. For example, excitation using
the blue line of an Ar þ laser at 454 nm will produce
essentially nine times more intense Raman scattering
than 785 nm excitation. Yet in recent years, there has
been an overwhelming trend toward using NIR ex-
citation. This trend stems from a single interference
source and frustration in Raman spectroscopy: flu-
orescence. Fluorescence occurs when the excitation is
able to promote an electron into an excited electronic
state. When impure samples or pure samples with
low energy transitions are placed in a Raman spec-
trometer, the spectrum recorded using visible excita-
tion often shows a broad fluorescence background,
frequently devoid of any Raman features. A good
rule-of-thumb is that the number of electronic states
decreases with energy. Thus NIR excitation is often
preferred since it will produce high quality Raman
spectra devoid of any fluorescence interference unat-
tainable with a higher energy visible excitation source.
A simple property of any spectrum can provide a
more quantitative picture to the problem of fluores-
cence in Raman spectroscopy. The signal-to-noise
ratio (S/N) describes the quality of the spectrum. A
high S/N is desirable for viewing a spectrum for both
qualitative analysis and for concentration determi-
nation with quantitative analysis. Obviously, the S/N
can be increased two ways: an increase in signal or a
decrease in noise. For example, if the Raman spec-
trum of cyclohexane is taken from a sample and has
been filtered to remove any particulates, the S/N ratio

will improve linearly with increasing laser power (for
relatively low laser powers) or as the square root
of the integration time. If one takes a spectrum of
cyclohexane with a fluorescent impurity, the S/N
will not increase linearly with laser power. Longer
integration times will be limited by the dynamic
range of the detector. This originates from the de-
pendency of the noise on the signal in spectroscopy.
When visible photons are detected, they produce a
noise equivalent to the variance of the signal. If a
sample has a background fluorescence that increases
with the laser power or integration time, noise will
also increase. This leads to the fundamental question
of whether Raman spectroscopy is a true zero-base-
line technique.
The answer is that Raman spectroscopy rarely fits
that description of zero-baseline. Most samples pos-
sess a background due to either inherent or impu-
rity fluorescence. If NIR excitation circumvents or
strongly reduces the background, the reduction in
intensity due to the n4 dependence of the Raman
signal becomes irrelevant compared to the noise pro-
duced by a large fluorescence background if visible
excitation is used.
A second problem arises from the shape of the
fluorescent background. If the background were
linear, it could be easily subtracted to produce a
high-quality spectrum, albeit noisy due to the noise
produced by the high background. However, in
addition to adding noise to the spectrum, the fluore-
scence creates broad features in the background that
are often very difficult to remove. This leads to
serious difficulties in quantitative analysis because
the Raman signal above the baseline is proportional
to the concentration, not the signal above a zero-
baseline. It also can lead to small frequency shifts in
qualitative Raman spectroscopy. NIR-Raman spec-
troscopy often leads to higher quality spectra due to
the minimization of background signals from the
sample.
The Difference in Detection
The S/N is a function of the excitation in Raman
spectroscopy and as already discussed above, Raman
spectroscopy can strongly benefit from using NIR
excitation. The detection method also affects the S/N
and is still perhaps a point of controversy among
Raman spectroscopists. The controversy stems from
the quasiphoton nature of NIR light. One may prefer
to consider the light particle in nature and count
the photons or one may prefer to consider the light
as a wave. To explain this duality and controversy, a
freshman chemistry experiment can be used as an
illustration.
RAMAN SPECTROSCOPY / Near-Infrared 107
Consider the simple process of weighing coins on a
balance. Assume the weight of three coins needs to be
determined using a balance. The common laboratory
balance has a standard error of e while a very good
analytical balance has an e of 0.0001 g. Thus, if
weighing 0.00170.0001 g or even 1070.0001 g, the
error is not dependent on the mass being weighed.
The three coins individually will weigh m17
0.0001 g, m270.0001 g, and m370.0001 g. The ac-
curacy can be improved by weighing the coins in
pairs. In this case, m1 þ m2 ¼ m1270.0001 g, m1 þ
m3 ¼ m1370.0001 g, and m2 þ m3 ¼ m2370.0001 g.
The three weights of each pair can be subtracted to
find the individual weights. For example, m1 ¼
2(m12 þ m13
m23)7eT. The total error, eT, can be
1
found from the propagation errors in the three meas-
urements. It is {[12(e1 þ e2
e3)]2}1/2. If the errors are
random, the equation can be simplified to m1 ¼
[14(e21 þ e22 þ e23)]1/2 and as stated above, the balance
has an error of e regardless of the mass. This leads to
m1 ¼ [14(3e2)]1/2 or 31/2/2e or 87% of e. In the actual
measurement, the weight of m1 would be m17
0.000087. In general, the more coins that are
weighed, the better the result. If N coins were
weighed, the error would be reduced by N1/2/(N
1),
or for a 1000 coins, the error would be reduced to
3% of its original value. This may be perceived of as
the basis of Fourier Transform Raman spectroscopy
(FT-Raman spectroscopy). In FT-Raman, the Raman
scattered light can be considered as waves that are
allowed to interfere with each other to produce an
interference pattern (interferogram) that can be
mathematically converted by an inverse FT into a
spectrum. The benefit, called Fellgett’s or Multiplex
Advantage, arises from putting the whole spectrum
at once onto the detector. This is exactly analogous
to placing the many coins on the balance at once to
find the weight of an individual coin.
The controversy arises from the fact that optical
detectors are not always like an analytical balance.
As the energy of a photon decreases, the prominent
detector noise becomes the thermal excitation of
electrons in the detector. When this noise source
dominates, the detector is similar to the analytical
balance. When the energy of the photon is large, the
thermal noise becomes insignificant. This transition
occurs in the NIR region. The noise from high-energy
photons arises from their quantum nature. If
individual photons were counted for 1-s intervals,
in one interval, 100 may be counted, in the next 95,
in the next 105, and so on. These differences occur
because the photons are random events and may
happen more often during one period than the next.
Statistics dictate that the error or standard deviation
of photons counted is the square root of the number
108 RAMAN SPECTROSCOPY / Near-Infrared
of photons being counted or as stated earlier, the
signal is equal to the variance in the signal.
Using the weight of the coins, the error of the bal-
ance can be calculated. The same calculations can be
made as described above, except the error over two
weighings is 21/2e. Calculations show that for m1, the
error is 7(3/2)1/2e or 1.22e. The error has increased
instead of decreasing. What was once an advantage
for one detector is now a disadvantage for another.
The result is two camps of Raman spectroscopists.
When a sample is very fluorescent, even in the far-red,
it is best to move further into the IR and use
FT-Raman spectroscopy. If fluorescence can be elimi-
nated or reduced, conventional NIR-excited dispersive
Raman spectroscopy provides the best option.
Instrumentation
For many years, NIR excitation of Raman spectra
was not considered viable; however, things changed
with several key instrumentation developments.
CCDs now allow parallel detection of Raman spec-
tra with high quantum efficiency and low noise. Vol-
ume-holographic filters and gratings improve overall
throughput and provide elastic light rejection for
Raman spectrographs. In addition, economical, long-
life laser diodes provide sufficient power for Raman
applications and optical fiber probes allow remote
chemical analysis by Raman scattering.
Laser Source
The sources used in modern Raman spectroscopy are
almost always lasers because their high intensity is
necessary to produce Raman scattering of sufficient
intensity to be measured with reasonable signal-
to-noise ratio. Argon ion lasers operate at 488
or 514.5 nm, krypton lasers operate at 530.0 or
647.1 nm, and helium/neon (He/Ne) lasers operate at
632.8 nm. In NIR-Raman spectroscopy, the laser
sources used are the diode laser (782 or 830 nm)
and Nd/YAG (1064 nm). NIR sources have two ma-
jor advantages over shorter wavelength lasers. The
first is that they can be operated at much higher
power (typically up to 50 W) without causing photo-
decomposition of the sample. The second is that they
are not energetic enough to populate a significant
number of fluorescence producing excited electronic
energy states in most molecules. Consequently, fluo-
rescence is generally much less intense or nonexistent
with these lasers. The Nd/YAG line at 1064 nm is
particularly effective in eliminating fluorescence. The
two lines of the diode array laser at 782 and 830 nm
also markedly reduce fluorescence in most cases.
Figure 2 provides an example where the diode
laser source completely eliminates background
Iodoanisole
785nm
Intensity
633nm
600 700 800 900 1000
Wavenumbers (cm−1)
Figure 2 Effect of different laser wavelengths, 633 nm (He/Ne
laser) and 785 nm (diode laser), on the reduction of fluorescence
in the analysis of iodoanisole.
fluorescence. The sample was iodoanisole and most
of recorded signal using visible excitation arises from
the fluorescence of that compound. The lower curve
was obtained with conventional Raman equipment
using the 633-nm line from a He/Ne laser for exci-
tation. The upper curve is for the same sample re-
corded with a spectrometer equipped with a diode
laser that emitted at 785 nm. Note the absence of
fluorescence background signal.
Two basic cavity configurations are used to con-
struct and stabilize NIR-Raman laser sources. The
first design, the Littrow configuration, collimates
light from the laser diode and sends it to the grating
at a specific angle of incidence. The first-order dif-
fracted light is sent directly back to the laser to
provide optical feedback across a narrow wavelength
range. The output beam is the light that is reflected
from the grating, thus the output is at an angle that is
approximately twice the angle of incidence from the
optical axis. Because of its simple design, the exter-
nal-cavity length can be quite short, resulting in wide
spacing of the external-cavity mode. This spacing
facilitates single-frequency operation where the goal
is to have only one external-cavity mode fall within
the bandwidth of the diffraction grating. The band-
width of the grating is relatively large because the
Littrow configuration uses a single-pass geometry.
In the second design, the Littman configuration,
the collimated laser light strikes the grating near the
grazing incidence so that the diffracted order does
not return to the laser directly. Instead, the diffracted

light is reflected by a mirror, diffracted by the grating
a second time and then returned as optical feedback
to the laser diode. The advantage of the double-pass
geometry is that the grating bandwidth is less than
half of what it is in the Littrow case but the external
cavity length tends to be longer. In practice, the Litt-
man configuration is usually selected when narrower-
frequency operation is desired.
Spectrometer
A Raman spectrophotometer analyzes the radiation
scattered by molecules when they are illuminated
with monochromatic exciting radiation. Currently,
most Raman spectrometers are either FT instruments
equipped with cooled germanium transducers or
multichannel instruments based upon CCDs. These
transducers, in contrast to photomultiplier tubes, are
sensitive to radiation at 785 nm produced by diode
lasers, which provide Raman excitation of many
compounds without significant fluorescence. CCDs
are not sensitive to the 1064-nm radiation from an
Nd/YAG laser.
Figure 3 is a schematic representation of a typical
NIR-Raman dispersing spectrometer with a CCD in
the Littman configuration. The source is a diode laser
and filter system that yields radiation at 785 nm. This
beam is focused on the end of the excitation filter by
means of a lens, and the Raman emission is trans-
mitted through the collimating lens. A diffraction
grating disperses the radiation and reflects it onto a
CCD that is made up of a large number of pixels.
These pixels store the charge produced by the light
quanta. Since the spectral information seen by the
individual pixels is accumulated simultaneously, an
Grating
Lens Beamsplitter Lens
Sample
Bandpass filter
½ Waveplate
Mirror
Figure 3 Typical Raman dispersing spectrometer with a CCD in the Littman configuration. (Courtesy of DeltaNu, Laramie, WY.)
RAMAN SPECTROSCOPY / Near-Infrared 109
array with n pixels is equivalent to n spectrometers
accumulating separately or one spectrometer wor-
king n times.
Applications
NIR-Raman spectroscopy has been used for a number
of applications and is particularly useful for biologi-
cal and biomedical uses. Fluorescence has been a
limiting factor for much Raman analysis of biologi-
cal samples, particularly whole-cell or whole-tissue
samples. NIR excitation reduces interference from
fluorescence and decreases photoinduced degradation
of the sample, enabling researchers to obtain spectra
for a variety of biomaterials and living cells.
Interest in both in vivo and in vitro use of NIR-
Raman spectroscopy of tissues for diagnostics con-
tinues to grow. NIR-Raman has been evaluated as a
diagnostic tool for cervical precancers. Although
spectra could not be collected in vivo due to long
integration times, NIR-Raman was found to have
specificity that was superior to standard method-
ologies, such as coloscopy and cytology, for differ-
entiating squamous intraepithelial lesions (SIL) from
non-SIL. Several research groups have described
NIR-Raman instrumentation with CCD detection
for biological tissue analysis, both in vivo and in
vitro. In a specific study, an 810-nm excitation
wavelength was used for in vitro laboratory analysis
of aorta tissues. Subsequent work showed that flu-
orescence interference in tissue Raman spectra could
be reduced even further using 830-nm excitation
without compromising CCD sensitivity. The useful-
ness of NIR-Raman coupled with CCD detection for
Focusing lens
CCD
Mirror
Collimating lens
Diode laser @ 785 nm
110 RAMAN SPECTROSCOPY / Surface-Enhanced
studying cancerous changes in the colon, urinary
bladder, breast, and soft tissue sarcomas has also
been examined.
NIR-Raman has found applications in the phar-
maceutical industry. McCreery and colleagues re-
ported the use of the technique for identification of
pharmaceuticals inside amber vials. Even with the
signal attenuation through the glass, adequate spec-
tra were obtained for determination of vial content
with 1–60 s integration times. Using a library of
spectra, identification of the pharmaceuticals in the
vials was performed and identification was found to
display accuracy between 88% and 96%. This work
demonstrated the potential of NIR-Raman for online
process monitoring.
The environmental community has also found uses
for NIR-Raman analysis. A continuous method was
developed by Weissenbacher and colleagues to detect
trace organic pollutants using flow injection analysis
and surface enhanced Raman spectroscopy (SERS).
This method uses NIR excitation and FT-SERS de-
tection to detect parts per million of pesticides in
aqueous solutions. In this study, the authors describe
the simultaneous detection of two pesticides, carbe-
nazim and metazachlorine.
A variety of industrial processes benefit from the
use of NIR-Raman spectroscopy. In a study exa-
mining unleaded petroleum gasoline, with excitation
at 514 nm from an argon laser, no Raman features
were observed above the strong fluorescence. With
He/Ne laser excitation at 633 nm, several Raman
features appeared above the background. With exci-
tation at 785 or 852 nm, a good quality Raman
spectrum was obtained. In the textile industry, NIR-
Raman was used to examine ready-made textiles for
the discrimination of the different raw materials.
NIR-Raman has been introduced for use into the
food processing industry. In the production of oils
and fats, the determination of the amount of unsat-
uration, such as cis and trans isomers, can be
important for food processing and food labeling.
NIR-Raman was reported to measure the total
unsaturation and the cis and trans isomers online
during the production process. Fluorescence inter-
ference with visible excitation of many fats was
Surface-Enhanced
R E Littleford, D Graham, and W E Smith, University
of Strathclyde, Glasgow, UK
I Khan, Imperial College, London, UK
& 2005, Elsevier Ltd. All Rights Reserved.
drastically reduced at NIR wavelengths. In addition
to food processing, NIR-Raman has been used for
determination of food quality as the main com-
ponents of food (carbohydrates, proteins, and lipids)
all show characteristic Raman lines using NIR
detection.
The pairing of SERS with NIR detection has
proved to be a powerful DNA detection technique.
NIR–SERS was reported to provide excellent dis-
crimination against fluorescent interference and was
nonresonant with most molecules. This allowed
greater excitation intensities without photobleaching
or destruction of the analyte. In one study, Kneipp
and colleagues used this technique in rapid DNA se-
quencing to detect single-molecule DNA bases or
nucleotides. NIR–SERS provided a method for de-
tection and identification of the single DNA base,
adenine, without any additional labeling. This study
reported trace detection levels of adenine and AMP
with well-resolved spectra.
See also: Raman Spectroscopy: Instrumentation;
Surface-Enhanced.
Further Reading
Hanlon EB, Manoharan R, Koo TW, et al. (2000) Pros-
pects for in vivo Raman spectroscopy. Physical and
Medical Biology 45: R1–R59.
Isola N, Stokes DL, and Vo-Dinh T (1998) Analytical
Chemistry 70: 1352–1356.
Keller S, Lochte T, Dippel B, and Schrader B (1993) Qual-
ity control of food with near-infrared excited Raman
spectroscopy. Fresenius Journal of Analytical Chemistry
346: 863–867.
Kneipp K, Kneipp H, Itzkan I, Dasari R, and Feld M
(1999) Ultrasensitive chemical analysis by Raman spec-
troscopy. Chemical Reviews 99: 2957–2975.
McCreery RL, Horn AJ, Spencer J, and Jefferson E (1998)
Journal of Pharmaceutical Science 87: 1–8.
Mahadevan-Jansen A, Mitchell MF, Ramanujam N, et al.
(1998) Photochemistry and Photobiology 68: 123–132.
Mulvaney SP and Keating CD (2000) Raman spectroscopy.
Analytical Chemistry 72: 145R–157R.
Weissenbacher N, Lendl B, Frank J, et al. (1997) Journal of
Molecular Structure 410–411: 539–542.
Introduction
Surface enhanced Raman scattering (SERS) is a sen-
sitive spectroscopic technique for the detection and