Fourier Transforms New Analytical Approaches and FTIR Strategies

Fourier Transforms - New Analytical Approaches and FTIR
Strategies
Edited by Prof. Goran Nikolic
ISBN 978-953-307-232-6
Hard cover, 520 pages
Publisher InTech
Published online 01, April, 2011
Published in print edition April, 2011
New analytical strategies and techniques are necessary to meet requirements of modern technologies and
new materials. In this sense, this book provides a thorough review of current analytical approaches, industrial
practices, and strategies in Fourier transform application.
How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:
Gür Emre Güraksın, Uçman Ergün and Ömer Deperlioğlu (2011). The Analysis of Heart Sounds and a Pocket
Computer Application via Discrete Fourier Transform, Fourier Transforms - New Analytical Approaches and
FTIR Strategies, Prof. Goran Nikolic (Ed.), ISBN: 978-953-307-232-6, InTech, Available from:
http://www.intechopen.com/books/fourier-transforms-new-analytical-approaches-and-ftir-strategies/the-
analysis-of-heart-sounds-and-a-pocket-computer-application-via-discrete-fourier-transform
InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
www.intechopen.com
1
Fourier Transform Infrared Imaging

Spectroscopy in Biomedicine –
Important Things to Consider When
Planning a New Experiment
Rieppo J.1, Rieppo L.2, Saarakkala S.3 and Jurvelin J.S.2
1Department
of Biomedicine, University of Eastern Finland, Kuopio,
2Department
of Physics and Mathematics, University of Eastern Finland, Kuopio,
3Department of Diagnostic Radiology, University of Oulu and Oulu University Hospital,
Oulu,
Finland
1. Introduction
Fourier transform infrared imaging spectroscopy (FT-IRIS) offers unique possibilities to collect
chemical information from biological samples with high spatial resolution (generally ~10 µm)
[1]. The development of FTIR instruments has introduced this technique also for biomedicine.
FT-IRIS is a demanding technique that requires a good understanding of the measurement and
analysis principles. Application of the technique can be divided into three phases: 1) sample
preparation 2) data collection and 3) data analysis. Each of the three steps is crucial for the
outcome of the study and significant sources of error may exist in all three steps. Biologists and
medical doctors are often not aware of the technical aspects of the measurement principle or
the data analysis methods. Therefore, lack of information may impair successful application of
this technique, and the full potential of the technique is not achieved.
This book chapter is written for an entry level user without the formal training for
spectroscopy or data mining techniques. This chapter addresses important issues which
should be considered when the FT-IRIS experiments are designed and data is analysed. We
cover briefly: 1) the overall potential of FT-IRIS in qualitative and quantitative biomedical
research, 2) essential issues to be taken into account in preparation and measurement of the
biological samples and 3) different methods for analysis of acquired spectral data.
2. Planning a new experiment

FT-IRIS studies require careful planning of the study protocol beforehand. The first task is to
define the aim of the study and to evaluate whether FT-IRIS is suitable for a planned
experiment. FT-IRIS is a diverse technique that can be applied to numerous applications.
Existing literature introduces numerous studies varying from simple qualitative
applications to highly sophisticated applications of complex neural networks. FT-IR
spectroscopy has been applied e.g. to cancer diagnostics [2-7], bone diseases [8-10],
osteoarthritis [11-13], neurological diseases [14-16] and atherosclerosis [17,18]. Each and
www.intechopen.com
4 Fourier Transforms - New Analytical Approaches and FTIR Strategies
every study is different and the used study protocols can be transferred only in rare cases
directly to a new study. In general, sample preparation has the least variation, and also the
measurement conditions can be standardized up to certain point, but the data-analysis has
to by designed for each study individually.
2.1 Sample preparation

FT-IRIS studies are carried out either using thin histological sections (transmission
measurement) or by using smooth, polished surfaces (reflection measurement). Histological
tissue sections can be produced from: 1) fixated samples embedded into paraffin or other
resins or 2) application of cryosectioning. Both techniques have their own advantages and
limitations, which are discussed next.
2.1.1 Chemically fixed histological samples

Chemical fixatives have been used for decades in histology for preparing thin tissue sections
for microscopy. Typical sample preparation involves formalin fixation and decalcification
process. Samples are further dehydrated with ascending series of alcohol, and finally lipids
are removed with xylene before sample is infiltrated with liquid paraffin [19]. Paraffin
embedding is needed to create support for fragile sections in order to be able to cut 3-7 µm-
thick histological samples. Embedded samples are easier to cut and, therefore, they have
been used most often in histological studies. However, it should be realized that chemical
fixatives introduce a few problems in FT-IRIS studies. Fixatives form chemical cross-links to
protein structures preventing tissue deformation during the section processing. This is
advantageous when only the morphology of the sample is considered, as is the case in
traditional histology. However, since chemical fixatives alter the proteins and carbohydrates
of the sample, they have potentially negative effects on spectroscopic analysis. It is known
that the effect of chemical cross-linking can be seen in IR spectra [20, 21]. This clearly
indicates that the fixation has permanently changed the tissue chemical properties.
However, chemical fixation is typically done in a standardized manner within one study,
and its effects can be assumed to be similar for each sample. Therefore, chemical fixation
does not necessarily hinder the use of embedded sections.
Another possible source of error is that residual traces of paraffin are evident in paraffin-
embedded sections even after chemical removal of paraffin [22]. Instead of chemical
removal of paraffin, it is also possible to subtract paraffin after the measurements by using
paraffin spectra [5]. Fortunately, a paraffin spectrum contains only a few narrow peaks. In
order to minimize the contribution of paraffin residues to the analysis, one might also
simply exclude the paraffin peak areas from the analysis.
In addition, the fact that lipids and part of the solubilized proteins are most likely lost from
the sample due to the sample processing must be taken into account. Thus, information
arising from these compounds cannot be studied from the fixated samples. Taken together,
these limitations should be taken into account when the use of chemically fixed tissue
sections are planned in new FT-IRIS studies.
2.1.2 Cryosections
When cryosections are used in FT-IRIS measurements samples are not treated with fixatives
prior to sectioning, and it requires only minimal sample preparation. Tissue is prepared and
embedded into cryo-embedding medium. Subsequently, sample is frozen with liquid
nitrogen and sectioned with a cryotome. However, here one should note that the accuracy of
www.intechopen.com
Fourier Transform Infrared Imaging Spectroscopy in Biomedicine –
Important Things to Consider When Planning a New Experiment 5
section thickness of cryotome is not at the level of paraffin or plastic embedded samples. On
the other hand, since the method does not require chemical processing of the sample, the
sample is maintained close to its original biological form. Embedding media used for
cryosectioning often stains the samples, which causes problems for data analysis. After
cutting tissue samples with a microtome, sections have to be rinsed with water, but the lack
of chemical cross-linking makes the thin sections pliant, and consequently sections wrinkle
easily during the preparation. In general, repeated sectioning needs to be done until the
good quality section is found for the final measurements.
Advice: Chemical treatment of the sample is not needed with cryosections. Tissue is
minimally altered compared to chemical fixation. Soluble proteins and lipids can also be
studied when cryosections are used.
Conclusions: Chemical fixatives alter the collected IR data. Different fixatives cause different
alterations and therefore sample processing must be standardized between the specimens.
Fixative-related alterations are not a problem with simple univariate analysis techniques, but
they may cause potential artefacts when sophisticated multivariate techniques are used.
Chemical fixation and embedding is also poorly suited for research problems when
solubilized proteins or lipids are main interests of the study. Furthermore, a significant
section thickness variation caused by a microtome itself exists both with embedded sections
and cryosections. This variation has to be kept in mind in quantitative analysis and
compensated, if necessary, with repeated measurements or with the reference sample
technique [23]. Preparation of the cryosections is fast but the overall time consumption is
considerable since the quality of cryosection is inferior to embedded sections.
2.1.3 The effect of the variable section thickness

Section thickness varies significantly between histological sections due to the various
uncertainties (e.g. temperature changes during cutting, inadequate embedding media
support, microtome-related inaccuracy and sample-related variation). In general, the
variation is smaller with embedded samples. Section thickness variation affects directly the
measurement results, since according to Beer-Lambert law, absorption is directly related to
section thickness. Normalization (e.g. vector normalization) of the spectral data is usually
conducted before qualitative multivariate analysis. Therefore, section thickness variation is
usually not a problem with qualitative analyses. However, quantitative analysis requires a
strict control of the section thickness since the thickness variation is one of the main sources
of errors in the FT-IRIS experiments. Variation is particularly harmful in sample-to-sample
correlation analysis where FT-IRIS measurements are correlated with the reference
technique. The nature of section thickness error is random, and therefore its distribution can
be assumed as Gaussian. Consequently, the section thickness error is not as significant when
group means of different sample groups are compared. However, it is essential to keep in
mind that the error caused by the section thickness variation can be greater than the
biological variation itself.
Advice: Evaluate whether the type of experiment you are conducting can be affected by the
variable section thickness. If quantitative analysis from FT-IRIS spectra is carried out and
compared with reference methods (correlation analysis), a good control of section thickness
is essential. On the other hand, qualitative (multivariate) analysis is least affected by the
section thickness variation. As a rule of thumb, section thickness artefact can be reduced by
averaging multiple measurements from single specimen or by using homogenous reference
material cut along with biological tissue, as described by Rieppo et al. (2004) [23].
www.intechopen.com
2.2 Data collection

Type of the experiment sets the requirements for data collection. Acquisition of spectral data
has to be optimized for signal quality and the measurement time. Measurement of a single
specimen can be very time consuming when high quality data is measured and large areas are
covered. A simple univariate data analysis can be carried out with low quality data, but when
sophisticated multivariate analysis is carried out, the signal-to-noise ratio has to be adequate.
The length of the data collection is determined by four factors: 1) measured spatial area, 2)
spectral resolution, 3) spatial resolution and 4) number of averaged scans. Spectral resolution,
spatial area and spatial resolution are directly proportional to the data collection time.
2.2.1 Spatial resolution

Modern imaging instruments allow the use of different spatial resolutions. Data collection
time increases significantly when spatial resolution is increased due to the decreased signal-
to-noise ratio and the increased amount of the measured pixels. Better spatial resolution
requires more repeated scans to obtain equal signal-to-noise ratio. It is important to notice
that spatial resolution is always limited by the diffraction due to the long wavelength used
in mid-IR measurements. Spatial resolution can be approximated to be one-half of the used
wavelength. Thus, best spatial resolution is achieved with the shortest wavelength of the
measured spectrum (approx. 2.5 µm with 4000 cm-1), and the resolution decreases gradually
as the wavelength increases. However, optimal spatial resolution requires synchrotron
operated devices due to the poor S/N-levels at diffraction limited resolutions.
Consequently, in practice with standard thermal globar IR light sources the resolution is
limited to ~7-15 μm depending on the used wavelength (1700-700 cm-1). A true spatial
resolution of the FT-IRIS device is defined as the minimum distance where two separate
features can be fully separated from each other. Many FT-IRIS devices offer a better pixel
resolution by magnification but that is only nominal resolution of the device (limited by
sample stage movement or magnification), as actual signal, limited by the physical
diffraction, arises from larger area. When planning FT-IRIS investigations one should
carefully think whether there is a need for the smallest pixel sizes. For example, increase
from 6.25 μm to 25 μm increases the number of measured pixels to 16-fold. Furthermore, if
an equal signal-to-noise ratio is wanted, the measurement time is increased even more.
Advice: High spatial resolution tremendously increases the measurement time. A very
critical evaluation has to be carried out whether the diffraction limited resolution is needed
for carrying the experiment. High quality spectral data is preferential to high spatial
resolution in most cases. Better spatial resolution increases the measurement time mainly
because the number of repeated scans has to be increased to simultaneously keep signal-to-
noise-ratio at an adequate level.
2.2.2 Spectral resolution

Spectral resolution is the accuracy (measured as wavenumbers) at which the spectral data is
acquired. Depending on the type of experiment, spectral resolution is typically 4-16 cm-1.
Sharp spectral features are only seen with a high spectral resolution. Therefore, it is
advantageous to increase the spectral resolution up to a certain point. On the other hand,
required measurement time is directly proportional to the spectral resolution. It is important
to remember that as the resolution increases, so does the noise. Thus, number of repeated
scans has to be increased if the S/N-ratio is kept unchanged, and therefore duration of the
measurement is considerably longer. It has been demonstrated that more spectral features
www.intechopen.com
can be seen with 4 cm-1 resolution compared to 8 or 16 cm-1. The effect is particularly evident
when 2nd derivative spectra are used in the analysis (Figure 1). Most of the reported FT-IRIS
studies have been conducted using either 4 or 8 cm-1 resolution. Application of the 2 cm-1
resolution is not typically used for two main reasons: 1) protein features generally lack very
sharp peaks, i.e., additional information is limited, and 2) measurement time becomes
impractical due to poor signal-to-noise-ratio.
2nd derivative absorbance

Absorbance
A B
2000 1800 1600 1400 1200 1000 2000 1800 1600 1400 1200 1000
Wavenumber (cm-1) Wavenumber (cm-1)
Fig. 1. An IR absorbance spectrum of articular cartilage measured with 4 cm-1 spectral

resolution (A) and second derivative spectra with 4 cm-1 (black) and 16 cm-1 (red) spectral
resolutions (B)
Advice: Simple univariate analysis can be carried out using spectral resolution of 8-16 cm-1.
If second derivative spectra are used for analysis, spectral resolution should be increased to
4-8 cm-1 to gain any advantage. From the theoretical point of view, a better spectral
resolution reveals more information from the sample, but in practice measurement time is
often a limiting factor. Signal-to-noise-ratio gets progressively worse as the spectral
resolution increases. Spectral resolution can be considered as the most important parameter
when a new FT-IRIS study is planned since it directly affects the subsequent possibilities for
data analysis. Thus, data analysis methods (univariate vs. multivariate analysis) required
need to be also planned beforehand. If numerous samples are measured, a careful balancing
between the optimal spectral resolution, signal-to-noise ratio and measurement time is
needed. Pilot studies are often essential to evaluate the spectral resolution needs against the
time consumption.
2.2.3 Standardized measurement conditions

IR measurements are significantly affected by the carbon dioxide and water vapour of the
atmosphere [24]. Humidity is substantially changed between winter and summer time
causing a significant variance to the atmospheric conditions. Altered measurement
conditions hinder the data quality when multivariate analyses are carried out. Standardized
measurement conditions are, thus, essential for FT-IRIS measurements. Equipment is
typically purged with N2-gas or dried air. For example, water vapour free air can be guided
into spectrometer, microscope and sample compartment. Furthermore, thin histological
www.intechopen.com
sections absorb moisture from the environment. Therefore, it is beneficial to store the
samples in a desiccator where steady environment can be maintained.
Advice: Each measurement system and sample type has to be separately tested for the
measurement stability. Temperature and humidity levels of the laboratory should be kept as
constant as possible. Separate purge system can be used to obtain low humidity and carbon
dioxide levels in the measurement chamber. It is good to keep in mind that the
measurement environment is altered each time the specimen is changed. It is a good idea to
measure the duration when steady measurement conditions are reached after the sample
compartment has been opened. A regular quality control also ensures that the
measurements are consistent with each other regardless of the measurement time.
3. Data analysis – a crucial step to get the answers for your research problem
Data-analysis is the most difficult part of the FT-IRIS experiments. A good quality data set is
a prerequisite for successful spectral analysis. Raw spectral data gives only a little
information without proper knowledge of spectral pre-processing and analysis.
Chemometric methods vary and offer different approaches to gather specific information
from the measured data. Analysis has to be designed to match the desired research
problems.
3.1 Spectral pre-processing

Spectral pre-processing is an essential procedure prior to the actual spectral analysis. Pre-
processing is particularly needed when data from several studies are combined. Data has to
be evaluated and filtered so that the non-biological variation between different
measurements is minimized. Pre-processing routines include data quality analysis (signal-
to-noise-ratio criteria, water vapour limitations, removal of corrupted data), measurement
area criteria (definition of the region of interest, removal of unwanted pixels), selection of
spectral region (data truncation, masked areas) and baseline corrections. In general, spectral
pre-processing should be conducted in a way that do not alter the actual biological
information but reduces the variance caused by the measurement itself or by the
measurement conditions.
There exists different kind of baseline correction methods. Offset correction is simply
conducted by setting, e.g., the minimum value of the spectrum (or alternatively a
wavenumber that should not have any absorbance) to zero level. Linear baseline correction
is done by fitting a line through two zero-absorbance points. The line is subsequently
subtracted from the spectrum. Also polynomial-based fitting is used, but too heavy
correction might alter not only the baseline but also the actual spectroscopic data. A more
realistic model-based approach has also been used for data pre-processing. In Extended
Multiplicative Signal Correction (EMSC), offset error, linear error and second order
polynomial error are fitted simultaneously. The model uses a good quality reference
spectrum when estimating the baseline errors, which makes EMSC a reliable method for
baseline correction [25-28]. Application of second derivative spectra for the analysis
eliminates the need for baseline correction as the differentiation removes most significant
offset and linear baseline errors [27]. Proper pre-processing is essential, as baseline
variations might hide the real biological variation if the baseline errors are not removed.
Advice: Pay attention for the spectral pre-processing before actual data-analysis. Pre-
processing steps and demands depend on the data analysis method.
www.intechopen.com
3.2 Univariate methods

Univariate methods are the simplest and most used data analysis techniques in IR
spectroscopy. Univariate analyses are usually carried out by using peak height, integrated
peak area or peak ratio for quantitative spectroscopic measurements [29]. These methods are
hindered by the fact that biological tissues are essentially composed of the same building
components regardless of the tissue type. Therefore, univariate methods are not suitable for
solving complex research problems. Biological tissues usually lack the distinctive spectral
features, which hinders the possibilities of simple univariate analysis. It seems impossible to
achieve specificity and sensitivity level of biochemical reference techniques with univariate
based parameters in most biological research problems [30]. Univariate analysis is, however,
useful if the tissue has major tissue constituents that can be isolated with a single parameter.
For example, bone and partially also cartilage, are good examples of tissues where separate
components can be measured with reasonable accuracy also with univariate methods [8, 9,
31, 32]. Application of second derivative spectroscopy offers enhanced spectral features that
potentially increase the parameter specificity. Second derivative spectroscopy increases
significantly the spectral features, but the noise related to measurements is also significantly
amplified. This sets higher demands for data quality.
3.3 Multivariate methods

Multivariate methods use more than one variable at a time. Multivariate methods offer
means to increase parameter specificity since a larger part of the spectral information is
exploited than in univariate analysis. Multivariate methods are suitable both for qualitative
and quantitative analysis. The main difference between the univariate and multivariate
methods is that the multivariate approach does not require accurate isolation of the
differentiating spectral feature before the analysis is conducted. Univariate analysis is
dependent on a priori knowledge of the spectral features and therefore the technique has
limited efficiency. Multivariate techniques can handle the complete spectral data set with
statistical means, and therefore the technique is more potential than user-limited univariate
methods. Multivariate approach is often needed when specificity of the IR parameters has to
be increased. This is commonly done by multivariate calibration, where a multivariate
model is calibrated against some actual reference information. The model can be then used
to obtain the same information from the spectroscopic data. Multivariate methods can also
be used to separate two or more sample groups by spectral means.
4. How to select appropriate analysis method for different research

problems?
A key element for successful use of the FT-IRIS is to be able to define the research question
before actual experiments are started. Potential and the limitations of the FT-IRIS technique
have to be kept in mind when study protocol is designed. Type of the research question
determines the demands for sample harvesting and for type of analysis methods best suited
for the particular research problem.
4.1 Quantitative measurement of tissue constituents

Quantitative measurement of different tissue constituents requires a specific parameter for a
given compound. This is not a trivial demand to meet and often univariate parameters are
not feasible. Multivariate models have a better chance to work since they do not require
www.intechopen.com
fully separated, specific spectral features evident for researcher to be noticed. Univariate
parameters are best suited in special situations where spectral differences between tissue
components are evident, e.g., bone mineral and matrix content [8-10]. Major tissue
constituents can probably be measured using univariate methods with reasonable
specificity. However, if the studied compound is present only in small quantities, then only
multivariate methods should be considered. Multivariate methods, such as principal
component regression (PCR) and partial least squares (PLS) regression, offer more efficient
means for quantitative analysis, since they can handle also overlapping spectral data. On the
other hand, multivariate analysis needs good understanding of the background of the
methods in use. Furthermore, in order to routinely use multivariate techniques to
quantificate the composition of different tissue constituents, a comprehensive reference data
set needs to be collected. This illustrates the complexity of FT-IRIS spectral analysis.
Example: Univariate vs multivariate analysis of proteoglycan distribution of articular
cartilage
-4
x 10
6
Absorbance spectrum analysis
r = 0.605 r = 0.701
Second derivative analysis
15 5
4
10
3
2
5
1
0
A 0
B
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
Reference information Reference information
3
Predicted content of PLSR model
r = 0.967
2.5
1.5
0.5
-0.5
C
0 0.5 1 1.5 2 2.5
Reference information
Fig. 2. FT-IRIS analysis of proteglycan content of articular cartilage. Univariate analysis from the
absorbance spectrum (A) produces a little worse result than the second derivative analysis (B).
However, a multivariate PLS regression model is clearly the most efficient analysis method
As a practical example, let us now consider univariate vs multivariate analysis techniques
for determination of proteoglycan distribution in articular cartilage tissue. Univariate based
solutions for FT-IRIS analyses of spatial proteglycan content in articular cartilage have been
www.intechopen.com
used in several studies [30-33]. We have demonstrated that specificity for cartilage
proteoglycans is significantly increased by taking advantage of the increased spectral
separation of the second derivative spectroscopy. Furthermore, we have compared
univariate results with the results of the PLS regression model. The PLS regression results
were calculated using the whole collected spectral region. The results demonstrated that the
PLS regression model is more consistent with the reference technique as compared to
univariate methods [34] (Figure 2).
4.2 Qualitative visualization of tissue morphology

Tissue morphology is traditionally investigated through light microscopy of stained thin
tissue sections. Morphological features are seen with specific staining patterns of different
tissue types. FT-IRIS can produce similar information, but it does not require any staining.
Image contrast is created with IR absorption of the tissue. Different tissue types absorb IR
energy differently, i.e., their absorption spectra are different. However, it is often difficult to
create contrast between different tissue types with univariate techniques. Therefore, if
morphological features are one of the main interests of the study, multivariate techniques
have to be considered. Application of the neural networks probably gives the most accurate
results, but building such a model requires a large data pool and is a very time consuming
procedure [3]. A large reference spectra data is collected and the model is trained to
recognise spectral features of different tissue types. Nevertheless, neural networks are
accurate and fast way to analyze specimens once the neural network is established. Simple
multivariate techniques can be used also within a single specimen by using cluster analysis.
The aim of cluster analysis methods is to minimize spectral differences within clusters while
maximizing differences between clusters. Therefore, tissue types can be classified into their
own groups by cluster analysis. Cluster analysis methods, such as K-means clustering or
fuzzy c-means clustering, arrange data into desired number of user-determined clusters
according to the spectral features [4]. Hierarchical cluster analysis allows unsupervised
clustering without pre-determined number of clusters [4].
Clustering methods are often used for isolating the regions of interest from the sample (e.g.
cancer area from surrounding tissue). Clustering methods can be used with a limited
number of samples but the calculations become very time consuming with large number of
spectra or with large number of different samples.
Advice: Morphological information can be gathered with multivariate clustering methods.
Data clustering can be done with various multivariate techniques depending on the research
application. Clustering methods become essential when large tissue sections are used and
only a part of the data is interesting. Proper spectral pre-processing is essential prior to
clustering. Any impurities or foreign material, e.g., such as embedding material, are likely to
produce a new cluster, which is particularly harmful when fixed number of cluster is used.
4.3 Classification studies

Multivariate techniques are particularly useful when research problem can be simplified
into few classes (disease vs. non-disease, tissue types 1,2,3,4 etc). Univariate based
parameters have only limited capability for clustering purposes. Multivariate clustering
methods are especially useful when different classes are sought (Figure 3). Spectral data can
be assigned into subsets according to its spectral features either in unsupervised or
supervised manner. In unsupervised techniques, only unlabeled data is used as an input.
Clustering is done blind without knowing any additional information of the studied sample.
www.intechopen.com
Data is re-arranged only on the basis of their spectral properties. In supervised clustering,
known class information is also included as an input, e.g. disease vs. non-disease. Known
information is used to seek spectral features that can be linked to the known information.
Once the model is built and verified, the data analysis can be done with great speed. This
type of application can be used if the specific feature is looked from the samples.
Advice: Multivariate methods are powerful to find even the smallest chemical differences
from the samples. However, analysis methods cannot automatically distinguish artefacts
from true differences. Therefore, maintaining stabile measurement conditions and a proper
spectral pre-processing is essential for application of multivariate techniques.
Fig. 3. Pictures of the FT-IRIS k-means clustering map (A) and a histological cartilage section
stained with type II collagen antibody (B) that shows a repaired cartilage defect and
surrounding intact cartilage. Clustering of the FT-IRIS map is done by mathematical
calculation without any user intervention or a priori information of the tissue properties. Grey
colour represents repair tissue, pink shows most likely tissue originating from the periosteal
flap used in repair surgery and green colour indicates the surrounding intact cartilage
5. Conclusion
Application of FT-IRIS offers new potential for biomedicine. Biologist and medical doctors
require training for data mining techniques since the methodology is still relatively new in
biomedicine. Multivariate methodology has increasingly been used in spectroscopy, and the
research questions are becoming more demanding all the time. Recent progress in
biospectroscopy has shown to be fast. From the theoretical point of view, FT-IRIS can be
used for the research problems that cannot be answered with traditional imaging
techniques. FT-IRIS links the tissue molecular information together with histological
imaging. FT-IRIS instruments have rapidly developed and become available for numerous
research groups worldwide. Utilization of the modern data mining techniques is likely to
further increase the development of the biospectroscopy.
6. References
[1] Lasch P, Naumann D. Spatial resolution in infrared microspectroscopic imaging of
tissues. Biochim Biophys Acta 2006; 1758: 814-829.
www.intechopen.com
[2] Bhargava R. Towards a practical Fourier transform infrared chemical imaging protocol
for cancer histopathology. Anal Bioanal Chem 2007; 389: 1155-1169.
[3] Lasch P, Diem M, Hansch W, Naumann D. Artificial neural networks as supervised
techniques for FT-IR microspectroscopic imaging. J Chemom 2007; 20: 209-220.
[4] Lasch P, Haensch W, Naumann D, Diem M. Imaging of colorectal adenocarcinoma using FT-
IR microspectroscopy and cluster analysis. Biochim Biophys Acta 2004; 1688: 176-186.
[5] Ly E, Piot O, Wolthuis R, Durlach A, Bernard P, Manfait M. Combination of FTIR
spectral imaging and chemometrics for tumour detection from paraffin-embedded
biopsies. Analyst 2008; 133: 197-205.
[6] Sahu RK, Mordechai S. Fourier transform infrared spectroscopy in cancer detection.
Future Oncol 2005; 1: 635-647.
[7] Zwielly A, Mordechai S, Sinielnikov I, Salman A, Bogomolny E, Argov S. Advanced
statistical techniques applied to comprehensive FTIR spectra on human colonic
tissues. Med Phys; 37: 1047-1055.
[8] Boskey A, Mendelsohn R. Infrared analysis of bone in health and disease. J Biomed Opt
2005; 10: 031102.
[9] Boskey A, Pleshko Camacho N. FT-IR imaging of native and tissue-engineered bone and
cartilage. Biomaterials 2007; 28: 2465-2478.
[10] Boskey AL, DiCarlo E, Paschalis E, West P, Mendelsohn R. Comparison of mineral quality
and quantity in iliac crest biopsies from high- and low-turnover osteoporosis: an FT-
IR microspectroscopic investigation. Osteoporos Int 2005; 16: 2031-2038.
[11] Bi X, Yang X, Bostrom MP, Camacho NP. Fourier transform infrared imaging
spectroscopy investigations in the pathogenesis and repair of cartilage. Biochim
Biophys Acta 2006; 1758: 934-941.
[12] Camacho N, West P, Yang X, Lin J, Bostrom M. An infrared fiber optic probe for
detection of degenerative cartilage. Trans Orthop Res Soc 2002; 27: 225.
[13] West PA, Bostrom MP, Torzilli PA, Camacho NP. Fourier transform infrared spectral
analysis of degenerative cartilage: an infrared fiber optic probe and imaging study.
Appl Spectrosc 2004; 58: 376-381.
[14] Lasch P, Beekes M, Schmitt J, Naumann D. Detection of preclinical scrapie from serum by
infrared spectroscopy and chemometrics. Anal Bioanal Chem 2007; 387: 1791-1800.
[15] Lasch P, Schmitt J, Beekes M, Udelhoven T, Eiden M, Fabian H, et al. Antemortem
identification of bovine spongiform encephalopathy from serum using infrared
spectroscopy. Anal Chem 2003; 75: 6673-6678.
[16] Peuchant E, Richard-Harston S, Bourdel-Marchasson I, Dartigues JF, Letenneur L,
Barberger-Gateau P, et al. Infrared spectroscopy: a reagent-free method to
distinguish Alzheimer's disease patients from normal-aging subjects. Transl Res
2008; 152: 103-112.
[17] Colley CS, Kazarian SG, Weinberg PD, Lever MJ. Spectroscopic imaging of arteries and
atherosclerotic plaques. Biopolymers 2004; 74: 328-335.
[18] Manoharan R, Baraga JJ, Rava RP, Dasari RR, Fitzmaurice M, Feld MS. Biochemical
analysis and mapping of atherosclerotic human artery using FT-IR
microspectroscopy. Atherosclerosis 1993; 103: 181-193.
[19] Király K, Hyttinen MM, Lapvetelainen T, Elo M, Kiviranta I, Dobai J, et al. Specimen
preparation and quantification of collagen birefringence in unstained sections of
articular cartilage using image analysis and polarizing light microscopy. Histochem
J 1997; 29: 317-327.
www.intechopen.com
[20] Aparicio S, Doty SB, Camacho NP, Paschalis EP, Spevak L, Mendelsohn R, et al.
Optimal methods for processing mineralized tissues for Fourier transform infrared
microspectroscopy. Calcif Tissue Int 2002; 70: 422-429.
[21] Pleshko NL, Boskey AL, Mendelsohn R. An FT-IR microscopic investigation of the
effects of tissue preservation on bone. Calcif Tissue Int 1992; 51: 72-77.
[22] Faolain EO, Hunter MB, Byrne JM, Kelehan P, Lambkin HA, Byrne HJ, et al. Raman
spectroscopic evaluation of efficacy of current paraffin wax section dewaxing
agents. J Histochem Cytochem 2005; 53: 121-129.
[23] Rieppo J, Hyttinen MM, Jurvelin JS, Helminen HJ. Reference sample method reduces
the error caused by variable cryosection thickness in Fourier transform infrared
imaging. Appl Spectrosc 2004; 58: 137-140.
[24] Martens H, Bruun SW, Adt I, Sockalingum GD, Kohler A. Pre-processing in
biochemometrics: correction for path-length and temperature effects of water in
FTIR bio-spectroscopy by EMSC. Journal of Chemometrics 2006; 20: 402-417.
[25] Bassan P, Byrne HJ, Bonnier F, Lee J, Dumas P, Gardner P. Resonant Mie scattering in
infrared spectroscopy of biological materials--understanding the 'dispersion
artefact'. Analyst 2009; 134: 1586-1593.
[26] Bassan P, Kohler A, Martens H, Lee J, Byrne HJ, Dumas P, et al. Resonant Mie scattering
(RMieS) correction of infrared spectra from highly scattering biological samples.
Analyst; 135: 268-277.
[27] Kohler A, Bertrand D, Martens H, Hannesson K, Kirschner C, Ofstad R. Multivariate image
analysis of a set of FTIR microspectroscopy images of aged bovine muscle tissue
combining image and design information. Anal Bioanal Chem 2007; 389: 1143-1153.
[28] Kohler A, Kirschner C, Oust A, Martens H. Extended multiplicative signal correction as
a tool for separation and characterization of physical and chemical information in
Fourier transform infrared microscopy images of cryo-sections of beef loin. Appl
Spectrosc 2005; 59: 707-716.
[29] Stuart B. Spectral analysis. In: Infrared Spectroscopy: Fundamentals And Applications,
Struart B Ed. Kent: John Wiley & Sons 2004.
[30] Rieppo L, Saarakkala S, Narhi T, Holopainen J, Lammi M, Helminen HJ, et al.
Quantitative analysis of spatial proteoglycan content in articular cartilage with
Fourier transform infrared imaging spectroscopy: Critical evaluation of analysis
methods and specificity of the parameters. Microsc Res Tech 2009.
[31] Camacho NP, West P, Torzilli PA, Mendelsohn R. FTIR microscopic imaging of collagen
and proteoglycan in bovine cartilage. Biopolymers 2001; 62: 1-8.
[32] Kim M, Bi X, Horton WE, Spencer RG, Camacho NP. Fourier transform infrared
imaging spectroscopic analysis of tissue engineered cartilage: histologic and
biochemical correlations. J Biomed Opt 2005; 10: 031105.
[33] Bi X, Yang X, Bostrom MP, Bartusik D, Ramaswamy S, Fishbein KW, et al. Fourier
transform infrared imaging and MR microscopy studies detect compositional and
structural changes in cartilage in a rabbit model of osteoarthritis. Anal Bioanal Chem
2007; 387: 1601-1612.
[34] Rieppo L, Jurvelin JS, Rieppo J, Saarakkala S. Partial Least Squares Modeling of FTIR
Microspectroscopy for Predicting Proteoglycan Content of Articular Cartilage:
Preliminary Results. In: SPEC: Shedding Light on Disease: Optical Diagnostics for the
New Millennium. Manchester 2010.
www.intechopen.com
2
Analysis of Bioactive Olygosaccharide-Metal

Complexes by Modern FTIR Spectroscopy:
Copper Complexes
Goran S. Nikolić and Milorad D. Cakić
University of Niš, Department of Organic Chemical Technology, Faculty of Technology,
Bulevar oslobodjenja 124, RS-16000 Leskovac
Serbia
1. Introduction
Spectroscopy of biomolecules has been founded at the end of last century by the researchers
working in the field of optical spectroscopy applied to biosystems. Since this time, the
interest of the activity has considerably grown up. Researchers have typically used
traditional spectroscopic techniques, such as Raman scattering, IR absorption, UV/Vis
absorption, circular dichroism, fluorescence, magnetic resonance, X-rays and neutron
scattering. Recently, particular attention has been devoted to the applications of
biomolecular spectroscopy in the fields of biomedical imaging, drug characterization for
pharmaceutical applications, drug delivery and nanobiotechnology.
Investigations of the bioactive metal complexes are very interesting in medicine and
pharmaceutical industry, with the aspects on therapy of different states of anemia or
metabolism disorder. On the other hand, polysaccharides and their derivatives, as the most
abundant class of biomolecules, are known to have a large variety of biological functions.
Through the interaction between these polyfunctional molecules and metal ions in living
organisms, the modification of the biological function of both counterparts may be expected.
The polysaccharide type compounds as ligands have received considerable interest. Simple
sugars and their derivatives with reduced and oxidized groups form metal ion complexes of
various composition and stability. One of the known roles of the oligo- or polysaccharide
complexes is the transport of metal ions through cell membranes. For example, the
commercial copper preparations based on polysaccharide dextran and its derivatives are
used for such purpose in both human and veterinary medicine.
In the field of biocoordination chemistry a lot of investigations are based on the synthesis
and characterizations of different metal complexes of ligands they present in biological
systems, or synthetic ligands, which will serve like the model-molecules for complex
biomolecular structures. Bio- or synthetic ligands are mainly natural chemical compounds of
macromolecular type. In this group of products of the special importance are chemical
compounds of olygosaccharide pullulan, dextran and inulin with cations of the different
biometals (Cu, Fe, Co and Zn). It is well known that raw microbiological exopolysaccharides
dextran and pullulan, are glucose polymers with the large molar mass from a few millions
g/mol, with own toxic and antigen characteristics so that they are not of pharmaceutical
www.intechopen.com
importance. For commercial reasons raw polysaccharides were depolymerized to the

products with adequate molar masses, with the aim of getting fractions with narrow molar
mass distribution. Synthesis procedures for the complex formation of biometals with poly-
or oligosaccharides are described in scientific and patent literature. However, the structure
of the bioactive metal complexes with oligosaccharides has not been explained in details yet,
despite a number of studies. The work represents further development in research of
complex structure and pharmacobiological activity of the complexes. Some new results that
are directly related to medical practice and structural physicochemistry of biomolecules
based on the oligosaccharide-metal complex are presented in this publication.
Different biometals (Cu, Fe, Co, Zn) complexes with inulin, pullulan and dextran
oligosaccharides, as well as reduced or oxidised derivatives, have been analyzed by IR
spectroscopy. Spectra-structure correlations of the complexes have been performed by using
modern spectroscopic techniques: FTIR microspectroscopy, ATR-IR, LNT-IR and D2O-FTIR.
The techniques are applying in the structure analysis of polysaccharide complexes, as well
as for the confirmation of suggested types of complex structure and for the testing of
homogeneities of samples. Results of IR microspectroscopic investigation shows that
structural form of complexes and metal content considerably depends of constitution and
ligands conformation, degree of crystallity, polymerization, polydispersity, and linearity of
macromolecules. Also, stability of the synthesized complexes, as well as their
pharmacological effect, depends of these parameters. FTIR investigation of the complexes by
D2O isotopic exchange proved to be a very sensitive method for determining OH group
coordination and is related to the hydrogen bond strength. Results of our investigations
points to the complexes are crystal hydrate molecules. Correlation of physicochemical and
spectroscopic investigations of these complexes, and structure of exopolysaccharide chain,
are suggesting different model structures of the synthesized complexes.
FTIR spectra and microscopy images were obtained by using an FTIR microspectroscopy
system, ATR–FTIR spectrometer Bruker Tensor-27 in conjunction with a FTIR Bruker
Hyperion-1000/2000 microscopy attachment equipped with the 4× viewing objective
(objective magnification 4×, visible magnification 57×) and 15×IR Schwarzschild objective
(objective magnification 15×, visible magnification 215×). The standard detector, a 250 μm
liquid nitrogen cooled, mid-band mercury–cadmium–telluride (MCT) detector (ATR
objective GMBH, Germany) with preamplifier, with the range of the IR spectrum from 7000
to 400 cm–1 was used. The spectra were measured with 2 cm–1 resolution and 200 scans co-
addition. The spectrometer was linked to a PC equipped with Bruker OPUS software to
allow the automated collection of IR spectra. The measurements were conducted in the
reflection mode. In the region from 4000–400 cm–1 all spectra were Interactive polynomials
baseline corrected and area normalized. A Kubelka–Munk arithmetic method was applied
to enhance the resolution in this spectral region. Deconvoluted spectra were smoothed by
the 40 point Fourier filter method. The IR spectra were imported to GRAMS/AI 7 (Thermo
Galactic, USA) for peak area integration.
Thus, various tests can be performed by the Bruker Hyperion microscope, such as
transmission, reflection, polarized, and ATR–IR measurements, the linear scan and mapping
techniques in terms of software, and optic video technology for true video analysis. In
addition, spatial resolution IR spectra and functional group imaging can also be acquired
and analyzed. For measuring IR spectra by FTIR microscopy accurately, several primary
parameters in the operation need to be selected and set first, which include aperture sizes,
number of scans, resolution, velocity of motional mirror, and sampling background.
www.intechopen.com
Analysis of Bioactive Olygosaccharide-Metal Complexes by Modern FTIR Spectroscopy:
Copper Complexes 17
2. Character and spectroscopy of bioinorganic compound

Metal ions in biological systems are divided into two classes (Nakamoto, 2009). The ions of
first class (K, Na, Mg and Ca) are important in maintaining the structure of proteins by
neutralizing negative charges of peptide chains and in controlling the function of cell
membranes that selectively pass certain molecules. In the second class, ionic forms of Fe, Co,
Cu, Zn, Mn, Mo, and so on exist in small to trace quantities, and are often incorporated into
proteins (metalloproteins). The latter class is divided into two categories: (A type) transport
and storage proteins and (B type) enzymes. Type A includes oxygen transport proteins such
as hemoglobin (Fe), myoglobin (Fe), hemerythrin (Fe), and hemocyanin (Cu), electron
transfer proteins such as cytochromes (Fe), iron–sulfur proteins (Fe), blue-copper proteins
(Cu), and metal storage proteins such as ferritin (Fe) and ceruloplasmin (Cu). Type B
includes hydrolases such as carboxypeptidase (Zn) and aminopeptidase (Zn, Mg),
oxidoreductases such as oxidase (Fe, Cu, Mo) and nitrogenase (Mo, Fe), and isomerases such
as vitamin B12 coenzyme (Co).
During the last decades laboratory (with animals) and clinical researches have shown that
many pathologic states of a body are accompanied by statistically reliable disturbances in
the metabolism of metals at the molecular and body levels. Any chronic disease, the cause of
which has not yet been established, can be due to abnormalities in metal metabolism. The
determination of the amount of biometals in the body is suggested as the earliest diagnostic
test of diseases (Grigorieva et al., 1983).
The metals participating in metabolism can be divided into the following groups: a) inherent
in a living body and involved in the sphere of essential biofunctions (Cu, Fe, Zn, Mn, Mo,
Co, Mg, Ca, K, Na); b) introduced, often toxic, whose physiological role has not been fully
elucidated and their presence in the body tissue and liquids is due to their abundance in
nature and wide application by people (Al, Cr, Cd, Ni, Pb, etc.). For the first group of metals
both positive and negative balances were detected in different pathologies, and for the
second group, as a rule, only the positive balance was observed. One of the reasons for the
abnormal accumulation and removal of metals from a human body may be the wide
application of drugs in clinics and which, by their chemical nature, are good ligand–
complexing agents (up to 80% of all used drugs). Using non-steroidal antiinflammatory
compounds (HL) and a copper-containing blood enzyme, ceruloplasmin (CuCPL), the
ligands (drugs) were shown to take away competitively the metals from metal-containing
and metal-activating enzymes: CuCPL + HL = CuL + CPL. Such an interaction results in a
“discomfort” of an enzyme system in the body which is indicative of a side effect of drugs,
i.e. complexing agents. For some diseases the shifts in metal metabolism are specific:
rheumatoid arthritis (−) Fe, Zn; (+) Cu, Al, Mn, Mo, Cr; atherosclerosis (−) Cr, Mn, Zn;
cancerogenesis (−) Cu, Fe, Mg; (+) Zn, Mn; diabetis (−) Cu, Mn, Cr; (+) Zn; etc. The
correction in the concentration of these metals results in a therapeutic effect. The complex
compounds of biometals with different types of drugs are the most promising tool for
introducing the required metal into the body. It has been established that the application of
antiinflammatory agents as complexes with some biometals decreases their toxicity and
increases and prolongs their therapeutic effect (chemico-therapeutic synergism);
antiulcerogenic, cytotoxic and other helpful properties, unusual to non-complexed agents,
appear.
To understand the roles of these metal ions in biological systems, it is first necessary to
know the coordination chemistry (structure and bonding) of metal ions in their active sites.
www.intechopen.com
Such information is difficult to obtain since these active sites are buried in a large and
complex protein backbone. Although X-ray crystallography would be ideal for this purpose,
its application is hampered by the difficulties in growing single crystals of large protein
molecules and in analyzing diffraction data with high resolution. As will be discussed later,
these difficulties have been overcome in some cases, and knowledge of precise geometries
has made great contribution to our understandings of their biological functions in terms of
molecular structure. In other cases where X-ray structural information is not available or
definitive, a variety of physicochemical techniques have been employed to gain structural
and bonding information about the metal and its environment. These include electronic,
infrared, resonance Raman, ESR, NMR, ORD, CD, Moossbauer spectroscopy, EXAFS, and
electrochemical, thermodynamic, and kinetic measurements.
Infrared spectroscopy has been used extensively for the study of bioinorganic compound. In
some cases, however, the vibrations of interest may not be enhanced with sufficient
intensity. Then, one must resort to IR spectroscopy, which exhibits all vibrations allowed by
IR selection rules. It should be noted, however, that IR measurements in aqueous media are
generally limited to the regions where water does not absorb strongly. Furthermore, it is
often necessary to use difference techniques to cancel out interfering bands due to the
solvent and some solute bands. In the following, we will review typical results to
demonstrate the utility of vibrational spectroscopy in deducing structural and bonding
information about large and complex bioinorganic molecules. Marked progress has been
made in chemistry of the bioinorganic complexes where the active site is modeled by
relatively simple coordination compounds. Thus, we compare vibrational spectra of
biological molecules and their model systems whenever appropriate or necessary. Since
biospectroscopy is one of the most exciting areas of modern research, the volume of
literature on biological compounds is increasing explosively. It is clearly not possible to
cover all important topics in a limited space. Several excellent monographs (Parker, 1983;
Nakamoto & Czernuszewicz, 1993) and review articles cited in each section should be
consulted for further information.
Infrared spectroscopy has been used particularly for the study of polysaccharide complexes
with metal ions, especially the active sites of the ions in the complexes. FTIR spectroscopy
opens up new possibilities for the fine structural analysis of polysaccharides and its
derivatives, the establishment of the type of bonding between the elementary links and their
rotational isomerism. Weak intermolecular interactions have a significant influence on the
specifically valuable properties of biological molecules and polymer compounds. We had to
restrict ourselves to a few examples of wide potentialities of the method of FTIR
spectroscopy in investigating the relationships between the structure and the properties of
extracellular polysaccharides and its complexes with different metal ions.
3. Copper(II) ion and its significance

Copper(II) ion is a biologically active, essential ion, creating ability and positive redox
potential allow participation in biological transport reactions. Cu(II) complexes possess a
wide range of biological activity and are among the most potent antiviral, antitumor and
antiinflammatory agents (Vosburg & Cooper, 1941). On the other hand, condensed triazoles
exhibit a range of pharamacological activities such as mitotic (Jackson & Polaya,1951),
hypotensive (Walker et al., 1951), CNS stimulant (Lepetil, 1975), antiflammatory
(Hardtmann & Kathawala, 1977) and analgesic activites (Kathawala, 1974; Clark et al., 1997).
www.intechopen.com
Copper Complexes 19
Copper(II) ion is an essential component of several enzymes such as ceruloplasmin,

cytochrome C oxidase, lysil oxidase, superoxid dismutase and tyrosinase that are required
to maintain the host homeostasis (Platonova et al., 2004; Gaiduk et al., 2009). At the same
time, copper ions can be involved in the reactions producing active radicals, which affect the
structure of all types of biomolecules. Therefore, it is not suprising that cells lack free copper
ions, while their safe transfer is realized by a special system, metabolic copper system, some
genes of which have been recently cloned.
Copper has an important role in the metabolism and transition of iron in the body.
Microcytic hypochromic anemia is one of the outcomes of copper deficiency. There is a great
number of hypocupremical drugs used commercially nowadays. As active substance, these
drugs contain CuSO4 (Expert group, 2002). In the literature is known that metal complex
with polysaccharides and their derivatives are of growing importance in medicine and
pharmacy. New blood substitutes with hemostimulating and antianemic function, which are
complexes of dextran and pullulan with Cu(II) ion differ from the existing analogues in
good bio- and hemocompatibility and more pronounced and prolonged action (Klimovich et
al., 1998; Gapanovich et al., 1998). These complexes are very stable during prolonged storage
and are not toxic. Copper(II) ion is used in the treatment of microcytic hypochromic anemia.
It is apsorbed from the lower part gastrointestinal tract. This active pharmaceutical
compound has a repetitive dose schedule (0.6-2 mg daily).
4. Bioactive copper-pullulan complex

Pullulan is a linear exopolysaccharide of α-D-glucopyranose that is often described as a α-
(1→6) linked polymer of maltotriose subunits. This unique linkage pattern gives pullulan
with distinctive physical properties. A number of potential applications have been reported
for this biopolymer as a result of its good film-forming properties; pullulan can form thin
films which are transparent, oil resistant and impermeable to oxygen. Pullulan may be used
as a coating and packaging material, as a sizing agent for paper, as a starch replacer in low-
calorie food formulations, in cosmetic emulsions, and in other industrial and medicinal
applications (Deshpande et al., 1992). Pullulan is derivatized easily to control its solubility or
provide reactive groups. Consequently, pullulan and its derivatives have numerous
potential food, pharmaceutical, and industrial applications.
and reported that α-D-glucopyranose is the major product of acid hydrolysis (Bernier, 1958).
Bernier isolated water-soluble polysaccharides from the cultures of Aureobasidium pullulans
polymer is a α-glucan in which α-(1→4) linkages predominate (Bender et al., 1959).

Based on the positive optical rotation and IR spectrum of pullulan was concluded that the
pullulan is essentially a linear glucan containing α-(1→4) and α-(1→6) linkages in a ratio of
Subsequent studies using IR, periodate oxidation, and methylation analysis established that
2:1 (Sowa et al., 1963). Partial acid hydrolysates of pullulan include isomaltose, maltose,
panose, and isopanose (Leathers, 2003). The discovery of the enzyme pullulanase provided a
specifically hydrolyzes the α-(1→6) linkages of pullulan and converts the polymer almost
critical tool for the analysis of the structure of pullulan (Wallenfels et al., 1961). Pullulanase
frequently described as a polymer of α-(1→6) linked maltotriose subunits (Fig. 1).

quantitatively to maltotriose (Wallenfels et al., 1965). Based on this result, pullulan is
However, pullulan can also be viewed as a polymer of panose or isopanose subunits, which
may reflect the biosynthetic origins of the molecule more accurately. Indeed, a number of
enzymes that produce panose or isopanose from pullulan have been described since. Catley
www.intechopen.com
(Catley et al., 1970) established that pullulan contains maltotetraose subunits (Fig. 2) in
addition to the predominant maltotriose subunits. The frequency of maltotetraose subunits
appears to vary on a strain-specific basis, from about 1% to 7% of total residues (Catley et
al., 1986). The evidence suggests that maltotetraose subunits are distributed randomly
maltotetraose residues are substrates for many α-amylases, and it has been proposed that
throughout the molecule (Carolan et al., 1983). Unlike the maltotriose subunits in pullulan,
hydrolysis of pullulan at these sites accounts for the decrease in molecular weight
commonly observed in late cultures.
Fig. 1. Molecular structure of a representative portion of pullulan, illustrating the primary

structure of repeating linkages: (a) 2D model, (b) 3D model stick and ball
Fig. 2. Molecular structure of the secondary (minor) repeating structure of pullulan,

occurring in about 1–7% of total linkage subunits: (a) 2D model, (b) 3D model stick and ball
www.intechopen.com
Copper Complexes 21
Many types of carbohydrate derivatives (reduced or oxidized) have been synthesized for
biomedical applications. In addition, polysaccharides such as chitin (Tanodekaew et al.,
2004), chitosan (Pan et al., 2003; Yin et al., 2003), heparin (Ishihara et al., 2003; Kweon et al.,
2003), alginate (Leonard et al., 2003; Perets et al., 2003), inulin (Nikolic & Cakic, 2007),
dextran (London, 2004; Serizawa et al., 2003; Lawrence et al., 1997) and pullulan (Ilic et al.,
2002; Kim et al., 2003; Nikolic at al., 2002) have been derivatized for biomedical applications.
Pullulan is a polysaccharide that has been used in a drug delivery because of its solubility
and biocompatibility. In addition, although the polysaccharides have many ionic groups,
both anionic and cationic, pullulan is nonionic (Shingel, 2004).
Reduced low-molar pullulan (RLMP), was chosen as a new material for complexing, and the
subsequent interactions with Cu(II) ions were investigated. The complexing process begins
in a weak alkaline solution (pH > 7), and involves OH groups in C(2) and C(3) or C(6)
pullulan monomer units (α-D-glucopyranose). Complexes of Cu(II) ion with reduced low-
molar pullulan were synthesized in the water solutions, at the boiling temperature and at
different pH values, ranging from 7.5–12. Cu(II) complexes were prepared from sodium
salts, and investigated in the solid state. Fourier transform infrared spectroscopic data of
synthesized complexes are rare in literature. FTIR spectroscopic characterization is now
widely used to study the composition of the complex carbohydrate systems, the molecular
interactions, a molecular orientation and conformational transitions of polysaccharides
(Zhbankov, 1972; Panov et al., 1976; Panov & Zhbankov, 1988; Shingel, 2002; Zhbankov et al.
1997). The major goal of this section is to use different FTIR spectroscopic techniques (FTIR,
LNT-FTIR, ATR-FTIR, and FTIR microspectroscopy) as the main tools to verify the
conformation and the structure of this type of ligand around the Cu(II) ions.
Experimental.
Pullulan of average molar mass 2 x 105 g mol-1 and reduced low-molar pullulan of average
molar mass 6000 g mol-1 was obtained from PCI ‘‘Zdravlje Actavis Co.” (Leskovac, Serbia).
CuCl2 x 2H2O was purchased from Merck (Darmstadt, Germany). Cu(II) complex synthesis
with RLMP have been described in detail by Nikolic (Nikolic et al., 2008). For FTIR sample
preparation the KBr pastille method was used. Fine pulverized, water-free samples (1 mg)
were mixed with potassium bromide (150 mg, Merck) stored at 80 0C for 6 h, and then
pressed at 200 MPa to obtain a transparent pellet. The reference measurement was
performed with pure KBr. The dryness of the pastille was controlled by the band at ca. 1640
cm-1, which is associated with the deformation vibrations of the O-H bond from water
molecules (Nikolic et al., 1996; Bellamy, 1954).
The FTIR spectra as an average of 40 scans were recorded at room (298 K) and liquid-
nitrogen (77 K) temperature on a BOMEM MB-100 FTIR spectrometer (Hartmann & Braun,
Canada) equipped with a standard DTGS/KBr detector in the range of 4000–400 cm-1 with a
resolution of 2 cm-1 by the Win–Bomem Easy software. The spectrometer was purged with
dry N2. A Specac P/N 21525 variable-temperature cell was used for the LNT measurements.
In the region all spectra were baseline-corrected and area-normalized. A Fourier self-
deconvolution based on the Griffiths/Pariente method was applied to enhance the
resolution in a spectral region of 4000–400 cm-1. A gamma factor of 12 corresponding to a
peak width of 24 cm-1 was used. Deconvoluted spectra were smoothed by the 30-point
Savitzky–Golay filter method.
In addition, FTIR microspectroscopy system, ATR-FTIR spectrometer Bruker Tensor-27 in
conjunction with a FTIR Bruker Hyperion-1000/2000 microscopy attachment equipped with
www.intechopen.com
a 15x objective and a 250 μm liquid-nitrogen cooled, narrow-band mercury–cadmium–

telluride (MCT) detector (ATR objective GMBH, Germany) with the range of the IR
spectrum from 4000 to 400 cm-1 was used in this analysis. The spectra were measured with 4
cm-1 resolution and 320 scans co-addition. The measurements were conducted in the
reflection mode. In the region from 4000–400 cm-1 all spectra were Interactive polynomials
baseline-corrected and area-normalized. A Kubelka/Munk arithmetic method was applied
to enhance the resolution in this spectral region. Deconvoluted spectra were smoothed by
the 40-point Fourier filter method.
Results and discussion.
following characteristic bands: ν(O-H) 3400 cm-1, ν(C-H) 2930 cm-1, δ(HOH) 1640 cm-1, δ(C-
The FTIR spectra of the RLMP and the synthesized Cu(II) complexes (Fig. 3) contain
H) 1450 and 1345 cm-1, δ(O-H) 1420 cm-1, a complex band ν(C-O) and ν(C-C) 1200–1000 cm-
1, γ(C-H) 1000–700 cm-1. Between FTIR spectrum of RLMP and FTIR spectra of the
synthesized complex on the different pH there is a clear difference in the area of vibrations
of all types of OH groups and molecules H2O (Fig. 3). That is, in the spectrum RLMP has
found the wide intensive band on the approx. 3400 cm-1 which is the result of valent
the 1640 cm-1 is the result δ(HOH) (Nikolic et al., 1996; Bellamy, 1954; Nikolic et al., 2007;
vibrations OH groups and valent vibration of H2O constitutional molecules. The band on
Nikolic et al., 2008).

The appearance of the spectrum in this region is different, as expected. In the spectrum of
the complex which is synthesized on the pH 7.5 (Fig. 3) the centroid of this band is shifted,
and decreased temperatures provoke a clear separation of two bands the frequency which
is 3378 cm-1 and 3246 cm-1 (data from LNT-FTIR). By the complex which is synthesized on
the pH 8 the frequencies of these bands have been 3453 cm-1 and 3333 cm-1, 3458 cm-1 and
at the decreasing temperatures so those, according to this criterion, need attribute ν(OH)
3348 cm-1 on the pH 10 and 3389 cm-1 and 3355 cm-1 on the pH 12. These bands are sensitive
vibrations. These changes in ν(OH) region are results of complexing i.e. the deprotonation
of the RLMP ligand OH group, most likely of different surroundings in the first
coordination sphere of the Cu(II) ion. Exactly let us say we know that by the complexing
Cu(II) ion with dextran (Dex) in the dependence on the pH form different types of the
complex (pH 8: Cu(II)(Dex)2(H2O)2, pH 10: Cu(II)(Dex)2(H2O)(OH), pH 12:
Cu(II)(Dex)2(OH)2) and the spectral picture in this region is very similar (Nikolic et al.,
2008). If in the case of the complex with RLMP would form similar complexes, bands in this
region would need the attribute valent vibrations of the OH group and coordinate
molecules H2O by the complex which is synthesized on the pH 7.5 with regard to the OH
ligand group and the OH group in the first coordination sphere of the Cu(II) ion by the
complex on the pH > 10.
from two previous quoted bands would originate from ν(HOH), whose correct position
In the spectrum presented in Fig. 4a, of the complex which synthesized on the pH 7.5, one
which in the spectrum of the complex was synthesized on the pH 12. In the area of δ(HOH)
could not be established and this is probably a low-frequent band 3246 cm-1 the absence of
spectrum of the complex which synthesized on the pH 7.5 in the area of δ(HOH) vibrations
vibrations unlike RLMP where spectrum has only one band on the 1645 cm-1, in the
have two bands (1657 and 1642 cm-1) which points to two different types of H2O molecules
(Fig. 4b). The higher frequency band 1657 cm-1 with the increasing pH diminishes the
intensity. In other words the complex on the pH 10 and pH 12 is absent (Fig. 4b).
www.intechopen.com
Copper Complexes 23
Fig. 3. FTIR spectra of RLMP, Mw = 6000 g mol-1 (1) and Cu(II) complexes with RLMP
synthesized at boiling point and pH 7.5 (2), 8.0 (3), 10.0 (4) and 12.0 (5)
www.intechopen.com
Fig. 4. Stretching (a), bending in plane (b) and bending in plane OH plane (c) region from
LNT-FTIR spectra of RLMP (1) and synthesized Cu(II)–RLMP complexes at boiling point
and pH 8.0 (2), 10.0 (3), 12.0 (4)
synthesized on the pH 10 and pH 12 intensity bands from ν(HOH) and δ(HOH) diminishes
The spectral picture favors the suggested structure in Fig. 5 (type I). The complex which is
with increasing pH. Expect the appearance band from δ(OH) from the first coordination
sphere of Cu(II) ion whose intensity with increasing of pH growth; this band in the
fact is that in the area of δ(HOH) by the complex which is synthesized on the pH 10 and pH
spectrum of the complex which synthesized on the pH 7.5 is absent. In addition to this, the
δ(OH) on the 1384 cm-1 (data from LNT-FTIR) (Fig. 4c).

12 only one band exists, which points to one type of H2O molecule (like by RLMP) and band
Spectroscopic FTIR study in a particular region of O-H (3400 and 1420 cm-1) and C-H (2900,
1460, and 1350 cm-1) vibrations indicates different binding between the central metal ion and
ligand, depending on pH and metal contents. Water protons take part in the formation of
relatively weak hydrogen bonds (Nikolic et al., 1996; Bellamy, 1954; Nikolic et al., 2007;
Nikolic et al., 2008). In the 1200–1000 cm-1 region, the spectra of RLMP and the complex
comprise a number of highly fused bands. The enhancement of the resolution by using a
Fourier self-deconvolution allows bands to be more accurately detected. The main bands
found in the deconvoluted spectra of RLMP and the complex at ca. 1154, 1108 1079, 1042,
and 1019 cm-1 are due to coupled valent vibrations of the C-O and C-C bonds and
www.intechopen.com
Copper Complexes 25
deformational vibrations of the CCH, COH, and HCO bonds. The band at about 1150 cm-1
has been assigned to valent vibrations of the C-O-C bond and glycosidic bridge. The broad
peak at 1108 cm-1 should most likely be ascribed to the vibration of the C-O bond at the C(4)
position of the glucopyranose units (Kacurakova et al., 1996). Complex vibrations involving
the stretching of the C(6)-O(6) bond with participation of the deformational vibrations of the
C(4)-C(5) bond result in the appearance of a band at 1079 cm-1 (Sivchik et al., 1979;
Nikonenko et al., 2005; Zhbankov et al. 2005; aZhbankov et al., 2003; bZhbankov et al., 2003;
Zhbankov et al., 2000). In the spectra of Cu(II) complex with RLMP band at 1079 cm-1 is less
pronounced than in the spectra of RLMP. In the case of pullulan complexes, part of C(6)
atoms participate in the formation of the C(6)-O-Cu(II) linkages; as a result, the band
intensity at 1079 cm-1 for the Cu(II) complex with RLMP is reduced more than in the case of
RLMP. The band at 1079 cm-1 in the FTIR spectra of RLMP is attributed to the antisymmetric
stretching vibration of C(6)-O-C(1) glycosidic bridge. These findings suggest that the 1079
cm-1 band for the Cu(II) complexes with RLMP can be considered as a characteristic for the
type of interunit links and for the Ligand–Metal [C(6)-O-Cu(II)] linkage (Shingel, 2002;
Zhbankov et al., 2000; aMitic et al., 2008; bMitic et al., 2008).
Cu(II)–RLMP linkages. As a result, the band frequency at 3404 cm-1 for ν(O-H) vibrations in
In the case of the Cu(II)–RLMP complexes O-H groups participate in the formation of the
RLMP is reduced to approximately 3340 cm-1 (Fig. 3) in Cu(II)–RLMP. These findings also
suggest that the band can be considered as characteristic for the type of Metal–Ligand links.
The band at about 1042 and 1019 cm-1 found for polysaccharide in the spectra of RLMP and
the complex were shown to relate to the crystalline and amorphous phases, respectively
(Smits et al., 1998). The changes in intensity of these bands are strongly associated with the
alterations in the macromolecular order. These bands in the spectra of RLMP and the
complex can be responsible for more and less ordered structures, respectively. The major
attention was focused on the bands in the 1160–1010 cm-1 region because the absorbance
pattern due to ring vibrations in this spectral range is known to be individual for each
carbohydrate structure.
Moreover, we attempted to obtain the information about the conformations of these
macromolecules in a solvent exhibiting a different influence on the system of intra- and
intermolecular interactions. Special interest in the IR range for structural investigation is
from 1000 to 700 cm-1. In the spectra of RLMP and the complexes, bands of negligible
intensity are found in the region (950, 916, 860, 760 cm-1). According to the normal
coordinate treatment on the RLMP model, these bands are interpreted as due to mixed CCH
deformation vibrations coupled with CCO, OCO, and COC bending (Buslov et al., 1998;
Zhbankov et al., 1997; Zhbankov, 1992; Zhbankov and Avsenev, 1984; Kiselev et al., 1977).
Both the number and frequencies of the bands in the IR range depend on the conformation
of the D-glucopyranose units. It is well known that the glucopyranose units exist in six
Komar et al., 1968). The similarities of the γ(C-H) range indicate that there is no difference in
different typical conformations (1C, C1, 1B, B1, 3B, and B3) (Panov & Zhbankov, 1976;
the conformation of the glucopyranose unit in the RLMP and complex molecules, and they
probably exhibit C1 chair conformation (916 and 850 cm-1).
It appears that the intensity of the 996 cm-1 band in the pullulan spectra may indicate the
extent of the interchain association. The band at 950 cm-1 belongs to the structure-sensitive
angles. The band at 935 cm-1 was recently used to discover the co-existence of α-(1→6) and
region, and together with the band at 935 cm-1, characterizes the type of interunit bonds and
www.intechopen.com
α-(1→4) glycosidic linkages in the pullulan structure (Shingel, 2002; Zhbankov et al., 2000).
A decrease of the band at 950 cm-1 indirectly confirms an occurrence of the conformational
the glycosidic bond. For pullulans the band at 900 cm-1 described α-(1→6) linkages. α-(1→4)
transitions in polysaccharide systems owing to rotational isomerism of pyranose rings about
linkages were observed at 925 cm-1 (Shingel, 2002). Ring deformations and scaffold
vibrations were observed at 710, 660, 600, 570, and 525 cm-1. In the experiment on the
influence of the medium pH on a binding Cu(II) ion with different polysaccharides (Mitic et
al., 2007; Norkus et al., 2002; Norkus et al., 2004), there is a possibility of gradual
complexing, where their reforming starts at pH 8. Degradation of the Cu(II)–RLMP complex
begins at pH values higher than 12. The Cu(II) ions form three different types of complexes
with the deprotonated monomeric RLMP unit. Different structural models of the Cu(II)–
RLMP complexes of tetragonal distorted Oh coordination in the function of pH synthesis pH
7–8 (type I), pH 8–10 (type II), and pH 10–12 (type III) are given in Fig. 5.
Fig. 5. Structure model of Cu(II)–RLMP complexes, with six O-donor atoms in tetragonal
distorted Oh environment of Cu(II) ions, with participation of: (a) C-2 and C-3 ligand OH
groups, (b) C-2 and C-6 ligand OH groups, (c) C-3 and C-6 ligand OH groups
www.intechopen.com
Copper Complexes 27
The reactivity of the RLMP depends primarily on the reactivity of the secondary,
equatorially oriented hydroxyl groups (OH-2, OH-3, OH-4 and OH-6). The contents of the
primary O-H groups in RLMP are slightly increased at lower Mw (about 2%). The reactivity
of the polysaccharide C-atoms was determined by 13C NMR spectroscopy for pullulan it
was C(6)>C(3)>C(2)> C(4) (Mahner et al., 2001). Carbohydrates without anchoring donor
groups form a very weak complex with Cu(II) ion in an aqueous solution. The availability of
more than one anchoring group can, however, prevent the coordination of the alcoholic O-H
groups fulfilling the coordination sphere of the metal ion. The metal interaction with the set
of the non-deprotonated OAH groups increases the complex stability. Mainly, the
complexes were shown to form, in different protonation states, the deprotonation processes
starting from pH 7 (Norkus et al., 2002). After deprotonation of one or more alcoholic O-H
groups, the Cu(II) ion complexes having anionic character are usually very stable. In some
cases, the formation of various amounts of alkoxo or hydroxo bridged dimeric (or
oligomeric) species can be detected. The disugars bound the metal ions less efficiently than
the monomeric units, while the trimetric ligands can probably simultaneously use terminal
subunits to coordinate the metal ion. In aqueous solution, the RLMP complexes are formed
by the displacement of the H2O molecules from the first coordination sphere of Cu(II) ion by
the alcoholic OAH groups. In general, it seems to be true that at least three O-H groups in a
favorable steric arrangement are required for the complex formation (Gyurcsik & Nagy,
2000). The general leading rule is that the sugars in pyranose form (in RLMP C1-chair
conformation) contain an equatorial–equatorial–equatorial (eq–eq–eq) sequence of three
adjacent hydroxyl groups. The possible coordination sites are depicted on the model given
in Fig. 5.
The characterization of metal ion coordination equilibrium of polyalcohols and other sugar-
type ligands, containing alcoholic and aldehyde (or ketone) oxygen donor atoms, is difficult
due to the low stability of the complexes in neutral or acidic aqueous solutions (Gyurcsik &
Nagy, 2000; Nagy et al., 2003). The low electron densities on these donor oxygens cause the
situation, in spite of their relatively large number in one ligand molecule, that they do not
readily substitute the water molecules bonded in the first coordination sphere of the metal
ions. With increasing pH, however, the hydrolysis of some metal ions prevents the
coordination of the organic ligands. Thus, complex formation can only be expected in
strongly alkaline solutions after deprotonation of alcoholic hydroxyl groups. The fact that in
solutions of carbohydrates the species are in anomeric and conformational equilibrium and
the isomers interact in different ways with metal ions makes the studies even more
complicated. Any shift in the above equilibrium due to the metal ion coordination, thereby
resulting in the changes in the fraction of the isomers having suitable positioned sequences
of alcoholic hydroxyl groups in the total concentration of the ligand, will also influence the
complex stability.
The methods, such as FTIR, NMR, ESR, X-ray and UV–Vis made it possible to assign the
binding hydroxyl or other groups and also to characterize the metal ion coordination of
carbohydrates monitoring the ligand conformation or/and configuration changes forced by
the complexation processes. FTIR spectra of the Cu(II)–RLMP complexes were recorded on
room (RT) and on low nitrogen temperatures (LNT) in order to detect bands which are
sensitive to the reduction temperatures respective bands which have originated from
vibrations of all types OH groups and H2O molecules. In Fig. 6 RT-IR and LNT-IR spectra of
the complex synthesized on the pH 7.5 have been presented in the comparison with RLMP.
www.intechopen.com
FTIR spectra show the correlation between O-H stretch frequency and the hydrogen bond
strength (3400–3200 cm-1) predict a red shift of the bonded O-H stretching band on cooling
(from 3350 to 3246 cm-1) (Fig. 6). It is expected that the non-interacting O-H group (at 3378
cm-1) is much less sensitive to cooling and consequently will show smaller red shifts. The
red shift of the band is an indication of the involvement of the appropriate O-H proton in a
weak hydrogen bond.
Fig. 6. Stretching and bending region from RT (a) and LNT-FTIR (b and c) spectra of Cu(II)–
RLMP complex (type I, pH 7–8) (a – 298 K, b – 173 K, c – 77 K)
3246 cm-1) are found in the region of ν(O-H) vibrations (Fig. 6). In the low-frequency region
In the LNT-FTIR spectra of the complex was synthesized on pH 7.5, two bands (3378 and
bands in the γ(OH) bending region from librations of coordinated water molecules on
on LNT-FTIR spectra were presented in Fig. 6, sensitive on the reduction of temperature
frequencies 847 cm-1 and 756 cm-1, show blue shift on cooling. The librations of the O-H
group in this region of the complex synthesized on pH 12 were much less sensitive to
cooling. The observation allowed one to suggest that the most probable water molecules are
coordinated around Cu(II) in the complex type I (Fig. 5). The number and shape of these
bands implies that in complexes type III there is the displacement of H2O molecules by the
O-H groups in the first coordination sphere of the Cu(II) ion. The LNT-FTIR results confirm
the structural models of complexes presented in Fig. 5. The results obtained from the
structural studies of the investigated complexes were based on other spectroscopic
techniques (Nikolic et al., 2005; Mitic et al., 2004; Nikolic et al., 2004; Nikolic et al., 2006;
Bartkowiak et al., 1998; Cakic et al., 2008).
region of ν(O-H) vibrations (3400 cm-1), δ(C-H) vibrations (1500–1300 cm-1) and ν(C-O)
The changes in number, frequencies, intensity, and width of the FTIR bands in the particular
vibrations (1200–1000 cm-1) (Fig. 3) were related to changes in the conformation and short-
range interactions of the RLMP. Very important changes can be observed in the range of
1500–1300 cm-1 by detailed empirical analysis. Otherwise, the FTIR range is specific of
bending vibrations of CH2-OH groups (Fig. 7). Namely, the exchange position and intensity
of complex bands can be registered in this range, where C-H and O-H bending vibrations
from the CH2-OH groups take part. The change of intensity on some bands was registered
www.intechopen.com
Copper Complexes 29
only in synthesized Cu(II)–RLMP complexes. An approximate effect exists in the stretching
about 1460 cm-1 and 1370 cm-1 from δ(C-H) vibrations and the band at about 1420 cm-1 from
of the FTIR range of C-H vibrations (3000–2800 cm-1, Fig. 3). The appearance of bands at
δ(O-H) vibrations are characteristic for one of more possible positions of the CH2-OH group,
rotating around the C(5)-C(6) bond of the glucopyranose unit. The change of the angle
decreases the intensity of the appropriate IR bands (ν(C-H) and δ(C-H) vibration). The
between the methylene CH2-OH group and the polysaccharide chain axes, consequently
Cu(II) ions in solution have a possible influence on the rotation of CH2-OH groups in the
complexes (Cakic et al., 2004; Mitic et al., 2007; Nikolic et al., 2006).
Fig. 7. Region of δ(CH) and δ(OH) CH2OH grop vibrations in FTIR spectra of: (a) RLMP;
(b) and Cu(II)–RLMP complexes
We also applied FTIR spectroscopy to determine spectral manifestation of the changes in the
complex structure caused by recrystallization from D2O and, thereby, complete the
structural investigation of this modified polysaccharide. The FTIR spectra of the Cu(II)
complexes with RLMP and recrystallized analogs from D2O were analyzed in order to find
the specific spectral peculiarities that allow one to obtain the information about the structure
and the conformation of these macromolecules in solvents that exhibit different influences
on the system of intra- and intermolecular interactions. No effect of the conformation
change was observed for the recrystallized Cu(II)–RLMP complex, especially in the range of
1000–700 cm-1. When the a-D-glucopyranose units with C1 chair conformations are present,
the FTIR spectra exhibit one band in the region between 925–885 cm-1 and another one
around 860–820 cm-1, which are assigned to mixed CCH deformation vibrations (Shingel,
2002; Nikolic et al., 2008). The results allowed one to suggest a predominant crystalline form
of the recrystallized Cu(II)–RLMP complexes.
Recently, FTIR spectroscopy was coupled with a microscope and a computer system,
capable of microanalysis of minute samples by using a dedicated MCT detector. The
resultant FTIR vibrational microspectroscopy can provide molecular information of samples
with a high spatial resolution at microscopic level. Samples with microscopic size can be
nondestructively analyzed by both vibrational microspectroscopies, particularly in the
application of biomedical sciences (Kacurakova et al., 2001; Lin et al., 2007; Chiu et al., 2004;
Nikolic et al., 2008). Thus, the use of vibrational microspectroscopy has extensively become
a great potential over other spectroscopic techniques for noninvasive investigation of
chemical components of ultrastructural samples (carbohydrates, lipids, proteins,
nucleotides) (Mousia et al., 2001; Yu et al., 2005).
www.intechopen.com
More recently, FTIR and/or Raman microspectroscopic imaging systems have also been
developed for applying to biosciences (Gierlinger & Schwanninger, 2007; Chenery &
Bowring, 2003). ATR-FTIR spectra may be simultaneously collected at a time in a stepwise
manner from different areas of a sample. The absorbance ATR-FTIR spectra of Cu(II)–RLMP
complex which was synthesized at pH 7.5 are shown in Fig. 8. The absorbance of a band
corresponding to a specific chemical component may be plotted as a map. ATR-FTIR spectra
were presented in Fig. 8(A–C) from different areas of Cu(II)–RLMP complex and show high
homogeneity of the sample.
Fig. 8. ATR–FTIR spectra of Cu(II)–RLMP complex synthesized at pH 7.5 (type I) from

different areas (A, B and C) of a sample
A new imaging capability has been established not only to image heterogeneous regions of
the samples and simultaneously provide spectroscopic and spatial information, but also to
show visually the concentrations of components and to highlight their effect from the three
dimensional plot. The application of microscopic FTIR imaging system to the ligand RLMP
and Cu(II)–RLMP complexes, were synthesized at pH 7.5–12, is shown in Fig. 9. FTIR
microscopy images of ligand RLMP, as well as images of the synthesized Cu(II)–RLMP
complexes differ which also indicates that the complexation process and the creation of
coordination compounds took place. FTIR microscopy images confirmed that the changes in
the intensity of the analyzed bands are strongly associated with the alterations in the
macromolecular order. These bands in the spectra of the complexes can be responsible for
more and less ordered structures, respectively (Fig. 9). The changes in color contour may
show the content and distribution of copper, and polysaccharides in Cu(II)–RLMP samples.
Conclusions.
The complexing process begins in a weak alkali solution (pH > 7.5), and involves OH
groups in C(2) and C(3) or C(6) pullulan monomer unit (α-D-glucopyranose). A part of FTIR
spectra, in the range on 1000–700 cm-1 of Cu(II) ion complexes with RLMP, indicates no
influence of complexing process on the conformation change of C1 glucopyranose units. The
IR band d(HOH) at the frequency of 1640 cm-1 indicated the existence of water molecules in
a complex structure.
www.intechopen.com
Copper Complexes 31
Fig. 9. FTIR microscopy images (250 μm x 300 μm) of RLMP, Mw = 6000 g mol-1 (1) and Cu(II)
complexes with RLMP synthesized at boiling point and pH 7.5 (2), 8.0 (3), 10.0 (4), 12.0 (5)
From LNT-FTIR it follows that non-interacting OH group is insensitive to temperature
variation whereas a bonded OH shows a significant red shift upon cooling. In the low-
www.intechopen.com
the γ(OH) bending region from librations of H2O molecules on frequencies 847 cm-1 and 756
frequency region on LNT-FTIR spectra, sensitive on the reduction of temperature bands in
cm-1, show blue shift on cooling.

Cu(II)–RLMP complexes are formed by the displacement of H2O molecules from the first
coordination sphere of Cu(II) ion by the OH groups. Copper(II) ion with RLMP unit (Glc)
forms three different types of complex (pH 7–8: Cu(II)(Glc)2(H2O)2, pH 8–10:
Cu(II)(Glc)2(H2O)(OH), pH 10–12: Cu(II)(Glc)2(OH)2).
The changes of the intensity on some bands were registered in RLMP complexes (in the
ranges of a stretching vibration at about 2930 cm-1 and a bending vibration at about 1400 cm-1).
The bands are characteristic of one of more possible positions of the CH2-OH group, rotating
around C(5)-C(6) bond of the pullulan glucopyranose unit.
FTIR microscopy images shows more and less ordered structures of the Cu(II)–RLMP
complexes. ATR-FTIR microspectroscopic data shows homogeneity of the Cu(II)–RLMP
samples. The results of the FTIR spectroscopic study by different techniques allowed one to
suggest a predominant crystalline form of Cu(II)–RLMP complexes.
5. Bioactive copper-dextran complex

Dextran H(C6H10O5)xOH is a complex, branched glucan composed of chains of varying
The straight chain consists of α-(1→6) glycosidic linkages between glucose molecules, while
lengths (from 10 to 150 kDa). It is a polysaccharide similar to amylopectin (Belder, 1985).
branches begin from α-(1→3) or α-(1→4) linkages (Fig. 10). Dextran is synthesized from
sucrose by certain lactic-acid bacteria, the best-known being Leuconostoc mesenteroides and
Streptococcus mutans. Dextran is an oral bacterial product that adheres to the teeth, creating a
film called plaque; dental plaque is rich in dextrans. Dextran is also formed by the lactic acid
bacterium Lactobacillus brevis to create the crystals of tibicos, or water kefir fermented
beverage which supposedly has some health benefits. Dextran is freely soluble in water,
methyl sulphoxide, formamide, ethylene glycol, glycerol, 4-methylmorpholine-4-oxide, and
hexamethylphosphoramide. Some dextran fractions may adopt a certain degree of
crystallinity and may only be brought into solution by strong heating.
Fig. 10. Molecular structure of a dextran: (a) 2D model, (b) 3D model stick and ball
Dextran and its derivatives have been studied extensively (Zhbankov, 1972; Skornyakov &
Komar, 1996). The Fourier-transform infrared spectra of a series of branched dextrans also
www.intechopen.com
Copper Complexes 33
were examined. The IR spectra of low molar dextran have been investigated in the range
between 4000 and 400 cm–1 (Nikolic et al., 1996). The FTIR spectrum of reduced low molar
dextran is presented in Fig. 11. Information on the glucopyranosyl units conformation in the
deformational γ(CH) vibrations bands. In this particular region, as it can be seen in Fig. 11, at
polysaccharide can be acquired in the 1000–700 cm–1 region in which we expect the
least two weak bands around 911 and another 850 cm–1 are observed, which is the proof for
C-1 conformation of glucopyranoside units of dextran. In the δ(OH) range of IR spectra one
band at about 1645 cm–1, which is sensitive to deuteration, has appeared. The band
originates from water molecules.
Fig. 11. FTIR spectra of reduced low molar dextran, Mw = 5000 g mol-1
Dextran is used medicinally as an antithrombotic (anti-platelet), to reduce blood viscosity,
and as a volume expander in anemia (Lewis et al, 2008). These agents are used commonly by
microsurgeons to decrease vascular thrombosis. The antithrombotic effect of dextran is
mediated through its binding of erythrocytes, platelets, and vascular endothelium,
increasing their electronegativity and thus reducing erythrocyte aggregation and platelet
adhesiveness. Dextran also increases blood sugar levels. Biological active polysaccharides
dextran have possibility to binding different ions and metals in the solution and making of
water-soluble complexes. These complexes have wide application in human medicine and
veterine. These compounds have great importance in investigations today.
Copper, essential biometal for living organisms, is a hematopoetical active element of some
metaloenzymes regulating the iron absorption in intestines, maintaining, at the same time,
the iron in a reduced state and influencing the iron incorporation into hemoglobin (Lewis,
1995). The copper amount necessary is usually supplemented by a normal diet in both
humans and animals. It is known that copper deficiency causes a number of pathological
states. Complex compounds of Cu(II) ion are important for prevention and treatment of
some anemia caused by iron deficiency.
The carbohydrate type compounds as ligands have been of a considerable interest. Simple
sugars and their derivatives with reduced and oxidized groups form metal ion complexes of
a various composition and stability. In both human and veterinary medicine commercial
copper preparations based on polysaccharide dextran and its derivatives are used for such
purpose (Ilic et al., 2003). According to literature data, dextran has the ability of complex
formation with various biometals (Co, Zn, Ca and Mg) (Mitic et al., 2007; Cakic et al., 2006;
Lugovaya et al., 1976; Gyurcsik & Nagy, 2000). Iron complexes with different polysaccha-
rides have special importance, and they have been described in detail (Ilic et al., 2002;
www.intechopen.com
Nikolic & Cakic, 2007; Pekic & Cvetkovic, 1988; Cakic & Nikolic, 2003; Nikolic et al., 2002).
The interaction of Cu(II), Ni(II) and Fe(III) ions with dextran may be used for their
speciation by ultrafiltration (Solpan & Sahan, 1993). Synthesis procedures for the complex
formation of Cu(II) ion with polysaccharides, including dextran, are described in scientific
literature (Nikolic et al., 2005; Mitic et al., 2004; Mitic et al., 2007).
In the section, we analyzed the IR spectra of Cu(II) ions complexes with reduced low molar
dextran (RLMD). Fourier-transform IR spectra of dextran and its compounds with copper(II)
ion, recorded at room temperature, were analyzed in order to obtain the information about
the structure and the conformation of these polymer compounds. For IR sample preparation
KBr pastille method was used. The IR spectra as an average of 40 scans were recorded at
room (298 K) temperature on FTIR spectrometer BOMEM MB-100 (Hartman–Braun)
equipped with a standard DTGS/KBr detector, in the range of 4000–400 cm−1 with the
resolution of 2 cm−1, by Win-Bomem Easy software. FTIR microspectroscopy system, ATR-
microscopy attachment equipped with a 15x objective and a 250 μm liquid-nitrogen cooled,
FTIR spectrometer Bruker Tensor-27 in conjunction with a FTIR Bruker Hyperion-1000/2000
narrow-band mercury–cadmium–telluride detector (ATR objective GMBH, Germany) was

used also in this analysis.
Results and discussion.
The plan for the synthesis of the Cu(II) complex with reduced dextran has required a
detailed analysis of the synthesis procedure; both from the aspect of the reaction conditions
of the synthesis and from the aspect of obtaining the stable and commercially applicable
preparation of the complex. The analysis of the synthesis of similar complexes has pointed
to the necessity of defining the physicochemical properties of commercial preparations. By
their correlation, the undesired effects can be eliminated and thus a considerably improved
pharmacological effect of the complex.
The reactivity of the dextran primarily depends on the reactivity of the secondary,
equatorially oriented hydroxyl-groups (OH-2, OH-3 and OH-4). The contents of the
primary OH groups in dextran are slightly increased at lower molar masses (about 2%). As
it is the case with the other glucans, the reactivity of the OH-2 group to the alkalizing
reagents is higher than in the OH-3 and OH-4 groups. This is rationalized in the context of
higher acidity of the OH-2 because of the proximity of anomeric centre (Gyurcsik & Nagy,
2000). When the OH-2 and OH-4 ionize, reactivity to OH-3 is reduced; however, the
substitution on the OH-2 and OH-4 abolish this action and induce the successive increase
in the reactivity of OH-3. On applying these generalizations, it is necessary to exercise
some precaution, since the potency of the base may affect the relative and absolute
reactivities of the hydroxyl. Through derivatization of the dextran to the reduced form the
large number of activation centers with will be at the disposal to the copper ions for the
purpose of binding the complex is expected. This creates the possibility of achieving the
considerably larger stability of the synthesized complex as well as of their pharmacological
effect.
For this reason, the choice and optimization of the low molar dextran in the capacity of the
ligand have been made. Considering the importance of physicochemical parameters on the
process and the synthesis results, the examination and optimization of ligands in relation to
molar mass (Mw), as well as the reaction conditions of the synthesis (pH, T and t) were
investigated and optimized (Mitic et al., 2007). The basic characteristics of synthesized Cu(II)
complexes with RLMD are given in Table 1.
www.intechopen.com
Copper Complexes 35
pH Complex Cu (%) Water

Sample
synth. color (by AAS) solubility
1 7.5 Light green 7.23 Soluble
2 7.5 Green 19.85 Very soluble
3 8.0 Green-blue 8.12 Soluble
4 10.0 Blue-green 8.20 Sparingly soluble
5 12.0 Dark blue 6.97 Slightly soluble
6 7.5 Dark green 4.08 Slightly soluble
Table 1. Characteristics of copper-dextran complexes with RLMD (Mw 5000 g/mol) as
ligand, synthesized at boiling temperature
On the basis of obtained experimental results (Table 1), a favorable result of Cu(II)-RLMD
complexes synthesis is obtained with dextran oligomers Mw 5000, at boiling temperature
and pH 7.5–8.0, within 7 min. Complexes obtained at pH > 8 present unfavorable effects of
synthesis. Comparing the obtained complexes of Cu(II) with RLMD, either in solid state or
in solution, it is obvious that, depending on pH values, various complex colors are obtained
(Table 1). The change of the solutions color during the synthesis may be an indicator
whether the syntheses of complexes were successful. The results obtained have shown that,
in the range of pH 7.5–12, the color can vary from light green to dark blue. This is confirmed
by the green solution color of the most stable complex of Cu(II) with RLMD (procedure 2,
Table 1), in comparison with an indigo-blue alkali solution of decomposed Cu(II) at pH>12,
where [Cu(OH)4]2− ions dominate.
Water solubility of synthesized complexes of Cu(II) with RLMD is different. The most water
soluble complex is obtained at pH 7.5 (Table 1). The solution is permanent and stable after a
longer period of time (6 months). The complexes that are synthesized at higher pH are less
soluble. The solution of the complexes obtained, following the procedure 5 (Table 1), after
resting for long period of time, start layering, precipitate and become opalescent. Medium
pH is changed after adding Cu–salts and Cu(II) content in a complex is much influenced by
it. Syntheses are performed at the same temperatures and within the same reaction period,
but at different pH values (Table 1). The highest Cu(II) content was got at pH 7.5. The
possibility of obtaining Cu(II)–RLMD complexes with a higher Cu(II) content has been
tested with the increased concentration of Cu–salts. The expected results have not been
obtained.
Solution pH probably has the influence on the way of binding of Cu(II) into a complex, i.e.
on the type of a bond because, due to the change of pH value, the stability (Nikolic et al.,
2006), the color and the solubility of the complex obtained are also changed (Table 1). Thus,
by the increase of solution pH values from 7.5 to 12, the percentage of the bounded Cu(II)
with RLMD in complex decreases. Some authors (Tolmachev et al., 1975), in the paper of
influence of medium pH on binding Cu(II) with dextran, point out the possibility of gradual
complexing, i.e. gradual forming of coordination bounds, where their reforming starts at pH
8. Thus, Cu(II) ions form of three different types of complexes with dextran (Norkus et al.,
2002). Decomposition of Cu(II)-dextran complex begins at pH values higher than 12.
The results of ATR-FTIR spectroscopic investigations show that spectra of Cu(II)–RLMD
complexes and ligand are basically similar (Fig. 12). The similarities of the bending (C–H)
range indicate that there is no difference in the conformation of the glucopyranose unit in
the dextran and the complex molecule (916 and 850 cm–1), and they probably exhibit C-1
chair conformation (Mitic et al., 2007).
www.intechopen.com
Fig. 12. ATR-FTIR spectra of Cu(II)–RLMD complexes synthesized at pH 7-8 and pH 10-12
Mid FTIR spectra of Cu(II)–RLMD complexes synthesized at different pH (pH 7–8, Fig. 13B,
and pH 10–12, Fig. 13C) recorded at 298 K, show that the correlation between the O–H
stretch frequency and the hydrogen bond strength. Spectroscopic IR study in particular
region of OH (3400, 1420 cm–1) and C–H (2900, 1460, 1350 cm–1) vibrations indicates different
binding between central metal ion and ligand, depending on pH and metal contents (Fig.
13). The difference in frequencies, intensity, and shape of these bands in the region 3600–
3100 cm−1, implies that in complexes which were synthesized at pH 10–12 there is the
displacement of H2O molecules by O–H groups in the first coordination sphere of the
copper(II) ion. Dextran and complexes with Cu(II) ion have one crystallographic type of
water molecule (1640 cm−1). Water protons take part in the formation of relatively weak
hydrogen bonds (Cakic et al., 2002; Nikolic et al., 2006; Nikolic et al., 2008).
The FTIR investigation corresponds with the results obtained by ESR spectrometry (Mitic et
al., 2004), as well as with the results obtained by UV-VIS investigations (Mitic et al., 2007).
ESR parameters of the spectra (A⎪⎪ = 187 × 10−4 cm−1, g⎪⎪ = 2.23 and g⊥ = 2.03), for the
complexes synthesized at higher pH values, were close to the values for the frozen Cu(II)–
ethylene glycol complex, thus indicating the square-planar coordination of Cu(II) ion with
four oxygen atoms. Although the Cu(II) ion content of complexes synthesized at lower pH
values was much higher (up to 18.95% for the complex synthesized at pH 7.5) the ESR signal
of these complexes was lacking due to strong spin–spin interactions of neighboring Cu(II)
ions. ESR spectrum of complex containing 6.97% of copper synthesized at pH 10 is
www.intechopen.com
Copper Complexes 37
presented to the left. The ESR spectra indicate the axial symmetry of synthesized complexes
and were typical for the Cu(II) ion with one unpaired electron in 3d subshell. Asymmetric
appearance of the hyperfine spectral lines originates from the unresolved spectral
contributions of two natural isotopes, 63Cu and 65Cu. ESR spectral parameters (A⎪⎪ = 187 ×
10−2 cm−1, g⎪⎪ = 2.23 and g⊥ = 2.03 ) point to the tetragonal coordination of Cu(II) with four
oxygen atoms from ligands in the same plane (Mitic et al., 2004; Nikolic et al., 2004).
Fig. 13. FTIR spectra of: LM dextran (A); stable Cu(II)–dextran complex, with high metal
content (≈ 18%), obtained at pH 7–8 (B); unstable Cu(II)–dextran complex, with low metal
content (≈ 5%), obtained at pH 10–12 (C)
In addition, depending on pH values, complexes of Cu(II) with RLMD also behave
differently considering wavelength at which they show absorption maximum. This range of
wavelengths in the VIS spectra is 650–700 nm (Mitic et al., 2007). Hypsochromic effect of
complexes absorption maximums with increase of pH solutions confirms the presence of
different types of complexes. Hexaqua copper(II) ion [Cu(H2O)6]2+ absorb at wavelength
812.7 nm, while synthesized complexes absorb within the ranges of 650–700 nm. With
increase in solution pH the absorption maximums change to shorter wavelengths
www.intechopen.com
comparing with [Cu(H2O)6]2+ ion. Complex, which has been decomposed at pH values
over 12, shows absorption maximum at 634 nm. Thereby, these spectrophotometric
criteria can be applied for the confirmation of the success of complex synthesis. On the
basis of the obtained results by spectroscopic investigations of this complexes, three
different types of Cu(II) complexes structure with deprotonized dextran monomer unit
(Glc–) are suggested depending on pH synthesis. At pH 7 to 8 [Cu(Glc)2(H2O)2] is formed;
at pH 8 to 10 [Cu(Glc)2(OH)(H2O)]– is formed and at pH values over 10 [Cu(Glc)2(OH)2]2–
is formed.
Fig. 14. FTIR microscopy images (250 μm x 300 μm) of RLMD, Mw = 5.000 g mol-1 (1) and
Cu(II) complexes with RLMD synthesized at boiling point and pH 7-8 (2), 8-10 (3) and 10-12
(4)
www.intechopen.com
Copper Complexes 39
The application of microscopic FTIR imaging system to the RLM dextran as ligand and
Cu(II)–RLMD complexes, were synthesized at pH 7.5–12, is shown in Fig. 14. FTIR
microscopy images of dextran, as well as images of the copper-dextran complexes differ
which also indicates that the complexation process and the creation of coordination
compounds took place (Mitic et al., 2010). FTIR microscopy images confirmed that the
changes in the intensity of the analyzed bands are strongly associated with the alterations in
the macromolecular order. These bands in the spectra of the complexes can be responsible
for more and less ordered structures, respectively. The changes in color contour may show
the content and distribution of copper, and polysaccharides in Cu(II)–RLMD samples.
After physicochemical standardization of the most stable complex obtained according to the
procedure 2 (Tab. 1), the preparation for the pharmacological test was provided. The
preparation was tested pharmacologically with the aim of determining systemic acute
toxicity expressed as a median lethal dose (LD50) and as an equivalent of Cu(II) dose per kg
of a mouse body weight (Cakic et al., 2008). None of the applied complex doses at the
concentration of 82–169 mg equivalent of Cu(II) per kg of a mouse’s body weight was lethal
in the tested mice. Therefore, in this case, a median lethal dose could not be determined. The
application of higher doses has caused the mortality of one part of experimental animals.
Thus, in this range, the preparation toxicity LD50 of 1419–1661 was determined, which
corresponds to the equivalent of Cu(II) dose of 281–329 mg per kg of the body weight, in the
concentration 5–20% of the complex solution. Toxicity investigations of various commercial
copper salts show a wide range of values for LD50. The level of the acute toxicity is higher
for more soluble than for less soluble Cu(II) salts. The results of our pharmacological
investigations point to the lower toxicity of Cu(II) complex with RLMD, what is much better
than in the case of commercially applicable inorganic copper salts.
6. References
Bartkowiak, A.; Jezierska, J.; Spychaj, T. (1998). Self-assembled monolayers and polymer
brushes in biotechnology. Polym. Bull., 41, 199-206.
Bellamy, L.J. (1954). The infrared Spectra of Complex Molecules, Methuen, London.
Belder, A.N. (1985). Dextran, Pharmacia AB, Uppsala, Sweden.
Bender, H.; Lehmann, J.; Wallenfels, K. (1959). Pullulan, an extracellular glucan from
Pullularia pullulans. Biochim. Biophys. Acta, 36, 309-316.
Bernier, B. (1958). The production of polysaccharides by fungi active in the decomposition of
wood and forest litter. Can. J. Microbiol. 4(3) 195-204.
of α-D-glucose Fourier transform infrared spectrum deconvolution: comparison

Buslov, D.K.; Nikonenko, N.A.; Sushko, N.I.; Zhbankov, R.G. (1998). Analysis of the results
with experimental and theoretical data. Spectrochim. Acta A, 55(1), 229-238.

Cakic, M.; Nikolic, G.; Ilic, Lj. (2002). FTIR spectra of iron(III) complex with dextran,
pullulan and inulin oligomers. Bull. Chem. Technol. Macedonia, 21(2), 135-146.
Cakic, M.D.; & Nikolic, G.S. (2003). Iron complexes of pharmaceutical interest, Antianemics.
Chem. Industr., 57(10), 437-443.
Cakic, M.D.; Nikolic, G.S.; Mitic, Ž.J.; Ilic, Lj.A. (2004). Conformational analysis of copper(II)-
dextran complexes by Fourier-Transform IR spectroscopy. Proceedings of 4th
International Conference of the Chemical Societies of the South–Eastern European
Countries, p. 72, Belgrade, Serbia, 18–21 July 2004.
www.intechopen.com
Cakic, M.D.; Nikolic, G.S.; Mitic, Ž.J.; Ilic, Lj.A. (2006). Synthesis and spectroscopic
characterization of zinc(II) complexes with dextran derivatives, Proceedings of 5th
International Conference of the Chemical Societies of the South-East European Countries,
p. 544, Ohrid, Macedonia.
Cakic, M.; Mitic, Ž. Nikolic, G.S.; Ilic, Lj.; Nikolic, G.M. (2008). The investigations of
bioactive copper(II) complexes with reduced low-molar dextran. Spectoscopy Int. J.,
22(2-3), 177–185.
Carolan, G.; Catley, B.J.; McDougal, F.J. (1983). The location of tetrasaccharide units in
pullulan. Carbohydr. Res., 114, 237-243.
Catley, B.J. (1970). Pullulan, a relationship between molecular weight and fine structure.
FEBS Letters, 10, 190-193.
Catley, B.J.; Ramsay, A.; Servis, C. (1986). Observations on the structure of fungal
extracellular-polysaccharide, pullulan. Carbohydr. Res., 153, 79-86.
Chenery, D. & Bowring, H. (2003). Infrared and Raman spectroscopic imaging in bioscience.
Spectrosc. Europe, 15(4), 8-14.
Chiu, H.C.; Hsiue, T.; Chen, W.Y. (2004). FTIR-ATR measurements of the ionization extent
of concentration-induced equilibrium swelling. Polymer, 45(5), 1627-1636.
Clark, R.L.; Pesolano, A.A.; Shen, Y.Y. (1997). S. Afr. Patent, 76, 03163, (Chem. Abstr. 1977, 87,
P189951a).
Deshpande, M.S.; Rale, V.B.; Lynch, J.M. (1992). Aureobasidium pullulans in applied
microbiology : a status report. Enzyme Microb. Technol., 14, 514-527.
Gaiduk, O.; Pantaler, R.; Grebenyuk, N.; Ostrovskaya, V. (2009). Rapid determination of
copper(I, II) ions using reagent indicator paper. J. Analyt. Chem., 64(2), 201-205.
Gapanovich, V.; Petrov, P.; Melnova, N. (1998). Human and drug. Abstracts of papers, V
Russian National Congr., Moscow, 1998, p. 480.
Gierlinger, N. & Schwanninger, M. (2007). The potential of Raman microscopy and Raman
imaging in plant research: review. Spectroscopy Int. J., 21, 69-89.
Grigorieva, A.S.; Konakhovich, N.F.; Budarin, L.I. (1983). Possible applications of
coordination compounds for correcting biometal metabolism in different
pathologies. Inorganica Chimica Acta, vol. 79, pp. 297-298.
Gyurcsik, B. & Nagy, L. (2000). Carbohydrates as ligands: Coordination equilibria and
structure of the metal complexes. Coord. Chem. Rev., 203, 81–149.
Hardtmann, G.G. & Kathawala, F.G. (1977). US. Patent 053, 800, (Chem. Abstr. 88, 2297).
Expert group on vitamins and minerals. (2002). Draft review of copper, EVM/99/19,
Secretariat, UK Government, London.
Ilic, Lj.; Ristic, S.; Cakic, M.; Nikolic G.; Stankovic, S. (2002). Polynuclear complex Fe(III)
with pullulan oligomers, process of its obtaining, and pharmaceutical preparations
on the basis of the complex, Pat. WO 02/46241 A2, Int. Cl. C08B 37/00, Int. Pub.
Date 13 June, 2002.
Ilic, Lj.A.; Mitic, Ž.J.; Nikolic, G.S.; Cakic, M.D. (2003). Synthesis and pharmacological
activity of copper(II) and cobalt(II) complexes with dextran oligomers, Proceedings
of XVII Ulusal Kimya Kongresi, p. 190, Istanbul, Turkey.
Ishihara, M.; Obara, K.; Ishizuka, T.; Fujita, M.; Sato, M.; Masuoka, K.; Saito, Y.; Yura, H.;
Matsui, T.; Hattori, H.; Kikuchi, K.; Kurita, A. (2003). Controlled release of
fibroblast growth factors and heparin from photocrosslinked chitosan hydrogels
and subsequent effect on in vivo vascularization. J. Biomed. Mater. Res., 64A, 551-59.
www.intechopen.com
Copper Complexes 41
Jackson, W.D. & Polaya, J.B. (1951). Aust. J. Soc., vol. 13, pp. 149.
Kacurakova, M. & Mathlouthi, M. (1996). FTIR and laser-Raman spectra of oligosaccharides
in water: Characterization of the glycosidic bond. Carbohydr. Res., 284, 145-157.
Kacurakova, M. & Wilson, R.H. (2001). Developments in mid-infrared FT-IR spectroscopy of
selected carbohydrates. Carbohydr. Polym., 44, 291-303.
Kathawala, F.G. (1974). US. Patent 3, 850, 932 (Chem. Abstr. 82, 140175).
Kim, S.; Chae, S.Y.; Na, K. ; Kim, S.W.; Bae, Y.H. (2003). Insulinotropic activity of
sulfonylurea/pullulan conjugate in rat islet microcapsule. Biomaterials, 24(26), 4843-
4851.
Kiselev, V.P.; Komar, V.P.; Skornyakov, I.V.; Firsov, S.P.; Virnik, A.D.; Zhbankov, R.G.
(1977). Spectroscopic study of various structural modifications of dextran. Polym.
Sci. USSR, 19(8), 2140-2146.
Klimovich, O.; Gapanovich, V.; Petrov, P. (1998). Human and drug. Abstracts of papers, V
Russian national Congr., Moscow, pp. 493.
Komar, V.P.; Zhbankov, R.G.; Khomyakov, K.P.; Zavada, O.A.; Virnik, A.D.; Rogovin, Z.A.
(1968). Dextran and derivatives. Polym. Sci. USSR, 10, 1351-1357.
Kweon, D.K.; Song, S.B.; Park, Y.Y. (2003). Preparation of water-soluble chitosan/heparin
complex and its application as wound healing accelerator. Biomaterials, 24, 1595-
1601.
Lawrence, R.; Lange, R.; Wu, C.; Helenek, M. (1997). Iron dextran formulations, US Pat. 5
624 668, Int. Cl A01N059/16, US Cl 424/078, Appl. No. 536 984.
Leathers, TD. (2003). Biotechnological production and applications of pullulan. Appl.
Microbiol. Biotechnol., 62(5–6), 468-473.
Leonard, M.; Rastello De Boisseson, M.; Hubert, P.; Dellacherie, E. (2003). Production of
microspheres based on hydrophobically associating alginate derivatives by
dispersion/gelation in aqueous sodium chloride solutions. J. Biomed. Mater. Res.,
68A (2003) 335-342.
Lepetil, G.S.P.A. (1975). Ger. Patent, 2. Vol. 424, 670 (Chem. Abstr. 1975, 83, 20628).
Lewis, A.G. (1995). The Biological Importance of Copper, Literature Review ICA Project, No.
223, British Columbia, pp. 7–17.
Lewis, S.L.; Heitkemper, M.M.; Dirksen, S.R.; O'Brien, P.G.; Bucher, L. (2008). Medical
Surgical Nursing : Assessment and Management of Clinical Problems, 7th Edition,
Mosby Elsevier.
Li, J.; Pan, J.; Zhang, L.; Guo, X.; Yu, Y. (2003). Culture of primary rat hepatocytes within
porous chitosan scaffolds. J. Biomed. Mater. Res., 67(3A), 938-943.
Lin, S.Y.; Li, M.J.; Cheng, W.T. (2007). FT-IR and Raman vibrational microspectroscopies
used for spectral biodiagnosis of human tissues. Spectroscopy, 21, 1-30.
London, E. (2004). The molecular formula and proposed structure of the iron-dextran
complex, imferon. J. Pharm. Sci., 93(7), 1838-1846.
Lugovaya, Z.A.; Tolmachev, V.N.; Naiden, V.I.; Valakhanovich A.I.; Zaretskaya, R.V. (1976).
Reaction products of low molecular-weight dextran with copper. Pharm. Chem. J.,
10(11), 1536–1538.
Mähner, C.; Lechner, M.D.; Nordmeier, E. (2001). Synthesis and characterization of dextran
and pullulan sulphate. Carbohydr. Res., 331, 203-208.
www.intechopen.com
Mitic, Ž.; Cakic, M.D.; Nikolic, R.; Ilic, Lj.; Nikolic, G.S. (2004). Polysaccharide complexes of
pharmaceutical interest. Water solutions stability of Cu(II) complexes with reduced
low-molecular dextran. World Polym., 7(5), 225–229.
Mitic, Ž.; Nikolic, G.; Cakic, M.; Nikolic, R.; Ilic, Lj. (2007 a). Synthesis and spectroscopic
characterization of copper(II) dextran complexes. Russ. J. Phys. Chem., 81(9), 1433–7.
Mitic, Ž.; Cakic, M.D.; Nikolic, G.S.; Nikolic, R.S.; Ilic, Lj.A. (2007 b). The investigation of
Co(II) dextran complexes. Chem. Ind., 61(5), 257–262.
Mitic, Ž.; Nikolic, G.S.; Cakic, M.D.; Nikolic, R.; Ilic, Lj. (2007 c). Fourier-transform IR
spectroscopy of Cu(II) ion complexes with different polyglucans, Proceedings of XII
European Conference on the Spectroscopy of Biological Molecules (ECSBM), p. 319,
Bobigny, Paris, France, 1–5 September 2007.
Mitic, Ž.; Nikolic, G.S.; Cakic, M.D.; Premovic, P.I.; Ilic, Lj. (2008 a). FTIR spectroscopic
characterization of Cu(II) coordination compounds with exopolysaccharide
pullulan and its derivatives. Proceedings of XXIX European Congress on Molecular
Spectroscopy (EUCMOS), Opatija, Croatia, 31 Aug–5 Sep 2008, p. 197.
Mitic, Ž.; Nikolic, G.S.; Cakic, M.D. (2008 b). Development of ATR-FTIR microspectroscopy
for analysis of bioactive copper(II)-polysaccharide complexes. Proceedings of
Conference SPEC2008, V International Conference of Spectroscopy, Shedding light on
disease : Optical diagnosis for the New Millennium, Sao Paulo, Brazil, 25–29 October
2008, p. 86-88.
Mitic, Ž.; Cakic, M.; Nikolic, G.; Ilic, Lj.; Stankovic, M. (2010). Spectroscopic characterization
of bioactive Cu(II) complexes with polysaccharides by FTIR microspectroscopy.
Chem. Ind., 64(1), 9–20.
Mousia, Z.; Farhat, I.A.; Pearson, M.; Chesters, M.A.; Mitchell, J.R. (2001). FTIR
microspectroscopy study of composition fluctuations in extruded amylopectin–
gelatin blends. Biopolymers (Biospectroscopy), 62(4), 208-218.
Nagy, L.; Yamaguchi, T.; Yoshida, K. (2003). Application of EXAFS and XANES methods in
coordination chemistry of carbohydrates and their derivatives. Struct. Chem., 14(1),
77-84.
Nakamoto, K. & Czernuszewicz, R.S. (1993). Infrared Spectroscopy, in J.F. Riordan and B.L.
Vallee, eds., Methods in Enzymology, Vol. 226, Part C, Academic Press, San Diego, CA.
Nakamoto, K. (2009). Infrared and Raman Spectra of Inorganic and Coordination Compounds. Part
B: Applications in Coordination, Organometallic, and Bioinorganic Chemistry, Sixth
Edition, John Wiley & Sons, Inc., Hoboken, New Jersey, pp. 333, ch 3.
Nikolic, G.M.; Ilic, Lj.; Cakic, M.D.; Nikolic, G.S.; Nikolic, R.; Mitic, Ž. (2004). Copper(II)
complexes of reduced low-molecular dextran: an ESR study, Proceedings of 7th
International Conference on Fundamental and Applied Aspects of Physical Chemistry, pp.
775–777, 21–23 September 2004, Belgrade, Serbia.
Nikolic, G.S.; Cakic, M.; Cvetkovic, D.; Ilic, Lj.; Premovic, P.I. (1996). A spectroscopic study
of polynuclear complex of iron(III) with inuline oligomers. Spectrosc. Lett., 29(7),
1437-1448.
Nikolic, G.S.; Cakic, M.D.; Ilic, Lj.; Ristic, S.; Cakic, Ž. (2002). Synthesis of some new
antianemics I. Iron pullulan complexes of pharmaceutical interest. Die Pharmazie
57(3), 155–158.
www.intechopen.com
Copper Complexes 43
Nikolic, G.S.; Cakic, M.D.; Mitic, Ž.; Nikolic, R.; Ilic, Lj. (2005). Correlation between UV–VIS
spectra and structure of Cu(II) complexes with hydrogenated dextran in alkaline
solutions. Chem. Industr., 59(1/2), 9–14.
Nikolic, G.S.; Mitic, Ž.; Cakic, M.D.; Ilic, Lj. (2006 a). Study of copper(II)–dextran complexes
stability by conductometry method, Proceedings of 8th International Conference on
Fundamental and Applied Aspects of Physical Chemistry, pp. 719–721, Belgrade, Serbia.
Nikolic, G.S.; Cakic, M.D.; Mitic, Ž.; Ilic, Lj. (2006 b). Copper(II) complexes of reduced low-
molar dextran: a deconvoluted LNT-FTIR study, Proceedings of 8th International
Conference on Fundamental and Applied Aspects of Physical Chemistry, pp. 722–724,
Belgrade, Serbia.
Nikolic, G.S. & Cakic, M.D. (2007). Physical investigation of the colloidal iron-inulin
complex. Colloid. J., 69(4), 464–473.
Nikolic, G.S.; Cakic, M.D.; Mitic, Ž.; Ilic, Lj. (2008). Deconvoluted Fourier-transform LNT-IR
study of coordination copper(II) ion compounds with dextran derivatives. Russ. J.
Coord. Chem., 34(5), 322-328.
Nikonenko, N.A.; Buslov, D.K.; Sushko, N.I.; Zhbankov, R.G. (2005). Spectroscopic
manifestation of stretching vibrations of glycosidic linkage in polysaccharides. J.
Mol. Struct., 752, 20-24.
Norkus, E.; Vaiciuniene, J.; Vuorinen, T.; Heikkilä, M. (2002). Interaction of Cu(II) with
dextran in alkaline solutions. Carbohydr. Polym., 50(2), 159-164.
Norkus, E.; Vaiciuniene, J.; Vuorinen, T.; Macaladay, D. (2004). Equilibria of Cu(II) in
alkaline suspensions of cellulose pulp. Carbohydr. Polym., 55(1), 47-55.
Panov, V.P. & Zhbankov, R.G. (1976). Conformations of Sugars, Nauka i Tekhnika, Minsk.
Panov, V.P.; Zhbankov, R.G. (1988). Intra- and intermolecular interactions in carbohydrates,
Nauka i Tekhnika, Minsk.
Parker, F.S. (1983). Application of Infrared Raman and Resonance Raman Spectroscopy in
Biochemistry, Plenum Press, New York.
Pekic, B. & Cvetkovic, D. (1988). Polynuclear complex Fe(III) with dextran and its
derivatives,YU Pat. 42384 B, Int. Cl. C08B37/02, Appl. No. 564.
Perets, A.; Baruch, Y.; Weisbuch, F.; Shoshany, G.; Neufeld, G.; Cohen, S. (2003). Enhancing
the vascularization of three-dimensional porous alginate scaffolds by incorporating
controlled release basic fibroblast growth factor microspheres. J. Biomed. Mater. Res.,
65A, 489-497.
Platonova, N.; Zhiguleva, E.; Tsymbalenko, N.; Mishchenko, B.; Vasin, A.; Zhivulko, T.;
Puchkova, L. (2004). Age-related features of ceruloplasmin biosynthesis and
distribution in rats. Russ. J. Develop. Biology, 35(3), 131-141.
Serizawa, T.; Yamaguchi, M.; Kishida, A.; Akashi, M. (2003). Alternating gene expression in
fibroblasts adhering to multilayers of chitosan and dextran sulfate. J. Biomed. Mater.
Res., 67A, 1060-1063.
Shingel, K.I. (2002). Determination of structural peculiarities of dexran, pullulan and
gamma-irradiated pullulan by Fourier-transform IR spectroscopy. Carbohydr. Res.,
337(16), 1445-1451.
Shingel, K.I. (2004). Current knowledge on biosynthesis, biological activity and chemical
modification of the exo-polysaccharide pullulan. Carbohydr. Res., 339, 447-460.
Sivchik, V.V.; Zhbankov, R.G.; Astreiko, M.V. (1979). Theoretical Investigation of the
Vibrational Spectrum of Dextran. Acta Polym., 30, 689-693.
www.intechopen.com
Skornyakov, I.V. & Komar, V.P. (1996). Infrared spectra of structural modifications of
dextran. J. Appl. Spectrosc., Volume 63, No. 3, 309-317.
Smits, A.L.M., Ruhnau, F.C.; Vliegenthart, J.F.; Van Soest, J.J.G. (1998). Ageing of starch-
based systems as observed with FTIR and solid-state NMR spectroscopy.
Starch/Staerke, 50, 478-483.
Solpan, D. & Sahan, M. (1993). Concentration and separation of metallic ions from aqueous
solutions by using EC-PEG 4000 alloy membranes. J. Appl. Polym. Sci., 48(12), 2091–
2093.
Sowa, W.; Blackwood, A.C.; Adams, G.A. (1963). Neutral extracellular glucan of Pullularia
pullulans (de Bary) Berkhout. Can. J. Chem., 41(9), 2314-2319.
Tanodekaew, S.; Prasitsilp, M.; Swasdison, S.; Thavornyutikarn, B.; Pothsree, T.; Pateepasen,
R. (2004). Preparation of acrylic grafted chitin for wound dressing application.
Biomaterials, 25(7), 1453-1460.
Tolmachev, V.N.; Lugovaya, A.Z.; Ischenko, I.K.; Valikhonovich, A.I.; Zaboronok, V.U.
(1975). Vysokomol. Soedin., Ser. A, 17(2), p. 419-425.
Vosburg, W.C. & Cooper, G.R. (1941). Complex Ions. J. Am. Chem. Soc., vol. 63, pp. 437-442.
Wallenfels, K.; Bender, H.; Keilich, G.; Bechtler, G. (1961). On pullulan, the glucan of the
slime coat of Pullularia pullulans. Angew. Chem., 73, 245-246.
Wallenfels, K.; Keilich, G.; Bechtler, G.; Freudenberger, D. (1965). Untersuchungen an
pullulan. Die klarung des strukturproblems mit physikalischen, chemischen und
enzymatischen methoden. Biochemische Zeitschrgt, 341, 433-450.
Walker, H.A.; Wilson, S.; Atkins, E.C.; Garret, H.E.; Richardson, A.R. (1951). J. Pharmacol.
Exp. Ther., vol. 101, No. 4, pp. 368-378.
Yin, Y.; Ye, F.; Cui, J.; Zhang, F.; Li, X.; Yao, K. (2003). Preparation and characterization of
macroporous chitosan-gelatin. J. Biomed. Mater. Res., 67A, 844-855.
Yu, W. & Xiong, L. (2005). Effect of various morphology and testing conditions on fiber
infrared spectrum. J. Appl. Polym. Sci., 96(4), 1003-1010.
Zhbankov, R. (1972). Infrared Spectra and Structure of Carbohydrates, Nauka i tekhnika, Minsk,
pp. 56–95, and pp. 180–227.
Zhbankov, R.G. & Avsenev, N.N. (1984). Study of the ordered conformational states of
dextran macromolecules in solutions with additions of alkali halide salts. Polym.
Sci. USSR, 26(2), 262-271.
Zhbankov, R.G. (1992). Vibrational spectra and structure of mono- and polysaccharides. J.
Mol. Struct., 275, 65-84.
Zhbankov, R.G.; Andrianova, V.M.; Marchewka, M.K. (1997). Fourier Transform IR and
Raman spectroscopy and structure of carbohydrates. J. Mol. Struct., 436, 637-654.
Zhbankov, R.G.; Firsov, S.P.; Korolik, E.V.; Petrov, P.T.; Lapkovski, M.P.; Tsarenkov, V.M.;
Marchewka, M.K.; Ratajczak, H. (2000). Vibrational spectra and the structure of
medical biopolymers. J. Mol. Struct., 555, 85-96.
Zhbankov, R.G.; Firsov, S.P.; Kolosova, T.E.; Prihodchenko, L.K.; Grinshpan, D.D.; Baran, J.;
Marchewka, M.K.; Ratajczak, H. (2003). Fourier-transform Raman spectroscopy of
biological molecules: Weak intermolecular interactions. J. Mol. Struct., 656, 275-286.
Zhbankov, R.G.; Zhbankova, M.R.; Baran, J.; Marchewka, M.; Ratajczak, H. (2005).
Anomalously high intensity of bands in the FTR-spectra of individual kinds of
polymers. J. Mol. Struct., 744/747, 585-587.
www.intechopen.com
3
Precision Quantitative Proteomics with Fourier-

Transform Mass Spectrometry
Qingbo Li
University of Illinois at Chicago
U.S.A.
1. Introduction
The Fourier-transform mass spectrometry (FTMS) instrument offers a mass resolution
higher than most of other mass spectrometers. This high resolution is in part due to the
better stability offered by a superconductor magnet in FTMS than a radio frequency voltage
utilized in many other mass spectrometers. The extremely high resolution of FTMS has very
important application in biomedical proteomics research. The high resolution not only
dramatically improves the reliability of protein identification but also the accuracy of
protein quantitation.
In this chapter, we present several examples of proteomics study that takes advantage of the
high resolution offered by FTMS. Particularly, we describe examples of proteome dynamic
study with isotopomer analysis, and precise peptide and protein label-free quantitation with
rigorous statistical assessments.
In protein dynamic studies, FTMS readily resolves all of the isotopomer peaks for peptides.
The well-resolved isotopomer peaks allow a direct integration of the intensities of different
isotopologue envelopes without the need of a deconvolution algorithm. With the high-
resolution data, we were able to show that protein turnover measurement revealed more
subtle changes in the dynamics of a proteome.
The high resolution of FTMS helps to reduce the interference of contaminant peaks (Fig. 1).
The ability to resolve the targeted peptide isotopomer peaks from interfering ones greatly
facilitates the implementation of a label-free quantitative proteomics method that relies on
peptide cross reference between liquid chromatography runs and the integration of
extracted ion chromatographic intensities (Lipton et al., 2002). With both high confidence in
peptide identification and quantitation, we showed that the major source of variability lies
more in sample preparation than in liquid chromatography/mass spectrometry analysis.
Such a result has a direct consequence in the statistical approaches utilized to assign
significance in label-free quantitative proteomics.
Two sections are presented in this chapter. The first one briefly discusses the general aspect
of a protein turnover analysis followed by examples of proteome dynamics study in acid
stressed and iron limited mycobacterial cells. The second one describes the utilization of
high-resolution FTMS data for label-free quantitation of proteins with a rigorous statistical
assessment of significance in differential protein abundances.
The spectra shown in the panels a to e of Fig. 1 are for a tryptic peptide from a superoxide
dismutase in M. smegmatis (MSMEG_6427; Mn) with a sequence of AFWNVVNWDDVQNR.
www.intechopen.com
The panels a and b are the ion-trap mass spectrometry (ITMS) MS2-scan spectra for a +2
precursor ion of the peptide in the nanoLC/LTQ-FT and the nanoLC/LTQ respectively.
Shown atop of panel a is the peptide sequence labeled with the detected y-series ions in the
MS2-scan spectra. For clarity, only y-series ion fragments are shown with labels. The panels
c to e show the MS-scan spectra obtained with a FTMS scan (c), an ITMS zoom MS-scan (d),
and an ITMS full MS-scan (e), respectively. Only an m/z range of 880 to 885 is shown for the
+2 peptide charge state. The three short arrows in the panels c, d, and e indicate the first
three isotopic peaks of the +2 peptide charge state. The asterisks in panel c indicate the
isotopic peaks probably not related to the peptide. The theoretical molecular weight of the
peptide is shown atop of panel c. The long dashed arrows point to the respective MS-scans
from which a precursor ion is selected for the MS2 scans. Adapted from (Li, 2010a).
Fig. 1. A resolution and a signal-to-noise ratio in MS-scan and MS2-scan spectra are
compared between a nanoLC/LTQ-FT and a nanoLC/LTQ mass spectrometry systems
2. Protein turnover analysis with FTMS

2.1 Protein turnover
Protein turnover is a fundamental cellular process in all cell types having important
implications in many aspects of biological science (Larrabee et al., 1980; Wilkinson, 2005).
Advancement of high resolution proteomic technologies has provided the possibility to
study protein turnover for multiple proteins simultaneously in complex cellular protein
extracts (Beynon, 2005; Cargile et al., 2004; Pratt et al., 2002; Rao et al., 2008a; Rao et al.,
www.intechopen.com
Precision Quantitative Proteomics with Fourier-Transform Mass Spectrometry 47
2008b; Vogt et al., 2005). We showed that a combination of protein abundance and turnover
data provides a highly interesting insight into the dynamic process of and interconnection
among protein synthesis, degradation, and secretion (Rao et al., 2008a).
Protein turnover shares an equally important role with gene transcription and protein
translation. Synthesis of new proteins and degradation of old ones form a dynamic process
in an organism. Turnover does not only help to clear old proteins but also aid in a fast
adaptation to a new condition or environment by adjusting the rate of protein synthesis and
degradation (Goldberg & Dice, 1974). Apart from this, turnover also brings new proteins
into action with reduced strain on the resources of an organism because preexisting cellular
materials are reused. Some early global protein turnover studies identified different E. coli
proteins that might have different turnover rates. One of those earlier studies showed that a
dynamic state for individual proteins existed in non-growing as well as growing cells
(Larrabee et al., 1980). Studies of turnover offer a dynamic view of the abundances of
proteins. When being applied to a larger scale of the proteome, protein turnover analysis
allows one to study the dynamic nature of the entire proteome (Li, 2010a).
Mass spectrometry continues to serve as a major approach for a protein turnover study due
to its wide availability and flexibility to analyze both single-cell cultures and multi-cellular
organisms. Except for the required administration of stable isotope-labeled metabolites,
amino acids, or water to the study subjects, mass spectrometry-based approaches do not
require any genetic manipulation of the study subjects. The avoidance of genetic
manipulation of the study subjects helps to minimize any unwanted perturbation to a
biological system and delivers the most physiologically relevant results. A range of
methodologies has been established to measure protein turnover based on stable isotope
labeling and mass spectrometry (Doherty & Beynon, 2006).
2.2 High-resolution mass spectrometry for protein turnover analysis

The advent of highly automated high-resolution mass spectrometry technology promises to
bring about in-depth insight into the dynamic nature of a proteome at the global level. The
work done by Pratt et al. (Pratt et al., 2002) demonstrated the determination of protein
degradation rate constants in a steady state population of yeast grown in a chemostat. The
authors used isotope labeling along with 2D gel analysis to study protein turnover and
advocated that protein turnover is ‘a missing dimension in proteomics.’ Another study done
by Cargile et al. (Cargile et al., 2004) labeled E. coli cells with 13C to study the relative
synthesis over degradation ratio (S/D). These earlier works demonstrated the global
analysis of protein turnover with individual protein identifications but their data were not
correlated with abundance values.
With one-dimensional SDS/PAGE fractionation and subsequent nanoLC/LTQ-FTMS
analysis, Rao et al. determined the global turnover profiles of Mycobacterium smegmatis, a
non-pathogenic surrogate of Mycobacterium tuberculosis, under acid-shock and iron-
limitation conditions (Rao et al., 2008b). A dynamic range of 3-orders of magnitude was
demonstrated for relative turnover measurements. The study provided direct evidence that
relative turnover in growing mycobacterium cells, with or without stress, was highly
heterogeneous. The results obtained in that study addressed the long-standing question
whether a ‘dynamic state’ exists in growing bacteria (Borek et al., 1958), and illustrated the
benefits and needs to study protein turnover at the global level with the most advanced
mass spectrometry technology.
www.intechopen.com
In the work by Rao et al. (Rao et al., 2008b), the cells were grown with two different
methodologies for both different types of stresses. For the pH stress the cells were initially
grown in [14N]-containing media at pH 7.0. Once the cells are in the initial log phase, the
cells were divided into two flasks and the media was doped with 50% [15N] and the pH was
reduced to 5.0 in one of the flasks. The cells were harvested after one doubling and analyzed
for protein turnover using LC/LTQ-FTMS. For low iron analysis, the cells were first grown
in [15N]-containing media until mid-log phase. The cells were then collected by
centrifugation and the media was then exchanged with [14N]-containing media. The cells
were then allowed to grow to one doubling and harvested to be analyzed by LC/LTQ-
FTMS.
In either the complete isotope swapping (iron-limitation experiments) or the partial isotope
labeling conditions, the isotopologue envelopes and the individual isotopomer peaks of a
peptide are clearly resolved. The complete resolution of the isotopomer peaks and the
isotopologue envelops for the old proteins and the de novo synthesized proteins facilitates
simple calculation of the abundances of the new and old fractions of a peptide (Fig. 2).
140 AL 35
AM Starved (low-iron)
Intensity (x10 counts)
Intensity (x10 counts)
120 Shocked (pH5) 30 Control (high-iron)

Control (pH7) AH
100 25
6
80
6
20
60 15 AL
40
10
20
5
0
0
957.03 959.03
961.02 963.01 957.03 959.03
965.00 961.02
967.00 963.01
965.00
a m/z 968.99
b m/z
967.00
968.99
Fig. 2. Calculation of isotopologue intensities for a representative tryptic peptide

ANLLGLSAPEMTTLVGGLR (MH22+, 23 N atoms) of protein KatG (MSMEG6346) in
shocked (pH5) and control (pH7) cultures (panel a), and in starved (low-iron) and control
(high-iron) cultures (panel b). AL, AH, and AM represent the isotopologue intensities with
light label (99.6At%[14N]), heavy label (99At%[15N]) and medium label (50At%[15N])
respectively. Red arrows and text labels indicate de novo synthesized proteins. Blue arrows
and text labels indicate old proteins. Adapted from (Rao et al., 2008b) with permission
Turnover analysis of M. smegmatis under both stressful conditions revealed two different
patterns (Rao et al., 2008b). In the low pH condition, many proteins had increased turnover
at pH 5.0 as compared to pH 7.0. It was an obvious reaction since the bacteria has to readjust
its proteome in order to counter the stress posed by increased proton concentrations. The
correlation coefficient for the low pH shock cells was small which indicated that the proteins
in the cells exposed to pH 5.0 underwent extensive readjustment in different directions. In
the low iron stress the correlation coefficient being high suggested that either there was not
much rearrangement of turnover values or all the proteins had changes in a similar
direction. KatG and Tpx, which are important for protection of mycobacterial cells against
oxidative stress, had low protein turnover values in both low iron as well as low pH
conditions. A study on M. tuberculosis Tpx suggested that it might be an important protein
www.intechopen.com
against oxidative stress because Tpx mutants were unable to survive in the macrophages in
an infected mouse model. However, it would be interesting to analyze how the low
turnover of Tpx correlates with the survival of mycobacteria in the cell.
60
a
50
AL,PCS AM,PCS
40
30
20
10
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
50
45 b
40 AL,PCS AM,PCS
35
30
25
Intensity (arb. unit)
20
15
10
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
35
30
c
AL,PCS AM,PCS
25
20
15
10
0
1 3 5 7 9 11 13 15 17 19 21 23 25
6
d
AL,PCS AM,PCS
5
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Isotopomer Number (M)
Fig. 3. Average isotopomer profiles and selected isotopomer ranges to calculate the
abundances of peptides in M. tuberculosis. AL,PCS and AM,PCS are peptide abundance for old
and new proteins respectively. Each panel was the stacked column graph of the normalized
isotopomer profiles of the detected peptide charge states (PCSs) having the same number of
N atoms (n). Profiles are shown for n equal to 11 (a), 16 (b), 20 (c), and 26 (d) respectively.
The blue and red arrows indicate the M ranges for calculating AL,PCS and AM,PCS respectively
An open question is how the protein turnover values correlate with protein abundances. To
investigate the correlation between protein abundance and protein turnover values in M.
tuberculosis, Rao et al. analyzed M. tuberculosis cells in an iron replete and iron depleted
condition using the high resolution LC/LTQ-FTMS instrument (Rao et al., 2008a). The
approach employed many large-scale quantitative proteomics techniques to make it readily
accessible for protein turnover studies at the global level. The concomitant measurement of
protein turnover and abundance was previously shown by Gerner et al. with 2D gel
www.intechopen.com
electrophoresis for separation and fluorography and autoradiography for quantitation of

individual gel spots without any protein quantification (Gerner et al., 2002). With the
advanced nanoLC/LTQ-FTMS system and a label-free quantitation approach, Rao et al.
demonstrated that protein abundance and turnover could both be measured with a dynamic
range of at least 3-orders of magnitude in an automated fashion (Rao et al., 2008a).
The study compared the sensitivity of relative turnover and relative abundance
measurements to detect a dynamic response of M. tuberculosis when the cell culture was
shifted from an iron-starved to iron-sufficient condition. An unlabeled iron-depleted M.
tuberculosis culture was grown to late-log phase and diluted with a fresh iron-replete
medium that was labeled with [15N]-labeled nitrogen source (Rao et al., 2008a). The
incorporation of the [15N]-labeled nitrogen source into the newly synthesized proteins
resulted in a complete separation between the old and the newly synthesized peptide
isotopologue profiles (Fig. 3). Similar to that shown in Fig. 2 for M. smegmatis, the complete
separation of the isotopologue profiles and the isotopomers allows the quantitation of the
abundances of old proteins, newly synthesized proteins, and the total proteins.
In this work, we are able to obtain both the protein abundance and turnover values to more
comprehensively assess the dynamic response of the H37Rv cells when they were shifted
from iron-starved stationary-phase to fresh low- and high-iron media. This is achieved by
applying both the isotope chasing and a label-free quantitation method (Rao et al., 2008a).
The results indicated that a relative turnover measurement was much more sensitive to
monitor the dynamic response of the M. tuberculosis cells. Meanwhile, a combination of
turnover and abundance measurements provided insight into the correlation of protein
synthesis, degradation, and secretion. A further principal component analysis of the M.
tuberculosis proteome dynamics reveals that protein relative turnover properties are
orthogonal to protein relative abundance properties (Rao & Li, 2009a). Thus, a study of
protein turnover at the global level would likely bring forward new findings that can be
missed with a protein abundance analysis alone.
The data obtained from the comparison of protein abundance and protein turnover values
showed that protein turnover is a much more sensitive measurement to discern the changes
in the proteome than abundance measurements. Upon the transfer of late-log phase cells
from a low iron to a high iron media, protein abundance measurements showed that out of
the 104 proteins that we identified, only 5 proteins were upregulated and 16 proteins were
downregulated in the HI media. Relative abundance of KatG was upregulated in cells
grown in the high iron media.
Protein turnover analysis of the proteins compared between cells grown in a low iron and a
high iron media showed that more proteins had increased synthetic activity in the high iron
grown cells. The S/D had increased for 24 proteins in the cells grown in the low iron media.
Eight proteins had decreased turnover. However, for cells grown in the high iron media, 56
proteins had increased S/D and 5 proteins had decreased S/D. A comparison of protein
abundance measurements to protein turnover measurements clearly suggests that protein
turnover does give more information to uncover the dynamic response of the proteome
(Fig. 4).
In addition to providing the information about synthesis and degradation of proteins, the
protein turnover analysis can also provide information regarding whether a protein has
been secreted when the protein turnover values are analyzed together with the protein
abundance measurements. In our study of proteome dynamics, we found that some proteins
had low changes in relative abundances even though their synthesis had increased
www.intechopen.com
significantly. As stated before the relative abundance of a protein in the cell can be affected
not only by synthesis or degradation but also by a secretion process. In our turnover
analysis, we found discrepancies between protein abundance values of certain proteins and
their turnover values. A previous proteomic study of M. tuberculosis culture filtrates showed
many of those proteins to be secreted into the culture filtrate. Proteins such as FbpC2, KatG,
and the mammalian cell entrance protein Rv0172 were also predicted to be secreted (Malen
et al., 2007).
a)
b) c)
Fig. 4. M-A plots representing the total protein abundances in high iron (HI) versus low-iron
(LI) cells (panel a), the newly synthesized protein abundances versus the old protein
abundances in LI cells (panel b) and in HI cells (panel c) respectively. The proteins with 2-
fold significant (p <.05) change in relative abundance are marked with black triangles and
diamonds. The M–axis represents the relative abundance values and the A-axis represents
the average abundance values. Adapted from (Rao et al., 2008a) with permission
These results support that protein turnover in combination with abundance analysis could
predict the secretion of proteins and reveal the interconnected roles of protein synthesis,
degradation, and secretion in determining the protein abundances in cells. These analyses
illustrate that protein turnover can divulge information that classical proteomics does not
provide. The integration of data from transcriptome studies, abundance measurements and
turnover analyses will likely provide a more complete picture of the dynamics associated
with the proteome. To some extent, it will probably reconcile the discordances between
transcriptome and proteome analyses.
www.intechopen.com
2.3 Implication of proteome turnover studies in mycobacteria

With the advent of high-precision and automated mass spectrometry instrumentation to
support large-scale proteomic studies, protein turnover analysis at the global level
potentially has an increasing importance for biomedical research. One example is the
application of protein turnover analysis to study M. tuberculosis, especially at its non-
replicating or dormant state. Whereas over a hundred research articles have been
published on mycobacterial proteomes, only a few dealt with non-replicating M.
tuberculosis (Cho et al., 2006; Rosenkrands et al., 2002). Information about protein turnover
in non-replicating and dormant M. tuberculosis is scarce in the literature. With the
potential importance of proteome dynamics in bacterial cell sporulation or dormancy
(Bernlohr, 1967; Bernlohr, 1972; Mandelstam, 1958; Spudich & Kornberg, 1968), a study of
mycobacterial protein turnover at the global level (Rao et al., 2008a; Rao et al., 2008b) will
likely help to advance our understanding of the molecular basis of M. tuberculosis
persistence (Rao & Li, 2009b).
Over the last century, a variety of control and eradication measures have been implemented
against tuberculosis such as vaccination, aggressive chemotherapy, and public health
surveillance. But tuberculosis still remains a major global health problem to continue to
cause nearly 2 million deaths and 9 million new infection cases per year. Ca. one third of the
world population is infected with M. tuberculosis. A majority of the infected individuals
remain asymptomatic whereas they carry a lifetime risk to develop an active disease i.e., the
latent tuberculosis infection. The state of latency represents the greatest obstacle to eradicate
the tuberculosis disease.
The metabolic requirement of M. tuberculosis in latency is unclear and difficult to study
because the bacilli presumably remain dormant in a granuloma (Pagan-Ramos et al.,
2006). The anti-tuberculosis drugs used today have their maximum effect against the
growing but not the dormant bacilli. The long therapeutic regime required to treat latent
tuberculosis infection is probably explained by the lack of a direct target that is specific to
dormant M. tuberculosis. There are new drug treatment regimes and several new anti-
tuberculosis drugs in a development pipeline that aim to shorten the treatment period
and to overcome multi-drug resistant strains (Murphy & Brown, 2008). Most of these new
drugs are still based on existing classes of antimicrobial compounds to target the
conventional pathways and molecular machinery that are critical for the growth of M.
tuberculosis. These drugs could still be countered by drug resistant strains that emerge
from the non-adherence of a prolonged regime against latent tuberculosis infection. Thus,
the need to discover novel drug targets, especially those against dormant bacteria, is
urgent.
A ‘simple but nonetheless vexing problem’ in target discovery against non-replicating M.
tuberculosis is that many methods rely on a growth-inhibition measurement to assess the
effect of drug treatment (Murphy & Brown, 2008). Rao et al. showed that a protein-
turnover measurement was much more sensitive than a protein relative abundance
measurement alone to uncover protein synthesis activities in M. tuberculosis (Rao et al.,
2008a); those data suggest that protein dynamics analysis with turnover and abundance
measurements could potentially add a valuable alternative to the drug target discovery
problem for non-replicating M. tuberculosis. The sensitive protein turnover analysis could
also be useful to detect and validate drug treatment effects at an early phase, during
which the most relevant drug effect can be isolated from other non-specific cell stress
responses.
www.intechopen.com
3. Label-free quantitation of proteins

There is an increased use of LC/MS instrumentation for proteomics studies at a large scale.
The depth of a proteomic analysis, i.e. the number of protein species that can be precisely
identified and characterized in an experiment, depends on the precision and sensitivity of a
mass spectrometry instrument. The impact of the precision and sensitivity of an LC/MS
instrument on a proteomic study is manifested in many areas of proteomic studies. For
example, in the analysis of intracellular bacterial proteomes, FTMS-based approach clearly
identified more proteins from scarce and complex intracellular bacterial samples compared
to many other proteomic methods.
The precise retrieval of biological information from a large LC/MS dataset critically
depends on algorithms for data interpretation, which remains a current bottleneck in the
rapid advance of proteomics technology (Mortensen et al., 2010). The quantitation of
differentially regulated proteins represents a major type of proteomics application in
biological studies. Protein quantitation with LC/MS data includes three conceptually
different methods i.e., spectral counting, differential stable isotope labelling, and label-free
LC/MS measurements by using extracted ion chromatographic intensities (Mueller et al.,
2008). Due to the increased time and complexity of sample preparation in stable isotope
labelling, cost of labelling reagents, and requirement of higher starting sample amount,
however, researchers are increasingly using label-free proteomics for faster and simpler
protein quantitation (Zhu et al., 2010).
Most of proteomics studies infer proteins with ≥2 identified peptides as reliable protein
identifications and usually disregard proteins with a single-peptide hit as unreliable for
quantitation. This “two-peptide” rule was recently challenged with the evidence that it
reduced protein identifications more in a target database than in a decoy database and thus
increased false discovery rates in protein identification (Gupta & Pevzner, 2009). Indeed, it
was shown that proteins with a single-peptide hit could represent 30% of the proteins
identified with ≥2 MS2 spectrum matches at p <.01 (Li & Roxas, 2009). Because those single-
peptide proteins had ≥2 MS2 spectrum matches (p <.01) in multiple LC/MS analyses under
the same condition, they had an adequate level of statistical confidence to be included for
quantitation.
But the inclusion of single-peptide proteins in a differential quantitative proteomics analysis
raises two issues. The first is that a conventional statistical test such as a t-test can not be
applied toward these single-peptide proteins when the t-test relies on multiple quantified
peptides as replicates to calculate the t-statistic for the protein relative abundance (Li &
Roxas, 2009). The second is that many single-peptide proteins are at a lower abundance and
thus noisy. More stringent thresholds are needed to control the false discovery rate when
these single-peptide proteins are included for the selection of differentially regulated
proteins.
Pavelka et al. applied a power law global error model (PLGEM) and the signal-to-noise ratio
(STN) statistic (Pavelka et al., 2004) to select differentially regulated proteins based on a
spectral counting quantitation method (Pavelka et al., 2008). The PLGEM-STN statistic
utilized a re-sampling approach to estimate the null distribution from replicates of a sample.
After the error model was calculated from a pool of re-sampling statistics that constituted
the null distribution, a set of STN thresholds was applied at a specified confidence level
toward samples with any level of replicates. The PLGEM-STN method is attractive in that it
could be applied toward samples with no replicates if several replicates for one sample are
www.intechopen.com
provided to estimate the null distribution. It is also applicable to proteins with any number
of identified peptides. The PLGEM-STN method, however, has not been demonstrated for
label-free quantitation with extracted ion chromatographic intensities.
In the work described in the following, the PLGEM-STN statistic was applied toward a
LC/MS dataset obtained with a high-resolution mass spectrometer (Roxas & Li, 2009). The
peptide and protein abundances were quantified with a label-free approach based on
extracted ion chromatographic intensities. The false discovery rate was estimated at
different confidence levels of the PLGEM-STN statistics.
The PLGEM-STN statistic alone did not provide a desired level of false discovery rate
control. Insufficient stringency in false discovery rate control was similar to the situation
when a t-test statistic was used alone (Li & Roxas, 2009). With the combination of a t-test
and the rule of minimum number of permuted significant pairings (MPSP), however, the
false discovery rate was significantly reduced in that study.
The combination of MPSP and PLGEM-STN was further tested to control the false discovery
rate. PLGEM-STN does not require that a protein have to have at least two detected
peptides for an assessment of statistical significance. Thus, the combination of MPSP and
PLGEM-STN has the potential to extend the selection of differentially regulated proteins to
those with lower fold-changes and to those with single-peptide hits. Similarly, a fold-change
threshold can be applied toward proteins with any number of detected peptides. With a
control, the statistical significance of a differentially regulated protein can also be assessed
based on a fold-change threshold. Therefore, the combination of MPSP and fold-change
thresholds was also tested and compared with the PLGEM-STN-MPSP approach.
It is important to use the high-resolution FTMS instrument to acquire data for the label-free
quantitation so that proteins with a single peptide hit can be quantified based on peptide
cross reference and extracted ion chromatographic intensity. Proteins with fewer than three
peptide hits are typically difficult to quantify by the spectral count method.
3.1 Purpose of the study

There are two purposes to investigate the combination of the PLGEM-STN statistic or a fold-
change threshold with the MPSP method to identify differentially regulated proteins. One
was to extend the selection of differentially regulated proteins to those that had single-
peptide hits. The other was to select differentially regulated proteins at smaller fold-changes
and at a false discovery rate ≤.05. The approaches to achieve this two-fold purpose were
investigated under a scenario where the number of sample replicates was small. When the
number of sample replicates is small, other typical statistics such as a t-test might not
perform well to provide the necessary specificity in the label-free quantitation of
differentially regulated proteins. Therefore, it was found necessary to insert an additional
measure, such as the MPSP rule to compensate for the lack of sample replicates (Li & Roxas,
2009). Even when more sample replicates are available, the additional use of MPSP might
still further increase the specificity although this possibility will need to be tested.
3.2 Results
With a null distribution built from the labelled control sample to establish thresholds,
different approaches were experimented with to select differentially regulated proteins by
using the combination of MPSP, PLGEM-STN, and fold-change methods. Differentially
regulated proteins were selected from the unlabeled sample pair SP and RP.
www.intechopen.com
The following two subsections of Results are described. Subsection 3.2.1 analyzes the source
of variability in the peptide and protein quantitation processes. Subsection 3.2.2 performs
multi-step extended selection of differentially regulated proteins.
3.2.1 Source of variability

An observed differential abundance of a PCS or protein between samples arose not only
from the difference in biological samples but also from measurement noise that included the
variability from multiple steps that involve LC/MS injection replicates, sample preparation
replicates, biological replicates, or the data processing method. The multi-step experimental
procedures are summarized in Fig. 5.
Internal
Operation pH 5 pH 7 Control
standard
culture (S) culture (R) culture (C)
culture (IS)
flowchart
(I) Cell culturing
S R C
IS
A B C A B C A
(II) Protein extraction

pooled pooled
(III) LC/MS samples

SP RP
(IV) LC/MS injections SP,1 SP,2 RP,1 RP,2
AAPCS,S
PCS,S
AAPCS,R
PCS,R
(V) PCS XIC intensities
AAPCS,cS
PCS,IS
AAPCS,cR
PCS,IS
(VI) Protein XIC AS A

AIS,S
cS
A
AIS,R
cR AR
intensities
For determining false positives
For determining positives
Fig. 5. Experimental outline of the label-free protein quantitation approach to assess the acid
stress response between the unlabeled stressed culture (S) and the unlabeled reference
culture (R) with the [15N]-labeled culture as control (C)
The biological sample model used in the study was the proteome response of an acid
stressed M. smegmatis culture (S) in reference to a neutral pH culture (R) (Roxas & Li, 2009)
(Fig. 5). Both S and R cultures were unlabeled. The proteins from a [15N]-labelled control
culture (C) were used as an internal standard to mix with the proteins from the unlabeled
cultures. Because the proteins from the control culture were analyzed repeatedly with two
other unlabeled samples, the repeated analyses of the labelled control provided replicates to
construct a null distribution.
In the null distribution, there were no true differentially regulated proteins. The null
distribution was thus useful to model the noise in the experiment. The error model was
derived from the null distribution that consists of at least two replicates of sample
preparation. The error model was preferred not to derive from the unlabeled protein
www.intechopen.com
samples SP and RP, because each of these two unlabeled samples had only LC/MS run
replicates but no sample preparation replicates. The use of only LC/MS replicates to model
the noise is likely to underestimate the noise level in the experiment.
The experimental procedures were divided into six stages (I-VI). Briefly, equal amounts of
protein extract from the S culture triplicates were pooled. Equal amounts of protein extract
from the R culture triplicates were also pooled. Into these two pooled unlabeled protein
samples, an equal amount of protein extract from the C culture was added. This resulted in
the two pooled samples i.e., SP and RP. The proteins differentially expressed between the S
and R cultures were determined based on comparison of the abundances of the unlabeled
proteins i.e., AS and AR, between samples SP and RP. For the purpose of false discovery rate
assessment, the abundances of the [15N]-labeled proteins i.e., AcS and AcR, were quantified
and compared between SP and RP in the same way as between AS and AR. The proteins found
differentially expressed between AS and AR were considered positives, because they
reflected the difference between the S and R cultures. The proteins found differentially
expressed between AcS and AcR in the labeled form were false positives, because difference
was not expected from the identical C sample that was run concurrently with two unlabeled
samples in separate runs.
To assist in the assessment of the source of variability in the label-free quantitation of the
LC/MS data, another three samples were used in addition to SP and RP. The three additional
samples were the biological replicates of the S culture sample, namely SA, SB, and SC. SP was
generated by pooling SA, SB, and SC.
The 3rd of the five fractions of an SDS/PAGE gel lane was processed for LC/MS analysis for
the protein samples SA, SB, SC, SP, and RP with duplicate injections for each sample (Li &
Roxas, 2009). The five samples with two LC/MS injections per sample resulted in 10 LC/MS
runs. These 10 LC/MS runs of the 3rd fraction allowed the quantitation of 349 proteins for
the 3rd fraction (Li & Roxas, 2009). Because a protein was quantified in both the unlabeled
form (for culture S or R) and the labelled form (for culture C), there were 20 quantitation
categories for each protein (Table 1). Thus, these 349 proteins and the 20 quantitation
categories formed a 349 x 20 matrix. The 349 x 20 matrix was examined by a clustering
analysis (Eisen et al., 1998). The clustering analysis provides an overview of the correlation
among the protein samples and LC/MS injections, thus reveals the major source of
variability.
From the dendrogram of the 20 quantitation categories shown in Fig. 6, it could be seen that
the distance between each pair of duplicate LC/MS injections was the shortest compared to
those between any other sample pairings. The closest distance of the duplicate LC/MS
injections for a sample indicated that the variability between LC/MS injections was the
smallest, which also indicated that the label-free data analysis methodology (Li & Roxas,
2009) did not introduce a more significant variability.
Unlabeled protein samples from [15N]-labeled protein samples

culture S or R from control culture C
SP RP SA SB SC cSP cRP cSA cSB cSC
Quantitation SP,1 RP,1 SA,1 SB,1 SC,1 cSP,1 cRP,1 cSA,1 cSB,1 cSC,1
category SP,2 RP,2 SA,2 SB,2 SC,2 cSP,2 cRP,2 cSA,2 cSB,2 cSC,2
Table 1. Twenty quantitation categories arising from the duplicate LC/MS analyses of the
3rd gel fraction for samples SA, SB, SC, SP, and RP along with the labeled control in them
www.intechopen.com
Fig. 6. Clustering of the 20 quantitation categories based on the 349 proteins quantified from
the 3rd gel fraction for the five protein samples SP, RP, SA, SB, and SC (Li & Roxas, 2009)
In Fig. 6, it was also apparent that the unlabeled and labelled quantitation categories were
separated into two distinct branches represented by nodes I and II, respectively. The
separation of the unlabeled and labelled quantitation categories into the two distinct clusters
indicated that the difference between cultures C and S or C and R was larger than the
difference between S and R. From the tree branch under node II, it could be seen that the
distance between the unlabeled protein samples SP and RP was larger than the distance
among the S culture replicates i.e., SA, SB, SC. The result indicated that the difference between
cultures S and R exceeded the difference among the S culture replicates, suggesting that the
variability in biological sample replicates was less than the actual difference between the
biological samples treated with different conditions.
Therefore, the clustering result in Fig. 6 indicated that the variability increased in the order
of LC/MS injections < sample preparation replicates (under node I) ~ biological replicates
(under node III) < biological samples (between nodes III and IV). Because these differences
were evaluated based on the proteomic quantitation data, a variability observed among
biological replicates also included the variability introduced during sample preparation for
LC/MS analysis. The similarity between the variability observed among the sample
preparation replicates and the variability observed among the biological replicates
suggested that the variability among biological replicates was not larger than the variability
among sample preparation replicates.
www.intechopen.com
3.2.2 Extended selection of differentially regulated proteins

This subsection describes the multiple steps leading to the extended selection of
differentially regulated proteins from all quantified proteins including those with only a
single-peptide hit. The proteins with a single-peptide hit represent 1/3 of the identified
proteins (Li & Roxas, 2009). Therefore, it is desirable to have a procedure to select regulated
proteins from all of the proteins including those with a single-peptide hit to maximize the
potential of the global protein expression profiling.
Establishing a null distribution
Based on the evaluation with the clustering analysis (Fig. 6), the variability among sample
preparation replicates appeared to be comparable with that among biological replicates.
Samples SP and RP represented the average of triplicate biological replicates for cultures S
and R respectively, because each of them was the pooled sample of three biological
replicates. The pooling process further reduced the biological variability between SP and RP.
Therefore, the [15N]-labelled control sample replicates (Table 1) were adequate to represent a
null distribution in which there was no differentially regulated protein.
100000 0.6
10000
0.5
1000
0.4
A S or A cS (arb. unit)
100
rSTD
0.3
10
0.2
1
0.1
0.1
0.01 0
0.01 0.1 1 10 100 1000 10000 100000
A R or A cR (arb. unit)
Fig. 7. APRO scatter plots, local variability, and thresholds for selecting differentially
regulated proteins. The blue dots represent the APRO scatter plot of AS vs. AR corresponding
to the unlabeled proteins in sample SP vs RP. AS is the average of AS,1 and AS,2. AR is the
average of AR,1 and AR,2. The red dots represent the APRO scatter plot of AcS vs. AcR
corresponding to the labeled proteins in control sample replicate cSP vs cRP. AcS is the
average of AcS,1 and AcS,2. AcR is the average of AcR,1 and AcR,2. AS,1, AS,2, AR,1, AR,2, AcS,1, AcS,2,
AcR,1, and AcR,2 were the APRO values for the eight quantitation categories defined in Table 1.
To evaluate the local noise of APRO measurement, the relative standard deviation (rSTD) for
each protein was calculated from its four unlabeled APRO values AS,1, AS,2, AR,1, and AR,2 (the
blue trace) or its four labeled APRO values AcS,1, AcS,2, AcR,1, and AcR,2 (the pink trace). The
rSTD- APRO traces were smoothed with a 100-point moving box. The grey straight lines
indicated a 3-fold (solid line) and a 2-fold (dashed line) change threshold. The solid red and
green curves represent the fold-change thresholds established with the PLGEM-STN
statistics based on the local variance in the null distribution (the pink-dot scatter plot)
www.intechopen.com
The null distribution afforded an estimation of measurement noise. The determined

measurement noise was then used to estimate the false discovery rate for the selected
differentially regulated proteins between samples SP and RP. The null distribution provided
a reference for setting thresholds to maximize the selection of differentially regulated
proteins (positives) while minimizing false positives. In Fig. 7, such a null distribution was
illustrated with the scatter plot represented by the pink dots.
To investigate the relationship between measurement variability and protein abundance
APRO, relative standard deviation (rSTD) was plotted against the mean APRO value for each
protein in the unlabeled protein samples (blue trace) or the labelled control protein samples
(pink trace) (Fig. 7). The rSTD-APRO trace in pink reflected the local noise of the null
distribution. The local noise of the null distribution was mainly due to the variability that
was introduced during sample preparation (Fig. 6). The rSTD-APRO trace in pink clearly
suggested that the APRO measurement noise had a reciprocal dependence on the APRO
amplitude. The rSTD-APRO trace in blue reflected both sample preparation variability and
biological sample difference between cultures S and R. Thus, the blue trace had higher rSTD
values than the pink trace throughout the APRO range.
Modelling local noise in the null distribution
Because of the reciprocal dependence of APRO rSTD on the APRO value, a universal 3-fold-
change cut-off missed some positives at higher APRO values where a <3-fold change was
already significantly different from the local noise. Missed positives at higher APRO values
could be observed in Fig. 7 by examining the spread of the two scatter plots in the high APRO
ranges. At APRO >1000, the rSTD was a few times smaller than that at APRO of ~100. From the
figure, it could be seen that it was possible to detect a < 2-fold change for the proteins with
APRO >1000. To the contrary, at APRO <10, a 3-fold change threshold was not sufficient to
eliminate many false positives. Therefore, a criterion adaptive to the dependence of APRO
noise on APRO values would uncover more differentially regulated proteins. This extended
selection of differentially regulated proteins could be achieved by penalizing proteins with
higher APRO values less than proteins with lower APRO values. Such an adaptive criterion,
however, requires a systematic modelling of the noise to establish the thresholds according
to local variability (Pavelka et al., 2004).
In this study, the PLGEM-STN statistic was experimented with for the selection of
differentially regulated proteins quantified with label-free proteomics based on protein
extracted ion chromatographic intensities. There were two reasons for the choice of the
PLGEM-STN method.
First, the PLGEM-STN method allowed statistical analyses of the proteins quantified with a
single PCS because the PLGEM-STN statistic did not rely on multiple PCSs of a protein like
a t-test (Li & Roxas, 2009). Because single-peptide proteins constituted a third of the
quantified proteins, being able to quantify these single-peptide proteins was important to
maximize the potential value of the data. Second, the PLGEM-STN method took into
account the dependence of APRO noise on APRO levels. A threshold adjustable to the local
dependence of APRO noise on APRO levels allowed the selection of differentially regulated
proteins with a smaller fold-change threshold at a higher APRO level.
Therefore, the PLGEM-STN method potentially could select more differentially regulated
proteins by applying a smaller fold-change threshold in the higher APRO range where the
variability was smaller. This possibility was tested as described below.
www.intechopen.com
Selecting differentially regulated proteins with PLGEM-STN
PLGEM-STN PLGEM-STN
FP, P, PLGEM-STN-
confidence Permuted sample pairings
and FDR MPSP
level Average
I II III IV
FP (cSP/cRP) 31 68 22 46 42 13
0.01 P (SP/RP) 141 155 134 148 145 101
FDR 0.22 0.44 0.16 0.31 0.29 0.13
FP (cSP/cRP) 6 15 3 9 8 2
0.002 P (SP/RP) 47 50 46 51 49 44
FDR 0.13 0.30 0.07 0.18 0.16 0.05
Table 2. Numbers of differentially regulated proteins selected with PLGEM-STN alone or in

combination with MPSP
120
PLGEM-STN
100
PLGEM-STN-MPSP C.L.=0.01
True positives (SP/RP)
80 C.L.=0.01
60 C.L.=0.003
C.L.=0.003
40
20
C.L.=0.0001
0
0 5 10 15 20 25 30 35 40 45 50
False positives (cSP/cRP)
Fig. 8. Receiver operating characteristic analysis of the PLGEM-STN approach with (red
curve) or without (blue curve) the combination with MPSP. Positives are the differentially
regulated proteins selected from the comparison of protein abundances between samples SP
and RP. False positives are the differentially regulated proteins selected from the comparison
of proteins abundances between samples cSP and cRP. True positives are estimated by
subtracting false positives from the positives. For each approach, i.e. PLGEM-STN-MPSP or
PLGEM-STN, 37 data points at different confidence levels (C.L.) are plotted in this figure,
starting from C.L.=0.0001 up to C.L.=0.01. The increment is 0.001 between C.L. of 0.0001 and
0.003 (30 data points). Between C.L. of 0.003 and 0.01, the increment is 0.01 (7 data points)
Table 2 shows the result of the PLGEM-STN analysis for the unlabeled samples SP and RP
and the labelled sample replicates cSP and cRP. cSP and cRP were the labelled control samples
analyzed concurrently with SP and RP, respectively. The differentially regulated proteins
www.intechopen.com
found between SP and RP were positives (P), and those found between cSP and cRP were
false positives (FP). Because each protein sample was analyzed with duplicate LC/MS
injections, permutation of the four LC/MS injections for a sample pair resulted in four
permuted sample pairings (Li & Roxas, 2009). These four permuted sample pairings were
numbered as I to IV in Table 2. In each column for a permuted sample pairing in Table 3, the
numbers of false positives and positives and the false discovery rate (FDR) were listed. The
false positives were determined as the differentially regulated proteins for the sample pair
cSP/cRP. The positives were determined as the differentially expressed proteins for the
sample pair SP/RP.
In Table 2, the positives and false positives were selected with the PLGEM-STN method at
the confidence level of 0.01 and 0.002, respectively. The results indicate that the numbers of
positives or false positives were not the same among the four permuted sample pairings. To
estimate an average false discovery rate, the numbers of positives and false positives were
respectively averaged among the four permuted sample pairings. The false discovery rate
was then calculated as the ratio of the average number of false positives divided by the
average number of positives. The false discovery rate was determined at two different
PLGEM-STN confidence levels (Table 2). With a receiver operating characteristic analysis,
the PLGEM-STN approach is examined over a broader confidence level range (Fig. 8) and
will be compared with another approach that is to be described below.
Incorporating the MPSP rule
Initially, the PLGEM-STN approach was carried out by comparing the duplicate LC/MS
injections from the two samples R and S without permutation pairings. But the false
discovery rate stayed high unless the sensitivity was severely compromised to reduce the
false discovery rate. For example, at a confidence level of 0.0001, only 16 differentially
regulated proteins were selected at 6% false discovery rate (data not shown). With all of the
permutation pairs and a combination of PLGEM-STN and MPSP, 44 differentially regulated
proteins were selected at a false discovery rate of 5% (Table 2). Therefore, a high sensitivity
is achieved to uncover differentially regulated proteins by utilizing all possible permutation
pairs with a combination of PLGEM-STN and MPSP.
Because of the variable numbers of positives and false positives among the four permuted
sample pairings, it was necessary to determine a consensus list of differentially regulated
proteins from the four permuted sample pairings. Previously, the rule of MPSP was applied
to determine the consensus list of differentially regulated proteins from four permuted
sample pairings (Li & Roxas, 2009). The MPSP rule required that only those proteins that
were found differentially regulated in a certain number of permuted sample pairings were
counted as positives (for SP/RP) or false positives (for cSP/cRP). When a sample pair such as
SP/RP had no sample replicates but had duplicate LC/MS injections, MPSP was found to be
optimum at four (Li & Roxas, 2009). Setting MPSP at four meant that a differentially
regulated protein had to be found differentially regulated in all of the four permuted sample
pairings.
Selecting differentially regulated proteins with a PLGEM-STN-MPSP approach
The application of the MPSP rule towards the PLGEM-STN results decreased both false
positives and positives (Table 2). But the false discovery rate was also decreased relative to
that when only the PLGEM-STN statistic was applied. From Table 2, it could be seen that the
number of true positives, which was estimated from the difference between the numbers of
www.intechopen.com
positives and false positives, remained about the same. Therefore, the combination of the
MPSP rule with the PLEGM-STN method reduced the false discovery rate by 2-3 times
without compromising the sensitivity.
As summarized in Fig. 8, the receiver operating characteristic analysis clearly shows that the
PLGEM-STN-MPSP approach significantly reduces false positives to improve the specificity
without significantly affecting the sensitivity. Compared to the use of the PLGEM-STN
statistic alone, the combination of PLGEM-STN and MPSP performs better in controlling
false discovery rates without compromising the sensitivity to select differentially regulated
proteins.
Selecting differentially regulated proteins with a fold-change-MPSP approach
Fold-change
Fold FP, P, Fold-change-
Permuted sample pairings
change and FDR Average MPSP
I II III IV
FP (cSP/cRP) 68 77 118 45 77 22
2 P (SP/RP) 171 154 186 147 165 104
FDR 0.40 0.50 0.63 0.31 0.47 0.21
FP (cSP/cRP) 30 33 47 20 33 9
3 P (SP/RP) 66 70 85 60 70 42
FDR 0.45 0.47 0.55 0.33 0.47 0.21
FP (cSP/cRP) 17 24 32 10 21 1
4 P (SP/RP) 42 50 53 35 45 26
FDR 0.40 0.48 0.60 0.29 0.47 0.04
Table 3. Number of differentially regulated proteins selected with a fold-change threshold

alone or in combination with MPSP
The use of MPSP with fold-change criteria was also examined (Table 3). With fold-change
criteria alone, the false discovery rate did not drop below 46% at 2- to 4-fold changes. With
the combination of MPSP and the fold-change criteria, the false discovery rate was reduced
from 46% to 21% at 2- and 3-fold changes. At a 4-fold change, the false discovery rate was
reduced to 4%. Compared to the combination of PLGEM-STN and MPSP, however, the
combination of fold-change and MPSP reduced more true positives at the similar false
discovery rate of 4-5%. Therefore, the application of MPSP with the fold-change criteria
reduced sensitivity. The reduced sensitivity was due to the increase in the fold-change
threshold.
With the 4-fold-change-MPSP and the PLGEM-STN-MPSP approaches, 26 and 44 proteins
were respectively selected as differentially regulated at a false discovery rate of 4% or 5%
(Tables 2 and 3). Among these 26 and 44 proteins, there were 55 unique proteins (Li, 2010b).
These 55 unique proteins included all of the 20 high-confidence differentially regulated
proteins identified previously with an empirical fold-change and abundance level cut-off
approach (Roxas & Li, 2009).
www.intechopen.com
100
B
No. of PCSs
10
A 1
0 10 20 30 40 50 60
100
C
4-fold- PLGEM- 10
Fold change
change- Common
ST N-MPSP 1
MPSP 0.1
15
0.01
11 29 0 10 20 30 40 50 60
100000
D
10000
1000
APRO
100
10
0
0 10 20 30 40 50 60
Protein number
Fig. 9. Comparison of the 26 and 44 differentially regulated proteins respectively selected by

the 4-fold-change-MPSP and PLGEM-STN-MPSP approaches at a 5% false discovery rate.
(A) Overlap of the two sets of differentially regulated proteins. Panels B-D show the
distributions of (B) the number of detected PCSs, (C) the fold changes, and (D) the
abundances of the quantified proteins. The blue square, the purple diamond, and the tan
triangle markers represent the differentially regulated proteins selected by 4-fold-change-
MPSP only, by both, and by PLGEM-STN-MPSP respectively. The protein number was from
1 to 55 on the x-axis representing the 55 unique proteins ranked according to their APRO in
each of the three groups (blue, purple, or tan)
Comparing the PLGEM-STN-MPSP and fold-change-MPSP approaches
Only 15 proteins were common between the two sets of differentially regulated proteins
selected with the 4-fold-change-MPSP and the PLGEM-STN-MPSP approaches (Fig. 9A).
The 4-fold-change-MPSP approach selected more single-PCS proteins than the PLGEM-
STN-MPSP approach (Fig. 9B). The PLGEM-STN-MPSP approach selected proteins with a
fold-change as low as 1.8-fold (Fig. 9C). However, these differentially regulated proteins
selected with PLGEM-STN-MPSP had a protein abundance higher than most of the
differentially regulated proteins selected with the 4-fold-change-MPSP approach (Fig. 9D).
Thus, the two approaches complement each other and could be used simultaneously.
3.3 Discussions
3.3.1 Motivation of the extensive label-free quantitative proteomics analysis
Despite the relative complexity in label-free proteomics data analysis and the demand of
more stringently controlled LC/MS experimental conditions, there are strong motivations
stemming from biological and experimental perspectives to use the label-free approach, as
discussed below.
As shown in Fig. 6, the unlabeled and labelled quantitation categories are separated into two
distinct clusters. One includes the quantitation categories from the labelled control culture C
www.intechopen.com
(under node I). The other includes the quantitation categories from the two unlabeled
cultures S and R (under node II). Thus, there was a larger difference between the labelled
(C) and either of the two unlabeled samples (S or R) than between the two unlabeled
cultures (S and R). The number of differentially regulated proteins between the labelled
culture and either of the unlabeled culture was about three times as many as that between
the two unlabeled cultures. Compared to the difference between the two unlabeled cultures,
the difference between the labelled culture and either of the unlabeled cultures was larger.
This larger difference was probably because the labelled culture was cultured in a synthetic
minimal medium while the two unlabeled cultures were grown in a commercial 7H9 broth
that was richer in ingredients. Another factor was that the acidic growth condition was a
relatively mild stress so that not many proteins were differentially regulated.
The apparent difference in proteome profile for cells cultured in different media is actually a
strong motivation for this study. In microbiological works, it is not always convenient to
make a [15N]-labelled medium with complex ingredients required to cultivate bacteria under
more physiologically relevant conditions. Even some of the stable-isotope-labelled media
are technically feasible to make, they often bear a costly price tag. For microbiological
works, one might not want to be restricted by the type of medium that can be used because
of the stable isotope labelling limitation. For example, some mycobacteria are difficult to
cultivate on simple synthetic media and prefer complex media. Thus, unlabeled media are
always convenient choices if the down-stream proteomic analysis is established to proceed
with the quantitation.
For such reasons, the focus of this study was on the comparison of protein expression
profiles between the two unlabeled cultures S and R. The labelled control culture C was
used as an internal standard to estimate false discovery rates.
3.3.2 The use of a [15N]-labeled internal standard for null distribution construction
The label-free quantitation scheme presented in this study incorporated a labelled internal
control to provide replicates for noise modelling without a requirement of other unlabeled
sample replicates. The inclusion of a labelled internal control facilitates the estimation and
control of false discovery rates.
Internal standards are commonly used to improve reliability of quantitative proteomics such
as to aid in removing outlier data and to detect fluctuation in instrument performance
(Mirzaei et al., 2009). Compared to other synthetic peptide internal standards(Mirzaei et al.,
2009; Winter et al., 2010), the [15N]-labelled control culture C provides more comprehensive
peptide internal standards. For most of the peptides, the extracted ion chromatographic
intensities can be matched among the three protein samples originated from the two
unlabeled (S and R) cultures and the labelled (C) culture. The C protein sample was mixed
and run together with either S or R protein sample, so that the reliability of the internal
standards was improved.
To construct the null distribution for the error model in PLGEM-STN, it would be ideal to
have the labelled internal standard identical to an unlabeled sample in protein composition.
As mentioned above, however, that requirement could restrict the culturing conditions
available for biological experiments. Thus, it is acceptable and sometimes necessary to use a
labelled protein mixture sample as internal standard, even though the internal standard
sample might be somewhat different from the unlabeled samples in protein abundance
profiles.
www.intechopen.com
Nevertheless, the null distribution is only utilized to establish the relation between the
signal-to-noise ratio and the peptide abundance in the PLGEM-STN method. There is no
requirement of direct one-to-one comparison between the labelled and unlabeled version of
a protein during this process. Therefore, the difference in proteome composition between
the labelled internal standard sample C and the two unlabeled samples S and R is not
expected to affect the modelling parameters derived from the null distribution constructed
from the labelled C sample.
One could choose to run multiple replicates of an unlabeled sample and use the replicates to
construct the null distribution. That approach would require more LC/MS runs as discussed
previously (Li & Roxas, 2009).
3.3.3 Label-free data analyses and selection of differentially regulated proteins

The LC/MS data used in this work was acquired with a high-resolution mass spectrometer
that resolved peptide peaks from a complex sample mixture to allow the determination of
the extracted ion chromatographic intensities of peptides and proteins. Repeated LC/MS
injections showed the highest reproducibility among several other types of replicates,
indicating that the major variability of the label-free quantitation did not lie within the
LC/MS separation and the data analysis method. Rather, sample preparation replicates
represented a major source of the variability. With a labelled control sample to run
concurrently with each of the unlabeled samples, replicates for the labelled control sample
were obtained. The control sample replicates provided data to model the noise in the label-
free quantitation with extracted ion chromatographic intensities.
We performed a two-step normalization procedure in which the information about the
abundance of a peptide or protein in a sample was preserved (Li, 2010b). The preservation
of the information about the abundance of a peptide or protein in the samples is critical for
performing the PLGEM-STN analysis. In addition, because protein extracted ion
chromatographic intensity was represented by the sum of the PCS extracted ion
chromatographic intensities belonging to that protein, the summation weighed the low-
intensity PCSs less than the high-intensity PCSs. Such a summation of PCS extracted ion
chromatographic intensities probably suppressed noise from lower-intensity PCSs. When a
protein abundance ratio is calculated as the average of PCS abundance ratios without
weighing, the noise from a lower-intensity PCS would be amplified. We have avoided this
potential issue by summing the PCS intensities to represent protein abundances before
calculating protein abundance ratios.
Single-peptide proteins made up about 35% of the quantified proteins (Li, 2010b). Selection
of differentially regulated proteins from these single-peptide proteins required a
significance assessment method that did not rely on multiple-peptide detection to calculate
a statistic about the confidence of a protein differential abundance. The use of a statistic that
does not rely on the detection of multiple peptides is especially useful when the sample
replicates were too low to use a typical statistical test such as a t-test. PLGEM-STN was a
method that fits this criterion.
However, PLGEM-STN alone was not strict enough to control the false discovery rate
without further diminishing the number of positives (Fig. 8). The lack of stringency by using
the PLGEM-STN method alone was similar to that by using the t-test alone (Li & Roxas,
2009). In that prior study, the lack of specificity with a t-test alone was overcome by
introducing the rule MPSP. The MPSP rule simply requires that a protein be selected as
differentially regulated only when it was repeatedly found so in certain number of
www.intechopen.com
permuted sample pairings. The MPSP rule was introduced to deal with datasets with small
replicates where other more sophisticated statistical tests could not be applied (Li & Roxas,
2009). Although the MPSP rule was originally used in combination with a t-test statistic and
a fold-change threshold, this study shows that it can be used in combination with other
types of statistical tests such as the PLGEM-STN method (Fig. 8).
The combination of the MPSP rule allowed the selection of differentially regulated proteins
at a false discovery rate <5%, which would have been impossible for a fold-change method
(Table 3). The MPSP rule significantly reduced false positives while keeping the number of
true positives relatively constant, thus effectively improving the statistical confidence of the
selected differentially regulated proteins by lowering the false discovery rate (Table 3). The
results suggest that MPSP is a rule that can be used in combination with different types of
statistics to select differentially regulated proteins.
The label-free quantitation simplified cell culturing and sample preparation. Another useful
aspect of the label-free quantitation is that peptide cross-reference could be used to increase
the number of proteins quantified in all of the samples run under the same condition
(Andreev et al., 2007). Lipton et al. introduced the concept of accurate mass and elution time
peptide tag for global protein quantitation using high resolution mass spectrometry (Lipton
et al., 2002). One advantage of this method over using the spectral counting method is that
the large number of identifications that occur in a LC/MS injection can be used as the basis
for improved quantitation of another LC/MS injection (Andreev et al., 2007; Fang et al.,
2006; Strittmatter et al., 2003). The accurate mass and elution time peptide tag approach uses
the extracted ion chromatographic intensities as the quantitative measurement of peptides
and proteins. The linear response of peptide extracted ion chromatographic intensities to
protein quantities was demonstrated (Hochleitner et al., 2005; Lundrigan et al., 1997; Wang
et al., 2006). This method was thus used to improve the comparability of proteins quantified
between samples, among LC/MS injections, and for different isotopic forms of a protein
(Rao et al., 2008a). The quantitation of 349 proteins from a single gel fraction for several
samples clearly demonstrated the power of the peptide cross-reference feature in extracted
ion chromatographic intensity-based label-free quantitative proteomics (Li & Roxas, 2009).
One drawback of extracted ion chromatographic intensity-based label-free quantitative
proteomics is that the success of an analysis critically depends upon the reproducibility of
LC/MS runs that have to be maintained across multiple samples. The reproducibility of
LC/MS runs across multiple samples is a prerequisite to reliable peptide cross reference
(Andreev et al., 2007). With the advancement in LC/MS instrumentation and the availability
of improved LC/MS chromatogram alignment methods (Fischer et al., 2006; Podwojski et
al., 2009), the reproducibility of LC/MS runs is unlikely to remain an obstacle for the
increasing use of label-free quantitative proteomics.
3.4 Summary
A label-free quantitative proteomics scheme was demonstrated to select differentially
regulated proteins with single-peptide hits and <2-fold changes at a 5% false discovery rate.
The scheme incorporated a labeled internal control into multiple unlabeled samples to
facilitate error modeling when there were no replicates for the unlabeled samples. The error
modeling allowed the use of the PLGEM-STN statistic to facilitate the selection of
differentially regulated proteins with single-peptide hits. The PLGEM-STN statistic also
facilitated the selection of differentially regulated proteins at different fold-change
www.intechopen.com
thresholds according to the local abundance level of the proteins. While the PLGEM-STN
statistic uncovered more differentially regulated proteins at higher abundance with smaller
fold-changes, the PLGEM error modeling of local variance versus abundance over-penalized
the proteins with lower abundance. With a constant fold-change threshold, however,
differentially regulated proteins with higher abundance were overlooked. Thus, the results
from this study showed that the PLGEM-STN and a constant fold-change threshold were
complementary to each other and could be used simultaneously. But, neither the PLGEM-
STN nor the 4-fold-change criterion alone was stringent enough for selecting differentially
regulated proteins at a 5% false discovery rate.
MPSP was introduced and shown to be a rule that could decrease false discovery rates when
being used in combination with the PLGEM-STN statistic or the 4-fold-change threshold.
The MPSP rule played a critical role in extending the selection of differentially regulated
proteins to those with single-peptide hit or with a lower fold-change in label-free proteomics
when sample replicates were limited. Although the approaches were demonstrated for a
representative replicate-limited scenario, they potentially can also be applicable to a
situation where more sample replicates are available.
4. Conclusion
This chapter presents several examples of proteomic studies utilizing the LTQ-FTMS
instrument. The instrument typically provides a high resolution (> 500,000), a large mass
range (one order magnitude in a single scan), and high mass accuracy (< 2 ppm) in many
experiments.
The protein turnover studies benefit greatly from the high resolution in a large mass range,
which allows the resolution of isotopomers and isotopologue profiles that could possibly be
generated by almost any degree and type of stable isotope labeling. The well resolved full-
range spectra simplify the data processing steps and improve data quality. Because a
spectral count method reports the MS2 events mostly on monoisotopic peaks, it is usually
not suitable for protein turnover study. Therefore, a high-resolution mass spectrometer such
as an FTMS instrument is highly desired for protein turnover studies.
The high mass accuracy is critical in the implementation of a label-free proteomics approach.
The label-free quantitative proteomics relies on a cross reference of peptides between runs
and an integration of extracted ion chromatographic intensities. The peptide cross reference
allows the quantitation of a peptide in a run in which the peptide is not identified but is
identified in another run. It is based on the accurate mass and reproducible liquid
chromatographic elution time of the peptide in multiple runs. The cross reference allows
more peptides to be quantified. It also improves the comparability of samples because the
same peptides, thus the same proteins, are quantified in multiple samples.
The label-free quantitation is useful not only for protein differential expression studies, but
also for proteome dynamics studies where the abundances of newly synthesized proteins
are to be quantified separate from those of the total proteins and the old proteins.
Thus, the label-free quantitation approach is universally applicable to different proteomic
studies. The assessment of statistical significance in such a quantitation approach is
challenging especially for the proteins with few peptides identified and when the replicates
are limited. A combination of the MPSP rule with a global error modeling method and a
fold-change threshold proves to be effective in controlling the false discovery rates in
protein differential analysis.
www.intechopen.com
5. References
Andreev, V. P.; Li, L.; Cao, L., et al. (2007). A new algorithm using cross-assignment for
label-free quantitation with LC-LTQ-FT MS. J Proteome Res, 6, 6, 2186-94.
Bernlohr, R. W. (1967). Changes in amino acid permeation during sporulation. J Bacteriol, 93,
3, 1031-44.
Bernlohr, R. W. (1972). 18 Oxygen probes of protein turnover, amino acid transport, and
protein synthesis in Bacillus licheniformis. J Biol Chem, 247, 15, 4893-9.
Beynon, R. J. (2005). The dynamics of the proteome: strategies for measuring protein
turnover on a proteome-wide scale. Brief Funct Genomic Proteomic, 3, 4, 382-90.
Borek, E.; Ponticorvo, L. & Rittenberg, D. (1958). Protein Turnover in Micro-Organisms. Proc
Natl Acad Sci U S A, 44, 5, 369-74.
Cargile, B. J.; Bundy, J. L.; Grunden, A. M. & Stephenson, J. L., Jr. (2004).
Synthesis/degradation ratio mass spectrometry for measuring relative dynamic
protein turnover. Anal Chem, 76, 1, 86-97.
Cho, S. H.; Goodlett, D. & Franzblau, S. (2006). ICAT-based comparative proteomic analysis
of non-replicating persistent Mycobacterium tuberculosis. Tuberculosis (Edinb), 86,
6, 445-60.
Doherty, M. K. & Beynon, R. J. (2006). Protein turnover on the scale of the proteome. Expert
Rev Proteomics, 3, 1, 97-110.
Eisen, M. B.; Spellman, P. T.; Brown, P. O. & Botstein, D. (1998). Cluster analysis and display
of genome-wide expression patterns. Proc Natl Acad Sci U S A, 95, 25, 14863-8.
Fang, R.; Elias, D. A.; Monroe, M. E., et al. (2006). Differential label-free quantitative
proteomic analysis of Shewanella oneidensis cultured under aerobic and suboxic
conditions by accurate mass and time tag approach. Mol Cell Proteomics, 5, 4, 714-25.
Fischer, B.; Grossmann, J.; Roth, V., et al. (2006). Semi-supervised LC/MS alignment for
differential proteomics. Bioinformatics, 22, 14, e132-40.
Gerner, C.; Vejda, S.; Gelbmann, D., et al. (2002). Concomitant determination of absolute
values of cellular protein amounts, synthesis rates, and turnover rates by
quantitative proteome profiling. Mol Cell Proteomics, 1, 7, 528-37.
Goldberg, A. L. & Dice, J. F. (1974). Intracellular protein degradation in mammalian and
bacterial cells. Annu Rev Biochem, 43, 0, 835-69.
Gupta, N. & Pevzner, P. A. (2009). False discovery rates of protein identifications: a strike
against the two-peptide rule. J Proteome Res, 8, 9, 4173-81.
Hochleitner, E. O.; Kastner, B.; Frohlich, T., et al. (2005). Protein stoichiometry of a
multiprotein complex, the human spliceosomal U1 small nuclear ribonucleoprotein:
absolute quantification using isotope-coded tags and mass spectrometry. J Biol
Chem, 280, 4, 2536-42.
Larrabee, K. L.; Phillips, J. O.; Williams, G. J. & Larrabee, A. R. (1980). The relative rates of
protein synthesis and degradation in a growing culture of Escherichia coli. J Biol
Chem, 255, 9, 4125-30.
Li, Q. (2010a). Advances in protein turnover analysis at the global level and biological
insights. Mass Spectrom Rev, 29, 5, 717-36.
Li, Q. (2010b). Assigning significance in label-free quantitative proteomics to include single-
peptide-hit proteins with low replicates. Int J Proteomics, 2010, 731582, 15 pages.
www.intechopen.com
Li, Q. & Roxas, B. A. (2009). An assessment of false discovery rates and statistical
significance in label-free quantitative proteomics with combined filters. BMC
Bioinformatics, 10, 43.
Lipton, M. S.; Pasa-Tolic, L.; Anderson, G. A., et al. (2002). Global analysis of the
Deinococcus radiodurans proteome by using accurate mass tags. Proc Natl Acad Sci
U S A, 99, 17, 11049-54.
Lundrigan, M. D.; Arceneaux, J. E.; Zhu, W. & Byers, B. R. (1997). Enhanced hydrogen
peroxide sensitivity and altered stress protein expression in iron-starved
Mycobacterium smegmatis. Biometals, 10, 3, 215-25.
Malen, H.; Berven, F. S.; Fladmark, K. E. & Wiker, H. G. (2007). Comprehensive analysis of
exported proteins from Mycobacterium tuberculosis H37Rv. Proteomics, 7, 10, 1702-
18.
Mandelstam, J. (1958). Turnover of protein in growing and non-growing populations of
Escherichia coli. Biochem J, 69, 1, 110-9.
Mirzaei, H.; Brusniak, M. Y.; Mueller, L. N., et al. (2009). Halogenated peptides as internal
standards (H-PINS): introduction of an MS-based internal standard set for liquid
chromatography-mass spectrometry. Mol Cell Proteomics, 8, 8, 1934-46.
Mortensen, P.; Gouw, J. W.; Olsen, J. V., et al. (2010). MSQuant, an open source platform for
mass spectrometry-based quantitative proteomics. J Proteome Res, 9, 1, 393-403.
Mueller, L. N.; Brusniak, M. Y.; Mani, D. R. & Aebersold, R. (2008). An assessment of
software solutions for the analysis of mass spectrometry based quantitative
proteomics data. J Proteome Res, 7, 1, 51-61.
Murphy, D. J. & Brown, J. R. (2008). Novel drug target strategies against Mycobacterium
tuberculosis. Curr Opin Microbiol, 11, 5, 422-7.
Pagan-Ramos, E.; Master, S. S.; Pritchett, C. L., et al. (2006). Molecular and physiological
effects of mycobacterial oxyR inactivation. J Bacteriol, 188, 7, 2674-80.
Pavelka, N.; Fournier, M. L.; Swanson, S. K., et al. (2008). Statistical similarities between
transcriptomics and quantitative shotgun proteomics data. Mol Cell Proteomics, 7, 4,
631-44.
Pavelka, N.; Pelizzola, M.; Vizzardelli, C., et al. (2004). A power law global error model for
the identification of differentially expressed genes in microarray data. BMC
Bioinformatics, 5, 203.
Podwojski, K.; Fritsch, A.; Chamrad, D. C., et al. (2009). Retention time alignment algorithms
for LC/MS data must consider non-linear shifts. Bioinformatics, 25, 6, 758-64.
Pratt, J. M.; Petty, J.; Riba-Garcia, I., et al. (2002). Dynamics of protein turnover, a missing
dimension in proteomics. Mol Cell Proteomics, 1, 8, 579-91.
Rao, P. K. & Li, Q. (2009a). Principal Component Analysis of Proteome Dynamics in Iron-
Starved Mycobacterium Tuberculosis. J Proteomics Bioinform, 2, 19-31.
Rao, P. K. & Li, Q. (2009b). Protein turnover in mycobacterial proteomics. Molecules, 14, 9,
3237-58.
Rao, P. K.; Rodriguez, G. M.; Smith, I. & Li, Q. (2008a). Protein dynamics in iron-starved
Mycobacterium tuberculosis revealed by turnover and abundance measurement
using hybrid-linear ion trap-Fourier transform mass spectrometry. Anal Chem, 80,
18, 6860-9.
www.intechopen.com
Rao, P. K.; Roxas, B. A. & Li, Q. (2008b). Determination of global protein turnover in stressed
mycobacterium cells using hybrid-linear ion trap-fourier transform mass
spectrometry. Anal Chem, 80, 2, 396-406.
Rosenkrands, I.; Slayden, R. A.; Crawford, J., et al. (2002). Hypoxic response of
Mycobacterium tuberculosis studied by metabolic labeling and proteome analysis
of cellular and extracellular proteins. J Bacteriol, 184, 13, 3485-91.
Roxas, B. A. & Li, Q. (2009). Acid stress response of a mycobacterial proteome: insight from
a gene ontology analysis. Int J Clin Exp Med, 2, 309-328.
Spudich, J. A. & Kornberg, A. (1968). Biochemical studies of bacterial sporulation and
germaination. VII. Protein turnover during sporulation of Bacillus subtilis. J Biol
Chem, 243, 17, 4600-5.
Strittmatter, E. F.; Ferguson, P. L.; Tang, K. & Smith, R. D. (2003). Proteome analyses using
accurate mass and elution time peptide tags with capillary LC time-of-flight mass
spectrometry. J Am Soc Mass Spectrom, 14, 9, 980-91.
Vogt, J. A.; Hunzinger, C.; Schroer, K., et al. (2005). Determination of fractional synthesis
rates of mouse hepatic proteins via metabolic 13C-labeling, MALDI-TOF MS and
analysis of relative isotopologue abundances using average masses. Anal Chem, 77,
7, 2034-42.
Wang, G.; Wu, W. W.; Zeng, W.; Chou, C. L. & Shen, R. F. (2006). Label-free protein
quantification using LC-coupled ion trap or FT mass spectrometry: Reproducibility,
linearity, and application with complex proteomes. J Proteome Res, 5, 5, 1214-23.
Wilkinson, K. D. (2005). The discovery of ubiquitin-dependent proteolysis. Proc Natl Acad Sci
U S A, 102, 43, 15280-2.
Winter, D.; Seidler, J.; Kugelstadt, D., et al. (2010). Minimally permutated peptide analogs as
internal standards for relative and absolute quantification of peptides and proteins.
Proteomics, 10, 7, 1510-4.
Zhu, W.; Smith, J. W. & Huang, C. M. (2010). Mass spectrometry-based label-free
quantitative proteomics. J Biomed Biotechnol, 2010, 840518, 6 pages.
www.intechopen.com
4
Application of Fourier Transform Mid-Infrared

Spectroscopy (FTIR) for Research
into Biomass Feed-Stocks
Gordon G. Allison
Institute of Biological, Environmental and Rural sciences, Aberystwyth University
United Kingdom
1. Introduction
Non-food biomass crops e.g. switchgrass (Panicum virgatum L.), Miscanthus x giganteus, and
short-rotation coppice poplar (Poplus spp.) and willow (Salix spp.) offer a sustainable source
of energy and platform chemicals (Sims et al., 2006). The majority of the energy stored in the
crop biomass is in the cell wall which constitutes the largest fraction of lignocellulosic
biomass. The three polymers that constitute the bulk of plant cell wall (cellulose,
hemicellulose and lignin) rank amongst the most abundant biopolymers on the planet and
their proportional concentrations range generally between 40 – 50%, 10 – 40% and 5 – 30% of
biomass by weight respectively (McKendry, 2002). The absolute and relative concentrations
of the components of the cell wall have a great influence on biomass quality i.e. its
suitability for conversion to heat, power and chemical products. However, because biomass
can be utilised by a number of conversion routes with differing feedstock demands
measures of feed-stock quality are often quite specific to how the material is to be used. For
example, biomass can be processed thermochemically. These routes include combustion or
co- combustion with coal to generate heat and electricity (Allison et al., 2010). Alternatively,
biomass can be converted by fast pyrolysis to bio-char, which is receiving much attention as
a soil improver and a means of sequestering carbon from the atmosphere into the soil (Laird,
2008; Woolf et al., 2010), and bio-oil, a liquid fuel (Bridgwater, 2003; Mohan et al., 2006).
Biomass can also be gasified to produce a combustible gas which has application for the
generation of heat and power, and for the chemical synthesis of liquid transport fuels and
industrial chemicals (Ptasinski et al., 2007). These thermochemical processes demand feed-
stocks with low moisture content and high energy density, often equating with high levels
of the poly aromatic polymer, lignin.
In contrast, non-thermochemical processes e.g. the production of bioethanol and industrial
platform chemicals by biological conversion processes are often inhibited by high levels of
lignin. High concentrations of lignin in the feedstock necessitate harsh chemical and heat
pre-treatments of the biomass prior to enzymic saccharification. This increases energy inputs
and often damages the polysachharide components of the cell wall giving rise to inhibitory
products (Carroll and Somerville, 2009; Chang, 2007; Fahmi et al., 2007; Fahmi et al., 2008;
Grabber, 2005). There is therefore considerable pressure to optimise feedstock composition
and at present the most feasible way to achieve this at a commercial scale is by breeding
www.intechopen.com
improvement (Clifton-Brown et al., 2008), although agronomic practice may also influence
composition (Hodgson et al., 2010). Crop breeding places huge demands on the analyst in
terms of large sample numbers and there is a demand for methods that can cope with high
rates of sample throughput whilst being of low unit cost.
Biomass is traditionally assessed in terms of cell wall fibre content measured by direct, or as
they are often called, gravimetric methods. The different types of fibre are isolated by
successively harsh chemical treatments which selectively and sequentially remove the
different classes of structural carbohydrates until only lignin remains. The proportion of
each fibre fraction in the biomass is quantified by sample weight change. Measurements of
this type are not entirely quantitative as treatments are not wholly selective, and it is often
impossible to directly compare cell wall parameters measured by different methods
(Hatfield et al., 1999). In addition, these procedures are time consuming, costly and of low
through put (Giger-Reverdin, 1995). The acetyl bromide method for the quantification of
lignin is an indirect method that has gained in popularity over recent years. The method
was first published by Johnson et al. in 1961 (Johnson et al., 1961) and a modified version of
the method has been used to analyse lignin in a wide range of species (Fukushima and
Dehority, 2000) from relatively small samples of tissue. Lignin is dissolved from purified cell
wall material by extraction into a solution of acetyl bromide in concentrated acetic acid at
50˚C, reacted with hydroxylamine and quantified by absorbance change at 280 nm.
Quantification requires reference to a standard curve that is set up using known amounts of
standard lignin extracted from similar plant material with acetyl bromide (Fukushima and
Dehority, 2000) or acidic dioxane (Fukushima and Hatfield, 2001). The concentrations of
lignin detected in samples by this method are comparable with those obtained using the
widely accepted gravimetric Klason method (Hatfield and Fukushima, 2005) and the
method has the advantage that it is more readily worked into a high-throughput scheme
(Foster et al., 2010). Adapting direct and indirect methods for high rates of sample
throughput is often complex and most likely requires investment in expensive robotised
laboratory equipment that may be beyond the resources of many groups (Foster et al., 2010).
Consequently many researchers have turned to less expensive technology for new high-
throughput methods.
Infrared (IR) spectrometry offers researchers and breeders an alternative approach that is
robust and rapid. The majority of the carbon based molecules in animals and plants are
highly active in the IR. At its simplest mid-IR spectroscopy is a useful analytical approach in
its own right that provides structural information on samples of pure compounds.
Generally, the approach taken is to correlate IR spectra with analytical data obtained from
the same samples using multivariate regression methods. This approach although
seemingly complex overcomes possible nonlinear relationships between absorption and
concentration that are encountered in the IR. Such deviations from the Beer-Lambert law
may be caused by spectral shifts due to hydrogen bonding, co-variance with other
components in the sample and poor design in older instruments (Hsu, 1997). The regression
models produced can be then applied to predict the concentrations of the cell wall
components in new samples. Properly executed, this approach has been shown to be rapid
and robust and therefore reduces the requirement for standard chemical analysis.
Two types of IR spectroscopy have found application for the measurement of chemical
composition in biomass: mid-infrared spectroscopy and near infrared reflectance
spectroscopy (NIRS). NIRS has a longer history of being used to predict chemical
composition in bulk plant samples as until recently the method was more applicable to bulk
www.intechopen.com
Application of Fourier Transform Mid-Infrared Spectroscopy (FTIR)
for Research into Biomass Feed-Stocks 73
plant samples. Mid-IR spectroscopy in contrast to NIRS informs on fundamental molecular

vibrations, rather than on harmonic and overtone absorptions, and for this reason gives a
better insight into the molecular bonds present in the sample. Over recent years the technique
has been revolutionised by the development of Fourier transform instruments (FTIR) and by
improvements in methods for sample presentation and many groups are choosing analysis
by FTIR in preference to NIRS. Several reports have shown that the combination of FTIR and
multivariate analysis is highly effective. For example, spectra of wood can be used to
discriminate between tree species (Huang et al., 2008). Recently, we reported on using FTIR
to predict the concentration of lignin and hydroxycinnamic acids (Allison et al., 2009),
nitrogen and alkali index (Allison et al., 2009) in samples of grasses by partial least squares
regression. In addition to the analysis of bulk samples, FTIR has long had application in the
field of microscopy. For example, FTIR microscopy has been used to identify and characterise
cell wall mutants and transgenic plants altered in cell wall biosynthetic genes (Chen et al.,
1998; McCann et al., 2007; Mouille et al., 2003; Stewart et al., 1997).
Although it is not intended to discuss FTIR microscopy in depth in this chapter because it is
not a method of choice for high-throughput screening it deserves some brief discussion and
comparison with allied methods. For the study of biological samples FTIR microscopy is
often used as a transmission technique where the IR beam passes through the sample.
Special objectives are available (see Section 4) which work through contact with the sample
but this results in damage to most biological specimens. FTIR microscopy is best suited to
the high resolution mapping of chemical composition at the level of the cell. In this type of
study high rates of sample through-put and the prediction of composition in bulk samples
are not usually experimental objectives. Earlier FTIR microscope instruments took a long
time to acquire spectra because the beam passing through the sample filled only part of the
detector. New developments in FTIR microscope detector design have decreased spectral
acquisition time substantially and many groups have taken advantage of focal plane array
(FPA) detectors. These were originally developed for military use and they allow the
simultaneous acquisition of many spectra. Over the last decade, FPA detectors have grown
from arrays of 8x8 elements to 16 x 16 and now 64 x 64 or even 128 x 128 element detectors
are available for groups with sufficient funds. FPA-FTIR microscopes allow very rapid
chemical mapping over large areas but the technique is not without fundamental problems.
FTIR microscopy is incompatible with water, which prevents the analysis of living samples,
and the analyst must be judicious in the use of chemical fixatives to preserve the structure of
sectioned materials as the absorption peaks from these substances may mask important
features in the spectral data. Furthermore, spatial resolution in the IR is limited to 2 – 5 µm
by wavelength and whilst this does not prevent resolution of most plant cells, which range
generally between 10 – 50 µm in diameter, cellular components and structures may be
substantially smaller than this size and therefore below the limit of resolution. Raman
microscopy, which will not be discussed in any depth here, offers a solution to many of
these problems. For many years Raman microscopy was considered an expensive and
unusual technique despite offering a spatial resolution of better than 1 µm (Schmidt et al.,
2010). In addition, Raman microscopy is not affected by water thus making possible the
analysis of living samples and samples where cellular structure has been preserved
cryogenically. The relatively low signal strength of Raman emission were problematic and
often led to poor signal and lengthy analysis times. This matter has been largely overcome
by improvements in instrument design and new sensitive Raman techniques e.g. surface
enhanced Raman (Knauer et al., 2010) may result in Raman soon becoming the best way to
www.intechopen.com
study and understand the structural organisation of the cell wall at the cellular level. The
resolution and flexibility of Raman imaging has been demonstrated recently by several
groups studying the ultra-structure and composition of cell wall in tree species and corn
stover (Agarwal, 2006; Gierlinger and Schwanninger, 2006; Sun et al., 2010).
In this review we discuss the principles underlying IR spectroscopy, the developments in
instrumentation and sample presentation methods that have led to Fourier transform
instruments becoming much more compatible with high through-put analysis, and we
present data showing how FTIR is being used to assist breeding improvement work in
Aberystwyth.
2. The interaction of matter and IR energy

To understand IR spectroscopy it is necessary to have some insight into how energy
interacts with the molecules present in living cells. The bonds in molecules are not rigid nor
of fixed length and both, bond length and their angle to each other vary about a mean
position and vibrate with specific frequencies. Covalent molecules absorb energy in the IR
when the frequency of the energy correlates exactly with the vibrational frequency of a
chemical bond within the molecule. This absorbance of energy by the molecule causes an
increase in the amplitude, but not the frequency of the bond’s vibration about its mean
centre. By definition, absorption of IR energy can only occur when it causes a change to the
bond’s electric dipole moment, i.e. the bond’s polarity, or the separation of positive and
negative electrical charges comprising the covalent bond. Simple homonuclear molecules
e.g. N2 and O2 cannot absorb in the IR because there is no dipole moment to change. In
contrast, it is common for many bonds in covalent heteronuclear molecules to exhibit some
degree of polarity e.g. H2O and CO2 and most organic compounds absorb strongly in the IR
with characteristic absorption frequencies.
The visible, IR and ultraviolet parts of the electromagnetic (EM) spectrum are defined by
frequency (υ; the number of wave-cycles passing through a point in one second) measured
in Hertz (Hz) and wave length (λ; the length of one complete wave-cycle). The mid-IR
region stretches between 3 x 10–4 and 3 x 10–3 cm and is of greatest practical use to the
organic chemist because this range corresponds to the fundamental frequencies of molecular
vibrations. Frequency and wavelength are related to each other by equation 1, where c is the
velocity of light in a vacuum, and has the value 2.997 x 108 m s-1:
c=λυ (1)
Experiments in the early part of the 20thCentury demonstrated that when interacting with
matter EM radiation is best described as discrete packets of energy called quanta. The
energy (E) of a quantum is directly proportional to its frequency and is calculated by the
Bohr equation (equation 2) where h is the Planck constant (h = 6.626 x 10-34 Joules-second).
Substitution of equation 1 into equation 2 shows that energy decreases with increasing wave
length.
E = hυ (2)
In IR spectroscopy wavelength is often expressed as its reciprocal, wave number (cm-1),
which equates to the number of waves per centimetre. In the mid-IR wave numbers span
from 4000 – 400 cm-1; this is a more convenient notation and has the advantage of being
proportional with energy.
www.intechopen.com
The infrared spectrum of a sample of a pure compound shows the position of absorptions
within the IR spectrum and therefore indicates the functional groups in the molecule. The
spectrum can be presented in terms of transmission, where the y axis ranges from 0 – 100%
transmission over a range of wave numbers, or in terms of absorption, in which percentage
transmission (%T) is converted to absorption (A) by the simple relationship shown in
equation 4.
A = log10 (100/%T) (3)

For qualitative analysis absorption has the advantage of being proportional to the
concentration of the analyte, which can be calculated by equation 4, the Beer-Lambert Law
in which c is the molar concentration of analyte, ε the molar extinction coefficient (L mol-1
cm-1) and l the light path through the sample (cm).
A = εcl (4)
Even simple compounds have relatively complex IR spectra showing many absorption
bands. This is because there may be many bonds within an individual molecule that are able
to absorb in the mid-IR. Bonds within a molecule can vibrate in two ways corresponding to
the movements of atoms sharing a bond. Atoms can move relative to each other causing the
bond to vary in length, this causes bond stretching, or one atom can move out of plane
relative to the other, causing bond bending. For reasons not explained here the maximum
number of absorptions in the mid-IR that might be expected from a pure compound
comprised of N atoms can be calculated by equation 5 for linear molecules such as CO2 and
by equation 6 for non-linear molecules such as H2O.
Linear = 3 N – 5 (5)
Non-linear = 3 N – 6 (6)
Not all possible vibrational modes are active in the IR, for example CO2 would be expected
to have 3 N – 5 absorption bands in its spectrum, in reality it has two as one of the possible
stretching modes is symmetrical and not IR active, furthermore the two bending modes are
degenerate and show as one combined band. Other reasons why fewer than the theoretical
number of IR bands are seen in the spectrum include the absorption not being in the 4000–
400 cm–1 range; an absorption being too weak to be observed and absorptions being too
close to each other to be resolved on the instrument. The vibration frequency of the
f
absorption in wave numbers ( v ) is proportional with the force constant or bond stiffness (k,
dyne cm-1) and the masses of the atoms (in grams) sharing the bond (equation 7).
k(m1 + m2)
v=
1
(7)
2 πc (m1m2)
3. The Fourier transform infrared spectrometer

Modern mid-IR spectrophotometers are almost exclusively Fourier transform instruments as
this design offers greater sensitivity and considerably faster scan speeds compared with the
older and now largely obsolete dispersive instruments. The main difference between the
www.intechopen.com
two designs is the manner in which the IR beam is produced which is passed through the
sample. Dispersive instruments are based on a monochromator; this device splits a
polychromatic beam produced by the source into a beam of very narrow wave number. As
the instrument scans over a wave number range the monochromator shifts the wave
number of the beam through the scan range. This of course means that the time required for
each scan is dependant on how quickly the monochromator can be made to work and at any
one point in the scan the beam transmitted through the sample is only a small portion of the
sources output. In contrast, a beam covering the entire wave number range of the source is
passed through the sample in a Fourier transform instrument. Therefore scan time and beam
intensity are considerably faster and brighter respectively. The heart of a Fourier transform
instrument is an interferometer. Most commonly this is a Michelson interferometer (Figure
1). In this device IR light from a Nernst or Globar source is split at a beam splitter into two
beams of equal intensity, one beam is reflected onto a stationary mirror whist the other is
reflected onto a moving mirror. The two beams are recombined to form the transmitted
beam at the beam splitter, and this beam is at 90˚to the input beam. The moving mirror
produces a varying optical path difference between the two beams resulting in constructive
and destructive interference when they are combined. A helium neon laser beam is included
in the FTIR spectrophotometer to provide a reference beam of known wave number that can
be used to measure precisely the displacement of the moving mirror. This laser is not shown
in Figure 1 for the purpose of simplicity. Two types of detector are commonly used in the
mid-IR: The deuterium triglycine sulphate (DTGS) detector works at room temperature and
has the advantage of great stability and ease of use. For more demanding work increased
sensitivity of the mercury cadmium telluride (MCT) may be necessary but this type of
detector requires cooling to near liquid nitrogen temperatures in order to work. The
resulting interferogram (Figure 2) contains the source’s frequency information modulated in
a time domain as a function of the moving mirror’s displacement.
The entire spectrum is therefore measured simultaneously in the interferogram and FTIR
instruments offer considerable speed benefits compared to dispersive instruments which
mechanically scan from one wave number to another. An interferogram can be obtained in
only a few seconds compared to many minutes so allowing many interferograms to be
collected in a comparatively short time. These are averaged allowing great improved signal
to noise ratios and increased sensitivity. The absorbance spectrum of the sample is produced
from the interferogram by Fourier-transformation; the process requires that the analyst also
records an interferogram where no sample is present in the beam path. The ratio of the
resulting absorbance spectra corresponds to the absorbance spectrum of the sample alone.
Key developments in technology were necessary for Fourier transform based instruments to
become practical. Firstly, the development of Fast Fourier transformation (FFT) by Cooley
and Tukey (Cooley and Tukey, 1965) offered an algorithm for Fourier transformation that
was several orders of magnitude simpler that previous calculation routes whilst only
slightly less accurate. Without FFT the lengthy computation would have required unfeasibly
powerful and expensive computers. Even the much shorter computation underlying FFT
however required the technology to wait for development of affordable desk-top computers
to supply the necessary computing power before FTIR based instruments could become
practical. In addition, not only does the computer handle the FFT allowing generation of
spectra it also allows the analyst the ability to manipulate spectra for integration, base-line
correction and averaging, and to analyse the spectra using multivariate approaches such as
principal components analysis (PCA) and multivariate regression.
www.intechopen.com
Fixed Mirror
Incident Beam Moving Mirror
IR Source
Transmitted Beam
Sample
Detector
Fig. 1. Schematic of a Michelson interferometer
Fig. 2. Typical interferogram (courtesy of Analytical Spectroscopy by R. P. W. Scott,

Essential Infromation for the Analytical chemist http://www.analyticalspectroscopy.net/)
4. The development of effective sample analysis techniques

Samples of plant material are generally prepared by oven or freeze-drying to eliminate
water which has high absorption in the IR; the dry samples are then finely ground.
Generally, the finer that samples can be ground the better but often the analyst finds that for
fibrous samples there is trade off between particle size and sample through-put and it is
often necessary to determine empirically the maximum size of particle that permits robust
analysis. Furthermore, the milled sample must be truly representative of the whole sample
and chemical differences between particles of different sizes must be accounted for. It is a
good idea to have access to a number of mills differing in design as each offers a
combination of through-put and effectiveness that is highly sample specific. The preparation
of samples by grinding is often the rate limiting step in IR analysis.
www.intechopen.com
Traditionally, analysis by IR spectroscopy was highly time-consuming. In this approach

small amounts of ground sample are ground with a larger quantity of an IR transparent salt
such as KBr in a pestle and mortar. The mixture is then pressed hydraulically into
transparent discs with between 5 – 10 tonnes of pressure (Kacurakova et al., 2000; Xu et al.,
2007). At all stages it is essential that the amount of water in the disk is kept to a minimum
and spectra must be taken quickly as the discs are hydroscopic and soon become opaque
due to the absorption of water from the atmosphere. This process is time-consuming but
allows spectra of a high quality to be obtained. It’s use has been reported by several groups
engaged on the structural study of various cell wall components including pectin and
hemicellulose; (Kacurakova et al., 2000; Sun and Tomkinson, 2002; Xu et al., 2006; Xu et al.,
2007), cellulose (Liu et al., 2006), lignin (Gosselink et al., 2004; Sun et al., 2002; Sun et al.,
2002; Xu et al., 2007), pyrolysis char (Hu et al., 2008) and lignin (Faix, 1991; Jung and
Himmelsbach, 1989; Monties, 1989; Nakanishi and Kawakami, 1991; Stewart et al., 1997).
Alternatively, samples can be ground with an oily mulling agent (typically Nujol) and
smeared onto salt disks prior to spectral acquisition.
In recent years a much more rapid method of analysing powdered samples has become
widely available and this has made the use of FTIR for larger sample numbers more
practicable (Allison et al., 2009; Gosselink et al., 2004). The method relies on crushing a
powdered sample against a flattened face of a trapezoid crystal of diamond, germanium or
zinc selenide mounted in a rugged plate. The accessory is most often placed in the
spectrophotometers sample bay and the IR beam is passed through the crystal by a series of
mirrors. Because the IR beam hits the walls of the crystal at a very shallow angle it is
reflected within the crystal and exits the far side where it passes to the detector. The
technique is known as attenuated total reflectance (ATR). Most ATR accessories are of the
single bounce design but for samples of only low absorption it is possible to purchase multi-
bounce ATR accessories. Spectra are obtained from the sample only when they are placed in
direct contact with the crystal and it is important that the particle size of the sample is
sufficiently small and uniform to allow spectral acquisition to be reproducible. Because the
IR beam only penetrates a few microns from the crystal surface it is possible to directly
analyse aqueous samples. All methods require for spectra to be collected in a two step
process; first, a blank spectrum is collected with no sample in place, next a sample spectrum
is collected and the spectrum of the sample is obtained by calculating the ratio of the two
measurements. Using ATR an operator using a manual spectrophotometer can comfortably
collect spectra from approximately 200 samples comfortably in one day.
5. Spectral interpretation and multivariate analysis

The handling and interpretation of the IR spectral data is as much a part of compositional
analysis as sample preparation and spectral acquisition and is discussed in this chapter in
order for the reader to have a clear insight into the entire process. From the information
provided in Section 2 it would appear relatively straight forward for the analyst to assign
absorption peaks in the spectrum of a pure sample to particular chemical moieties or
molecular structures. In practice, IR spectra are more complex than might be expected.
Often additional absorption bands are present in the spectrum due to overtones or
combinations of the fundamental vibrations. In addition, the spectra of larger molecules
have an additional level complexity caused by the coupling of vibrations over part or the
entire molecule. When two oscillating bonds share a common atom, the vibrations of the
two bonds are coupled and as one bond contracts, the other bond contracts or expands, as in
www.intechopen.com
asymmetrical and symmetrical stretching. When coupling occurs, the spectral position of
bands due to particular bonds become shifted rather than superimposed as might be
expected. These shifted vibrations are known as skeletal vibrations and give information on
the entire molecule rather than on specific functional groups within the molecule. However,
despite these complexities spectral interpretation can generally be achieved without
resorting to first principles by remembering the principles underlying equation 7, and by
reference to either a text book on the subject e.g. Stuart (2005) or to one of the collections of
IR spectral fingerprints e.g. Movasaghi et al. (2008).
Spectra from samples of complex mixtures such as plant cell wall preparations or ground
samples of dried lignocellulosic crops, present an additional level of difficulty to interpret as
it is next to impossible to deconvolute the superimposed spectra of the individual
components in the sample. Prior knowledge of the likely composition of the sample often
allows the identification of spectral features known to be associated with likely components
e.g. in plant material the presence of bands at 1590 cm-1 and 1610 cm-1 are correlated with
lignin (Monties, 1989), and if no knowledge of sample composition exists IR spectra
provides clues as to the kind of compounds which may be present. The analyst can thus
make an informed decision on the best methods to use for initial chemical analysis of the
sample.
Whilst useful for qualitative analysis, these approaches are not sufficiently robust to allow
the prediction of composition in spectra from large numbers of samples as they are too
easily influenced by co-variance between chemical components in the samples and by
random differences in the spectral data. The goal of using IR spectra to predict chemical
composition in complex samples has been made possible by the adoption of multivariate
approaches to data analysis and the availability of powerful and affordable desk top
computers to handle the data processing and file storage. The simplest and possibly the
most commonly used multivariate approach is PCA. This was first described in theoretical
terms by Karl Pearson in 1901 (Pearson, 1901) and later made into a practical reality by
Hotelling in 1933 (Hotelling, 1933). PCA allows the variance in the spectra data at hundreds
of wave numbers to be condensed into a much smaller number of new principal
components which explain most of the variance in the original data. Because there are fewer
variables the data can be more easily explored graphically or statistically for correlation with
experimental, environmental and sample effects. PCA works best with highly correlated
data and spectral data is highly correlated. Any single data point in a spectrum is highly
influenced by the value of its immediate neighbours. PCA is of great value for the biologist
and analytical chemist hoping to understand how differences between samples in
geographic location, growth year, species or some experimental treatment such as the
addition of fertiliser, have an effect on chemical composition. This is illustrated by Figure 3A
which shows a plot of the first and second principal components obtained from 194 mid-IR
spectra of reed canary grass and switchgrass samples collected at the end of two consecutive
growth years. The two components plotted in this figure represent the majority of the
variance in the data set. The data points are coded according to species and it is quite
apparent that spectral differences exist between the two grass species and these differences
are accounted for primarily by the variance explained by principal component 2. The
separation of the data points is however only partial showing that chemical differences
between samples from the two species occur in only a proportion of cases. The analyst
would usually wish to explore this further and establish whether this was the result of
processing and sample preparation e.g. differences in sample moisture content, or whether
www.intechopen.com
these differences reflected or factors influencing the growth and development of the plants
during their growth in the field. Re-plotting these data but coding the data points by growth
year would result in an identical distribution of data points but show no separation of the
data according to growth year by these components (data not shown).
PC Axis 2 (19.9.)% var) 2 A

1
-1
-2
-3
-4
-5
-6
-7
-8 -6 -4 -2 0 2 4 6 8
PC Axis 1 (33.9% var)
2.5
B
NC Predicted (% dry wt.)
1.5
0.5
0
0 0.5 1 1.5 2 2.5 3
NC Measured (% dry wt.)

Fig. 3. (A) PCA of 194 mid-infrared attenuated total reflectance absorbance spectrums of
samples of reed canary over the spectral range of 500- 4000 cm-1 (□ = switchgrass, ● = reed
canary grass). (B) Scatter plots showing measured vs. PLS model predicted values for
nitrogen content in the training (●) and test (+) spectral data. (Data from Allison et al., 2009)
PCA is obviously a highly useful tool for evaluating differences and similarities between
samples. In addition, the analyst may identify the wave numbers where these differences
occur. This information is contained in the PCA loadings, a matrix of data produced during
the PCA process that shows the influence of each of the original variables i.e. absorbance’s at
groups of wave numbers, on each of the new components. Analysis of the loadings for
principal component 2 would in this case identify the discriminatory wave numbers
between the two species and perhaps make it possible to identify the class of chemical that
www.intechopen.com
differed most between the two species. This very powerful way to look at chemical data will
also reveal not just increases or decreases in given compounds but changes in the
concentrations of several chemical components simultaneously.
PCA and other methods of this kind e.g. canonical correlation, cluster analysis and
multidimensional scaling are therefore excellent tools for exploring structure relationships
within the spectral data but they cannot be used to determine the concentration of
components within the sample mixture. For that it is necessary to employ a multivariate
regression approach. Many approaches have been developed over the years e.g. see Otto
(2007), and currently many analytical chemists rely on regression by partial least squares
(PLS). Other methods such as principal components regression and multivariate linear
regression offer subtle differences in capability but the way in which they are employed is
similar. Multivariate regression requires the analyst to provide a training set of data from
samples that have been also been analysed using standard chemical methods. The variance
in these training data is arranged by PLS into new orthogonal components (latent variables)
but unlike PCA these new latent variables not only capture variance but also achieve
correlation with the analytical data. To ensure that the correlation is genuine and based on
the chemical components being measured rather than to some selective fitting of noise the
process is controlled by cross validation. The mathematics underlying these calculations is
complex and many analytical scientists rely on software from specialist suppliers or
instrument manufacturers, although it is also possible to find free-ware packages on the
internet. The PLS regression is judged for predictive accuracy using an independent test set of
spectral data from chemically analysed samples that were excluded from the PLS regression
fitting process. This gives the user confidence that predictions made with the model will be
accurate. Confidence in predictions made using the model is obtained by chemically
analysing a small percentage of all samples and checking the agreement between te real and
predicted values, and by monitoring the variance presented by new samples to ensure that
they are adequately explained by the existing model. Figure 3B shows a plot of predicted vs.
measured nitrogen content for the 194 samples of reed canary grass and switchgrass.
Theoretical perfect fit is displayed as a dotted line in the figure.
6. Example of process
In this example using previously unpublished data we show how FTIR spectra can be used
to develop predictive PLS models for several combustion parameters in samples of coal.
This may seem to be a curious choice for illustration but the procedure by which the data
are analysed and models developed are essentially identical to those employed for
developing models to predict aspects of cell wall composition in less ancient biomass. The
purpose of this study was to explore the possibility of using FTIR for the process monitoring
of coal and biomass destined for co-firing. Two similar studies were published in 2009
which used NIRS (Kim et al., 2009) and FTIR (Geng et al., 2009) and these provide a useful
comparison for these findings. Sixty nine samples of powdered coal were provided by
E.ON. It was known that the coal samples were from a number of locations but details of
their origins were not disclosed. The coal samples had been subjected to proximate analysis
and values were supplied of the mass fractions of moisture, volatiles, ash and sulphur, see
Table 1. With only minor exceptions the data were normally distributed. Triplicate IR
spectra were taken from each sample from 600 cm-1 to 4000 cm-1 using a Golden Gate ATR
accessory fitted with a single bounce diamond crystal (Specac, U.K.) placed in the sample
compartment of an Equinox 55 FTIR spectrophotometer (Bruker Optik GmbH, Germany).
www.intechopen.com
Each spectrum represented an average of 32 scans at a resolution of 4 cm-1. Spectra were

averaged, smoothed and derivatised to the first Savitsy- Golay derivative using a window of
25 wave numbers and a second order polynomial fit before being normalised to unit
variance. All spectral processing was performed using MatLab (version 7.10; Mathworks)
and the PLS toolbox (version 6.0; Eigenvector Research, Inc.).
Xa Xm Xs Xv
Mean 10.81 11.93 1.124 31.01
Minimum 2.30 4.00 0.120 24.10
Maximum 41.10 25.90 2.390 35.60
SD 4.70 3.72 0.648 2.56
Table 1. Values, range and standard deviation (SD) of Q, calorific value (corrected for ash
and moisture content); Xa, ash mass fraction; Xcl, chlorine mass fraction; Xm, moisture mass
fraction; Xs, sulphur mass fraction and Xv, volatile mass fraction for 69 samples of
powdered coal. Data supplied by E.ON
Figure 4A shows the complete set of 207 spectra taken from the 69 samples before
averaging, derivatisation and normalisation. Derivatisation is a commonly employed
practice in spectroscopy as it improves the separation of non-resolved peaks. In NIRS it is
not unusual to derivatise spectra to the 3rd or 4th derivative. In FTIR however, the peaks are
better resolved and derivatisation beyond the 1st or 2nd derivative typically offers no
advantage. The spectra shown in Figure 4B are much more tightly grouped and it is
apparent that this regime of pre-processing has improved the noise in the data considerably.
Other processes which might have been applied include scatter correction, baseline
correction offsets and general least squares regression which serves to decrease noise
between chemically similar data points. The 69 spectra were divided into two groups; one
group of 59 spectra that were used to develop the partial least squares regression model and
an independent test set of 10 spectra which were excluded from the model but which were
used to assess model predictive accuracy. The PLS modes were developed using SIMPLS
algorithms (de Jong, 1993) and the PLS toolbox software. Models were cross validated using
venetian blinds cross validation protocol (7 data splits) and developed to an optimal number
of latent variables to ensure that the models were based on variance explaining the
parameter of interest rather than on noise in the data. In each case the best number of latent
variables to include was indicated by a minimal value of the root mean square error of cross
validation (RMSECV). This measure of error can be interpreted as the standard deviation of
the unexplained variance in the cross validated regression, and has the useful property of
being in the same units as the response variable. This measure of error is a much better
indication of model fit than root mean square error of calibration (RMSEC) as the latter does
not indicate when the model is over-fitted to the data. The models were tested for predictive
accuracy by measuring the root mean square error of prediction (RMSEP) using the
independent data test set. For RMSECV, n relates to the number of spectra in the training
data set, yi is an observed value obtained by chemical analysis and yˆ l a value predicted from
the cross validated regression model. RMSEC can be calculated using the same equation by
substituting the values of yˆ l with the predicted values from the non-cross validated.
Similarly, RMSEP is calculated by using values from the independent data test set.
www.intechopen.com
RMSECV =
∑ i =1 ( yi − yˆ l )2
n
(8)
n
0.7
0.6
A
0.5
Absorbance
0.4
0.3
0.2
0.1
0
1000 1500 2000 2500 3000 3500 4000
Wave Number cm -1
8
BB
Rate of Change of Absorbance
-2
-4
-6
-8
1000 1500 2000 2500 3000 3500 4000
Wave Number cm -1
Fig. 4. (A) Triplicate ATR-FTIR spectra (207 spectra) of 69 the samples of powdered coal. (B)
First derivative averaged spectra after Savitsky-Golay smoothing and normalisation (69 spectra)
Plots of regression fits for these four models are shown in Figure 5 and the parameters of the
four models in Table 2. All of the models are sufficiently good to be used for prediction of
these parameters in unknown samples. Ideally, the variance in spectra from the unknown
samples would be compared with those used to develop the model in question. This ensures
that the model is capable of making an accurate prediction. Some of the samples detected as
dissimilar would be analysed chemically and these data incorporated into a new model better
able to make predictions. In all circumstances a proportion of the unknown samples, perhaps 5
– 10% of the total number should be analysed chemically to validate model predictions.
www.intechopen.com
30 2.5
A C
25
2
Predicted Xm
Predicted Xs
20
1.5
15
1
10
0.5
5
0 0
0 5 10 15 20 25 30
0 0.5 1 1.5 2 2.5
Measured Xm Measured Xs
36 18
34
B 16 D
14
32
12
Predicted Xv
Predicted Xa
30 10
28 8
6
26
4
24
2
22 0
24 26 28 30 32 34 36 2 4 6 8 10 12 14 16 18 20
Measured Xv Measured Xa
Fig. 5. PLS regression fits (red line) for the spectral data from the coal samples for A,
moisture; B, volatiles; C, sulphur and D, ash mass fractions. Black and red symbols indicate
samples in the training and test data sets respectively. The green line denotes a 1:1
regression fit
Model Calibration Validation

LV % Var % Var R2 RMSEC R 2cv RMSECV PredBias R 2 RMSEP
p
x y
Xm 9 78.7 99.5 0.995 0.228 0.781 1.802 -0.08 0.888 1.181
Xv 7 76.3 97.4 0.974 0.435 0.714 1.490 -0.35 0.869 0.782
Xs 5 62.3 95.9 0.935 0.137 0.827 0.286 0.05 0.887 0.290
Xa 5 69.5 93.2 0.932 0.785 0.762 1.504 0.02 0.928 0.831
Table 2. Summary of the PLS models developed to measure moisture (Xm), volatiles (Xv),
sulphur (Xs) and ash (Xa) mass fractions. Details are shown of the number of latent variables
in each model (LV), the percentage variation included from the spectral (% Var x) and non-
spectral data (% Var y), predictive bias (Pred Bias) and model goodness of fit is shown for
correlation, cross validation and prediction as root mean square error (RMSEC, RMSECV
and RMSEP) and as determination coefficients (R2, R 2cv and R p2 )
www.intechopen.com
7. Conclusions
Mid-IR spectrophotometry has become a potentially useful predictive tool for the plant
breeder, biologist and engineer as it offers high rates of sample through-put, low unit cost
and robust and accurate analysis. Key to this approach has been the development of modern
Fourier transform spectrophotometers, the advent of modern personal computers and the
availability of chemometrics software. Whilst many still prefer to use near-IR spectral
analysis for the prediction of compositional parameters mid-IR gives the analyst additional
information on the molecules within the sample which may be more difficult to discern
using NIRS.
8. Acknowledgements
This work was funded by the Engineering and Physical Sciences Research council (EPSRC)
and carried out as part of the Supergen Consortium in Biomass, Biofuels and Energy Crops
(GR/S28204). The author wishes to thank Catherine Morris for making FTIR spectral
analysis of the coal samples and Hugh Burnham-Slipper of E.ON for supplying the samples.
9. References
Agarwal, U.P. (2006). Raman imaging to investigate ultrastructure and composition of plant
cell walls: distribution of lignin and cellulose in black spruce wood (Picea mariana).
Planta, 224, 1141-1153.
Allison, G.G., Morris, C., Hodgson, E., Jones, J., Kubacki, M., Barraclough, T., Yates, N.,
Shield, I., Bridgwater, A.V., Donnison, I.S. (2009). Measurement of key
compositional parameters in two species of energy grass by Fourier transform
infrared spectroscopy. Bioresource Technology, 100, 2428-2433.
Allison, G.G., Robbins, M.P., Carli, J., Clifton-Brown, J., Donnison, I. (2010). Designing
biomass crops with improved calorific content and attributes for burning: a UK
perspective. in: Plant Biotechnology for Sustainable Production of Energy and Co-
Products, P. Mascia, J. Schefrran, J.M. Widholm (Eds.). Springer, pp. 25-56, IBSN
978-3-642-13439-5.
Allison, G.G., Thain, S.C., Morris, P., Morris, C., Hawkins, S., Hauck, B., Barraclough, T.,
Yates, N., Shield, I., Bridgwater, A.V., Donnison, I.S. (2009). Quantification of
hydroxycinnamic acids and lignin in perennial forage and energy grasses by
Fourier-transform infrared spectroscopy and partial least squares regression.
Bioresource Technology, 100, 1252–1261.
Bridgwater, A.V. (2003). Renewable fuels and chemicals by thermal processing of biomass.
Chemical Engineering Journal, 91, 87-102.
Carroll, A., Somerville, C. (2009). Cellulosic biofuels. Annual Review of Plant Biology, 60, 165-
182.
Chang, M.C.Y. (2007). Harnessing energy from plant biomass. Current Opinion in Chemical
Biology, 11, 677-684.
Chen, L., Carpita, N.C., Reiter, W.-D., Wilson, R.H., Jeffries, C., McCann, M.C. (1998). A
rapid method to screen for cell-wall mutants using discriminant analysis of Fourier
transform infrared spectra The Plant Journal, 16, 385-392.
www.intechopen.com
Clifton-Brown, J., Robson, P., Allison, G., Lister, S., Sanderson, R., Hodgson, E., Farrar, K.,
Hawkins, S., Jensen, E., Jones, S., Huang, L., Roberts, P., Youell, S., Jones, B.,
Wright, A., Valantine, J., Donnison, I. (2008). Miscanthus: breeding our way to a
better future. in: E. Booth, M. Green, A. Karp, I. Shield, D. Stock, D. Turley (Eds.),
Biomass and Energy Crops III, pp. 199-206
Cooley, J.W., Tukey, O.W. (1965). An algorithm for the machine calculation of complex
Fourier series. Mathematics of Computation, 19, 297-301.
de Jong, S. (1993). SIMPLS: an alternative approach to partial least squares regression.
Chemometrics and Intelligent Laboratory Systems, 18, 251-263.
Fahmi, R., Bridgwater, A.V., Darvell, L.I., Jones, J.M., Yates, N., Thain, S., Donnison, I.S.
(2007). The effect of alkali metals on combustion and pyrolysis of Lolium and
Festuca grasses, switchgrass and willow. Fuel, 86, 1560-1569.
Fahmi, R., Bridgwater, A.V., Donnison, I., Yates, N., Jones, J.M. (2008). The effect of lignin
and inorganic species in biomass on pyrolysis oil yields, quality and stability. Fuel,
87, 1230-1240.
Faix, O. (1991). Classification of lignins from different botanical origins by FTIR
spectroscopy. Holzforschung, 45, 21-27.
Foster, C.E., Martin, T.M., Pauly, M. (2010). Comprehensive Compositional Analysis of Plant
Cell Walls (Lignocellulosic biomass) Part I: Lignin. Journal of Visualised Experiments,
37, http://www.jove.com/index/Details.stp?ID=1745.
Foster, C.E., Martin, T.M., Pauly, M. (2010). Comprehensive Compositional Analysis of Plant
Cell Walls (Lignocellulosic biomass) Part II: Carbohydrates. Journal of Visualised
Experiments, 37, http://www.jove.com/index/Details.stp?ID=1837.
Fukushima, R.S., Dehority, B.A. (2000). Feasibility of using lignin isolated from forages by
solubilization in acetyl bromide as a standard for lignin analyses. Journal of Animal
Science, 78, 3135-3143.
Fukushima, R.S., Hatfield, R.D. (2001). Extraction and Isolation of Lignin for Utilization as a
Standard to Determine Lignin Concentration Using the Acetyl Bromide
Spectrophotometric Method. Journal of Agricultural and Food Chemistry, 49, 3133-
3139.
Geng, W., Nakajima, T., Takanashi, H., Ohki, A. (2009). Analysis of carboxyl group in coal
and coal aromaticity by Fourier transform infrared (FTIR) spectrometry. Fuel, 88,
139-144.
Gierlinger, N., Schwanninger, M. (2006). Chemical imaging of poplar wood cell walls by
confocal Raman microscopy. Plant Physiology, 140, 1246-1254.
Giger-Reverdin, S. (1995). Review of the main methods of cell wall estimation: interest and
limits for ruminants. Animal Feed Science and Technology, 55, 295-334.
Gosselink, R.J.A., Abacherli, A., Semke, H., Malherbe, R., Kauper, P., Nadif, A., van Dam,
J.E.G. (2004). Analytical protocols for characterisation of sulphur-free lignin.
Industrial Crops and Products, 19, 271-281.
Grabber, J.H. (2005). How do lignin composition, structure, and cross-linking affect
degradability? A review of cell wall model studies. Crop Science, 45, 820-831.
Hatfield, R.D., Fukushima, R.S. (2005). Can lignin be accurately measured? Crop Science, 45,
832-839.
Hatfield, R.D., Ralph, J., Grabber, J.H. (1999). Cell wall structural foundations: molecular
basis for improving forage digestibilities. Crop Science, 39, 27-37.
www.intechopen.com
Hodgson, E.M., Fahmi, R., Yates, N., Barraclough, T., Shield, I., Allison, G., Bridgwater,
A.V., Donnison, I.S. (2010). Miscanthus as a feedstock for fast-pyrolysis: Does
agronomic treatment affect quality? Bioresource Technology, 101, 6185-6191
Hotelling, H. (1933). Analysis of a complex of statistical variables into principal components.
Journal of Educational Psychology, 24, 417-441; 498-520.
Hsu, C.-P.S. (1997). Infrared spectroscopy. In: Handbook of instrumental techniques for analytical
chemistry Prentice Hall, F.A. Settle (Ed.), pp. 247-283, ISBN 0-131-77338-0.
Hu, S., Xiang, J., Sun, L., Xu, M., Qiu, J., Fu, P. (2008). Characterization of char from rapid
pyrolysis of rice husk. Fuel Processing Technology, 89, 1096-1105.
Huang, A., Zhou, Q., Liu, J., Fei, B., Sun, S. (2008). Distinction of three wood species by
Fourier transform infrared spectroscopy and two-dimensional correlation IR
spectroscopy. Journal of Molecular Structure, 883-884, 160-166.
Johnson, D.B., Moore, W.E., Zank, L.C. (1961). The spectrophotometric determination of
lignin in small wood samples. TAPPI, 44, 793-798.
Jung, H.-J.G., Himmelsbach, D.S. (1989). Isolation and characterisation of wheat straw lignin.
Journal of Agricultural and Food Chemistry, 37, 81-87.
Kacurakova, M., Capek, P., Sasinkova, V., Wellner, N., Ebringerova, A. (2000). FTIR study of
plant cell wall model compounds: pectic polysaccharides and hemicelluloses.
Carbohydrate Polymers, 43, 195-203.
Kim, D., Lee, J., Kim, J. (2009). Application of near infrared diffuse reflectance spectroscopy
for on-line measurement of coal properties. Korean Journal of Chemical Engineering,
26, 489-495.
Knauer, M., Ivleva, N.P., Liu, X., Niessner, R., Haisch, C. (2010). Surface-Enhanced Raman
Scattering-Based Label-Free Microarray Readout for the Detection of
Microorganisms. Analytical Chemistry, 82, 2766-2772.
Laird, D.A. (2008). The charcoal vision: A win-win-win scenario for simultaneously
producing bioenergy, permanently sequestering carbon, while improving soil and
water quality. Agronomy Journal, 100, 178-181.
Liu, C.F., Xu, F., Sun, J.X., Ren, J.L., Curling, S., Sun, R.C., Fowler, P., Baird, M.S. (2006).
Physicochemical characterization of cellulose from perennial ryegrass leaves
(Lolium perenne). Carbohydrate Research, 341, 2677-2687.
McCann, M.C., Defernez, M., Urbanowicz, B.R., Tewari, J.C., Langewisch, T., Olek, A.,
Wells, B., Wilson, R.H., Carpita, N.C. (2007). Neural network analyses of infrared
spectra for classifying cell wall architectures. Plant Physiology, 143, 1314-1326.
McKendry, P. (2002). Energy production from biomass (part 1): overview of biomass.
Bioresource Technology, 83, 37-46.
Mohan, D., Pittman, C.U., Steele, P.H. (2006). Pyrolysis of wood/biomass for bio-oil: A
critical review. Energy & Fuels, 20, 848-889.
Monties, B. (1989). Lignins. Methods in Plant Biochemistry: Plant Phenolics, J.B. Harborne (Ed.),
1, pp. 113-157.
Mouille, G., Robin, S., Lecomte, M., Pagant, S., Hofte, H. (2003). Classification and
identification of Arabidopsis cell wall mutants using Fourier-Transform InfraRed
(FT-IR) microspectroscopy. The Plant Journal, 35, 393-404.
Movasaghi, Z., Rehman, S., Rehman, I.u. (2008). Fourier transform infrared (FTIR)
spectroscopy of biological tissues. Applied Spectroscopy Reviews, 43, 134 - 179.
www.intechopen.com
Nakanishi, R., Kawakami, H. (1991). Determination of lignin in needles by diffuse

reflectance Fourier-transform infrared spectrometry. Mokuzai Gakkaishi, 37, 841-846.
Otto, M. (2007). Chemometrics. Wiley-VCH, IBSN 3-527-31418-0.
Pearson, K. (1901). On lines and planes of closest fit to a system of points in space.
Philosophical Magazine, 2, 557-572.
Ptasinski, K.J., Prins, M.J., Pierik, A. (2007). Exergetic evaluation of biomass gasification.
Energy, 32, 568-574.
Schmidt, M., Schwartzberg, A.M., Carroll, A., Chaibang, A., Adams, P.D., Schuck, P.J.
(2010). Raman imaging of cell wall polymers in Arabidopsis thaliana. Biochemical
and Biophysical Research Communications, 395, 521-523
Sims, R.E.H., Hastings, A., Schlamadinger, B., Taylor, G., Smith, P. (2006). Energy crops:
current status and future prospects. Global Change Biology, 12, 2054-2076.
Stewart, B. (2005). Infrared spectroscopy: Fundamentals and applications John Wiley & Sons,
Ltd., ISBN 0-470-85428-6.
Stewart, D., Yahiaoui, N., McDougall, G.J., Myton, K., Marque, C., Boudet, A.M., Haigh, J.
(1997). Fourier-transform infrared and Raman spectroscopic evidence for the
incorporation of cinnamaldehydes into the lignin of transgenic tobacco (Nicotiana
tabacum L.) plants with reduced expression of cinnamyl alcohol dehydrogenase.
Planta, 201, 311-318.
Sun, L., Simmons, B.A., Singh, S. (2010). Understanding tissue specific compositions of
bioenergy feedstocks through hyperspectral Raman imaging. Biotechnology and
Bioengineering.
Sun, R.C., Sun, X.F., Fowler, P., Tomkinson, J. (2002). Structural and physico-chemical
characterization of lignins solubilized during alkaline peroxide treatment of barley
straw. European Polymer Journal, 38, 1399-1407.
Sun, R.C., Sun, X.F., Wang, S.Q., Zhu, W., Wang, X.Y. (2002). Ester and ether linkages
between hydroxycinnamic acids and lignins from wheat, rice, rye, and barley
straws, maize stems, and fast-growing poplar wood. Industrial Crops and Products,
15, 179-188.
Sun, R.C., Tomkinson, J. (2002). Characterization of hemicelluloses obtained by classical and
ultrasonically assisted extractions from wheat straw. Carbohydrate Polymers, 50, 263-
271.
Woolf, D., Amonette, J.E., Street-Perrott, F.A., Lehmann, J., Joseph, S. (2010). Sustainable
biochar to mitigate global climate change. Nat Communications, 1, 56.
Xu, F., Geng, Z.C., Sun, J.X., Liu, C.F., Ren, J.L., Sun, R.C., Fowler, P., Baird, M.S. (2006).
Fractional and structural characterization of hemicelluloses from perennial ryegrass
(Lolium perenne) and cocksfoot grass (Dactylis glomerata). Carbohydrate Research,
341, 2073-2082.
Xu, F., Sun, J.X., Geng, Z.C., Liu, C.F., Ren, J.L., Sun, R.C., Fowler, P., Baird, M.S. (2007).
Comparative study of water-soluble and alkali-soluble hemicelluloses from
perennial ryegrass leaves (Lolium perenne). Carbohydrate Polymers, 67, 56-65.
Xu, F., Zhou, Q.-A., Sun, J.X., Liu, C.-F., Ren, J.L., Sun, R.-C., Curling, S., Fowler, P., Baird,
M.S. (2007). Fractionation and characterization of chlorophyll and lignin from de-
juiced Italian ryegrass (Lolium multifolrum) and timothy grass (Phleum pratense).
Process Biochemistry, 42, 913-918.
www.intechopen.com
5
Applications of Fourier Transform Infrared

Spectroscopy to Study Cotton Fibers
Noureddine Abidi, Eric Hequet and Luis Cabrales
Texas Tech University
USA
1. Introduction
Fourier Transform Infrared microscopy (FTIR) has become an essential analytical tool
available to scientists to study various materials. Specifically FTIR has been increasingly
used to study cell wall developments in plants, investigate the efficiency of the surface
modification of polymers, identifying contaminants, and predicting the physical
properties of certain polymers and biopolymers, etc. This chapter reviews some selected
application of the FTIR to study cellulose development in cotton fibers and to predict
cotton fiber physical properties. Cotton fiber maturity is a major yield component and an
important fiber quality trait that is directly linked to the quantity of cellulose deposited
during the secondary cell wall (SCW) biogenesis. Cotton fiber development consists of
five major overlapping stages: differentiation, initiation, polar elongation, secondary cell
wall development, and maturation. The transition period between 16 and 21 dpa (days
post anthesis) is regarded to represent a major developmental stage between the primary
cell wall and the SCW. Fourier Transform Infrared spectroscopy was used to investigate
the structural changes that occur during the different developmental stages. The IR
spectra of fibers harvested at different stages of development (10, 14, 17, 18, 19, 20, 21, 24,
27, 30, 36, 46, and 56 dpa) show the presence of vibrations located at 1,733 cm-1 (C=O
stretching originating from esters or amides) and 1,534 cm-1 (NH2 deformation
corresponding to proteins or amino acids). The results converge towards the conclusion
that the transition phase between the primary cell wall and the secondary cell wall occurs
between 17 and 18 dpa in fibers from TX19 cultivar, while this transition occurs between
21 and 24 dpa in fibers from TX55 cultivar. The Universal Attenuated Total Reflectance
FTIR (UATR-FTIR) was used to evaluate the cotton fiber properties. One hundred and
four cotton samples were selected. Thirty FTIR spectra were acquired from each sample
and analyzed. Partial Least Square (PLS) analysis of the FTIR spectra was performed and
the results showed that micronaire and surface area (calculated from the AFIS data) could
be predicted from the FTIR measurements with very high coefficients of determination.
However, the prediction of fiber maturity is probably not possible with the UATR-FTIR.
It was concluded that, to be able to predict the fiber maturity with the FTIR, it would be
necessary to perform the measurements in the transmission mode rather than in the
reflectance mode.
www.intechopen.com
2. Fourier transform infrared spectroscopic approach to the study of the

secondary cell wall development in cotton fiber
“With kind permission from Springer Science + Business Media: Cellulose, Fourier
transform infrared spectroscopic approach to the study of the secondary cell wall
development in cotton fiber, 17, 2010, 309-320, Noureddine Abidi, Luis Cabrales, Eric
Hequet”.
2.1 Introduction
Fiber maturity is a major yield component and an important fiber quality trait that is
directly linked to the quantity of cellulose deposited during the Secondary Cell Wall (SCW)
biogenesis, and to the organization and orientation of crystalline microfibrils. It is inuitively
obvious to hypothesize that immature fibers (having a thin, poorely developed secondary
wall) will be fragile, and therefore, are likely to break during the multiple mechanichal
stresses involved in transforming fibers into yarns. Immature fibers generate short fibers
and neps (entanglement of fibers) that result in yarn defects and decreased productivity in
the spinning mills. Therefore, studying cotton fiber maturity and understanding the link
between SCW biogenesis and cotton fiber maturity is very important. This study was
designed to investigate the structural changes occurring during the growth and
development of cotton fibers.
Cotton fiber development consists of five major overlapping developmental stages
(Wilkin & Jernstedt, 1999): differentiation, initiation, polar elongation, secondary cell wall
deposition, and maturation. The day of flowering is referred to as anthesis and the term
“days post-anthesis” (dpa) is often used to describe the cotton fiber development. Fiber
initiation, which commences at 0 dpa, signals the onset of fiber morphogenesis. Fiber
growth is characterized by the synthesis of the primary cell wall and an increase in fiber
length up to ~30 mm within 3 weeks after anthesis. The stage of secondary cell wall
development commences in general around 21 dpa and continues for a period of ~3 to 6
weeks post-anthesis. This phase is marked by a massive deposition of a thick cellulosic
wall (Wilkin & Jernstedt, 1999). The transition period between 16 and 21 dpa is considered
to represent a developmental switch in emphasis from primary to secondary cell synthesis
during cotton fiber development. During these developmental stages, important
structural changes occur leading to cellulose macromolecules formation (β(1→4)
glucopyranose).
Tokumoto et al. (2002) reported on the changes in the sugar composition and molecular
mass distribution of matrix polysaccharides during cotton fiber development. The results
showed that the extractable matrix (pectic and hemicellulosic) polysaccharides accounted
for 30 - 50% of total sugar content during the elongation stage and less than 3% during the
cell thickening stage. With respect to the amount of cellulose present during fiber
development, it was reported that the secondary wall thickening and maturation stages are
characterized by a dramatic increase in the amount of cellulose (Tokumoto et al., 2002). The
primary cell wall was reported to contain between 35 and 50% cellulose (Timpa & Triplett
1993; Meinert & Delmer, 1977). However, the secondary cell wall is composed of nearly
100% cellulose (Haigler et al., 2005).
The composition of the cotton fiber cell wall exhibits a continuous change throughout the
development of the fiber (Meinert & Delmer, 1977). Several studies have been focused on
www.intechopen.com
Applications of Fourier Transform Infrared Spectroscopy to Study Cotton Fibers 91
the analysis of the composition of the cell wall extracts (Tokumoto et al., 2002; Timpa &
Triplett 1993; Meinert & Delmer 1977; Maltby et al., 1979; Huwyler et al., 1979, Gokani et al.,
1998). However, limited research has been conducted on intact fibers. In previous research,
we reported on the usefulness of Fourier Transform Infrared (FTIR) and Thermogravimetric
Analysis (TGA) to investigate cotton fiber development (Abidi et al., 2008). The results
showed that these two analytical techniques could be used to evaluate the cell wall
composition and structure during fiber development. In addition, because of the
nondestructive character of the FTIR analysis other testing could be performed on the same
set of samples.
Fourier Transform Infrared spectroscopy (FTIR) has emerged as a key technique for the
study of plant growth and development (McCann et al., 2007; Yong et al., 2005; Carpita et
al., 2001; Zeier & Schreiber, 1999; Chen et al., 1998; McCann et al., 1997; Séné et al., 1994;
McCann et al., 1993; McCann et al., 1992). Zeier and Schreiber (1999) used FTIR to
characterize isolated endodermal cell walls from plant roots and assigned FTIR frequencies
to functional groups present in the cell wall, including the relative amounts of the cell wall
biopolymers suberin and lignin, as well as cell wall carbohydrates and proteins. FTIR
absorption spectra indicated structural differences for three developmental stages of the
endodermal cell wall under study. The authors concluded that FTIR could be used as a
direct and non-destructive method suitable for the rapid investigation of isolated plant cell
walls. The approach has since been successfully applied to screen large numbers of mutants
for a broad range of cell wall phenotypes using FTIR of leaves of Arabidopsis thaliana and flax
(Linum usitatissimum) (Chen et al., 1998). In this study, Chen and co-workers reported that
principal component analysis (PCA) of FTIR spectra can distinguish between mutants that
are deficient in cell wall sugars. Also, FTIR and Fourier-Transform Raman spectroscopy
have been successfully used to investigate the primary cell wall architecture at a molecular
level (Séné et al. 1994). Dynamic changes in cell wall composition of hybrid maize
coleoptiles (Zea mays) were investigated by FTIR (McCann et al., 2007). The authors reported
that neural network algorithms could correctly classify infrared spectra from cell walls
harvested from individuals differing at one-half-day interval of growth.
The aim of the work reported in this paper has been 2-fold: 1) to investigate the structure
and composition of fiber during different phases of development; and 2) to elucidate the
effect of cultivar on the developmental stages of the cotton fiber.
2.2 Experimental
2.2.1 Materials
For this study, two independent replications (10 plants each) of two cotton cultivars
(Gossypium hirsutum L. cv. TX19 and TX55) were planted in a greenhouse with day/night
cycles varying from 13/11 to 11/13 hours and day/night temperatures of about 31oC /
24oC. Plants were grown in 20-L (5 gallons) pots of Sungrow SB 300 potting mix that had
been amended with Peters 15-9-12 slow release fertilizer prior to potting. Plants were
watered as needed. On the day of flowering (0 dpa), individual flowers were tagged, and 14
developing bolls per cultivar and per replication were harvested at 10, 14, 17, 18, 19, 20, 21,
24, 27, 30, 36, 46, and 56 dpa. The pericarp was immediately removed (excised with scalpel)
and isolated ovules were transferred into cryogenic vials and stored in a Cryobiological
Storage System filled with liquid nitrogen for analyses. Each replication was tested
independently.
www.intechopen.com
2.2.2 Methods
2.2.2.1 Sample dehydration
Frozen cotton fiber samples were dehydrated using the procedure described in (Abidi et al.,
2008; Muller & Jackset al., 1975; Rajasekaran et al., 2006). Frozen samples were first rinsed
with water and then washed with acidified solution of 2,2-dimethoxypropane (one drop of
HCl in 50 ml of 2,2-dimethoxypropane), followed by five exchanges for 15 minutes each in
100% acetone. In a slightly acidic solution, 2,2-dimethoxypropane is instantly hydrolyzed
by water to form methanol and acetone (Muller & Jacks, 1975).
2.2.2.2 FTIR measurements
The FTIR spectra of cotton fiber samples were recorded in an environmentally-controlled
laboratory maintained at relative humidity of 65±2% and 21±1oC using the Spectrum-One
equipped with an UATR (Universal Attenuated Total Reflectance) accessory (Perkin-
Elmer, USA). The UATR-FTIR was equipped with a ZnSe-Diamond crystal composite that
allows collection of FTIR spectra directly on a sample without any special preparation.
The instrument is equipped with a “pressure arm” which is used to apply a constant
pressure to the cotton samples positioned on top of the ZnSe-Diamond crystal to ensure a
good contact between the sample and the incident IR beam and prevent the loss of the
IR beam. The amount of pressure applied is monitored by the Perkin-Elmer FTIR
software.
Thirty FTIR spectra per sample were acquired for each developmental stage to produce a
total of 180 spectra (30 spectra x 3 replications per dpa x 2 greenhouse replications). All
FTIR spectra were collected at a spectrum resolution of 4 cm-1, with 32 co-added scans over
the range from 4,000 to 650 cm-1. A background scan of clean ZnSe-Diamond crystal was
acquired before scanning the samples.
2.2.2.3 FTIR spectra analysis
The Perkin-Elmer software was used to perform spectra normalization, baseline corrections,
and peak integration. FTIR spectra were then exported to Excel and were subjected to
Principal Component Analysis (PCA) with leverage correction and mean-center cross
validation boxes checked using Unscrambler V. 9.6 Camo Software AS (CAMO Software AS,
Norway).
2.2.2.4 Scanning electron microscope
Hitachi Scanning Electron Microscopy (TM-1000, Hitachi Japan) with an accelerating voltage
of 15kv was used to visualize the frozen and dehydrated samples. Fibers were placed either
on a carbon disc or on a microscopy slide and no coating was performed prior to
visualization.
2.3 Results and discussion

2.3.1 Fourier transform infrared analysis
Figures 1a and b show a series of FTIR spectra of fibers respectively from TX19 and TX55
cultivars during the elongation stage (10, 17, and 19 dpa), during the SCW formation stage
(20, 24, and 30 dpa), and during the maturation stage (56 dpa). Compared to the FTIR
spectra of mature cotton fibers, several additional vibration bands are noticed in the FTIR
spectra of developing cotton fibers.
www.intechopen.com
Fig. 1a. FTIR spectra of developing cotton (Gossypium hirsutum L. cv. TX19) fibers at different
days post-anthesis (dpa)
The vibrations located at 2,918 and 2,850 cm-1 are attributed to –CH2 asymmetric vibrations
and could originate from the presence of wax substances present on the surface of the
primary cell wall (Abidi et al., 2008). The intensities of these two peaks start decreasing at
19 dpa for both cultivars. This result is in agreement with our previous findings (Abidi et al.,
www.intechopen.com
2008). We reported that the percent contribution of the primary cell wall to the total weight
of the fiber decreased as wall thickness increases (increased fiber maturity, thus secondary
cell wall). Therefore, the relative importance of the vibration bands attributed to
noncellulosic substances (e.g. waxes which are located essentially on the primary cell wall)
is less.
Fig. 1b. FTIR spectra of developing cotton (Gossypium hirsutum L. cv. TX55) fibers at
different days post-anthesis (dpa)
www.intechopen.com
In addition to O-H bending vibration of water molecules located at 1,627 cm-1, the FTIR
spectra of fibers showed strong absorption between 3,000 cm-1 and 3,600 cm-1 (which is
attributed to O-H stretching vibration). This absorption band is composed of two vibrations
located at 3,285 cm-1 (attributed to intermolecular hydrogen bonds) and 3,335 cm-1
(attributed to intra-molecular hydrogen bonds) (Liang & Marchessault, 1959). The intensity
of the vibration 3,280 cm-1 decreased in the FTIR spectra of fibers older than 17 dpa from
TX19 cultivar while for fibers from TX55 the decrease occurred at 19 dpa. The decrease in
intensity of 3,280 cm-1 band is associated with an increase in the intensity of the vibration
located at 3,335 cm-1.
The vibration located at 1,733 cm-1 is attributed to C=O stretching vibration and could
originate from esters or amides (Abidi et al., 2008). The integrated intensity of this peak
(I1733) was calculated between 1,780 cm-1 and 1,701 cm-1 and was reported as function of
dpa for both cultivars TX19 and TX55 (Fig. 2a). As exhibited in this chart, for fibers from
TX19 cultivar I1733 decreased sharply between 10 and 18 dpa and leveled-off at 19 dpa.
However, for fibers from TX55 cultivar a continuous decrease is observed between 10 and
24 dpa. The statistical analysis (analysis of variance) showed significant effects of the
developmental stage (dpa) and the cultivar on the integrated intensity (Table 1). These
results indicate that the structural changes that occur during fiber development are
influenced by the cultivar.
Fig. 2a. Evolution of the integrated peak intensity (I1733) as function of developmental stages
(dpa)
www.intechopen.com
Parameter df F Probability Integrated intensity I1733 ¥
Intercept 1 574.4213 0.000001

Cultivar 1 71.1905 0.000001
dpa (days post anthesis) 12 60.0900 0.000001
10 9.5049 a
14 7.7578 b
17 4.7894 c
18 3.5184 d
19 2.7784 de
20 2.1827 de
21 1.5740 ef
24 0.6320 f
27 0.4391 f
30 0.4388 f
36 0.2263 f
46 0.1651 f
56 0.0881 f
Cultivar *dpa 12 7.0895 0.000016
Error 26
different with α = 5% (according to Newman-Keuls tests).

df, degrees of freedom; F, variance ratio; ¥ Values not followed by the same letter are significantly
Table 1. Variance Analysis: Effect of developmental stage (day post-anthesis) and cultivars
on the integrated intensity of the peak 1,733 cm-1 (I1733)
The vibration located at 1,627 cm-1 is attributed to O-H bending of adsorbed water
molecules (Abidi et al., 2008). The integrated intensity of this peak (I1627) was calculated
between 1,701 cm-1 and 1,576 cm-1. The change in the integrated intensity I1627 as function of
dpa is reported in Fig. 2b for both cultivars. The statistical analysis (analysis of variance)
showed a significant effect of the developmental stage and the cultivar on the integrated
intensity (Table 2). For fibers from TX19 cultivar, the amount of adsorbed water decreased
between 10 and 18 dpa, no significant changes are observed between 19 and 56 dpa.
However, for fibers from TX55 cultivar the amount of adsorbed water decreased linearly
until the fibers reached 24 dpa. The decrease in the amount of adsorbed water could be
attributed to the decrease of the surface area and to reduced accessibility of water molecules
to the internal hydroxyl groups to establish hydrogen bonding. This could be the result of
increased polymerization reactions of glucose units to form cellulose macromolecules
followed by increased crystallinity. Hsieh et al. reported that the degree of crystallinity of
two cotton fiber cultivars (Maxxa and SJ-2) increases beginning 24 dpa (Hsieh et al., 1997).
The results indicated that the big change in crystallinity occurred between 24 and 28 dpa,
and no significant changes occurred thereafter. These results are in agreement with our FTIR
results, which indicate that no change in the amount of adsorbed water is noticed between
27 and 56 dpa. In this developmental stage, fibers from both cultivars have nearly the same
amount of adsorbed water. This could indicate that the remaining water molecules are those
which are strongly bonded to cellulose macromolecules via hydrogen bonding.
www.intechopen.com
Fig. 2b. Evolution of the integrated peak intensity (I1627) as function of developmental stage
(dpa)
Intercept 1 1755.569 0.000001

Cultivar 1 52.774 0.000001
10 30.7341 a
14 31.4025 a
17 23.1798 b
18 17.8545 c
19 15.6207 cd
20 13.8401 cde
21 12.3703 ed
24 6.8814 fg
27 5.7146 g
30 5.3511 g
36 11.6973 def
46 10.3251 defg
56 8.7280 efg
Cultivar*dpa 12 4.916 0.000334
Error 26

df, degrees of freedom; F, variance ratio, ¥ Values not followed by the same letter are significantly
on the integrated intensity of the peak 1,627 cm-1 (I1627)
www.intechopen.com
The vibration located at 1,534 cm-1 is attributed to NH2 deformation and likely indicative of
proteins or amino acids (Abidi et al., 2008). The integrated intensity of this peak (I1534) was
calculated between 1,575 cm-1 and 1,487 cm-1 and the evolution of I1534 as function of dpa for
both TX19 and TX55 showed a behavior similar to I1733 (Fig. 2c). An abrupt decrease is
observed between 10 and 18 dpa for fibers from TX19 cultivar and a continuous decrease
between 10 and 24 dpa for fibers from TX55 cultivar. The statistical analysis (analysis of
variance) showed a significant effect of both the developmental stage (dpa) and the cultivar
on the integrated intensity (Table 3).
Fig. 2c. Evolution of the integrated peak intensity (I1534) as function of developmental stage
(dpa)
The vibration located at 1,236 cm-1 is attributed to C=O stretching or NH2 deformation
(Abidi et al., 2008). This vibration decreased in intensity during fiber development. The
disappearance of this band is accompanied by the appearance of a vibration at 1,204 cm-1 at
20 dpa for fibers from TX55 cultivar and as sharp band at 17 dpa for fibers from TX19
cultivar. It is important to point out that the vibration located at 1,204 cm-1 appears only
beginning 17 dpa for fibers from TX19 cultivar and beginning 20 dpa for fibers from TX55
cultivar. This vibration has been attributed by Ilharco et al. (1997) to C-O-C stretching mode
located at 900 cm-1 (attributed to β-linkage). Consequently, these two vibrations could be
of the pyranose ring. This vibration appears almost simultaneously with the vibration
considered as the secondary cell wall’s fingerprint.

Other vibrations located at 1,148, 1,096, and 1,048 cm-1 are present only as shoulders in the
spectra of fibers from TX19 cultivar at 10 and 14 dpa and become sharper beginning at 17
dpa. It is also noteworthy the shift at 17 dpa of the vibration 1,148 cm-1 to 1,161 cm-1, the
vibration 1,096 cm-1 to 1,105 cm-1, and the vibration 1,048 cm-1 to 1,056 cm-1. However, for
fibers from TX55 cultivar the shift of these vibrations to higher wavenumbers occurred at 20
www.intechopen.com
dpa. The vibration located at 1,161 cm-1 is assigned to the anti-symmetric bridge C-O-C
stretching vibration (Ilharco et al., 1997). The vibration located at 1,105 cm-1 is assigned to
anti-symmetric in-plane ring stretching band (Ilharco et al., 1997). The vibration located at
1,056 cm-1 is attributed to C-O stretching mode (Abidi et al., 2008).
Intercept 1 474.3505 0.000001

Cultivar 1 49.8492 0.000001
10 9.3057 a
14 8.0100 a
17 5.4175 b
18 3.8020 c
19 3.1142 c
20 2.7056 c
21 2.1114 c
24 0.7674 d
27 0.3300 d
30 0.0455 d
36 0.0000 d
46 0.0000 d
56 0.0000 d
Cultivar*dpa 12 4.3089 0.000888
Error 26

df, degrees of freedom; F, variance ratio, ¥ Values not followed by the same letter are significantly
on the integrated peak intensity of the peak 1,534 cm-1 (I1534)
The vibration located at 1,017 cm-1, attributed to C-O stretch, is shifted to 1,031 cm-1 at 17
dpa for fibers from TX19 cultivar. However, for fibers from TX55 cultivar, this shift occurs
at 19 dpa.
The vibrations located at 1,003 cm-1 and 985 cm-1 appeared at 30 dpa in the spectra of fibers
from TX19 cultivar but only at 56 dpa in the spectra of fibers from TX55 cultivar. These
The vibration located at 900 cm-1 is attributed to β-linkage (Abidi et al., 2008). This vibration
vibrations are attributed to C-O and ring stretching modes (Liang & Marchessault, 1959).
is present only as a small shoulder in the FTIR spectra of fibers from both cultivars at 10 dpa
but becomes sharper in the FTIR spectra of fibers from TX19 cultivar at 17 dpa and in the
FTIR spectra of fibers from TX55 cultivar at 19 dpa.
The vibration located at 710 cm-1 is attributed to CH2 rocking vibration in cellulose Iβ
(Schwanninger et al., 2004). Using 2D FTIR spectroscopy, Salmén et al. (2005) reported that
the peak around 710 cm-1 (which is characteristic of cellulose Iβ found in native cotton) has a
linear correlation with the percentage of cellulose Iβ of the crystalline part. We calculated the
peak height of the vibration 710 cm-1 from all FTIR spectra and we reported the data as
function of dpa for both genotypes as shown in Fig. 2d. The statistical analysis shows
significant effects of the type of cultivar [F(1,26)=122.318, p=0.000001] and the
developmental stage [F(12,26)=121.334, p=0.00001]. There is also an interaction
www.intechopen.com
developmental stage * type of cultivar [F(12,26)=5.986, p=0.00008]. For fibers from TX19
cultivar the peak height of the vibration 710 cm-1 increases linearly between 10 and 30 dpa.
No major change is noticed between 30 and 56 dpa. This indicates that the cellulose
synthesis during the secondary cell wall development is accompanied by structural
organization (increased crystallinity). However, for fibers from TX55 cultivar, the major
change of the peak height occurred only beginning 21 dpa. This indicates that for fibers
from TX55 cultivar, the structural organization of the cellulose is initiated few days later
than in fibers from TX19 cultivar.
Fig. 2d. Evolution of the peak height for 710 cm-1 as function of developmental stage (dpa)
Figures 2e and f show the relationships between the integrated intensity of the peak 1,627
cm-1 (corresponding to adsorbed water) and the peak height of the vibration 710 cm-1 for
fibers from TX19 and TX55, respectively. This relationships show that the decrease of the
amount of adsorbed water is associated with an increase in the structural organization of the
cellulose (increased crystallinity).
Principal Components Analysis (PCA) was performed on the FTIR spectra in order to
identify distinct groups of spectra. The effect of this process is to concentrate the sources of
variability in the data into the first 2 PCs (PC1 and PC2). The plots of PC1 against PC2
scores are depicted in Fig. 3a and b, respectively for fibers from TX19 and TX55 cultivars.
For fibers from TX19 cultivar, two groups of FTIR spectra could be identified: group 1
includes FTIR spectra of fibers at 10, 14, and 17 dpa; and group 2 includes FTIR spectra of
fibers from 18 to 56 dpa. For fibers from TX55 cultivar, two groups of FTIR spectra could
also be identified: group 1 includes FTIR spectra of fibers from 10 to 21 dpa and group 2
includes FTIR spectra of fibers from 24 to 56 dpa. It is important to note that the transition
between the two groups of FTIR spectra is different for these cultivars: for TX19, the
transition occurs at 17 dpa and for TX55 the transition occurs at 21 dpa. Several studies
have reported that the transition period between 16 and 21 dpa represents the
www.intechopen.com
developmental switch in emphasis from primary to secondary cell wall synthesis during
cotton fiber development (Wilkin & Jernstedt, 1999).
Fig. 2e. Relationship between the integrated peak intensity (I1627) and the peak height for 710
cm-1 (Gossypium hirsutum L. cv. TX19)
Fig. 2f. Relationship between the integrated peak intensity (I1627) and the peak height for 710
cm-1 (Gossypium hirsutum L. cv. TX55)
www.intechopen.com
Fig. 3a. Principal Component Analysis of FTIR spectra of fibers from TX19 in the range 4,000
– 650 cm-1
Fig. 3b. Principal Component Analysis of FTIR spectra of fibers from TX55 in the range 4,000
– 650 cm-1
The FTIR data seem to indicate that the secondary cell wall synthesis in fibers from TX19
cultivar could start between 17 and 18 dpa. However, it could start between 21 and 24 dpa
in fibers from TX55 cultivar. These data indicate that the FTIR spectroscopy technique has a
potential use as screening tool for cotton fiber cultivars that have potentials for early
maturation.
2.3.2 Scanning electron microscopy

Scanning electron microscopy micrographs of single fibers from selected developing stages
are exhibited in Fig. 4. As illustrated in these micrographs, single cotton fibers at 17 dpa
www.intechopen.com
from both cultivars appear transparent consisting mainly of primary cell walls. Single fiber
from TX19 at this developmental stage appears slightly thicker than that of TX55. At 20 dpa,
fibers from TX19 are thicker than those from TX55, signaling that the secondary cell wall
synthesis in TX19 cultivar is well underway at this stage. Fibers from TX55, however,
appear thicker only beginning at 24 dpa.
Fig. 4. Scanning electron microscopy micrographs of developing cotton fibers (A) TX19 17
dpa, (B) TX19 20 dpa, (C) TX19 24 dpa, (D) TX55 17 dpa, (E) TX55 20 dpa, (F) TX55 24 dpa
www.intechopen.com
2.4 Conclusion
Fourier Transform Infrared spectroscopy was used to investigate the structural changes that
occur during cotton fiber development starting at 10 days post-anthesis. FTIR spectra show
pronounced differences during fiber development. The evolution of the integrated
intensities of specific vibration bands located at 1,733, 1,534, and 1,627 cm-1 as function of
developmental stages could be used to monitor the development of the secondary cell wall.
FTIR results indicate that the two cultivars investigated (TX19 and TX55) exhibited different
structural evolution. The results converge towards the conclusion that the transition phase
between the primary cell wall and the secondary cell wall occurs between 17 and 18 dpa in
fibers from TX19 cultivar, while this transition occurs between 21 and 24 dpa for fibers from
TX55 cultivar. FTIR findings were supported by thermogravimetric characterization of
fibers at different stages of development, changes in sugar composition as measured by
High Performance Liquid Chromatography, and cellulose content as determined by the
anthrone method (data not shown).
3. Cotton fiber physical properties prediction using FTIR

3.1 Introduction
There is universal interest in measuring cotton fiber properties that are useful to predict the
performance of the fiber as an industrial raw material. In addition, significant efforts are
being made to develop, through breeding and biotechnology, new cotton varieties that
provide superior fiber properties. The “missing link” in these efforts is a scientific
understanding of the relationships between the desired properties and the fiber
structure/morphology. Among fiber properties, maturity is paramount for several reasons.
The effect of maturity on the dye uptake is well known and constitutes the basis of the
Goldthwait test (Goldthwait et al., 1947). Similarly, it is known that fine and mature fibers
make it possible to spin a finer yarn. But maturity and fineness of cotton fibers are also
essential qualitative characteristics if one wants to better understand the propensity of
rupture of fibers when they are subjected to stress. It is intuitively obvious to hypothesize
that immature fibers (having a thin, poorly developed secondary wall) will be fragile. Thus,
they are likely to break during multiple mechanical stresses involved in transforming the
fibers from the field to the yarn. These generate short fibers and neps (entanglements of
fibers), which will result in yarn defects and decreased productivity.
The dominant tools used today to evaluate fiber properties are the High Volume
Instruments (HVI) and the Advanced Fiber Information System (AFIS). The HVI is a fast
instrument while the AFIS is relatively slow. Therefore, most of the cotton breeders rely
only on HVI measurements. The HVI provides micronaire, length, strength, and color of the
lint while the AFIS provides length, maturity, and fineness distributions as well as the nep
count and the trash content. While the information provided by the HVI is necessary, it is
not sufficient to provide answers about the structure/morphology that are needed to
achieve new breakthroughs. The AFIS would be more appropriate but its relatively slow
speed and high cost makes it non affordable by most of the cotton breeding programs. In an
earlier work, Thermogravimetric Analysis was used to study the relationship between
cotton fiber thermal properties and maturity and fineness (Abidi et al., 2007). The results
showed that, low micronaires (immature and/or finer fibers), and low maturity ratios
(immature fibers) are associated with high weight losses in the region 225 - 425oC. Both low
micronaire and low maturity fibers exhibit high surface area for a given weight of fibers.
www.intechopen.com
Therefore, the quantity of the non-cellulosic compounds, mostly present on and in the
primary cell wall, is quite large. This leads to higher weight losses in the region 225 - 425oC
because in addition to the pyrolysis reactions of cellulose, the decomposition of non-
cellulosic compounds also takes place between 225 - 425oC. Furthermore, it was showed
that it is possible to estimate the width of the primary cell wall, by comparing the weight
losses of two cotton fibers that are identical except for having different maturities (i.e.
different degrees of secondary cell wall development).
Our hypothesis is that the FTIR spectroscopy has the potential to assess quickly and
accurately some fiber properties related to cotton fiber maturity and fineness. Ramey (1982)
used Near-infrared reflectance (NIR) to estimate some quality components of natural fibers.
The author explored the use of the NIR to predict the cross-sectional area, the specific
surface, micronaire, and causticaire maturity index of cotton fibers. A Neotec Model 41
1.53, 1.97, and 2.32 μm central wavelength filters. The approximate wavelengths for data
Grain Quality Analyzer was used to conduct the study. This instrument was equipped with
collection were 1.49 to 1.51, 1.90 to 1.93, 2.16 to 2.19, and 2.26 to 2.30 μm. The author
concluded that meaningful estimates of the four mentioned fiber properties could be
obtained with near-infrared reflectance. In addition, the author indicated that, except for
micronaire, when performing near-infrared reflectance measurements, technician and
instrument time per specimen is less than the usual procedures.
Montalvo and Von Hoven (2004) made a comprehensive review on both the reference
methods and the application of NIR to predict cotton fiber properties (fineness, maturity,
and micronaire). The authors indicated that successful application of NIR spectroscopy to
cotton fiber quality measurements has been limited by shortcuts in the development and
validation of NIR instruments and methodologies.
In this study, we used the Universal Attenuated Total Reflectance Fourier Transform
Infrared (UATR-FTIR) to predict cotton fiber properties. The FTIR spectra were acquired in
the mid-infrared range (4,000 – 650 cm-1). After baseline correction and spectra
normalization, the Partial Least Square was used to predict micronaire, maturity, and
surface area.
3.2 Materials
One hundred and four cotton bales representing the two principal cultivated species were
selected. These cotton samples were the same cotton samples used in our previous study
(Hequet et al., 2006). The vast majority of the bales originated in the U.S.A., but some
foreign-grown cotton bales were also selected (Egypt, Uzbekistan, Pakistan, Cameroon,
Syria, Benin, and Australia). The bales were opened and ten samples per bale were taken.
Each sample was tested using a HVI Uster 900A. For each bale, a total of 40 micronaire tests
and 100 length and tenacity tests were done. This allowed us to conclude that the intra-bale
variability was acceptable and that we had a wide range of fiber properties (Table 4). The
same samples were also tested on the AFIS with 5 replications and 3,000 fibers tested per
replication. This totaled 150,000 fibers per bale.
A representative sample of approximately 30 kg (70 pounds) was taken from each bale.
Each sample was homogenized according to the protocol used by the ICCSC (International
Cotton Calibration Standard Committee, 1999 (USDA, 1999) to produce reference cottons.
From the card web produced, 20 samples were taken. Samples 1 to 5 were re-sampled (8
pinches per sample). This new sample was delicately mixed manually then 2 fibrograph
www.intechopen.com
combs were formed. We chose to sample with the fibrosampler because this method,
unlike Lord’s method, is not length biased. This procedure was repeated for samples 6-
10, 11-15 and 16-20. A sample was then taken from each of the 8 combs produced. For
each comb, a minimum of 500 cross sections were analyzed following the procedure
described in (Hequet et al., 2006). When the CV% of the averages between combs was
higher than 2%, eight other combs were produced and 500 additional cross sections (per
comb) were analyzed. The original CV% was confirmed in almost every case. From this
point forward the 104 cottons homogenized according to the ICCS protocol will be called
reference cottons.
UHML Tenacity
Micronaire UI (%) Elongation (%)
(inch) (cN/tex)
Average : 4.2 1.08 81.9 29.4 5.4
Minimum 2.6 0.96 78.6 21.6 3.6
Maximum 5.7 1.31 84.5 41.3 8.3
Intra-bale CV% :
Average 0.7 0.9 0.6 1.9 3.5
Minimum 0.0 0.4 0.2 0.8 1.4
Maximum 2.2 2.2 1.1 3.1 7.7
Table 4. HVI fiber properties of the 104 bales: Basic statistics
3.3 Methods
3.3.1 FTIR measurements
Spectrum-One equipped with an UATR (Universal Attenuated Total Reflectance) accessory
was used to acquire the FTIR spectra of the cotton fiber samples (PerkinElmer, USA). The
UATR-FTIR was equipped with a ZnSe-Diamond crystal composite that allows collecting
the FTIR spectra directly on a sample without any special preparation. The FTIR
spectrometer was placed in a conditioned laboratory at 65±2%RH and 21±1oC and all the
FTIR spectra were acquired in this environment. Cotton samples were placed on top of the
ZnSe-Diamond crystal and a pressure was applied on the sample to ensure a good contact
between the sample and the incident IR beam and to prevent loss of the IR beam. The
amount of pressure applied was kept the same and it was monitored through the FTIR
software included in the Perkin-Elmer software package. Therefore, all samples were
subjected to the same pressure.
A representative sample was taken from each reference cotton. Thirty spectra from each
sample were acquired and analyzed. All the FTIR spectra were collected at a spectrum
resolution of 4 cm-1, with 32 co-added scans, over the range of 4,000 - 650 cm-1. A
background scan of clean ZnSe-Diamond crystal was acquired before acquiring the spectra
of the sample. The Perkin-Elmer software was used to perform the baseline corrections of
the spectra.
www.intechopen.com
3.3.2 FTIR spectra analysis

The raw FTIR spectra were normalized (centered and scaled) and averaged. Then, the PLS1
(Partial Least Square) was run with a full cross-validation (The Unscrambler V. 9.6 Camo
Software AS, USA). An uncertainty test was performed for each test. Full cross validation
means that the same samples are used both for model estimation and testing. One sample is
left out from the calibration data set and the model is calibrated on the remaining data
points. Then the value for the left-out sample is predicted and the prediction residuals are
computed. The process is repeated with another subset of the calibration set, and so on until
every sample has been left out once; then all prediction residuals are combined to compute
the validation residual variance and RMSEP (Root Mean Square Error of Prediction).
RSMEP can be interpreted as the average prediction error. Therefore, we can give the
predicted Y-values an estimated precision of two times the RSMEP.
3.4 Results and discussion

For each sample in addition to the FTIR measurements, the following fiber properties were
selected: HVI micronaire, image analysis of the cross-sections (perimeter and theta), and the
AFIS data (calculated Area mm2/mg). PLS1 (Partial Least Square) procedure, was used to
predict these fiber maturity-related parameters. Partial Least Square regression is a method
that relates the variations in one response variable (micronaire for example) to the variations
of several predictors (X variables = absorbance for the different wavelengths) with
explanatory or predictive purposes. This method performs particularly well when the
various X-variables express common information, i.e. when there is a large amount of
correlation, or even collinearity between them, which is the case with the FTIR spectra.
Partial Least Square regression is a bilinear modeling method where information in the
original FTIR data is projected onto a small number of underlying (“latent”) variables called
PLS components. The Y-data are actively used in estimating the “latent” variables to ensure
that the first components are those that are most relevant for predicting the Y-variables.
Interpretation of the relationship between X-data and Y-data is then simplified as this
relationship is concentrated on the smallest possible number of components.
3.4.1 Prediction of the micronaire

Cotton fiber micronaire is a function of fineness and maturity. It is based on the
measurement of an air flow that passes through a porous plug of cotton fibers. The
micronaire is proportional to the inverse of the specific surface of the cotton fibers. It should
be noted that the definition of fiber fineness in cotton does not relate directly to fiber
perimeter. Indeed, fiber fineness (expressed in millitex) is the weight in mg of 1,000 meters
of fibers. Therefore a fine fiber may have a small perimeter and a high maturity ratio.
Conversely, a fine fiber may have a large fiber perimeter and a low maturity ratio (which
implies a large lumen in the fiber). In a similar manner, high micronaire readings indicate
coarse fibers (high weight per unit length), while low micronaire readings indicate fine
fibers (low weight per unit length).
Figure 5a shows the relationship between the FTIR and the HVI micronaire. With a
coefficient of determination of 0.9252, the FTIR measurements lead to very good prediction
of the micronaire (FTIR prediction = 0.9253*Micronaire + 0.3349). It should be noted that the
coefficient of determination of the validation (R2 = 0.8677) set is nearly as good as the
coefficient of determination of the calibration set (fig. 5b).
www.intechopen.com
Fig. 5a. FTIR HVI micronaire prediction (calibration) versus HVI micronaire
Fig. 5b. FTIR HVI micronaire prediction (validation) versus HVI micronaire
3.4.2 Prediction of fiber perimeter and theta

Fiber perimeter and theta were determined by image analysis (Hequet et al., 2006). Theta,
degree of secondary cell wall thickening, is defined as the ratio of the area of the cell wall to
the area of a circle having the same perimeter as the fiber cross-section. The measurement of
theta leads to the determination of the maturity ratio M (θ = M*0.577). Figure 6a shows the
prediction of the fiber perimeter with FTIR (FTIR prediction = 0.6529*Perimeter + 18.308).
www.intechopen.com
The coefficient of determination (R2 = 0.6529) of the calibration set is not very good. For the
validation set, the coefficient of determination (R2 = 0.4548) is much lower (Fig. 6b). These
results reveal that the PLS1 model is not valid.
Fig. 6a. FTIR Perimeter Image Analysis prediction (calibration) versus Perimeter Image
Analysis
Fig. 6b. FTIR Perimeter Image Analysis prediction (validation) versus Perimeter Image
Analysis
www.intechopen.com
Figure 7a shows the prediction of theta (maturity) with FTIR (FTIR prediction =
0.8107*Theta + 0.0955). The coefficient of determination (R2 = 0.8106) of the calibration set is
acceptable. However, the coefficient of determination (R2 = 0.6981) of the validation set is
much lower than the one of the calibration set, revealing that a good prediction of maturity
with this method (reflectance) is probably not possible (Fig. 7b).
Fig. 7a. FTIR Theta Image Analysis prediction (calibration) versus Theta Image Analysis
Fig. 7b. FTIR Image Analysis Theta prediction (validation) versus Theta Image Analysis
www.intechopen.com
3.4.3 Prediction of the surface area

The results obtained with the AFIS for both Standard fineness and maturity are very similar
to the one obtained with image analysis. From AFIS data, the surface area (expressed in
mm2/mg) was calculated. Fig. 8a and b shows the prediction of the surface area by FTIR
(FTIR prediction = 0.9191*surface area + 26.264). Both the coefficient of determination of
calibration set (R2 = 0.9191) and the validation (R2 = 0.8645) are good.
Estimating the specific surface with two drastically different methods (the micronaire and
the AFIS) led to the same conclusion. It is possible to predict with the FTIR the Area in mm2
per mg of fibers.
Fig. 8a. FTIR Area (mm2/mg) prediction (calibration) versus Area (calculated from AFIS data)
Fig. 8b. FTIR Area (mm2/mg) prediction (validation) versus Area (calculated from AFIS data)
www.intechopen.com
3.5 Conclusion
In this study we investigated the use of the Universal Attenuated Total Reflectance Fourier
Transform Infrared spectroscopy to evaluate the cotton fiber properties. One hundred and
four cotton samples were tested and 30 FTIR spectra were acquired and analyzed from each
sample. The Partial Least Square (PLS) procedure was performed on the FTIR spectra. The
results showed that the micronaire and the surface area (calculated from the AFIS data)
could be predicted from the FTIR measurements with very high coefficient of determination.
Two drastically different techniques to estimate the surface area of the cotton fiber
(micronaire and AFIS data) led to the same conclusion. However, the prediction of fiber
maturity is probably not possible with FTIR because of the low coefficient of the
determination. It was concluded that, to be able to predict the fiber maturity with the FTIR,
it is necessary to perform the measurement in the transmission mode.
4. Acknowledgments
The authors would like to thank the Texas Department of Agriculture/Food and Fibers
Research Grant Program, Cotton Incorporated, and the International Cotton Research
Center/USDA for the financial support.
5. References
Abidi, N., Hequet, E., & Ethridge, D. (2007). Thermogravimetric analysis of cotton fibers:
relationships with maturity and fineness. J. Appl Polym Sci , 103: 3476-3482
Abidi, N., Hequet, E., Cabrales, L., Gannaway, J., Wilkins, T., & Wells, LW. (2008)
Evaluating cell wall structure and composition of developing cotton fibers using
Fourier transform infrared spectroscopy and thermogravimetric analysis. J Appl
Polym Sci 107:476-486
Carpita , NC., Defernez, M., Findlay, K., Wells, B., Shoue, DA., Catchpole, G., Wilson, RH.,
& McCann, MC. (2001) Cell wall architecture of the elongating Maize Coleoptile.
Plant Physiol, 127: 551-565
Chen, L., Carpita, NC., Reiter, W-D., Wilson, RH., Jeffries, C., & McCann, MC. (1998) A
rapid method to screen for cell-wall mutants using discriminant analysis of Fourier
transform spectra. Plant J16(3): 385-392
Goldthwit, C.F.; Smith, H.O.; & Barnett, M.P. New dye technique shows maturity of cotton.
Textile World, July 1947, p 105.
Gokani, SJ., Kumar, R., & Thaker, VS. (1998). Potential Role of Abscisic Acid in Cotton Fiber
and Ovule Development. JPlant Growth Regul 17:1-5
Haigler, CH., Zhang, D., & Wilkerson, CG. (2005). Biotechnological improvement of cotton
fibre maturity. Physiol Plant 124:285-294
Hequet, E., Wyatt, B., Abidi, N., & Thibodeaux, D.P. (2006). Creation of a set of reference
materials for cotton fiber maturity measurements. Text Res J 76(7): 576-586.
Huwyler, HR., Franz, G., & Meier, H. (1979). Changes in the composition of cotton fibre cell
walls during development. Planta 146:635-642
Hsieh, Y-L., Hu, X-P., & Nguyen, A. (1997). Strength and crystalline structure of developing
Acala cotton. Textile Res J67(7):529-536
www.intechopen.com
Ilharco, LM., Garcia, AR., Lopez, da Silva J., & Vieira Ferreira, LF. (1997), Infrared approach
to the study of adsorption on cellulose: influence of cellulose crystallinity on the
adsorption of benzophenone. Langmuir 13:4126-4132
Liang, C-Y., & Marchessault, RH. (1959). Infrared spectra of crystalline polysaccharides. II.
Maltby, D., Carpita, NC., Montezinos, D., Kulow, C., & Delmer, DP. (1979). β-1,3-glucan in
Native celluloses in the region from 640 to 1700 cm-1. JPolym Sci XXXIX:269-278
developing cotton fibers, structure, localization, and relationship of synthesis to

that of secondary wall cellulose. Plant Physiol 63:1158-1164
McCann, MC., Chen, L., Roberts, L., Kemsley, EK., Sene, C., Carpita, NC., Stacey, NJ., &
Wilson, RH. (1997). Infrared microspectroscopy: sampling heterogeneity in plant
cell wall composition and architecture. Physiol Plant 100:729-738
McCann, MC., Hammouri, N., Wilson, R., Belton, P., & Roberts, K. (1992). Fourier transform
infrared microspectroscopy is a new way to look at plant cell walls. Plant Physiol
100:1940-1947
McCann, MC., Defernez, M., Urbanowicz, BR., Tewari, JC., Langewisch, T., Olek, A., Wells,
B., Wilson, RH., & Carpita, NC. (2007). Neural network analyses of infrared spectra
for classifying cell wall architectures. Plant Physiol 143:1314-1326
McCann, MC., Stacey, NJ., Wilson, R., & Roberts, K. (1993). Orientation of macromolecules
in the walls of elongating carrot cells. JCell Sci 106:1347-1356
Meinert, MC., & Delmer, DP. (1977). Changes in biochemical composition of the cell wall of
the cotton fiber during development. Plant Physiol. 59:1088-1097
Montalvo, J G.& Von Hoven, T. (2004). Analysis of Cotton. In Near-Infrared Spectroscopy in
Agriculture. Agronomy Monograph no. 44. 2004. Crop Science Society of America.
Muller, LL. & Jacks TJ. (1975). Rapid chemical dehydration of samples for electron
microscopic examinations. JHistochem Cytochem 23(2):107-110
Rajasekaran, K., Muir, AJ., Ingber, BF., & French, AD. (2006). A dehydration method for
immature or wet cotton fibers for light and electron microscopy. Textile Res. J.
76(6):514-518
Ramey, H.H. (1982) Estimating quality components of natural fibers by Near-infrared
reflectance: Part 1: Cotton fiber cross-sectional area and specific surface. Text Res J,
52, 20-25
Timpa, JD. & Triplett, BA. (1993). Analysis of cell-wall polymers during cotton fiber
development. Planta 189:101-108
Tokumoto, H., Wakabayashi, K., Kamisaka, S., & Hoson, T. (2002). Changes in the sugar
composition and molecular mass distribution of matrix polysaccharides during
cotton fiber development. Plant Cell Physiol 43(4):411-418
Salmén, L., Åkerholm, M., & Hinterstoisser, B. (2005). Two-dimensional Fourier transform
infrared spectroscopy applied to cellulose and paper. In: Dumitriu S (ed)
Polysaccharides, structural diversity and functional versatility, 2nd edn. Marcel
Dekker, New York, pp159-187
Schwanninger, M., Rodrigues, JC., Pereira, H., & Hinterstoisser, B. (2004). Effects of short-
time vibratory ball milling on the shape of FT-IR spectra of wood and cellulose. Vib
Spectrosc 36:23-40
Séné, CFB., McCann, MC., Wilson, RH., & Grinter, R. (1994). Fourier-transform Raman and
Fourier-transform infrared spectroscopy an investigation of five higher plant cell
walls and their components. Plant Physiol 106:1623-1631
www.intechopen.com
USDA, International Calibration Cotton Standards Program, USDA-AMS, Cotton Program,

Memphis, TN, March 1999.
Wilkins, TA., & Jernstedt, JA. (1999). Molecular genetics of developing cotton fibers. In: In:
Basra AS (Ed) Cotton fibers, developmental biology, quality improvement, and
textile processing, edn. Food Products Press, New York, pp 231-269
Yong, W., Link, B., O’Malley, R., Tewari, J., Hunter, CT., Lu, C-An., Li, X., Bleecker, AB.,
Koch, KE., McCann, MC., McCarty, DR., Patterson, SE., Reiter, W-D., Staiger, C.,
Thomas, SR., Vermerris, W., & Carpita, NC. (2005). Genomics of plant cell wall
biogenesis. Planta 221: 747-751
www.intechopen.com
6
Medical Image Registration

with Fourier basis Functions
Roberto A. Isoardi1, Amílcar R. Osorio2 and Germán Mato3
1Fundación Escuela de Medicina Nuclear, Mendoza
2FundaciónCentro Diagnóstico Nuclear, Buenos Aires
3Grupo Física Estadística, Centro Atómico Bariloche, Bariloche
Argentina
1. Introduction
Registration is one of the most interesting, yet challenging computer-aided tasks in medical
image processing, aimed at bringing two or more data sets into spatial and/or temporal
correlation. If the represented data are medical images, there are countless situations where
it is of interest to attain such correlation, as it has become routine practice in many
diagnostic and image-guided therapeutic procedures.
For example, one of the most frequent clinical applications is to align two scans of a given
patient, e.g. for easy identification of equivalent structures on both registered images,
follow-up of disease, etc. Another use of image registration is commonplace during brain
activation studies, where several functional Magnetic Resonance (fMRI) or Positron
Emission Tomography (PET) scans are repeated on the same or different subjects while
receiving sensorial or cognitive stimulations. In order to perform a statistical analysis on the
brain images, it is often necessary to register all data sets with respect to a brain atlas. In this
way, both statistical power and signal-to-noise ratio are enhanced in what is known as intra-
modality registration (Friston et al, 2006).
One of the most exciting applications is inter-modality registration, e.g. correlating PET and
MR (Magnetic Resonance) scans of a same subject, or PET vs. CT (Computed Tomography).
PET imaging provides distinctive functional and metabolic information, but lacks the high
anatomical resolution which is in turn provided by conventional MR or CT. Thus PET/CT
registration is synergic, since it facilitates the location of malignancies in their anatomical
context. Moreover, this modality combination is often useful for Radiotherapy Planning
(Townsend, 2008).
An automatic image registration algorithm normally includes a floating image to be aligned
to the coordinate system of a reference image. To do this, a spatial transformation function –
containing a number of parameters-, must be proposed and applied to the former data set.
The parameters are chosen in such a way that a proposed similarity measure between both
data sets is optimized. This measure may be –for example-, the Normalized Mutual
Information (NMI) or the Cross Correlation Coefficient.
Methods which apply rigid-body transformations perform reasonably well for registering
images of the head (brain) and extremity portions. In a 3D space, a rigid-body
transformation involves only six parameters, e.g. three translations (Δx, Δy, Δz) and three
www.intechopen.com
rotation angles (θx, θy, θz). A natural extension of the rigid approach is the use of affine
transformations that include scaling and shearing (Maintz & Viergever, 1998).
However, those simple transformations often yield unacceptable results when dealing with
deformable body regions such as thorax and abdomen. Even with a careful repositioning on
the scanner table, the body region of interest may displace between different scans with
respect to the scanner coordinate system, due to various reasons, which may include
voluntary and involuntary motion (e.g. respiration, cardiac function), misplacement errors,
variations in morphology and physiology, etc.
Several non-rigid or deformable methods have been widely explored to treat this problem
(Crum et al, 2004)(Rueckert & Aljabar, 2010). This is an open field of research, both for its
complexity and situations of interest. Unlike rigid models, deformable registration is
particularly difficult to validate with clinical images, where internal fiducial markers are not
available. Besides, the optimization of the objective function may be non-unique or even
physically meaningless.
Nevertheless, the main complexity of a deformable system lies on its many degrees of
freedom. In an extreme situation, it would be necessary to propose a 3D independent
transformation for each image voxel. In this case, the number of transformations would be
three times the number of voxels. To treat this problem a set of approximations is
implemented, otherwise the optimization would be very costly in such a parameter space.
Most methods apply a rigid transformation as a first approximation, and then improve
registration with a non-rigid approach with a small set of basis functions.
These functions form the basis for a vector space, where any function can be represented as
a linear combination of those basis functions. They can be polynomial {1, t, t2}, just like
those used to build splines (Bookstein 1989). For instance, spline basis functions are used
when fiducial markers are available or with physical model-based algorithms (e.g. elastic
deformations, viscous fluids, etc) (D’Agostino et al, 2003).
{sin(nπx), cos(nπx)}, where n ≥ 0 is the order of the basis function. In this chapter we present
One particular kind of functions that make up an orthonormal basis are trigonometric
a systematic analysis for the deformable registration problem using trigonometric Fourier
basis functions (Ashburner & Friston, 1999). With these functions, the typical size of the
deformation field can be easily controlled, ranging from long to short wavelengths. By
combining this approach with a volume subdivision scheme, we expect to apply a small set
of basis functions at a given stage of the algorithm. In principle, given that Fourier functions
make a complete set, any arbitrary deformation may be approximated only by increasing
the number of basis functions.
In our study, we analyze intra-modality registration (CT-CT, MR-MR, PET-PET) in 3D. For
each modality, we determine the optimum number of coefficients (transformation order) for
the basis functions and the number of subvolumes to attain a satisfactory registration within
a reasonable computing time.
As a similarity measure, we calculate the Normalized Mutual Information (NMI) for
different transformation orders and applying the algorithm in the thoraco-abdominal
region. Each clinical volume data set was artificially deformed using a known displacement
field, in order to simulate an inhalation expansion, and then co-registered to its original
counterpart, taken as the reference image. In order to evaluate its clinical usefulness, we
also apply this method to co-register two scans of the same subject, each one acquired on
different dates. In a previous work, we established and tried this methodology in 2D (Osorio
www.intechopen.com
Medical Image Registration with Fourier basis Functions 119
et al, 2007). In this chapter, we extend it to 3D, including parameter optimization analysis
(Osorio et al, 2010).
2. The image registration process

The aim of a registration procedure is to attain spatial alignment of one or more images –
which we shall call “floating” (B), with respect to another one taken as the “reference”
image (A). In a very general framework, given image A with coordinates rA, and image B
with coordinates rB, we search the transformation:
T : rB → rA
such that: T(rB) = rA and maxT[S(A,B)]

where S(A,B) is a similarity measure between A and B, usually represented by an objective
function. For most cases, the similarity measure is critical for the success and general
performance of the registration process. Its choice depends on the use of extrinsic or
intrinsic image properties (e.g. fiducial markers or pixel intensity values), body region,
modalities, matrix dimensions, etc. If the registration algorithm is iterative, another essential
component is the optimizer for function S, which necessarily involves multiple evaluations in
a multidimensional parameter space (Hill & Batchelor, 2001). The process is complete when
a termination criterion is met.
Transformation T describes a spatial mapping from rB to rA. In terms of elasticity,
transformations may be rigid, affine, projective or curve (van den Elsen et al, 1993). The
simplest transformation is the rigid one, requiring only 6 parameters in 3D (3 translations +
3 rotation angles). An affine transformation keeps parallelism between lines, introducing
scaling and shearing (12 parameters in 3D). After a general projective transformation, line
straightness is conserved, though parallelism is lost. Finally, a curved transformation
produces a deformation that may be arbitrarily complex, though some constraints may
apply to preserve smoothness and topology (e.g. each voxel should keep the same
neighbours after applying the deforming transformation). As introduced in the previous
section, the rigid model provides satisfactory results when the body region is limited to the
head and some limb portions. To account for non-rigid displacements that may happen in
the body, we adopt the scheme described in the next section.
3. Fourier-based registration: theoretical grounds

We have devised the following algorithm for non-rigid registration of image volumes A and
B, each consisting of a stack of tomographic slices:
1. Pre-processing: segmentation, resampling, filtering.
2. An initial registration is carried out by applying an affine transformation
(rigid+scaling), B ->A, optimizing the Normalized Mutual Information as a similarity
measure (to be described later on).
3. B is divided into k sub-volumes Br (k=8 in 3D).
4. An independent transformation is applied to each portion Br , with a rigid and a non-
5. The global transformation Tglobal=S (T1 ∪ T2...∪ Tk ) is obtained after assembling the
rigid component.
transformations Tr .
www.intechopen.com
Fig. 1. Flowchart of the whole registration algorithm

Fig. 1 shows a diagram of this algorithm. After the initial affine approach in steps 2 and 4,
the non-rigid stage is performed by a n-order Fourier expansion along each dimension:
Tx ( r ) = x + ∑ ⎡ aijkϕijk ( r ) + bijkψ ijk ( r )⎤

⎣ ⎦
n
(1)
i , j ,k =1
Ty ( r ) = y + ∑ ⎡cijkϕijk ( r ) + dijkψ ijk ( r )⎤

⎣ ⎦
n
(2)
i , j ,k =1
Tz ( r ) = z + ∑ ⎡ eijkϕijk ( r) + f ijkψ ijk ( r)⎤

⎣ ⎦
n
(3)
i , j , k =1
where:
⎛ π ix ⎞ ⎛ π jy ⎞ ⎛ π kz ⎞
ϕ ijk ( r ) = sin ⎜ ⎟ sin ⎜ ⎟ sin ⎜ ⎟,
⎝ X ⎠ ⎝ Y ⎠ ⎝ Z ⎠
www.intechopen.com
and:
⎛ π ix ⎞ ⎛ π jy ⎞ ⎛ π kz ⎞
ψ ijk ( r ) = co s ⎜ ⎟ co s ⎜ Y ⎟ co s ⎜ Z ⎟ .
⎝ X ⎠ ⎝ ⎠ ⎝ ⎠
initial condition is randomly chosen with a Gaussian distribution N(0,σ2).

X,Y,Z are the image dimensions and a, b, c, d, e, f the coefficients to find up to order n. The
With this approach, functions ϕijk represent the transformation subset that keeps the volume
boundary invariant, whereas functions ψijk represent the transformations with null gradient
on that boundary.
4. Subdivision scheme
Once the rigid transformations are applied, the independent subvolumes were assembled
using quaternion interpolation (Walimbe et al, 2004). Since there is no standard
interpolation method adopted for non-rigid transformations, in this work we propose the
strategy shown on Fig. 2. It consists of a hierarchical scheme where each volume is divided
into 8 equal subvolumes, this process being repeated s times. In what follows, we shall refer
to the variable s as the subdivision number. In this way we build up a global transformation
which is smooth, continuous and differentiable. A detailed description of this strategy is
developed in the Appendix of Osorio et al (2010).
Fig. 2. a) Sub-volume assembling method. b) subvolumes before and c) after assembling.

Figures show a 2D section of the 3D volume
5. Similarity measure and optimization

The metric adopted for this project is the Normalized Mutual Information, widely explored
in the literature (Pluim et al, 2003), (Maes et al, 1997), (Studholme et al, 1999):
www.intechopen.com
∑ pi log( pi ) + ∑ p j log( p j )
NMI =
∑ pij log( pij )
i j
(4)
ij
where pi , pj are the marginal probability distributions and pij is the joint distribution, using
the partial volume method as interpolator (Chen et al, 2003).
The extended Downhill Simplex method was chosen as optimizer of the objective function
(Press et al, 1992), (Zagrodsky et al, 2001). The program terminates when the change in the
NMI is less than 10-4 between consecutive iterations.
6. Image data
Three tomographic modalities were studied: CT, MR and FDG1-PET of thorax and abdomen.
For each one of those, we carried out a systematic analysis of the performance of intra-
modality registration. CT studies were acquired with a HiSpeed scanner (GE, Milwaukee,
USA) (matrix size: 512x512x47, voxel size: 0.7x0.7x7 mm3). MRI studies were performed
with a Signa Advantage 0.5 scanner (GE, Milwaukee, USA) (matrix size: 256x256x24) (voxel
size 1.7x1.7x9 mm3). The PET scanner used was a Quest 250 (UGM, Philadelphia, USA)
(matrix size: 128x128x50, voxel size: 2x2x4 mm3). All studies were completed at the
Fundación Escuela de Medicina Nuclear (Mendoza, Argentina).
With the purpose of evaluating algorithm registration performance in a systematic way and
for different initial conditions, we devised the following strategy. Each selected data set was
slightly deformed using TPS (Thin-Plate Splines) (Bookstein, 1989)(Rohr et al, 2001) with a
regular grid of 432 control points and average displacement of 27 mm (max. 40 mm),
simulating a moderate thorax expansion during inhalation. This artificially deformed
volume was selected as the floating image to be registered to the reference image (original,
non-deformed version). This procedure was repeated for each modality, in order to find the
parameters which maximize the NMI in a reasonable computing time.
Once the optimal Fourier order and subdivision number were found, we applied the
algorithm to co-register pairs of thoraco-abdominal studies of a selected subject, which were
acquired on different dates. In this way we had two data sets corresponding to the same
body region, but displaying so evident anatomical discrepancies that would justify the
application of a non-rigid model.
7. Error estimation
The error measure was chosen as the absolute average displacement over the whole image
volume:
ε= ∑ T (ui ) − vi
1 N
(5)
N i =1
where ui and vi are the voxel coordinates for the floating and reference images, respectively,
and N is the total number of voxels.
1 FDG: 18 F-labelled FluoroDeoxiGlucose
www.intechopen.com
8. Registration of TPS-deformed image pairs

Fig. 3 shows the results corresponding to CT-CT registration after artificial deformation of
the floating image. Shown are overlays of initial rigid registration and Fourier registration
for transaxial slices. Coronal and sagittal overlays are shown in Fig. 4, which provide a
visual assessment of our non-rigid approach. Bone structures remain unaltered during the
process, which is consistent with a real situation.
(c) NMI (circles) and error ε (squares) vs. Fourier order for s=1. (d) NMI and error ε vs.
Fig. 3. CT vs. CT deformed with TPS (a) Rigid registration. (b) Fourier registration, n=2, s=4
number of subdivisions for n=2. (e) Computing time for Fig. 3c. (f) Computing time for
Fig. 3d
www.intechopen.com
Fig. 4. Coronal and sagittal views of CT vs. TPS-deformed CT before (a) and after (b) Fourier
registration
The graphs in Fig. 3 show the NMI, ε and computing time as a function of both the
transformation order n and the subdivision number s. For all graphs, n,s=0 and n,s=0.5
mind that non-rigid registration comes into play when n ≥ 1. As expected, the NMI
refer to initial centre-of-mass alignment and rigid registration, respectively. Let us keep in
increases and ε decreases at higher n. Looking for further optimization in the parameter
space in a reasonable computing time, we set n=2 and plotted NMI vs. s. In this way,
maximum NMI was attained at (n=2, s=4) for CT-CT registration. Each registration cycle
was run 12 times, starting with different deformations, and the graphs show average
values with their corresponding error bars. The computing time shows an exponential
increase for n>1.
higher n values, error ε hardly decreases at the expense of long execution times. For the
From the above results, we chose n=2 as the most appropriate transformation order, since at
same reason, a suitable subdivision number was chosen at s=3,4. These results were also
confirmed by visual inspection of the registered data sets.
www.intechopen.com
s=4 (c) NMI (circles) and error ε (squares) vs. Fourier order for s=1. (d) NMI and error ε vs.
Fig. 5. MRI vs. MRI deformed with TPS (a) Rigid registration. (b) Fourier registration, n=2,
Fig. 5d
Similar results were obtained for the MR-MR situation (Fig. 5). As for PET vs. PET, since it is
a smaller dataset, no further improvement is obtained for s>3, when few voxels remain in
each subvolume to calculate the NMI with enough statistical power.
www.intechopen.com
s=4 (c) NMI (circles) and error ε (squares) vs. Fourier order for s=1. (d) NMI and error ε vs.
Fig. 6. PET vs. PET deformed with TPS (a) Rigid registration. (b) Fourier registration, n=2,
Fig. 6d
9. Clinical studies
Once the optimum parameters were chosen from the previous analysis with TPS-deformed
datasets, we selected three pairs of clinical studies to evaluate our method for intra-subject,
intra-modality registration in real situations.
For each modality, the first study was taken as reference to which the second study was
registered. Fig. 7 shows the results for all modalities, which were qualified by expert
www.intechopen.com
radiologists as “acceptable” by visual inspection. The reader should keep in mind that in
this case we lack from a “gold standard” for error calculation for obvious reasons, so this
evaluation cannot be considered a validation test.
10. Error estimation with digital fiducial markers

Our choice for error estimation as described in Section 7, includes large portions of the field-
of-view that are not relevant to registration process itself, such as the background that
surrounds the body region of interest. To check if this has a significant impact on error
calculation, 10 digital spheres were inserted in selected anatomical locations inside the CT,
MR and PET volume data sets, so as to simulate easily identifiable internal fiducials. Next
we applied a TPS deformation on such “marked” volumes and carry out the registration
cycle as described in Section 8.
Fig. 7. Intra-modality registration of two clinical scans (same subject, different sessions).
From left to right: Rigid registration, Fourier registration, Image difference after rigid
registration, Image difference after Fourier registration. From top to bottom: CT, MRI and
PET
After registration of both data sets, we measured the distance between center-of-masses for
homologous spheres. Finally, the Mean Root Square Error (MRSE) was calculated using Eq.
5. In this case the summation is limited to the available sphere centers. In order to control
registration, two different sphere sizes were studied: big (8 mm φ) and small (2 mm φ). The
for the effect of the additional deformation induced by the adjustments of the spheres on the
results are summarized in Table I, showing that the effect of the sphere size on the global
error is quite small. This means that the spheres themselves are not affecting the registration
process in a significative fashion.
We can see from Table 1 that the typical errors are between 1.5 mm and 2.5 mm, which are
slightly smaller than the average displacements calculated for the whole volumes, as shown
www.intechopen.com
in Figs. 3d, 5d and 6d. These results are of the same order as the ones found in recent
approaches (Andronache et al, 2008), (Sohn et al, 2008) , using more complex methods such
as a combination of cross-correlation and mutual information in the former and local
matching of anatomical features in the latter. The computing time (as seen in Figs, 3, 5 and
6) is between 400 secs and 1000 secs, which is also of the same order as the time reported in
the former reference.
2 mm φ Sphere 8 mm φ Sphere Error whole image

MRSE (CT) 2.55 mm 2.44 mm 2.62 mm
MRSE (MRI) 1.29 mm 1.55 mm 2.99 mm
MRSE (PET) 1.98 mm 1.58 mm 2.80 mm
Table 1. Mean Root Square Error (MRSE) for the three modalities (CT, MRI, PET). The errors
in the second and third columns are evaluated after the registration with images that
includes a set of 6 to 10 spheres. Sphere diameters are 2 mm and 8 mm. The error in the last
column was evaluated over the whole image. The parameters of the algorithm were n=2, s=4
for CT and n=2, s=3 for MRI and PET
11. Discussion and conclusions

Our systematic analysis shows how the similarity measure (NMI) behaves with both the
order of the transformation n and the number of subdivisions s. For intra-modality
for n ≥ 3 (s=1). For n=2 the NMI is only slightly inferior, however the computing time
registration, and for the three modalities studied, we found that maximum NMI is attained
becomes an order of magnitude longer if calculation is performed up to n = 3 (Figs. 3e, 4e,

5e), as the number of Fourier parameters to be optimized is 6n3 in 3D (Eqs. 1-3).
accompanied by a decrease in error ε (Figs. 3d-5d). The improvement in registration quality

As expected, the similarity measure increases with the number of subdivisions s,
results using s=3 versus s=4. For PET-PET, NMI and ε does not get better for s>3, because
was also confirmed by visual assessment for both CT-CT and MRI-MRI by comparing
slight improvements in the NMI and ε were recorded for n>3 and s>3, but at a very high
further subdivision results in sub-volumes with too sparse data for that modality. Only
computational cost and providing negligible visual improvements. One issue of concern is
that rigid structures such as bone in CT should remain so after registration. Since the
characteristic size of the deformation applied is greater than typical bone structures, they do
not deform noticeably (Fig. 4).
Regarding calculation time, by setting optimal parameters (n=2, s=3), and using an ordinary
computer2, the time for co-registering two CT volumes (matrix dimensions 512×512×47) is ~
600 secs. (~ 200 secs. if s=2). For MRI-MRI, somewhat shorter times were measured, whereas
for PET-PET, execution took slightly over 400 secs. (Figs. 3f,5f,6f). Such computation times
were attained without any specific optimization technique. Let us note that the algorithm
leads naturally to parallelization because the subvolumes can be processed independently.
In that way, the computing time can be substantially reduced.
In principle, the use of Fourier basis functions allows arbitrary deformations on any given
image volume; the combination of this method with a subdivision scheme allows to
2 Core 2 Duo, 3GHz, no multithreading (2008).
www.intechopen.com
accommodate small image portions in a progressive fashion, without affecting the rest of the
image data (Walimbe et al, 2004), (Likar & Pernus, 2001). Obviously, as the subdivision
creates small subvolumes with fewer and fewer pixels, similarity measures like the NMI or
the CCC become affected in their performance (Andronache et al, 2008).
In general, the proposed registration method rendered acceptable results for small and
moderate deformations (~ 25mm). A preliminary study suggests that it is fairly robust, even
in the presence of Gaussian noise (Osorio et al, 2007). We evaluated its performance using
clinical images after deformation with Thin-Plate Splines, as well as image pairs
corresponding to different scan sessions for a same subject. The selected studies were
thoracic and abdominal scans for three common tomographic modalities. Obviously, not
only organ deformations and displacements may come about between scan sessions, but
also significant variations in anatomy and function, due to normal or pathological
conditions. In these cases, the outcome of any non-rigid registration method offers an
approximation whose usefulness must be assessed for each particular situation.
12. References
Andronache A.; von Siebenthal M.; Székely G. & Cattin P. (2008). Non-rigid registration of
multi-modal images using both mutual information and cross-correlation. Med.
Imag. Anal. 12, pp. 3-15.
Ashburner J. & Friston K. (1999). Nonlinear spatial normalization using basis functions.
Hum Brain Mapp. Vol. 7, June 1999, pp. 254–266.
Bookstein F. (1989). Principal warps: Thin-plate splines and decomposition of deformations.
IEEE transactions on Pattern Analysis and Machine Intelligence, 11:567-585.
Chen M. & Varshney P. K. (2003). Mutual information-based CT-MR brain image
registration using generalized partial volume joint histogram estimation. IEEE
Trans. Med. Imaging 22, pp. 1111-1119.
Crum W.; Hartkens T. & Hill D. (2004). Non-rigid image registration: theory and practice. Br
J Radiol. Vol. 77: S140–S153.
D'Agostino E.; Maes F.; Vandermeulen D. & Suetens P. (2003). A viscous fluid model for
multimodal non-rigid registration using mutual information. Med Image Anal. Vol.
7, December 2003, pp. 565-575.
Friston K.; Ashburner J.; Kiebel S.; Nichols T. & Penny W. (2006). Statistical Parametric
Mapping: The Analysis of Functional Brain Images. ISBN-10: 0123725607 Academic
Press, London.
Hill D. & Batchelor P. (2001). Registration Methodology: Concepts and Algorithms. En:
Medical Image Registration (eds: Hajnal J.V. et al), CRC Press.
Likar B. & Pernus F. (2001). A Hierarchical Approach to Elastic Registration Based on
Mutual Information. Image Vis. Comput. 19, pp. 33-44.
Maes F.; Collignon A.; Vandermeulen D.; Marchal G. & Suetens P. (1997). Multimodality
image registration by maximization of mutual information. IEEE Trans. Med.
Imaging 16, pp. 187-198.
Maintz J. & Viergever M. (1998). A survey of medical image registration. Med Image Anal.
Vol. 2, March 1998, pp. 1–36.
Osorio A.; Isoardi, R. & Mato G. (2007). Non-rigid registration of tomographic images using
Fourier transforms. Journal of Physics: Conference Series. Vol. 90, Issue 1, 2007, 012058
www.intechopen.com
Osorio A.; Isoardi, R. & Mato G. (2010). Deformable CT registration using Fourier basis
functions in 3D. In: XXIII Brazilian Symposium on Computer Graphics and Image
Processing (SIBGRAPI ’2010), Gramado, IEEE Computer Society Press (in press,
September 2010).
Pluim J. P. W.; Maintz J. B. A. & Viergever M. A. (2003). Mutual-Information-Based
Registration of Medical Images: A Survey. IEEE Trans. Med. Imaging 22, pp. 986-
1004 .
Press W. H.; Flannery B. P.; Teukolsky S. A. & Vetterling W. T. (1992). Numerical Recipes in
C: The Art of Scientific Computing, 2nd ed. (Cambridge University Press,
Cambridge).
Rohr K.; Stiehl H. S.; Buzug T. M.; Weese J. & Kuhn M. H. (2001). Landmark-Based Elastic
Registration Using Approximating Thin-Plate Splines. IEEE Trans. Med. Imaging 20,
pp. 526-534.
Rueckert D. & Aljabar P. (2010). Non-rigid registration of medical images: theory, methods
and applications. IEEE Signal Proc. Mag. Vol. 27, July 2010, pp. 113-119.
Sohn M.; Birkner M.; Chi Y.; Wang J.; Yan D.; Berger B. & Alber M. (2008). Model-
independent, multimodality deformable image registration by local matching of
anatomical features and minimization of elastic energy. Med. Phys. 35, pp. 866-878.
Studholme C.; Hill D. L. G. & Hawkes D. J. (1999). An overlap invariant entropy measure of
3D medical image alignment. Pattern Recognit. 32, pp. 71-86.
Townsend D. (2008). Multimodality imaging of structure and function, Phys. Med. Biol., Vol.
53, February 2008: R1-R39.
van den Elsen P.; Pol E-J. D. & Viergever M. (1993). Medical image matching-a review with
classification. IEEE Eng. Med. Biol. Mag. Vol. 12, March 1993, pp. 26-39.
Walimbe V. S.; Zagrodsky V.; Raja S.; Bybel B.; Kanvinde M. & Shekhar R. (2004). Elastic
registration of three-dimensional whole body CT and PET images by quaternion-
based interpolation of multiple piecewise linear rigid-body registrations,
Proceedings of the SPIE (SPIE, San Diego, CA, 2004), pp. 1191–1228.
Zagrodsky V.; Shekhar R. & Cornhill J. F. (2001). Multi-function extension of simplex
optimisation method for mutual information based registration of ultrasound
volumes. Proc. SPIE 4322, pp. 508-515.
www.intechopen.com
7
Fourier Transform Infrared Microspectroscopy

for Cancer Diagnostic of
C6 Glioma on Animal Model
Beljebbar A. and Manfait M.
Unité MéDIAN, MEDyC, CNRS UMR 6237, IFR 53, UFR de Pharmacie,
Université de Reims Champagne-Ardenne
France
1. Introduction
Cancer is one of the leading causes of death in the world. Currently, the gold standard in
most cancer diagnosis is histopathological evaluation, which involves the removal of tissue
biopsies and examination by pathologists. This process includes tissue staining and
morphological pattern recognition. During tissue transformation, it is expected that
substantial modifications occur at molecular level before visible morphological changes
become apparent. Histological examination requires extensive human observations to
recognize both the constitutive histologic entities and the pathologic state. Early detection of
cancer is the most important factor in the prevention of cancer and a guarantee in most cases
of an effective treatment and in some cases for a complete cure. Moreover, the histological
assessment must be performed while surgery is ongoing to determine whether the tissue
has been removed or spared. For example, brain tissue cannot be removed with a large
safety distance from the tumor. The resection of brain tumors is strongly limited to the
border of the tumor. The main objective is a detection of the source of the pathological
variation at molecular level in order to further understand the molecular carcinogenic
process in a range of cancers. Indeed, tumor tissues are mostly heterogeneous in nature, and
this heterogeneity further depends on the stage of disease and its aggressiveness. The
emergence of a novel technique, complementary to histopathology and
immunohistochemistry, can thus help in the early diagnostic of tissue transformation during
carcinogenesis.
Fourier-transform infrared microspectroscopy (FTIRM) has emerged as a powerful tool to
study molecular structure and structural interactions in biological systems. When this
technique is applied to tissues, the resulting spectra is composed of characteristic absorption
bands originating from all infrared-active vibrational modes of biological macromolecules
present in the tissue, such as proteins, lipids, and nucleic acids (Parker, 1971). Each of these
molecules provides a unique absorption spectral pattern named fingerprint through the
entire infrared spectrum. This property offers a way to identify the molecule type
(qualitative analysis) and the amount or quantity of this molecule in the sample
(quantitative analysis) (beljebbar et al., 2008). This method can be used as a diagnostic tool,
complementary to histopathology or immunochemistry (Fernandez et al., 2005). As the
www.intechopen.com
image contrast is based on the intrinsic vibrational signature of the tissue components,
spectral images does not require the use of added dyes or labelling methods for
visualization of different chemical components in the sample (Bates, 1976). Indeed, FT-IRM
imaging combined a high spatially resolved morphological and biochemical information
that offer a number of advantages for ex-vivo assessment of tissue and aid the
histopathologist in the identification and classification of subtle biochemical changes related
to carcinogenesis (Petibois & Déléris, 2006; Cohenford & Rigas, 1998; Kneipp et al., 2000;
Yano et al., 2000). With the fast image acquisition provided by modern mid-infrared
imaging systems, it is now envisaged to analyze tumor biopsies in delays compatible with
surgery (Levin & Bhargava, 2005). Other advantages of this method are that it is objective
and provides reproducible diagnosis, minimize inter-observer variability. Indeed, IR
spectroscopy can detect and monitor characteristic changes in molecular composition and
structure that accompany transformation from normal to cancerous state (Afanasyeva et al.,
1998, Diem et al., 1999, Franck et al., 1998). The identification and quantification of these
specific molecular changes within tissues can provide diagnostic information for aiding in
early detection of diseases and their optimized treatment. Correlations of morphologic and
biochemical tissue differences could be used to identify variations that occur between
healthy and diseased tissues. The development of clinical protocols for the routine
examination of tissue histology or of localized tumors using IR microspectroscopic methods
has been largely used in medical diagnostics to identify neoplasia in breast, (Ci et al, 1999)
cervix, (Wong et al, 1991) colon, (Rigas et al, 1990) lung, (Yano et al, 2000) stomach, (Li et al.,
2005) and glioma. (Krafft , 2006, 2007; Amharref et al, 2006; Beljebbar et al., 2008).
Infrared spectra contain many overlapping bands and so data interpretation cannot be made
by simple visual inspection and alternative approaches are needed. Because of the high
complexity of the FTIR spectra obtained from tissues, multivariate statistical methods are
required to extract biochemical information related to tissue. This would permit to
objectively differentiate distinct tissue structures and for identifying origin that gave rise to
the specific tissue pathology. These methods have had a major impact on the quantitative
and qualitative analysis of infrared spectral data. They have been shown to improve
analysis precision, accuracy, reliability, and applicability for infrared spectral analyses
relative to the more conventional univariate methods of data analysis. Rather than
attempting to find and use only an isolated spectral feature in the analysis of spectral data,
multivariate methods derive their power from the simultaneous use of multiple intensities
(i.e. multiple variables) in each spectrum (Mourant, et al., 2003). During the last decade, it
has been recognized that FT-IR, in combination with the appropriate multivariate analysis
strategies, has considerable potential as a metabolic fingerprinting tool for the rapid
detection and diagnosis of disease or dysfunction (Goodacre et al, 2004; Diem et al., 1999).
Multivariate imaging techniques including Unsupervised Hierarchical Cluster Analysis
(UHCA) (Jackson et al., 1998; Mohlenhoff et al., 2005), K-means clustering (Lasch et al., 2004;
Zhang et al., 2003), Principal Components Analysis (PCA) (Lasch and Naumann, 1998),
Linear Discriminant Analysis (LDA) (Mansfield et al, 1999), Fuzzy C-means clustering
(Lasch et al., 2004; Mansfield et al., 1997) and neural networks (Lasch & Naumann, 1998)
have proven to be invaluable in the identification of spectral groups or "clusters" which can
be directly compared to stained tissue sections. In multivariate methods, the information of
the entire spectrum can be utilized for the analysis. The high correlation of spectral clusters
with anatomical and histopathological features has been conclusively demonstrated for a
number of different tissue types including.
www.intechopen.com
Fourier Transform Infrared Microspectroscopy for
Cancer Diagnostic of C6 Glioma on Animal Model 133
The aim of this chapter was the monitoring and interpretation of molecular changes
associated to C6 glioma growth and invasion by micro-FTIR imaging. Micro-FTIR maps
were recorded on normal brain tissue and on glioma growth after injection of C6 glioma cell
suspension in brain parenchyma. Multivariate statistical analysis were used to i) identify the
molecular changes associated with the development of the glioma tumor ii) definition of the
tumor and peritumoral margins, iii) grading of malignancy and prognosis based on the
presence of necrosis. We have investigated the spatial distribution of molecular changes
associated with C6 glioma progression using integrated intensity ratios of some specific
bands associated to lipids and proteins in order to determine spectroscopic markers to
successfully monitor the changes in the molecular composition associated to C6 glioma
progression.
2. FTIR characterization of normal brain tissues and C6 glioma progression

using cluster analysis
To investigate the potential of FTIR spectroscopy for clinical application, experimental
animals are necessary. Rat C6 glioma cells are an experimental cell line that when injected
into neonatal Wistar rats grow into an intracerebral tumor with pathological similarities to
human glioblastoma (GBM) (Auer et al., 1981). This glioblastoma model was used in a
variety of studies related to brain tumor biology including tumor growth (Nagano et al.,
1993; San-Galli et al., 1989), invasion (Nagano et al., 1993; Bernstein et al., 1991; Chicoine &
Silbergeld, 1995), and evaluation of the therapeutic efficacy of cancer treatments (Barth
1998). The glioma tumors were obtained by injection of C6 glioma cells suspension in brain
parenchyma as described elsewhere (Grobben et al., 2002). All animals with implanted C6
cells developed tumors with reproducible localization and size around the site of injection.
These groups were sacrificed after 5, 7, 9, 12, 15, 19, days post-implantation (PI). After brain
excision, tissue samples were snap-frozen by immersion in methyl-butane cooled down in
liquid nitrogen and stored at - 80°C. Two adjacent sections were cut from each sample using
a cryomicrotome. One section, 10 µm thick, was placed onto infrared transparent calcium
fluoride (CaF2) slides for infrared imaging. The second section, 7 µm thick, was placed on a
microscope glass slide and stained with hematoxylin and eosin (H&E) for histopathological
image. Spectra were collected using an FTIR imaging system (SPOTLIGHT, Perkin-Elmer,
France) coupled to a FTIR spectrometer (Spectrum 300, Perkin-Elmer, France). This system is
equipped with a liquid N2 cooled Mercury-Cadmium-Telluride MCT line detector
comprised of 16 pixel elements. The microscope was equipped with a movable, software-
controlled x, y stage. In this study, FTIR images were collected from selected sites with a
spatial resolution of 25 µm/pixel, in transmission mode, in the 4000–720 cm-1 range, with a
final spectral resolution of 4 cm-1, and 16 scans per pixel. After atmospheric correction, data
were cut to high wavenumber fingerprint region (2600 to 3700 cm-1), converted to their first
derivative, and smoothed using a seven point Savitzky-Golay algorithm in order to
minimise the influence of background scatter in the spectra (Savitzky & Golay, 1964). The
resulting spectra were then normalized using a Standard Normal Variate (SNV) procedure.
A multivariate statistical analysis (Principal Component Analysis (PCA) and K-Means
(KM)) was performed on this dataset. K-means clustering was performed on these principal
component scores. Pseudo-color maps based on cluster analysis were then constructed by
assigning a color to each spectral cluster. The cluster spectra were calculated by averaging
absorbance spectra associated to each group and used for the interpretation of the chemical
www.intechopen.com
or biochemical differences between clusters. All data measured on normal brain and tumor
development (from 7 to 21 days growth) were pooled in one dataset, processed at the same
time and the results were displayed as pseudo-color maps with the same color scale. In this
way, we can easily determine all their common and discriminating features by comparing
their infrared maps. The rat brains analysed before 5 days after tumor implantation did not
show any visible changes in the brain tissue (data not shown). Indeed, all animals died
beyond 25 days post-implantation and therefore no data are available beyond this time
point.
Histological and FTIR analysis revealed that the tumor size dependent on the period
elapsed after injection of glioma cells. Fig. 1 displays FTIR pseudo-color maps of brain tissue
and their histopathological images. 13 clusters describing both normal brain and cancer
features were extracted and pseudo FTIR maps were constructed with the same color scale.
Different clusters in the FTIR images were correlated with features of the histopathological
images. White color represents the area where no tissue was present. In the pseudo-color
map obtained from glioma tumor obtained 7 days after C6 cells injection (fig. 1B). This
pseudo-color map displays some normal structures associated to white matter and grey
matter. Cluster 4 described white matter corresponding to corpus callosum (CC) and
commissura anterior (CA). CC and CA present important lipid content due to high myelin
level in these structures involved in communication within and between hemispheres.
Several clusters (6, 1, 2, 8, and 11) describe the transition from white matter to gray matter
(cortex). In fact, the CC is the largest white matter structure in the brain, consisting mainly
of interhemispheric fibers. Gray matter is distributed at the surface of the cerebral
hemispheres (cerebral cortex) and of the cerebellum (cerebellar cortex). It is predominantly
composed of neuron cell bodies and unmyelinated axons. Comparison between pseudo-
color maps and the histopathology images (Fig. 1A and 1B) shows that the FT-IRM image
provides more information on the cortex than standard histopathology. In fact, six layers
were identified from the cortex in the pseudocolor FTIR map (fig. 1B), whereas, H&E
staining did not allowed to discriminate between these layers in the cortex (fig. 2A). Luxol
fast blue (LFB) staining was then used to visualize myelin distribution into brain tissues and
to map particular sections within the cortex. In fig. 2B, LFB staining shows a gradation color
density between brain structures. In stained preparations, myelin is intensely blue, so that
white matter is well differentiated from gray matter. In fact, large fiber tracts like CC and
some bundles in CP can be easily recognized. However, even with higher magnification it
was very difficult to distinguish between all cortex layers because this staining is mainly
restricted to fibers.
Thus, the individual cell type cannot be recognized within the tissue. This LFB staining was
then combined with cresyl violet (CV) coloration to stain not only rough endoplasmic
reticulum (Nissl substance) but chromatin and nucleoli as well. This coloration was used in
our study to visualize all different cell types present in the cortex (Fig. 2C). This pseudo-
color FTIR map (fig. 2D) was correlated with that obtained with LFB-CV staining (fig. 2C).
With this coloration, six different layers were then identified in the cortex instead of five
layers in the FT-IRM map (fig. 2D). On the other hand, when multivariate statistical analysis
was applied only on FT-IRM data measured from normal brain tissue we were able to
distinguish between five cortex layers. The cortex consists of a thin layer of gray substance
which covers the two hemispheres and whose thickness varies between 2 and 4 mm.
According to the cortical histological organization, one distinguishes 6 layers numbered
from I to VI from top to bottom (Burwell, 2001). The outermost molecular layer (I)
www.intechopen.com
www.intechopen.com
Fig. 1. Photomicrographs (H&E staining) associated to glioma progression (A, C, E, G, I) from respec
days post-implantation. All data measured on several brain tissues during tumor evolution were po
at the same time to extract all features describing both normal and tumor tissues. Data were cut into
3700 cm-1) and cluster analysis was carried out on the first derivative spectra (to enhance the resoluti
K-means was calculated the cluster-membership of spectra by assigning each color to one class. Pseu
corresponding histological images were constructed on tumor progression (maps B, D, F, H, and J) r
Fig. 2. Photomicrography of a brain tissue section (A) H & E staining, (B) stained with LFB;
in this image CC and CP appears more lightly stained, (C) stained with LFB-CV, pointing
out the cortical layers of the rat cortex. Pseudo-color FTIR map (D) based on K-means cluster
analysis applied only on FTIR data measured from normal brain structures
containing non-specific fibers, corresponds to the yellow cluster. The external granular
layer (II) is a rather dense layer composed of small cells. The external pyramidal layer (III)
contains pyramidal cells, frequently in row formation. These three layers were encoded by
cluster 11. The internal granular layer (IV) is usually a thin layer with cells similar to those
in the external granular layer (cluster 8). The ganglionic layer (V) contains, in most areas,
pyramidal cells that were fewer in number but larger in size than those in the external
pyramidal layer (clusters 1 and 2). The fusiform layer (VI) consists of irregular fusiform cells
www.intechopen.com
whose axons enter the adjacent white matter known as the CC (cluster 1 and 6). By
correlating FT-IRM spectral maps with histopathology (H&E, LFB, and LFB-CV) of the
adjacent tissue sections, we highlighted the potential of FT-IRM to identify the morphologic
origin that gave rise to the specific spectral features found in this study. In fact, with standard
staining (H&E), we were not able to discriminate between the different cortex layers. On the
other hand, FT-IRM pseudo-color maps were clearly similar to LFB-CV staining for
visualizing myelin distribution in healthy brain tissues (white and gray matters).
Fig. 1B exhibit a particular structure (cluster 7) located in the caudate putamen (CP)
corresponding to C6 cells injection site. This feature was associated to C6 cell growth.
Histological interpretation however showed small abnormalities in rats sacrificed seven
days after implantation of C6 cells (fig. 1A). Indeed, this abnormality is characterized by a
diffuse structure within the brain parenchyma. The tumor cells appear scattered in the form
of cells or grouped in clusters in the vicinity of the blood vessels. At day 9, the viable tumor
started to be visible in the implantation site, destroys the corpus callosum and grew into all
the cortical layers (fig. 1D). The tumor area was encoded by clusters 12 and 13. On the other
hand, clusters 5 and 7 were detected around the tumor. These clusters were associated to the
proliferative and invasive character of glioma tumors. In fact, in our previous study,
immunohistochemical Ki-67 and MT1-MMP staining were used to visualize the proliferative
and invasive activities of glioma and were clearly correlated with the cluster that encoded
the surrounding tumor area (Amharref et al., 2007). Histopathological staining confirms this
FTIR result (fig. 1C), In fact, a Large focus of invasion (6mm) were well separated from the
surrounding brain tissue. Tumor was hypercellular with cellular and nuclear pleomorphism
and mitotic figures observed. From day 12, the tumor is fairly large and deeply situated
within the cortex with massive infiltration into the brain tissue (fig. 1F, 1H, 1J). Most of
cerebral cortex is destroyed by tumor tissue. At day 15, we observe the appearance of
clusters 9 and 5 associated to the formation of necrosis and perinecrosis (fig. 1H).
Histological image displayed oedematous zones as well as zones of necrosis with a
pseudopalissading cells (> 8mm) (fig. 1G). Around necrotic zones, tumor shows a large cell
density with an increase of the proliferation of endothelial cells and haemorrhagic zones.
This necrotic zone increased until day 19 post-implantation (fig. 1J). Cluster 5 observed in
the border of the necrotic zone seems to correlate with the pseudopalisading formation.
Cluster 9 correlates to the center of the necrosis (full necrosis) of the tumor. The presence of
necrosis is important for grading tumors and is often linked to a poorer clinical prognosis
(Barker et al., 1996). Indeed, the most characteristic finding of glioblastoma is the necrotic
foci surrounded by tumor cells (Kleihues et al., 1993). Pallisading cells delineate the foci of
necrosis and lymphocytic infiltration, with the occasional formation of edema fluid (Auer et
al., 1981). At day 19, we observed the transformation of the structure of the tumor (fig. 1J). In
fact, cluster 12 was replaced partially by cluster 13. On the other hand, the tumor occupied
almost the entire hemisphere as visualized on the tissue of section (fig. 1I).
Fig. 3 shows class average spectra associated to normal brain structures and tumor
development. The lipids spectra contain a large proportion of methyl, methylene and
carbonyl bands in the region 2600-3700 cm-1. The bands at 2852 cm-1 and 2924 cm-1 were due
to the symmetric CH2 stretching mode of the membrane lipid which is directly related to the
lipid acyl, primarily saturated, chains. The band at 2956 cm-1 is associated to asymmetric
CH3 stretching mode of the methyl group and less contribution of proteins. The =C–H
stretching bands due to unsaturated acyl chains are found at 3014 cm−1. The bands at 3292
cm-1 and 3080 cm-1 were linked to the Amide A (mainly N–H stretching) and Amide B (N–H
www.intechopen.com
2924
3292
2956
2852
2874
3080
3014
1
2
3
Absorbance (A.U.)
5
6
7
8
9
10
11
12
13
2600 2800 3000 3200 3400 3600
Wavenumber (cm-1)
Fig. 3. Representative cluster mean FTIR spectra extracted from pseudocolor maps. Cluster
averaged spectra were obtained by meaning absorbance spectra associated to each group. 13
models describing normal and glioma brain development. Each cluster averaged spectrum
assigned to one class was plotted with the same color than in pseudo-color map
stretching) of proteins respectively. The band at 2874 cm−1 was related mainly to the
symmetric stretch of proteins. The changes in frequencies, intensities and band shapes of
these bands may provide further information about the structural changes associated with
malignancy. In the malignant tissues, i) the intensity of the symmetric CH2 band decreased
compared to the corresponding bands in normal brain structures (white matter and grey
matter) and ii) the band ratio 3292/2852 cm-1 was higher in the glioma tumor and reduced in
the normal brain tissues. Indeed, the intensity ratio 2924/2956 cm-1 was higher in the spectra
associated to white matter and decreased from grey matter to C6 glioma tumors. Band 2874
cm-1 became better resolved in the spectra of malignant tissues because of diminished
intensities of CH2 bands in this region. Therefore, the largest variances from spectra to
spectra in IR spectroscopic maps of normal tissue were assigned to spectral contributions of
lipids reflecting cell differentiation. This result confirms that the development of tumor was
characterized by a reduction in total lipid content. This reduction was also observed in the
invasive area (clusters 5 and 7), which is composed of healthy and tumoral cells. This result
is in agreement with those obtained in brain diseases (Krafft et al., 2006; Kneipp et al., 2000).
www.intechopen.com
This lipid reduction in malignant tissues could be related to the fast growth of tumor cells
which need more energy (Wang et al., 2003). Indeed, it is known that, in developing brain
tumors, structural and functional cell changes take place in which lipids play a crucial role.
Yet, qualitative and quantitative aspects of lipid changes in brain tumors of different degree
of malignancy are still the subject of numerous studies (Steiner et al., 2003; Krafft et al., 2006;
Campanella, 1992).
13
Tumor and necrosis
12
Tumor, invasion, and
necrosis structures 9
10
Cluster averaged spectra

Peritumor 5
3
White matter
Normal brain 4
features Intermediate layers 6
Grey matter 2
11
External layers
8
20 18 16 14 12 10 8 6 4 2 0
Heterogeneity
Fig. 4. Dendrogram obtained from hierarchical cluster analysis on spectral cluster averages
associated to different tissue types. Heterogeneity represents the discriminating distance
given by arbitrary units (au)
To distinguish between normal, tumor, and necrotic brain structures, cluster averaged
spectra obtained from pseudo-color maps were input in the hierarchical cluster analysis
using Ward’s clustering algorithm and the square Euclidian distance measure. The result, as
shown on the dendrogram in Fig. 4, showed a clear distinction between all normal and
tumor brain structures. Indeed, the class related to normal brain structures was divided to
two sub-clusters associated to white matter (clusters 4 corresponding to higher lipid
content) and the second sub-cluster related to grey matter described by clusters 1 and 6
(intermediate cortex layers) and clusters 2, 8, and 11 (external cortex layers). We were also
able to discriminate between glioma tumor (clusters 12, and 13) and necrosis (cluster 9) from
peri-necrosis (clusters 3, 5, 7, and 10) in the second groups. In general, an increase in
www.intechopen.com
malignancy is accompanied by a reduction in total lipids that involves all main classes of
lipids found in plasma membranes (Campanella, 1992). Changes in lipid and in
phospholipids contents, as seen in glioblastoma as compared with adjacent tissue, could
indicate an evolution in the undergoing pathological process. This loss of lipids, correlated
with demyelination, observed in different disorders, could be used as a spectroscopic
marker. Increased levels of cholesterol esters (cholesterol oleate and linoleate) have also
been reported in glioma tissue (Koljenovic et al., 2002). Koljenoviv et al. demonstrated that
the difference between meningioma and dura is mainly related to lipids, cholesterol
linoleate and linoleic acid levels. Steiner et al. studied the discriminating constituents
between normal and tumoral tissues (astrocytoma and glioblastoma) by infrared
spectroscopy (Steiner et al., 2003). They demonstrated that changes mainly arise from
differences in lipid constituents. The potential use of lipid measurements for judging the
stage, and hence the prognosis, of low grade tumors is suggested by the apparent gradual
increase in lipid content over time. This increase, believed to be associated with necrosis,
could thus be used in low grade tumors as an early marker of disease prior to the patient
becoming symptomatic (Krafft et al., 2006; Koljenovic et al., 2005; Steiner et al., 2003). During
tumor development, tissue composition and concentration of lipids decreased. Kraft et al.
have investigated the lipid content of the white matter of human brain tissue using near
infrared Raman spectroscopy (Krafft et al., 2005). They reported that the brain lipids can be
divided into three principal classes: neutral lipids, phospholipids and sphingolipids.
3. Distribution of molecular changes in brain constituents associated to

tumor growth and invasion
We have investigated the spatial distribution of molecular changes associated to C6 glioma
progression using FT-IR micro-spectro-imaging in order to better understand the tissue
transformation during carcinogenesis. Integrated intensity ratios bands were used in the
region of 3700–2800 cm−1 to characterize differences between healthy and pathological brain.
Maps of absorbance intensity ratios of bands in the region from 3000 to 2800 cm−1 due to
CH2 and CH3 stretching vibrations (mainly due to membrane lipid which is directly related
to the lipid acyl, primarily saturated) and those due to CH2 and NH stretching vibrations
(due to the protein-to-lipid ratios) were calculated and pseudo color maps scores were
constructed (Fig. 5). These integrated absorbance intensity were correlated to molecular
changes associated to tissue transformation. The comparison between the pseudo color
scores maps and histological image shows that high scores described the white matter
structures such as CC and CA (colorbar). These scores decreased in the grey matter and
become null in the tumor tissues. This intensity ratio was correlated to myelin content. In
fact, white matter presents important lipid content due to high myelin level. The
concentration of myelin decreased from CC to cortex. In general, an increase in malignancy
is accompanied by a reduction in total lipids that involves all main classes of lipids found in
plasma membranes (Krafft et al., 2006). This loss of lipids was correlated with
demyelination, observed in the scores maps associated to tumor. We have study the
distribution of the intensity ratio 3292/2852 cm-1 corresponds to NH stretching vibration in
proteins vs symmetric CH2 stretching mode of the membrane lipid. The results show that
high score value was related to tumor at days 7, 12, 15, and 19 days PI. These scores
decreased from invasion zone to normal brain structures (white matter and grey matter).
The common underlying effects of malignancy are changes in the constituents that lead to
www.intechopen.com
www.intechopen.com
Fig. 5. Biochemical distribution of the changes in the molecular composition of tissues through the
Maps of absorbance intensity ratios of bands in the region from 3000 to 2800 cm−1 due to CH2 and C
(mainly due to membrane lipid which is directly related to the lipid acyl, primarily saturated) and
stretching vibrations (due to the protein-to-lipid ratio) were constructed and used to identify whic
be more potential indicators of such variations between normal and tumor development
an altered metabolism and biochemical composition in the malignant tissues. This study
confirms the absorption intensity ratios were correlated with the histological state of the
sample. Such a result may indicate the change of lipid and protein distribution in
pathological tissues with respect to healthy one. There are some papers considering the
pathological state of tissues by analysis of the changes in band intensities in the region from
3700 to 2800 cm−1. Eckel et al. have analyzed breast cancer tissues (Eckel et al., 2001). They
have considered the 3300 cm−1 band which corresponds to NH stretching vibration in
proteins as a good criterion for distinguishing between non-cancerous and cancerous parts
of tissue. Liu et al. applied IR absorption of human breast tissues in vitro and studied the
CH stretching region of lipids and NH absorption region from 3600 to 2700 cm−1 (Liu et al.,
2006). Gazi et al. used Synchrotron-FTIRM imaging to study prostate cancer cells and
analyzed the distribution of lipid absorption intensities (CH2 and CH3) in the region from
3000 to 2800 cm−1 (Gazi et al., 2005) Krafft et al. have evaluated the usefulness of the lipid-to-
protein ratio (2850/1655 cm−1) as a spectroscopic marker to discriminate between normal
and tumor tissue, as well as between low- and high-grade glioma tissues. They
demonstrated that this ratio is maximal for normal brain tissue and decreases with the
progression of the disease (Krafft et al., 2007). The intensity ratio 3014/2874 cm-1 was higher
in the invasion zone and peritumor part of the necrosis and decreased in the tumor and
normal brain tissues. This intensity ratio can be used to provide an objective method to
delineate lesion margins that would reduce unnecessary tissue excisions.
4. Conclusion
This study demonstrated that FT-IRM imaging, with high spatially resolved morphological
and biochemical information can be used as a diagnostic tool, complementary to
histopathology in order to understand the molecular changes associated to C6 glioma
progression. Cluster analysis allowed investigation of C6 glioma progression (from day 7 to
day 19 post implantation). Different clusters in the FTIR images were correlated with
features of the histopathological images such as white and grey matters, tumor, peritumor,
and necrosis. Our results showed that 7 days after tumor implantation, FTIR investigations
displayed a very small abnormal zone associated with the proliferation of C6 cells in the
caudate putamen. From this day, rats developed solid and well-circumscribed tumors.
Additionally, we have identified one peculiar structure all around the tumour. This
structure was attributed to infiltrative events, such as peritumoral oedema observed during
tumor development. The presence of necrotic areas was visible from day 15. In fact, the
grade of malignancy and prognosis, in particular GBM multiforme, is based on the presence
of necrosis. By combining intensity ratios of specific bands with imaging technique, we were
able to take in account the variance due to the heterogeneity of brain tissues. We have
monitored the changes in the intensity ratios of specific bands related to lipid and proteins.
Our results reported that by correlating pseudocolor map scores with H&E staining it was
possible to screen histological changes associated with tissue transformation. In fact, the
integrated intensity in the 2800 to 3000 cm-1 spectral region described normal brain
structures such as white matter (CC and CA) and some cortex layers (grey matter). The
intensity decreased in the tumor tissues. Intensity ratio 3292/2852 cm-1 allowed the
identification of tumor part of the tissue. The invasion zone was described by the 3014/2874
cm-1 ratio. These constituents can be used as spectroscopic markers for early detection of
tissue abnormality and discrimination among normal, invasion, tumor and necrosis.
www.intechopen.com
5. Acknowledgement
This work was supported by La Ligue de la Marne contre le Cancer, France.
6. References
Afanasyeva, N. I., Kolyakov, S.F., Artjushenko, S.G., Sokolov, V.V. & Frank, G.A. (1998).
Minimally invasive and ex vivo diagnostics of breast cancer tissues by fiber optic
evanescent wave Fourier transform IR (FEW-FT-IR) spectroscopy. Proceedings of
SPIE - The International Society for Optical Engineering, pp. 140-146. 0277786X, San
Jose, CA,
Amharref, N., Beljebbar, A., Dukic, S., Venteo, L., Schneider, L., Pluot, M., Vistelle, R.. &
Manfait, M. (2006). Brain tissue characterisation by infrared imaging in a rat glioma
model. Biochim. Biophys. Acta, Vol. 1758, No. 7, (07.2006) 892-899, ISSN 00052736
Amharref, N., Beljebbar, A., Dukic, S., Venteo, L., Schneider, L. Pluot, M., & Manfait, M.
(2007). Discriminating healthy from tumor and necrosis tissue in rat brain tissue
samples by Raman spectral imaging. Biochim. Biophys. Acta, Vol. 1768, No. 10, (10
2007) 2605-2615, ISSN 00052736.
Auer, R.N., Del Maestro, R.F., Anderson, R. (1981). A simple and reproducible experimental
in vivo glioma model. Can J Neurol Sci., Vol. 8, No. 4, (11 1981), pp. 325-331, ISSN
0317-1671.
Barker, F.G., Davis, R.L., Chang, S.M., & Prados, M.D. (1996). Necrosis as a prognostic factor
in glioblastoma multiforme. Cancer, Vol. 77, No. 6, (03 1996), 1161-1166, ISSN
0008543X.
Barth, R.F. (1998). Rat brain tumor models in experimental neuro-oncology: the 9L, C6, T9,
F98, RG2 (D74), RT-2 and CNS-1 gliomas. Neurooncol., Vol. 36, No. 1, (01 1998), pp.
91-102, ISSN 0167-594X.
Bates, J.B. (1976). Fourier transform infrared spectroscopy. Science, Vol. 9, No. 191, 4222,
(01.1976), pp. 31-37, ISSN 00368075
Beljebbar, A., Amharref, N., Lévèques, A., Dukic, S., Venteo, L., Schneider, L., Pluot, M., &
Manfait, M. (2008). Modeling and quantifying biochemical changes in C6 tumor
gliomas by Fourier transform infrared imaging. Anal. Chem., Vol. 15, No 80,
(11.2008), pp. 8406-8415, ISSN 00032700
Bernstein, J.J., Laws, E.R. Jr, Levine, K.V., Wood, L.R., Tadvalkar, G., Goldberg, W.J. (1991).
C6 glioma-astrocytoma cell and fetal astrocyte migration into artificial basement
membrane: a permissive substrate for neural tumors but not fetal astrocytes.
Neurosurgery, Vol. 28, No. 5, (5 1991), pp. 652-658, ISSN 0148-396X.
Burwell, R.D. (2001). Borders and cytoarchitecture of the perirhinal and postrhinal cortices
in the rat, J. comparative neurol.. Vol. 437, No. 1, (08 2001), pp. 17-41, ISSN 00219967.
Campanella, R., (1992). Membrane lipids modifications in human gliomas of different
degree of malignancy. J. Neurosurg. Sci., vol. 36, No. 1, (1992), pp. 11–25, ISSN
00264881.
Cherayil, G.D.& Scaria, K.S. ( 1970). Thin-layer chromatography of tissue lipids without
extraction. J. Lipid Res., Vol. 11, No. 4, (07 1970), pp. 378-381, ISSN 00222275.
Chicoine, M.R., & Silbergeld, D.L. (1995). Invading C6 glioma cells maintaining
tumorigenicity. J Neurosurg Vol. 83, No. 4, (10 1995), pp. 665-671, ISSN 0022-3085.
www.intechopen.com
Ci, Y., Gao, T., Feng, J., & Guo Z. (1999). Fourier transform infrared spectroscopic
characterization of human breast tissue: implications for breast cancer diagnosis.
Applied Spectrosc., Vol. 53, No. 3, (03.1999), pp. 312–315, ISSN 00037028
Cohenford, M.A. & Rigas B. (1998). Cytologically normal cells from neoplastic cervical
samples display extensive structural abnormalities on IR spectroscopy: implications
for tumor biology. Proc. Natl. Acad. Sci. U S A., Vol. 95, No. 26, (12.1998), pp. 15327–
15332, ISSN 00278424
Diem, M., Boydston-White, S. & Chiriboga, L. (1999). Infrared spectroscopy of cells and
tissues: shining light onto a novel subject. Appl. Spectrosc., Vol. 53, No. 4, pp. 148-
161, ISSN 00037028
Eckel, R., Huo, H., Guan, H.W., Hu, X., Che X., & Huang W.D. (2001).Characteristic infrared
spectroscopic patterns in the protein bands of human breast cancer tissue. Vib.
Spectrosc., Vol. 27, No. 2,, (12 2001), pp. 165–173, ISSN 09242031.
Fernandez, D.C., Bhargava, R., Hewitt, S.M.& Levin, I.W. (2005). Infrared spectroscopic
imaging for histopathologic recognition. Nature Biotechnology, Vol. 23, No. 4, pp.
469-474, ISSN 10870156
Franck, P., Nabet, P. & Dousset, B. (1998). Applications of infrared spectroscopy to medical
biology. Cell. Mol. Biol., Vol. 44, No. 2, (03.1998), pp. 273- 275, ISSN 01455680
Gazi, E., Dwyer, J., Lockyer, N.P., Miyan, J., Gardner, P., Hart, C.A., Brown M.D., & Clarke
N.W. (2005). A study of cytokinetic and motile prostate cancer cells using
synchrotron-based FTIR microspectroscopic imaging. Vib. Spectrosc., Vol. 38, No. 1-
2, (07 2005), pp. 193–201, ISSN 09242031.
Goodacre, R., Vaidyanathan, S., Dunn, W.B., Harrigan, G.G. & Kell, D.B. (2004).
Metabolomics by numbers: Acquiring and understanding global metabolite data,
Trends Biotechnol., Vol. 22, No. 5, (05.2004), pp. 245–252, ISSN 01677799.
Grobben, B, De Deyn, P.P., & Slegers, H. (2002). Rat C6 glioma as experimental model
system for the study of glioblastoma growth and invasion. Cell Tissue Res., Vol. 310,
No. 3,, pp. 257-270, ISSN 0302766X.
Jackson, M., Ramjiawan, B., Hewko, M., & Mantsch, H.H.. (1998). Infrared microscopic
functional group mapping and spectral clustering analysis of hypercholesteramic
rabbit liver. Cell. Mol. Biol., Vol. 44, No. 1, (02.1998), pp. 89–98, ISSN 01455680
Kleihues, P., Burger, P.C ., Scheithauer, B.W. (1993). The new WHO classification of brain
tumours. Brain Pathol., Vol. 3, No. 3, (03 1993), pp. 255-68, ISSN 10156305.
Kneipp, J., Lasch, P., Baldauf, E., Beekes, M., & D. Naumann, (2000). Detection of
pathological molecular alterations in scrapie-infected hamster brain by Fourier
transform infrared spectroscopy. Biochim. Biophys. Acta, Vol. 1501, No. 2-3, (06
2000), pp. 189-199, ISSN 09254439.
Koljenovic, S., Choo-Smith, L.P., Bakker Schut, T.C., Kros, J.M., van den Berge, H.J., &
Puppels, G. (2002). Discriminating vital tumor from necrotic tissue in human
glioblastoma tissue samples by Raman spectroscopy. Lab. Invest., Vol. 82, No. 100,
(10 2002) pp. 1265-1277, ISSN 00236837.
Koljenovic, S., Schut, T.B., Vincent, A., Kros, J.M., Puppels, G.J. (2005). Detection of
meningioma in dura mater by Raman spectroscopy. Anal. Chem., Vol. 77, No. 24, (12
2005), pp. 7958-7965, ISSN 00032700.
www.intechopen.com
Krafft, C., Neudert, L., Simat, T., & Salzer, R. (2005). Near infrared Raman spectra of human
brain lipids. Spectrochim. Acta (Part A), Vol. 61, No. 7, (05 2005), pp. 1529-1535, ISSN
13861425.
Krafft, C., Sobottka, S.B., Geiger, K.D., Schackert, G., & Salzer, R. (2007). Classification of
malignant gliomas by infrared spectroscopic imaging and linear discriminant
analysis. Anal. Bioanal. Chem., Vol. 387, No. 5, (03 2007), pp. 1669-1677, ISSN
16182642
Krafft, C., Sobottka, S.B., Schackert, G., & Salzer R. (2006). Raman and infrared spectroscopic
mapping of human primary intracranial tumors: A comparative study. J. Raman
Spectrosc., Vol. 37, No. 1-3, (01 2006), pp. 367-375, ISSN 03770486.
Krafft, C., Thummler, K., Sobottka, S.B., Schackert, G., & Salzer, R. (2006). Classification of
malignant gliomas by infrared spectroscopy and linear discriminant analysis.
Biopolymers, Vol. 82, No. 4, (07 2006), pp. 301-305, ISSN 00063525
Lasch, P & Naumann D. (1998). FT-IR microspectroscopic imaging of human carcinoma thin
sections based on pattern recognition techniques. Cell. Mol. Biol., Vol. 44, No. 1,
(02.1998), pp. 189–202, ISSN 01455680.
Lasch, P., Haensch, W., Naumann, D., & Diem, M. (2004). Imaging of colorectal
adenocarcinoma using FT-IR microspectroscopy and cluster analysis. Biochim
Biophys Acta., Vol. 1688, No. 2, (03.2004), pp. 176–186, ISSN 09254439.
Laws, E.R. Jr & Shaffrey, M.E. (1999). The inherent invasiveness of cerebral gliomas:
Implications for clinical management. Int. J. Dev. Neurosci., Vol. 17, No. 5-6, (08
1999), pp. 413-420, ISSN 07365748
Levin, I.W. & Bhargava, R. (2005). Fourier transform infrared vibrational spectroscopic
imaging: Integrating microscopy and molecular recognition. Annu. Rev. Phys.
Chem., Vol. 56, pp. 429-474, ISSN 0066426X
Li, Q.B., Sun, X.J., Xu, Y.Z., Yang, L.M., Zhang, Y.F., Weng, S.F., Shi, J.S & Wu, J.G (2005).
Diagnosis of gastric inflammation and malignancy in endoscopic biopsies based on
Fourier transform infrared spectroscopy. Clin. Chem., vol. 51, No. 2, (02.2005), pp
346–350, ISSN 00099147
Liu, C., Zhang, Y., Yan, X., Zhang, X., Li, C., Yang W., & Shi, D. (2006). SR-FTIR
spectroscopic preliminary findings of non-cancerous. J. Lumin., Vol. 119–120,
SPEC, (07 2006), pp. 132–136, ISSN 00222313.
Mansfield, J.R., McIntosh, K., Crowson, A.N., Mantsch, H.H. & Jackson, M. (1999). LDA-
guided search engine for the non-subjective analysis of infrared microspectroscopic
maps. Appl. Spectrosc., Vol. 53, No. 11, (11.1999), pp. 1323–1330, ISSN 00037028.
Mansfield, J.R., Sowa, M.G., Scarth, G.B., Somorjai, R.L. & Mantsch, H.H. (1997). Analysis of
spectroscopic imaging data by fuzzy C-means clustering. Anal. Chem., Vol. 69, No.
16, (08.1997), pp. 3370–3374, ISSN 00032700.
Mohlenhoff, B., Romeo, M., Diem, M., & Wood, B.R. (2005). Mie-type scattering and non-
Beer-Lambert absorption behaviour of human cells in infrared microspectroscopy.
Biophys J., Vol. 88, No. 5, (05.2005), pp. 3635–3640, ISSN 00063495.
Mourant, J. R., Yamada, Y.R. Carpenter, S., Dominique, L.R. & Freyer, J.P. (2003). FTIR
spectroscopy demonstrates biochemical differences in mammalian cell cultures at
different growth stages. Biophys. J., Vol. 85, No. 3, (09.2003) 1938–1947, ISSN
00063495.
www.intechopen.com
Nagano, N., Sasaki, H., Aoyagi, M., & Hirakawa, K. (1993). Invasion of experimental rat
brain tumor: early morphological changes following microinjection of C6 glioma
cells. Acta Neuropathol., Vol. 86, No. 2, (1993), pp. 117-125, ISSN 00016322.
Parker, F. S., (1971). Application of infrared spectroscopy in biochemistry, biology and medicine
Plenum, New York.
Petibois, C. & Déléris, G. (2006). Chemical mapping of tumor progression by FT-IR imaging:
towards molecular histopathology. Trends in Biotechnology, Vol. 24, No. 10,
(08.2006), pp. 455-462, ISSN 01677799
Rigas, B., Morgello, S., Goldman, I., & Wong, P. (1990). Human colorectal cancers display
abnormal Fourier-transform infrared spectra. Proc. Natl. Acad. Sci. USA, Vol. 87, No.
20, (10.1990), pp. 8140–8144, ISSN 00278424
San-Galli, F, Vrignaud, P., Robert, J., Coindre, J.M., Cohadon, F. (1989). Assessment of the
experimental model of transplanted C6 glioblastoma in Wistar rats. J. Neurooncol.,
Vol. 7, No. 3, (9 1989), pp. 299-304, ISSN 0167594X.
Savitzky, A. & Golay, M.J.E. (1964). Smoothing and differentiation of data by simplified least
squares procedures. Anal. Chem., Vol. 36, No. 8, (08 1964), pp. 1627-1639
Steiner, A., Shaw, G., Choo-Smith, L.P., Schackert, G., Steller, W., Abuid, H.M., & Salzer, R.
(2003). Distinguishing and grading human gliomas by infrared spectroscopy.
Biopolymers, Vol. 72, No. 6, (2003), pp. 464-471, ISSN 00063525.
Wang, J.S., Shi, J.S., Xu, Y.Z., Duan, X.Y., Zhang, L., Wang, J., Yang, L.M., Weng, S.F., & Wu,
J.G. (2003). FT-IR spectroscopic analysis of normal and cancerous tissues of
esophagus. World J. Gastroenterol., Vol. 9, No. 9, (09 2003), pp. 1897-1899, ISSN
10079327.
Wong, P.T., Wong, R.K., Caputo, T.A., Godwin, T.A. & Rigas, B. (1991). Infrared
spectroscopy of exfoliated human cervical cells: Evidence of extensive structural
changes during carcinogenesis. Proc. Natl. Acad. Sci. USA, Vol. 88, No. 24, (12.1991),
pp. 10988–10992, ISSN 00278424
Yano, K., Ohoshima, S., Gotou, Y., Kumaido, K., Moriguchi, T.& Katayama H. (2000). Direct
measurement of human lung cancerous and noncancerous tissues by Fourier
transform infrared microscopy: can an infrared microscope be used as a clinical
tool? Anal. Biochem. Vol. 287, No. 2, (12.2000), pp. 218–225, ISSN 00032697
Zhang, L., Small, G.W., Haka, A.S., Kidder, L.H., & Lewis, E.N. (2003). Classification of
Fourier transform infrared microspectroscopic imaging data of human breast cells
by cluster analysis and artificial neural networks. Appl. Spectrosc., Vol. 57, No. 1,
(01.2003), pp. 14–22, ISSN 00037028.
www.intechopen.com
8
Statistical Analysis for Automatic Identification

of Ovarian Cancer Protein-Biomarkers Based on
Fast Fourier Transform Infrared Spectroscopy
Marcano A.1,2,3, D. Pokrajac1,4, A. Lazarevic1,
M. Smith3, Y. Markushin1,2 and N. Melikechi1,2,3
1Centerfor Research and Education in Optical Sciences and Applications,
2Center for Applied Optics for Space Science,
3Department of Physics and Pre-Engineering
4Department of Computer and Information Sciences,
Delaware State University, 1200 North Dupont Highway, Dover, DE 19901

United States of America
1. Introduction
Fast Fourier transform infrared (FTIR) spectroscopy has been used widely for the study of
vibrations of protein molecules (Arrondo et al. 1993; Goormaghtigh et al. 1994; Haris &
Chapman, 1994; Siebert, 1995; Barth & Zscherp, 2002; Petibois et al., 2006). Valuable
information can be obtained of the secondary structure of the protein since peak positions
and their relative amplitude are affected by the number of hydrogen bridges that sustain
this secondary structure (Byler & Sussi, 1989; Fabian et al., 2001; Fabian et al., 2002).
However, the spectral lines of proteins are usually broadened due to different molecular
interactions thus making the identification of the structure difficult. Furthermore,
identification of a particular protein within a complex matrix like a blood or a serum sample
based on FTIR spectra is particularly challenging. Namely, direct application of automatic
classification techniques is not a simple task, due to large numbers of attributes
(measurements at different wavenumbers). Recently, principal component analysis (PCA)
has been used as a statistical method for the feature extraction in the analysis of
spectroscopic data aimed at detection of several complex organic samples (Hybl et al., 2003;
Melikechi et al.; 2008, Lazarevic et al., 2009). In these methods, the spectroscopic data can be
represented in a three-dimensional (or arbitrary dimension) space of eigenvector projections
of the matrices corresponding to a series of experimental data measured for different
selected wavelengths (Massart et al., 2003). In this regard, each point of this space represents
a full set of spectroscopic measurements corresponding to one sample. Differences between
the spectra can be then visualized graphically as different points in the space of
eigenvectors. Linear discriminant analysis (LDA) or support vector machines (SVM), an
advanced machine learning technique, can be subsequently used for automatic observing
these differences between spectra. LDA generates linear models that separate classes based
on the assumption that class-wise distributions are multivariate Gaussian with the same
www.intechopen.com
covariance matrix (independent of the class label). SVM are classification algorithms that
automatically assign a class label to a vector of data with theoretically best generalization
(ability to predict the class outside the training data), independently of the data distribution.
SVM generate a hyperplane in the transformed feature space (a non-linear transformation
applied to the original data) such that the separation plane is as far from the data closest to it
as possible. By using non-linear transformation, the likelihood that the training data can be
separated by a hyperplane increases. By maximizing the distance between the data and the
hyperplane, we achieve the smallest complexity of the classifier and hence, according to
computational learning theory, maximize the generalization capability of the classification
model. In this study, we propose to use the output of PCA analysis as input of LDA and
SVM and to perform an automatic identification of protein molecules based on their FTIR
spectra.
We use the proposed methodology to distinguish among the fast Fourier transform infrared
(FTIR) spectra of proteins reported as possible biomarkers of ovarian cancer: monoclonal
antibodies (MAB) and antigens (AG) of ovarian cancer marker CA125, Osteopontin (OPN),
Leptin and insulin-like growth factor II (IGF2) (Mor et al., 2005; Schorge et al., 2004;
Sutphen et al., 2004). We also complete a similar study on the common protein Bovine
Serum Albumin (BSA) and human plasma samples for comparison purposes. We show that
despite the presence of broadening mechanisms and evident similarities in the FTIR spectra
of these proteins, the proposed method provides an automatic and effective identification of
the proteins with almost perfect accuracy. This statistical procedure can also be applied to
other spectroscopic methods such as fluorescence, NIR-VI absorbance spectroscopy and
laser-induced breakdown spectroscopy.
As an important application we also perform deuteration of proteins and study the
differences in the FTIR spectra introduced by this process using the PCA and LDA
methods. FTIR spectra of deuterated versions of the proteins have been used extensively
for the study of the secondary structure (Baenziger & Methot, 1995; Dave et al., 2002; Nie
et al., 2005). Deuteration occurs by simple dilution of proteins in heavy water that
contains the deuterium isotope of hydrogen (2H). We have studied in details the changes
induced by deuteration in the FTIR spectra of BSA and ovarian cancer biomarkers
referred above. We have also explored the use of temperature and ultrasound to increase
the changes. We use PCA and LDA methods to differentiate undeuterated and deuterated
versions of the same protein. We propose that these methods can be used for
identification of proteins within a matrix containing a large variety of proteins like a
blood or serum sample. Furthermore, we propose a FTIR based immunoassay that uses
the developed data analysis method and deuterated versions of the corresponding
monoclonal antibodies for detection of protein biomarkers contained in a complex matrix
like blood, plasma or serum samples.
2. Experimental method
For measuring the FTIR spectra we use an attenuated total reflection (ATR) FTIR
spectrophotometer NICOLET 6700 (Thermo Industries, Inc). Drops of the samples are
deposited over an aperture on the top of the device. This aperture connects to the surface of
a diamond prism where the total reflection occurs. Samples under study are distilled and
deionized water, heavy water (99.8% purity Deuterium oxide from Alfa Easer) and high
www.intechopen.com
Statistical Analysis for Automatic Identification of Ovarian Cancer Protein-Biomarkers
Based on Fast Fourier Transform Infrared Spectroscopy 149
purity proteins (Sigma): BSA, 15 mM saline solution of MAB to AG CA125, AG CA125,

Leptin, OPN and IGF2. Usually water masks most of the contribution from the proteins. To
before collecting data. A drop of 5 μL of the solution is deposited over the aperture of the
eliminate water peaks the samples are dried through simple evaporation of the solvent
spectrophotometer. The samples are then left to dry at room temperature during 30 minutes.
The drying process is monitored by taking spectra every 5 minutes until solvent (water of
heavy water) contribution is depleted. When the drying process is complete the spectra do
not show further changes. The dried protein sample forms a film over the aperture of
several tens of micrometers good enough for total reflection spectroscopy. The spectra are
collected with a resolution of 4 cm-1. One hundred scans are averaged for each spectrum.
The spectra show high reproducibility and a signal to noise value usually larger than 100. To
collect data for the data analysis we repeat the spectroscopy experiment 40 times for each
specimen. The deuteration of the proteins is performed using the dilution method at
different concentrations and different dilution times. For solid samples like BSA we prepare
directly a heavy water solution of the protein. For the solution samples we mix equal
volumes of D2O and the original protein solution. Deuteration can be improved by adding
additional drops to the previously dried sample. Deuteration can also be improved by
changing the temperature or using ultrasound. For this purpose we use an ultrasound
cleaner with temperature control (Fisher Scientific FS20). The temperature is monitored with
an independent thermocouple.
3. Classification methodology
We propose a statistical framework for automatic classification of the FTIR spectra of
different proteins. The framework is illustrated in figure 1.
FTIR data
Dimensionality reduction (PCA)
principal components
Classification (SVM, LDA)
classified FTIR spectra
Fig. 1. Statistical framework for automatic FTIR spectra classification

The first step in the framework corresponds to dimensionality reduction, in which we
reduce the number of frequencies in FTIR spectra using PCA. Principal components
obtained through PCA are them used as an input to the classification module, which
provides final classification of particular FTIR spectra.
www.intechopen.com
3.1 Principal component analysis (PCA)

PCA is a powerful technique for dimensionality reduction in machine learning and data
mining (Jolliffe, 2002). The central idea of PCA is to reduce the dimensionality of a data set
consisting of a large number of interrelated variables (with non-diagonal covariance matrix),
while retaining as much as possible of the variation present in the data set. This reduction is
achieved by transforming to a new set of variables, which are called the principal
components (PC). The PC are uncorrelated, and ordered in such a way that the first few
retain most of the variation present in all of the original variables.
Suppose that X is a N-dimensional matrix of k-dimensional random variables [x1; x2; xN],
and that the variances of the k random variables and the structure of the covariances or
correlations between the k variables are of interest. Assume that we intend to approximate
( )
assume that we would like to determine xˆ i = ∑ j = 1 a ij v j , such that the mean square error
the vector xi as a linear combination of m<k predetermined variables. In other words,
m
E xˆ i − x i
2
is minimized. It can be proven that the mean square error is minimized when
vi, i = 1,…,m are eigenvectors corresponding to m largest eigenvalues of the covariance
matrix of X, and when aij are principal values projections of vector xi with respect to its
mean and first m eigenvectors. The vector x̂ i contains m variables and thus it is typically
stated that m “most significant” features are extracted out of k original coordinates.
The covariance matrix C of x can be estimated as:
C=
1
N−1
XTX , (1)
Its eigenvectors (column vectors) and corresponding eigenvalues satisfy the following
condition:
C vj=λj vj , (2)
We can formally define eigenvector matrix and the diagonal matrix of eigenvalues
respectively as:
V=[v1 v2… vk] , (3)
Λ = diag ( λ1 ,..., λk ) . (4)
Therefore we can compute the coefficients aij as aij=xi vj. Computation of PCA for high-
dimensional data by definition may be very cumbersome, since it has O(k3) complexity,
where k is the number of dimensions. The reason for such computational complexity is the
requirement to compute eigenvalues and eigenvectors of a k*k matrix
C=
1
N−1
X T X . To make computation feasible, we will follow recently proposed approach
(Bishop, 2006) that can extract up to N-1 principal components with the largest eigenvalues,
when N < k.
Define C * = XX T , and let U and Λ * be respectively matrices of eigenvectors and
1
N−1
eigenvalues of C* such that:
C * U = UΛ * , (5)
www.intechopen.com
If we assume V = X T U and consider the estimated covariance matrix C, we can easily

obtain that:
CV = X T XX T U = X T C * U = X T UΛ * = VΛ * ,
1
N−1
(6)
In other words, V is the eigenvector matrix of C and Λ * is corresponding diagonal matrix of

eigenvalues. Since C* is of size N*N, using this technique would allow huge computational
savings when N<<k. Note that vectors in V are not necessary normalized (to have a unit
norm). Hence, to achieve orthonormal eigenvectors, an additional normalization step is
required.
3.2 Linear Discriminant Analysis (LDA)

LDA (Krzanowski, 1988; Seber, 1984) is a statistical technique that classifies objects by
computing the logarithm of the likelihood function (likelihood is the probability of the class
given the observed data). Here, data from each class is assumed to belong to a multivariate
Gaussian distribution. The Gaussian distributions corresponding to different classes are
assumed to have the different means but the same covariance matrix, leading to the linear
discrimination.
Formally, given the estimates of the prior probabilities pj, and means µj for each class j, and
the estimate of the covariance matrix C, the logarithmic likelihood for a sample specified by
a vector yi can be computed as
l( j) = − ln C + ln p j − ( y i − μ j ) C −1 ( y i − μ j ) j = 1,..., c ,
1 1 T
, (7)
2 2
where c is the total number of classes.

Using eq. (7), the classification of an example from a test set, specified by vector y new , is
performed according to:
⎛ ⎞
c new = arg max j l(j) = arg max j ⎜ ln p j − μ jC −1μ Tj + μ jC −1 y Tnew ⎟ .
1
⎝ ⎠
(8)
2
Hence, the separation plane between classes i and j can be described as a hyperplane:
( ) ⎛
⎝ 2
1
2
⎞
fij ( y ) = μ i C −1 − μ jC −1 y T + ⎜ ln pi − ln p j − μ i C −1μ iT + μ jC −1μ Tj ⎟ = 0 .
1
⎠
(9)
For each class, we can define a decision margin as a minimal distance between a sample
from a class and the separation planes.
Let yi,j, i =1,…,nj be row feature vectors from the training set belonging to class j and let nj
be the number of vectors in class j. We estimate the class priors, means, and the covariance
matrix as:
pj =
∑ j '= 1,…,c n j '
nj
www.intechopen.com
μj = ∑ i =1,…,n
1
j
yi , j (10)
nj
C= ∑ j '= 1,…,c n
∑ j '= 1,…,c n j ' − 1
1 C j'
j' − 1
where:
Cj = ∑ ( y i , j − μ j ) ( yi , j − μ j ) .
1
n j − 1 i = 1,…,n j
T
(11)
3.3 Support Vector Machines (SVM)

LDA provides optimal classification with linear decision boundaries if its assumption of
class-specific Gaussian distributions with identical covariances is satisfied. However, if this
assumption is not satisfied, the optimal decision boundaries could be obtained by using
SVM (Vapnik, 2000). The main idea of SVM is to construct a separation hyperplane, which
optimally separates data examples belonging to two classes, such that the minimal distance
between points and the separation hyperplane is maximized. Such constructed hyperplane
provides the best generalization of unknown examples. SVM use structural risk
minimization principle and aim to achieve zero training error while minimizing the
complexity of the model. However, if linear separation is not possible, SVM work towards
minimization of the number of misclassified examples on the training set by introducing the
slack variables and regularization. Formally, SVM learning can be represented as the
following quadratic programming problem (Bishop, 2006):
⎛1 ⎞
min ⎜ w T w + C ∑ ξ i ⎟ s.t.
N
w , ξ i ,do ⎝ 2 ⎠
(w y )
i =1
T
i + d0 ⋅ ci ≥ 1 − ξ i ,i = 1,...,N (12)
ξ i ≥ 0,i = 1,...,N.
where, w is vector defining the separation hyperplane, d0 is the intercept of the separation
hyperplane, ci∈{-1,1} is a class label of the ith example determined by attribute vector xi, ξ i is
the slack variable corresponding to the ith example, C is a preset regularization constant, and
y i= f ( xi ) is a vector representing a non-linear function of xi . In general, xi and w can be
infinitely dimensional.
represented as the optimization in dual space of Lagrangian multipliers λi. In this case, the
Using the Karush-Kuhn-Tucker (KKT) theorem (Karush, 1939), SVM learning can be
learning phase reduces to the following optimization problem:
⎛N ⎞
max ⎜⎜ ∑ λ i −∑∑ λ i λ j c i c j y iT y j ⎟⎟ s.t.
N N
⎝ i =1 ⎠
(13)
λ l =1 j=1
∑ λ i ci = 0
N
i =1
λ i ≥ 0,i = 1,...,N
λ i ≤ C,i = 1,...,N.
www.intechopen.com
An example xnew from the test set is subsequently classified according to the following
( )
equation:
c new = sign w T y new + d0 , (14)
where y new =f (xnew), which can be expressed using the Lagrangian multipliers as:
⎛ 1 ⎛1 ⎞⎞
c new = sign ⎜ ∑ λ i ci y iT y new + ⎜⎜ − ∑ λ j c ji y i y j ⎟⎟ ⎟ ,
⎜ i∴λi > 0 N s ⎝ ci j∴λ j > 0 ⎟
⎠⎠
T T
⎝
(15)
where Ns denotes the number of support vectors—points closest to the separation

hyperplane (i.e., number of non-zero Lagrangian multipliers).
Similar as in the case of LDA, with SVM we can explicitly calculate the separation planes in
the transformed space specified by:
w T y + do = 0 , (16)
where
w= ∑ λ i y i ci , (17)
i∴λ i > 0
⎛1 ⎞
d0 = ∑ ⎜ c − w y i ⎟.
1
i∴λ i > 0 ⎝ i ⎠
T
Ns
If we select features in the transformed space to be proportional to eigenvalues and
( )
eigenfunctions of a symmetric non-negative definite kernel, then, due to the Mercer’s
theorem, we can write y iT y j = K x i , x j where K is symmetric non-negative-definite
function of two vectors (Bishop, 2006). Then, due to the Mercer’s spectral theorem for non-
negative definite symmetric kernels, SVM learning and classification can be stated directly
using original (non-transformed) feature vectors as:
⎛N ⎞
max ⎜⎜ ∑ λ i −∑∑ λ i λ j ci c jK ( x i , x j ) ⎟⎟ s.t
N N
λ
⎝ i =1 l =1 j=1 ⎠
∑ λ i ci = 0
N
i =1
λ i ≥ 0,i = 1,...,N (18)

λ i ≤ C,i = 1,...,N.
⎛ 1 ⎛1 ⎞⎞
c new = sign ⎜ ∑ λ i ci K ( x i , x new ) + ⎜⎜ − ∑ λ j c ji K ( x i , x j ) ⎟⎟ ⎟ ,
⎜ i∴λi > 0 N s ⎝ ci j∴λ j > 0 ⎟
⎠⎠
T
⎝
(19)
This makes possible using implicit and infinitely dimensional transformation f. Popular
Linear kernel: K ( u , v ) = u T v ;
choices of kernel function include:
•
www.intechopen.com
• (
Polynomial kernel (p is a prespecified parameter): K ( u , v ) = 1 + u T v ; )
p
Exponential kernel (σ is a pre-specified parameter): K ( u , v ) = e

− u−v
•
1 2
2 σ2
.
The original SVM technique is designed for a two-class problem. For a multiclass problem
(i.e., c>2) we use Directed Acyclic Graph SVM (DAG-SVM) method (Platt et al., 2000)
which for a c-class problem trains c(c-1)/2 two-class support machines and the class
decision is performed based on successive elimination of classes as a result of a two-class
comparison. In comparison to one-to-rest classifiers, the application of DAG-SVM is more
practical, since it does not result in imbalanced training sets (Hsu & Lin, 2002; Jiang et al.
2005).
3.4 Classification accuracy evaluation

To validate the accuracy of the classification model on the data unseen during the
learning process, we use a four-fold cross validation, that can be described as follows
(Bishop, 2006): 1) split randomly the dataset into four subsets; 2) set aside one of the
subsets as the test set while the other three subsets are chosen to form the training set; 3)
Utilize the training set to learn the classification model and employ the test set to
evaluate the accuracy of classification on data unseen during the learning process; 3)
repeat the process four times so that each of the four subsets has a chance to be a test set;
4) use averaged results from the four classification experiments as an overall measure of
the model performance.
As a measure of performance, we utilize overall classification accuracy, the ratio of correctly
classified samples for all classes versus the number of all classified samples in the test set
(Bramer, 2007), defined as:
Overall Accuracy = × 100[%].

correctly classified samples from all classes
(20)
total number of samples
4. Classification of proteins using PCA and SVM analysis of their FTIR

spectra
Figure 2 depicts the FTIR spectra from dried MAB to CA125, BSA, human plasma, MAB to
ILGF2, MAB to Leptin and MAB to OPN. All spectra exhibit a similar structure. The origin
of the peaks has been well documented in the literature (reviewed by Barth & Zscherp,
2002). The spectra have several distinctive regions. The first region corresponds to the
interval 2800-3500 cm-1. A NH2 region around 3200 cm-1 is strongly overlapped with OH
stretching band. The region 1800-2700 cm-1 is relatively free of peaks. Amide bands are
characteristics in the region 1200-1700 cm-1. Those arise from the amide bonds that link the
amino acids. The amide I centered about 1740 cm-1 corresponds to the stretching mode of the
C=O bond of the amide. It may have some contribution from CN stretching and CCN
deformations. The amide peak II centered around 1550 cm-1 corresponds to the bending
mode of the NH bond of the amide with contributions from C=O in plane bending and NC
stretching. Amide III mode is the in-phase combination of NH in-plane bending and CN
stretching. Other smaller peaks corresponding to CC stretching and CO bending are
observed in this region. The characteristics of these peaks provide information about the
www.intechopen.com
secondary structure of the proteins since the hydrogen bonds that establish this structure,
are mostly associated to the CO and NH bonds. The wide peak in the region 400-800 cm-1
corresponds to librations with contribution from other rotational and low energy vibrational
lines. In figure 3 we show the results of the use of the first two PCA variables to represent
the data presented in figure 2. Despite the evident similarities between the spectra the data
are perfectly separable even with the use of only the first two PCA variables. For MAB to
Leptin, ILGF2 and OPN the separation is larger.
0.6 0.8
0.6
MAB AG CA125
BSA Human plasma
0.4 0.4
Absorbance
Absorbance
Absorbance
0.4
0.2 0.2
0.0 0.0 0.0

1000 2000 3000 4000 1000 2000 3000 4000 1000 2000 3000 4000
-1
-1
Wavenumbers (cm ) Wavenumbers (cm )
Wavenumbers (cm-1)
0.4
MAB ILGF2
MAB LEPTIN MAB OPN
0.04
Absorbance
Absorbance
0.2 Absorbance
0.2
0.02
0.0
0.00 0.0
1000 2000 3000 4000 1000 2000 3000 4000 1000 2000 3000 4000
-1
Wavenumbers (cm-1)
-1
Fig. 2. FTIR spectra of MAB to AG CA125, BSA, human plasma, MAB to ILGF2, MAB to
Leptin and MAB OPN
Fig. 3. Two-dimensional PCA of the data presented in figure 2
www.intechopen.com
From figure 4 we can see that even the first two principal components are sufficient to
achieve perfect separation of protein classes. Therefore, even the application of linear SVM
is able to provide perfect accuracy on both training and test sets (100% accuracy). The result
does not depend on the number k of principal components used (k>1). The average number
of support vectors per class is relatively small (Figure 4) and practically does not depend on
the number of principal components used, which indicates good generalization and stability
of the proposed technique. The results demonstrate the possibility of automatic classification
of proteins using PCA and linear SVM with accuracy of nearly 100%. Hence, this justifies the
application of the conceptually simpler LDA technique. Namely, LDA is also capable of
achieving 100% accuracy using as little as 2 principal components. Hence, below we discuss
the use of LDA to separate the FTIR spectra of proteins and their deuterated versions aimed
at the development of a FTIR based immunoassay.
Fig. 4. Support vector per class as a function of the number of PCA components after
Lazarevic et al. (2009). Reproduction authorized by the International Society for Optical
Engineering SPIE
5. FTIR of deuterated proteins

Deuterium is a stable isotope and can be used as a labeling agent. Deuteration occurs by
simple dilution of proteins in heavy water that contains the deuterium isotope of
hydrogen (2H). Hydrogen atoms on the surface of the protein are exposed to a fast
exchange with deuterium atoms while hydrogen atoms deeply buried within the protein
molecule exchange at a low pace. As an effect by substituting hydrogen atoms by
deuterium atoms vibration modes of OH (hydroxyl peaks), NH2 (amide peaks), NH
and/or CH can be affected. Deuteration also induces the appearance of a strong peak in
the region around 2400 cm-1. This region is usually free of peaks for most of the proteins.
Besides its evident advantages and extensive use for the study of the secondary structure
deuteration of proteins can have another important application still not considered in the
www.intechopen.com
literature. Indeed, the FTIR spectrum of a deuterated protein is different from the non-
deuterated one, and this can be used for their identification within a matrix containing a
large variety of proteins, e.g., a blood, plasma or serum sample. Furthermore, as we
demonstrate, the use of PCA and LDA can identify these differences automatically with
high accuracy.
BSA in H2O
0.6
Absorbance
0.3
BSA in D2O
0.0
1000 2000 3000 4000
-1
Wavenumbers (cm )
Fig. 5. FTIR spectra of non-deuterated BSA and partially deuterated BSA

As it is well known, the spectrum of water exhibits the stretching symmetric and
antisymmetric modes at 3253 cm-1 and 3315 cm-1, respectively, the bending mode at 1647 cm-
1 and libration peaks observed below 700 cm-1. The spectrum also exhibits a small peak at
2094 cm-1 that corresponds to the interaction between bending and libration modes. The
heavy water spectrum exhibits a similar structure of peaks but all the peaks are shifted by a
factor of 1.37 in close correspondence to the factor of 1.41 calculated from the differences of
masses between hydrogen and deuterium atoms. The spectral widths of the D2O lines are
also reduced by a factor of 1.35 comparing to those of water. Of special interest are the peaks
of the stretching vibration of deuterium oxide molecule which are centered at 2401cm-1 and
2471 cm-1. Proteins FTIR spectra are usually free of peaks in this area (see figure 2).
Deuterated proteins can have peaks in this region, a feature that can be used for protein
identification and calibration. As a consequence, shifts of spectral lines, changes in relative
amplitudes and changes in the spectral widths are expected for deuterated proteins
(Marcano et al., 2008).
www.intechopen.com
In figure 5 we show the FTIR spectra of the dried BSA sample from solutions of distilled
μg/mL. The amide I peak (1644 cm-1) remains almost unaltered while the amplitude of the
double dionized water (dash line) and D2O (solid line) prepared at concentration of 42
amide II peak (1530 cm-1) decreases. This peak corresponds to NH bending vibrations
which are strongly affected by substitution of hydrogen by deuterium atoms. Amplitude
increase is observed for other peaks in the region 1200-1300 cm-1. Remarkable is the
presence of the peaks in the region 2400-2500 cm-1 which is free of peaks not only for BSA
but also for a number of antibody proteins (see figure 2). These peaks correspond to
stretching OD vibration. Correspondingly, the hydroxyl peaks in the region 3000-3500 cm-1
are reduced.
1.2
Normalized Absorbance
a)
o
25 C
o
75 C
b) 75 oC
0.4
0.8 o
65 C
o
45 C
0.4 0.2 55 oC
65 oC
25 oC
o
35 C
o
55 C
o
35 oC 45 C
0.0 0.0
1200 1400 1600 2200 2400 2600
-1 -1
Fig. 6. Details of the FTIR spectra of BSA as a function of temperature in the regions
1200-1700 cm-1 and 2200-2600 cm-1
Increase in temperature deepens the changes observed. In figure 6 we show details of the
FTIR spectra of BSA in the regions 1200-1700 cm-1 (a) and 2200-2600 cm-1 (b), respectively,
after heating the solution from 25 oC up to 75 oC. High temperature breaks the hydrogen
bonds opening the protein molecule and exposing it to wide deuteration. The effect is small
up to a certain temperature (60 oC in our case) but when the thermal energy is enough to
break the hydrogen bonds the effect increases substantially. For the sample heated up to 75
oC the peaks at 2400 cm-1 are more than 5 times larger than the one for 55 oC (see figure 6b).
The amide II peak is depleted as well as other peaks at 1200 cm-1(see figure 6a). The
depletion is also remarkable for the hydroxyl peaks at 3500 cm-1. Deuteration can be
significantly increased by the use of ultrasound. Ultrasound shakes the molecule exposing
its hydrogen bonds to deuteration. In figure 7 we show the results of deuteration of BSA in
D2O at concentration of 500 mg/mL by using ultrasound during 120 minutes at room
temperature. After exposing to ultrasound the solution is put to rest for long term dilution
(1 week) at 6 oC. The changes in the absorbance FTIR are remarkable. The peak at 2466 cm-1
dominates the center of the spectrum. The amide II peaks is almost depleted and the peak at
1434 cm-1 triples its amplitude value.
www.intechopen.com
0.6
Absorbance
-1
1434 cm
-1
2466 cm
0.3
0.0
1000 2000 3000 4000
-1
Wavenumbers (cm )
Fig. 7. FTIR spectrum of highly deuterated BSA obtained using ultrasound
MAB to AG CA125
MAB to Leptin 1.0
1.0
a) b)
No D2O
No D2O After dilution in D2O

0.5 0.5
After dilution
in D2O -1
2407 cm
0.0 0.0
1000 2000 3000 4000 1000 2000 3000 4000
-1 -1
Fig. 8. FTIR spectra of deuterated MAB to Leptin and MAB to AG CA125

We have studied the effects of deuteration on MAB to Leptin, MAB to AG CA125, AG
the supplier of chemicals. For deuteration we use 5 μL of the sample and diluted it into 5 μL
CA125, Leptin, OPN and IGF2. All these proteins are originally diluted in saline solution by
of heavy water. One drop of this diluted solution is then deposited to dry over the
spectrometer. In figure 8 we show the FTIR spectra of MAB to Leptin (8a) and MAB to AG
CA125 (8b) from the original saline solution (dot lines) and heavy water dilution (solid
lines). The spectra are normalized with respect to the amplitude of the amide I peak. Again
we observe the surge of the DO peak in the region around 2400 cm-1 and also changes in the
relative amplitudes of hydroxyl and amide II peaks. Reduction in the spectral width of the
peak is also observed.
www.intechopen.com
6
797 cm-1
Relative amplitudes of peaks

6
No D O
2
1 drops of D O
2
a) 1018 cm-1
b)
2 drops of D O
3 2
3 1085 cm-1
4 drops of D O
2
-1
3283 cm
1259 cm-1
2962 cm-1
0 0
1000 2000 3000 4000 0 1 2 3 4
-1
Wavenumbers (cm ) Number of D2O drops
Fig. 9. FTIR spectra of AG CA125 exposed to different amount of D2O

We have also studied the effect of deuteration of ovarian cancer AG CA125, OPN, IGF2 and
leptin. In figure 9 we show the effect of deuteration over the ovarian cancer AG CA125 after
adding drops of D2O (figure 9a) subsequently. The sample is left to dry after addition of
each drop and before recording the spectra. We observe decrease in the relative amplitude
of several peaks all over the spectrum. In figure 9b we plot the relative amplitudes of five of
these peaks as a function of the amount of D2O used. Similar results are obtained for leptin,
OPN and IGF2.
Figures 5-9 demonstrate that deuteration of proteins is relatively easy to achieve by simple
dilution in heavy water for both the monoclonal antibody proteins and their corresponding
antigens. If required the impact of deuteration can be increased by increasing the
temperature or by applying ultrasound.
6. Use of PCA method for detection of deuterated proteins

As expected, using PCA to perform feature extraction can lead to distinguish with high
efficiency between deuterated and undeuterated versions of the same protein. In figure 10
we plot the tridimensional principal projection of the FTIR data from AG Leptin exposed to
1, 2 and 3 drops of D2O. A good separation between the data is obtained. The absolute
distance between the data increases with the number of D2O drops as correspond to larger
deuteration effect as suggested by figure 10. However, the effect depends on the type of
protein. In figure 11 we show the result for BSA. In this case, the distance between the
unexposed sample PCA data and the exposed ones does not change monotonically with the
numbers of D2O drops. This may be related to parasitic D-H exchange with water of the
surrounding the sample atmosphere over the time of the experiment. Nevertheless, a very
clear separation between BSA samples with different numbers of D2O drops is achieved
with 100% classification accuracy using LDA with four-fold cross-validation.
www.intechopen.com
Non exposed
PCA 3
0.5 3 drops
0
2 drops
-0.5
1 drop
-1
4
2
PCA 2 0 4
0 2
-4 -2
-2 -6 PCA 1
Fig. 10. Tridimensional PCA representation of the FTIR data from AG Leptin exposed to
successive drops of D2O. Non-exposed to D2O data are included for comparison
0.3
2 drops
0.2
0.1
1 drop
PCA3
-0.1 Non exposed

-0.2
3 drops
-0.3
-0.4
1
0.5 1.5
1
0 0.5
0
-0.5 -0.5
-1
PCA2 -1 -1.5 PCA1
Fig. 11. Tridimensional PCA of FTIR spectra of BSA exposed to 1, 2 and 3 drops of D2O. The
non-exposed to D2O samples is included for comparison
Different regions of the spectra contribute differently to the separation. In figure 12 we show
the plot of the PCA loadings (absolute value of components for eigenvectors v1-v3 from Eq. 2
as functions of the wave-numbers. The larger the absolute value of the PCA loading is the
larger is the importance of the corresponding spectral region for a more efficient separation.
www.intechopen.com
In the figure we see the importance of the low wavenumbers region (400-600 cm-1), the
amide peaks region (1300-1600 cm-1) and the DO peak region (2400-2500 cm-1).
0.14
0.12
Loadings magnitude
PCA 2
PCA 1
0.1
0.08
0.06 PCA 3
0.04
0.02
0
500 1000 1500 2000 2500 3000 3500 4000
Wavenumbers (cm-1)
Fig. 12. PCA variables as functions of the wavenumbers corresponding to the data of
figure 11
7. Detecting a protein antibody in a complex matrix

The proposed methods can be used to detect the presence of a particular protein in a
complex matrix, such as blood, plasma, serum or other biomedical samples. The efficiency
of separation can be increased by using deuterated versions of the proteins. In figure 13a
we show the FTIR spectra of pure human plasma (stars), plasma with added non-
μg/ml normalized by the amplitude of the peak at 3275 cm-1. The differences are more
deuterated (diamonds) and deuterated (squares) MAB to CA125 at a concentration of 50
remarkable for the deuterated samples comparing to the non deuterated ones. In figure 13b
we show the tridimensional PCA plot corresponding to these data. Despite the similarities
of spectra, the data are separable in the PCA coordinates space. The separation is larger for
the deuterated version of the protein. The absolute distance between the centers of data
cluster in the PCA coordinates space corresponding to plasma and the cluster
corresponding to deuterated MAB to CA125 is more than twice larger than the distance of
the center of the plasma cluster to the non-deuterated protein data cluster. Although the
concentration used is relatively high, the result demonstrates possibilities of detection of
proteins embedded in a complex matrix and the increase in sensitivity when using
deuterated versions of the proteins.
www.intechopen.com
The spectral changes in deuterated proteins and the statistical and data mining methods
used for their analysis can be applied to develop new kinds of immunoassays for detection
of antigen proteins. The immunoassays are aimed at detection of a particular antigen
protein embedded in a complex matrix, such as blood, serum or a plasma sample. Figure
14 describes two such immunoassays. Figure 14a depicts an immunoassay where non-
deuterated protein antibodies are deposited over a glass substrate (step 1). The plate is
then exposed to the sample. The antibody proteins on the plate capture their
corresponding antigens from the sample (step 2). Finally, the system is exposed to the
presence of the deuterated versions of the antibody proteins. These deuterated antibodies
are then attached to the trapped antigens forming a sandwich structure which is wash
away to remove non captured proteins (step 3). This sandwiched structure can then be
analyzed using an FTIR spectrophotometer. In a second type of immunoassay the first
step is the same (see figure 14b). Then, the plate is exposed to the sample which has been
previously diluted in heavy water. Deuterated antigens can then be captured by their
antibody protein deposited on a plate. The rest of the sample can be washed away.
Finally, the presence of the antigens can be detected by performing the FTIR experiment
over the treated plate.
2.0
a) b)b)
Plasma
Nondeuterated
CA125
CA125 0.2
1.5 non-deuterated
Plasma
0
PCA 3
CA125
1.0 -0.2 deuterated
Deuterated -0.4
0.5 CA125
-0.6
1
2
0.0 0.5
1
1000 2000 3000 4000 0
PCA 2 0
-1 PCA 1
Wavenumbers (cm ) -0.5 -1
Fig. 13. a) Normalized FTIR spectra of plasma and plasma containing deuterated and non-
deuterated MAG AG CA125, b)Tridimensional PCA plot of FTIR data from plasma and
deuterated and non-deuterated versions of the protein MAB to CA125
Several steps need to be completed before developing a FTIR-based immunoassay of
practical use. Deuteration can affect the bioactivity of the proteins. In this regard, the affinity
constant between the antibody and antigen can depend on the level of deuteration. The level
of deuteration can be also affected by parasitic D-H exchanges that can mask the real results.
Practical comparison with well established immunoassays such as ELISA must be
completed. However, we show that the use of deuteration techniques in combination with
statistical methods, such as PCA, LDA and SVM, will play a crucial role in the developing of
this new kind of FTIR-based immunoassays aimed at detection of a targeted protein in a
complex biosample.
www.intechopen.com
a)
D D D
1 2 3
b) D D D
1 2
Fig. 14. Proposed FTIR based immunoassays for detection of a given antigen
8. Conclusions
In this study, we demonstrate that using combination of PCA analysis and statistical and
data mining classification techniques (LDA and SVM) it is possible to automatically
determine a class of the FTIR sample of an unknown protein using a small number of
principal components. In such a case, using conceptually simpler LDA analysis (that relies
on stronger assumptions about the data than SVM) is justified. The full advantage of non-
linear SVM could, however, be expected in case of more complex and noisy spectroscopic
data. The proposed data analysis technique is computationally fast and can in principle be
applied in on-line learning classification framework. Work in progress includes testing the
proposed technique on larger datasets to exclude the small variability of samples (the
sample bias) as a potential reason for extremely high classification accuracy. We show that
the techniques distinguish between different proteins with similar FTIR spectra and
between deuterated and non-deuterated versions of the same protein. Furthermore, we
demonstrate the use of the method for separation and identification of proteins embedded
in a complex matrix of proteins such as plasma. We show that deuteration increases the
sensitivity of the method. Finally, we propose an immunoassay that is aimed to utilize the
demonstrated sensitivity of the methodology to detect a particular antigen protein in a
complex biosample.
9. Acknowledgements
This research has been possible thanks to the support of the National Science Foundation
(NSF-CREST grant No 0630388 and NSF-MRI grant No 0922587), National Institute of Health
(NIH Grant No P20 RR016472), Department of Defense (DoD/DoA Grants No 45395-MA-
ISP, and No 54412-CI-ISP) and of the National Aeronautics and Space Administration
(NASA URC 5 grant No NNX09AU90A). We also would like to thank Blood Bank of
Delmarva, Delaware, for providing the human plasma samples.
www.intechopen.com
10. References
Arrondo, J. L. R.; Muga, A.; Castresana, J. & Goñi, F. M. (1993). Quantitative studies of the
structure of proteins in solution by Fourier-transform infrared spectroscopy. Prog.
Biophys. Molec. Biol., Vol. 59, No. 1, 23–56, ISSN 0079-6107.
Baenziger, J. E. & Methot, N. (1995). Fourier transform infrared and hydrogen/deuterium
exchange reveal an exchange-resistant core of alpha-helical peptide hydrogens in
the nicotinic acetylcholine receptor. J. Biol. Chem., Vol. 270, No. 49, 29129-29137,
ISSN 0021-9258.
Barth, A. & Zscherp, C. (2002). What vibrations tell us about proteins. Quarterly Reviews of
Biophysics, Vol. 35, No. 4, 369–430. Cambridge University Press, DOI:
10.1017/S0033583502003815. ISSN 0033-5835.
Bishop, C. M. (2006). Pattern recognition and machine learning. Springer + Business Media
LLC, ISBN-10 0-38731073-8, Singapore.
Bramer, C. M. (2007), Principles of Data Mining (Undergraduate Topics in Computer Science),
Springer-Verlag, ISBN-10: 1-84628-765-0, London.
Byler, D. M. & Susi, H. (1989). Examination of the secondary structure of proteins by
deconvolved FTIR spectra. Biopolymers, Vol. 25, 469-487, ISSN 1097-0282.
Dave, N.; Lórenz-Fonfrı a, V. A.; Villaverde J.; Lemonnier R.; Leblanc G. & Padrós, E.
(2002). Study of Amide-proton Exchange of Escherichia coli Melibiose Permease by
Attenuated Total Reflection-Fourier Transform Infrared Spectroscopy: Evidence of
structure modulation by substrate binding. J. Biol. Chem. Vol. 277, 3380-3387, ISSN
0021-9258.
Fabian H. & Shultz, C. P. (2001). Encyclopedia of Analytical Chemistry, R. A. Meyers (Ed.),
Wiley, 5779-5803. ISBN: 978-0-471-97670-7, Chichester.
Fabian H. & Mantele W. (2002). Handbook of Vibrational Spectroscopy, J. M. Chalmers, P. R.
Griffiths (Eds.) Wiley, 3399-3425, ISBN: 978-0-471-98847-2, Chichester
Goormaghtigh, E.; Cabiaux, V. & Ruysschaert, J. M. (1994). Determination of soluble and
membrane protein structure by Fourier transform infrared spectroscopy I.
Assignments and mode compounds. Subcell. Biochem. Vol. 23, 329–362, ISSN -0306-
0225.
Haris, P. I. & Chapman, D. (1994). Analysis of polypeptide and protein structures using
Fourier transform infrared spectroscopy. Methods in Molecular Biology, Microscopy,
Optical Spectroscopy, and Macroscopic Techniques, vol. 22 (eds. C. Jones, B. Mulloy &
A.H. Thomas), 183–202. Humana Press Inc. ISBN 0-89603-232-9, Totowa, NJ.
Hsu, C. W & Lin, C. J. (2002). A comparison of methods for multi-class support vector
machines. IEEE Trans. Neural Networks, vol. 13, No. 2, 415-425, ISSN 1045-9227.
Hybl, J. D.; Lithgow, G. A. & Buckley, S. G. (2003). Laser-induced breakdown spectroscopy
detection and classification of biological aerosols. Appl. Spectros. Vol. 57, 1207-1215.
Jiang, Z. Q.; Fu, H. G. & Li, L. J. (2005). Support vector machine for mechanical faults
classification. Journal of Zhejiang University Science (Online) 6A(5).
http://www.zju.edu.cn/jzus/2005/A0505/A050513.pdf.
Jolliffe, I. T. (2002). Principal Component Analysis. Springer Verlag, ISBN 0-387-95442-2, New
York.
Karush, W. (1939). Minima of Functions of Several Variables with Inequalities as Side Constraints.
M.Sc. Dissertation. Dept. of Mathematics, Univ. of Chicago, Chicago, Illinois.
Available from http://wwwlib.umi.com/dxweb/details?doc_no=7371591.
www.intechopen.com
Krzanowski, W. J., 1988). Principles of Multivariable Analysis: A User’s Perspective. Oxford

University Press, ISBN 0 198 8507089, New York.
Lazarevic, A.; Pokrajac, D.; Marcano, A. & Melikechi, N. (2009). Support vector machine based
classification of fast Fourier transform spectroscopy of proteins, Proc. SPIE, Vol. 7169,
71690C; DOI: 10.1117/12.809964, ISBN 9780819474155, San Jose, California, USA,
January 2009, SPIE, Bellingham.
Massart, D. L.; Vandenginste, B. G.; Deming, S. N.; Michotte, Y. & Kaufman, L. (2003).
Chemometrics: A Textbook. Elsevier, ISBN 0-444-42660-4, Amsterdam.
Marcano A.; Markushin, Y.; Melikechi, N. & D. Connolly (2008). Fourier Transform
Spectroscopy of Deuterated Proteins. Linear and Nonlinear Optics of Organic Materials
VII by Rachel Jakubiak, Proc. SPIE, Vol. 7049, pp. 70490z-1-8. ISBN 9780819472694,
San Diego, California, USA, August 2008, SPIE Bellingham.
Melikechi, N.; Ding, H.; Rock, S.; Marcano, A. & Connolly, D. (2008). Laser-induced
breakdown spectroscopy of whole blood and other Liquid organic compounds.
Optical Diagnostic and Sensing VIII, Editors G. Cote and A. V. Priezzhev, Proc. SPIE,
Vol. 6863, 68630O1-7, DOI:10.1117/12.761901, ISBN 9780819470386, San Jose,
California, USA, January 2008, SPIE, Bellingham.
Mor, G.; Visintin, I.; Lai, Y., Zhao, H.; Schwartz, P. Rutherford, T.; Yue, L.; Bray-Ward, P. &
Ward D. C. (2005). Serum protein markers for early detection of ovarian cancer.
PNAS, Vol. 102, No. 21, 7677-7682, on line ISSN 1091-6490.
Nie, B.; Stutzman, J. & Xie, A. (2005). A vibration spectra maker probing the hydrogen-
bonding status of protonated Asp and Glu residues. Biophys. J., Vol. 88, No 4, 2833-
2847, ISSN 0006-3495.
Petibois, C.; Gionnet, K.; Goncalves, M.; Perromat, A.; Moenner, M. & Deleris, G. (2006).
Analytical performances of FT-IR spectrometry and imaging for concentration
measurement within biological fluids, cells, and tissues. Analysis, Vol. 131, 640-647,
ISSN 0003-2638.
Platt, J.; Cristianini, N. & Shawe-Taylor, J. (2000). Large Margin DAGs for Multiclass
Classification. Advances in Neural Information Processing Systems, Vol. 12, pp. 547-
553, ISBN-13 978-0-262-56145-7.
Schorge, J.O.; Drake, R. D.; Lee, H.; Skates, S. J.; Rajanbabu, R.; Miller, D. S.; Kim, J. H.;
Cramer, D. W.; Berkowitz, R. S. & Mok, S. C. (2004). Osteopontin as an adjunct to
CA125 in detecting recurring ovarian cancer. Clin. Cancer Res., Vol. 10, No. 10, 3474-
3478, ISSN 1557-3265.
Siebert, F. (1995). Infrared spectroscopy applied to biochemical and biological problems.
Methods Enzymol., Vol. 246, 501–526, ISBN 978-0-12-182147-0.
Sutphen, R.; Xu, Y.; Wilbanks, G. D.; Fiorica, J.; Grendys Jr., E. C.; LaPolla, J. P.; Arango, H.;
Hoffman, M.S.; Martino, M.; Wakeley, K.; Griffin, D.; Blanco, R. W.; Cantor, A. B.;
Xiao, Y. J. & Krischer, J. P. (2004). Lysophospholipids are potential biomarkers of
ovarian cancer. Cancer Epidemiol. Biomarkers Prev. Vol. 13, No. 7, 1185-1191, ISSN
1055-9965.
Vapnik, V. (2000). The Nature of Statistical Learning Theory, Springer Verlag, ISBN 0-397-
98780-0, New York.
www.intechopen.com
9
The Analysis of Heart Sounds and a

Pocket Computer Application via
Discrete Fourier Transform
Gür Emre Güraksın1, Uçman Ergün2 and Ömer Deperlioğlu2
1Afyon Kocatepe University, Engineering Faculty
Computer Engineering Department
2Afyon Kocatepe University, Engineering Faculty
Biomedical Engineering Department

Afyonkarahisar
Turkey
1. Introduction
The heart is one of the two crucial centers for human life. Thus, any disorder concerning the
heart which may be able to occur is of prime importance to human health. The mortality rate
stemming from heart diseases took the second lead after brain embolism in the world from
1985 to 2006 (Jiang & Choi, 2006). Being such significant to humans, the heart organ is
consisted of two cycles. At the moment of the closure of mitral and tricuspid valves,
ventricular contraction comes out, alias one cycle begins with systole and ends with diastole
(Sharif et al., 2000). Auscultation with stethoscope is a preferential method that the doctors
use in order to differentiate normal cardiac systems from the abnormal ones that come out
(Sinha et al., 2007). The listened heart sounds are formed through the flow of blood entering
and exiting the heart and with the movements of cardiac valves connected to this flow. The
sounds comprised of this blood flow are listened by the physicians via stethoscope. The
heart sounds are interpreted and it is determined whether the patient has any disorder
about the heart. On the other hand, auscultation method has some limitations. Auscultation
depends on the physician’s interpretation of different heart sounds, hearing skill, experience
and expertize (Kandaswamy et al., 2004). The required experience and expertize are
achieved as a result of long examinations and diagnoses made by the physicians. Even
though one physician has the necessary training so as to conduct auscultation and to
diagnose cardiac disorders, he is still in need of clinical experience. Lacking experience and
expertize are especially a difficulty for the newly graduates and medical interns. However,
unsuitable conditions of the environment and incompatibility of the patient can also lead to
deficiency in the diagnosis process. Because of these difficulties which may be experienced,
auscultation method that is listening with a stethoscope has been insufficient in the
exploration of heart abnormalities.
As the auscultation method fail to meet the needs of the physicians, they also make use of
Electrocardiography (ECG) data along with stethoscope. ECG method, generated through
the advancement of the technology, is a procedure helping physicians in the diagnosis of the
www.intechopen.com
heart disorders. ECG is the wave form taking the record of the electrical activity of the heart
via electrodes attached to the skin. A simple and cheaply method, ECG is constantly used by
the physicians. ECG records and their analysis that are used to detect defects in the heart are
relatively a good method. On the other hand, if the likely generating system of heart is
utterly acting with small heart defects, it may look very difficult to diagnose these
abnormalities with the analysis of ECG records as nearly no change in the ECG will be
discovered. However, it was determined that this situation has almost led to changes among
the sounds produced by the heart (Sinha et al., 2007). As it is not possible to detect some
heart abnormalities with ECG, listening to heart sounds with auscultation method has
earned more importance. Owing to the reasons explained above, there is a need for faster
and more accurate diagnosis in the record and analysis of these heart sounds.
The very first phase of developing the system that will help the physicians interpret the heart
sounds accurately is the signal processing methods. When the studies based on these methods
are searched, it can be seen that Fourier analysis has been used in many of the studies. Thanks
to the Fourier analysis, it is possible to examine sound signals in the frequency space. In
studies especially done on the classification of heart sounds, the frequency analysis of these
digitized heart sounds are processed and then passed to classification stage with various
artificial intelligence methods. If we survey some of the studies in literature that used heart
sounds and Fourier transform together, in one study Segaier et al. developed a digital
algorithm in order to detect the first heart sound (S1) and the second heart sound (S2) and to
characterize systolic murmurs. The study done on the pediatry patients, Short-Time Fourier
Transform (STFT) was used so as to carry out the analysis of the heart sounds taken from the
patients (El-Sagaier et al., 2005). In another study, Folland et al., in the moment of auscultation,
applied Fast Fourier Transform (FFT) and Levinson-Durbin algorithm to heart sounds to
analyse the abnormalities in the heart sounds; and applied the data to the Multilayer
Perceptron (MLP) and Radial Basis Function (RBF) artificial neural networks to classify
abnormal sounds. Eventually, it was determined that in the classification of the heart sounds,
the sensitivity levels that the MLP and RBF neural networks have gained were 84% and 88%
consecutively (Folland et al., 2002). In another study, Debbal ve Breksi-Reguig have done the
time-frequency analysis of S1 and S2 heart sounds, and applied Wigner distribution and
wavelet transform techniques to the heart sounds. They made a comparison between FFT and
STFT and with these techniques they implemented (Debbal & Breksi-Reguing 2007). In one of
the study by Abdel-Alim et al., they made a classification of heart sounds that belonged to
different cardiac valve disorders with the use of a feed forward artificial neural network. In the
course of analysis of these sounds, discrete wavelet transform, FFT, and linear prediction
coding methods were employed. In the end, the classification success was found out to be
%95.7 (Abdel-Alim et al., 2002).
In this study, the heart sounds achieved through a stethoscope were initially computed, and
they were subjected to Discrete Fourier Transform (DFT) and next the graphics and the
frequency spectrum in the time domain that belonged to the heart sounds were drawn on
the pocket computer. Thus, frequency spectrum of normal and abnormal heart sounds was
obtained via DFT, so it was aimed to prepare the physicians with some more data in the
course of auscultation.
2. Structure of the heart and heart sounds

Heart is an empty muscle that pumps the blood to the blood vessels in the whole body
(Sharif et al., 2000). The most significant and primary duty of the heart is to dispatch and
pump the blood to the circulation system like a pump (Ahlström, 2006).
www.intechopen.com
Pocket Computer Application via Discrete Fourier Transform 169
As it can be seen in figure 1, the heart is composed of two parts: the right heart and the left
heart. The right heart pumps the blood into lungs. This cycle is called as pulmonary cycle. On
the other side, the left heart is the part that provides all the organs and the whole body with
the oxygen and nutrients (Sharif et al., 2000). In addition, there are four chambers in the heart
that are known as the right and the left atriums and ventricles. These two atriums are the
places where the blood entering the heart is stored. The ventricles, on the other hand, convey
the blood to the whole body like a pump. As the heart contracts, the blood makes a pressure
towards the valve and moves from the atrium to the ventricle (Barschdorff et al., 1990).
Fig. 1. Structure of the Heart

The relation among the volume, pressure and flow of the blood in the heart determines the
opening and the closing of the valves. Normal heart sounds occur in the course they close.
Besides, the sounds occurring in the heart and in the vessels with the flow are constituent of
the heart sounds, too. But, as a matter of fact, how they occur is still a matter of discussion
(Ahlström, 2006).
The abnormalities in the heart structure mostly reflect to the sounds that the heart produce
(Leung et al., 2000). The formation of heart sounds and murmurs are generally developed
through the actions of myocardial walls, the opening and closing of cardiac valves, and the
blood flow into and out of the ventricles (Kemaloğlu & Kara, 2002).
While the sinoatrial node contracts itself and the contraction spreads to the atria, the
pressure of left atrium surpasses the pressure of left ventricle. This contraction is then
extended to the ventricles. This moment is the time when ventricle starts to depolarize. The
contraction spreading into the ventricle contracts the muscles of the ventricle and leads to a
contraction in the ventricle. The pressure in the left ventricle starts to rise and as soon as it
reaches to the pressure of the left valve, the atrium and the valve ventricles close. At that
moment, S1 comes out. Mitral valve normally closes earlier than the triscuspid valve. For
that reason, S1 has two elements namely mitral and triscuspid. The frequency band is 20–45
Hz and the period is 50–100 ms.
The ventricle pressure goes on rising, and as the pressure goes over the pressure of aorta,
firstly aorta valve and next pulmonary valve open. Soon after starts the period of sending the
blood out of ventricles. As the ventricle muscles relax, the ventricle pressure starts to decrease.
At the moment that the internal pressure in the ventricle goes below the aorta pressure, the
aorta valve closes. The pulmonary valve closes respectively. The closure of these two valves
forms the S2 sound. The frequency band is 50–70 Hz and the period is 25–50 ms.
www.intechopen.com
When the internal pressure of the ventricle drops below the pressure of atrium, mitral and
triscuspid valves open and the ventricles are filled with blood. As the ventricles are filled
with blood, the vibrations of the ventricle muscles form the third heart sound (S3). It could
be heard in the young while it is an indicator of disorder of myocard function in the
overaged. This sound arises almost after 150 ms after the closure sound of aorta.
In the last parts of blood fulfillment of the ventricles, the flow of blood that was again
quickened with depolarization of the atrium revibrates the ventricle walls and in some
pathologic cases, this causes the fourth heart sound (S4). S4 is normally not heard in the
adults but could be taken in children.
Four sounds seen in figure 2 are known as the simple heart sounds. Apart from these
sounds, some sounds as murmur may occur in some heart disorders. These sounds are in
the frequency band of 100–600 Hz and are long-time compared to the simple heart sounds
(Kemaloğlu & Kara 2002).
Fig. 2. The First, Second, Third and Fourth Heart Sounds

The sounds that are named as murmur and caused by the blood passing through the
cardiovascular system loudly are the significant examples of abnormal sounds. The timing
of the murmur and the level of height have remarkable significance concerning the situation
of the heart. For instance; in the course of diastole a murmur marks erroneous functionality
of the heart valve. On the other hand, in the course of systole, the murmurs may be related
with the healthy and pathological heart depending on the acoustic character of the murmur
(Ölmez & Dokur, 2003).
The murmurs are formed through irregular blood flow and as a result of narrowing and
leaking valves or the existence of the abnormal passages in the heart. Eventually, the blood
flow that comes out leads to steady and irregular vibrations that are transmitted from the
cardiac and tissues belonging to the chest to surface of the chest (Ahlström, 2006).
While defining the heart sounds, we need to draw attention to the frequency, density and
the quality of the sound. For that reason, it is a must to listen to the heart sounds carefully
that are used as reference S1 and S2 and to determine the place accurately in the course of
listening. These sounds experience the systole and diastole phases as the heart operates, and
the sound differentiations in these places provide related information in connection with the
disorders in the heart (Say, 2002).
3. Fourier transform
When the signals in the real world are searched, it could be said that the signals that are
encountered practically are the time domain signals and the size measured is the function of
the time. For that reason, the signal needs to be transferred to a different domain through an
application of mathematical transform and some information from the constituents that
represent the signal in the domain can be obtained about the signal. For example, the
frequency spectrum (frequency constituents) of the signal is gained with Fourier transform.
The information hidden in the time domain is brought out in the frequency domain (Say,
2002).
www.intechopen.com
The Fourier display of the signals plays an extremely substantial role in signal processing in
both continuous time and discrete time. Thus, it supplies a method to match the signals to a
domain and makes it possible to study on them. Fourier transform allows for a distinctive
way to interpret the signal and systems (Hayes, 1999).
Fourier transform is a method of analysis that was developed by Jean B. Joseph Fourier, a
French physicist and a mathematician, in 1987 when he was studying on the Fourier’s
research about heat and dispersion and it plays an important part in the signal processing
(Swanson, 2000). In the study that Fourier conducted in 1807, he said that it could be
achieved through selecting and gathering the sinus and cosine waves among the continuous
and periodical signals (Smith 1997).
The Fourier transform that is used to determine the frequency constituent of the raw signal
in the time domain can be defined with the two equities as follows:
∫ x(t ).e
∞
X(ω ) = − jωt
.dt (1)
−∞
∫ X(ω ).e
∞
x(t ) = jωt
.dω ω = 2π f
2π
1
(2)
−∞
With Fourier transform, the signal is extricated into complex exponential functions that have
various frequencies. In the equations t, specifies time; ω, specifies angular frequency and f
specifies the frequency. x, specifies the signal in the time domain, X specifies the signal in
the frequency domain. In the equation 1 given above, the Fourier transform of x(t) and in
the equation 2, inverse Fourier transform are displayed.
When the equation 1 is viewed, it is seen that x(t) signal is multiplied with an exponential
term in a certain f frequency and the integral of this multiplication is taken from minus
infinity to plus infinity all over the time. It should be taken into account that the effect of the
f frequency constituent to the integration will be the same no matter when it may come out
among these times. The integration result will not change whether the f frequency
constituent comes out in the course of t1 or t2. Fourier transform only indicates whether a
certain frequency constituent exists or not. Just a spectral constituent of a signal can be
gained through Fourier transform (Say, 2002).
Today, given that, the signal processing is held by computer algorithm and the computers
may work in limited length and with discrete signals, it is noticed that the Fourier transform
to be used will be discrete time Fourier , in other words, DFT (Smith, 1997).
3.1 Discrete Fourier Transform

On the contrary to some arrays defined theoretically, the Fourier transform of real arrays
may not be calculated. Thus, it is not convenient for the digital signals to use Fourier
transform. That the frequency is displayed analogically and it requires infinite number
samples are some of the basic reasons of this ineptitude.
Due to these difficulties, if the importance of Fourier transform in signal processing is taken
into account, a more practical transform needs to be defined. The new transform defined as
normal spaced N frequency point (Ωk) around a unit disc and N sample of x(n) arrays is
called as DFT (Kayran & Ekşioğlu, 2004).
www.intechopen.com
DFT is in essence a kind of transform such as Fourier arrays transform and Fourier integral
transform. The transform feature is very powerful for the time arrays, which also allows for
inverse transform. As can be noticed from its name, it owns utterly similar characteristics
with Fourier integral transform. It especially defines the spectrum of a time array (Cochran
et al., 1967).
This transform that also allows inverse transform includes distinctive qualities. The
principal quality is the equivalent of multiplication of two DFT in the time domain is the
total convolution of arrays. In addition, many spectrum analysis methods are based on DFT
(Kayran & Ekşioğlu, 2004).
DFT is defined through the equation 3 given below:
∑ X k exp( −2π jrk / N )

N −1
Ar = r = 0,..., N − 1 (3)
k =0
Here, Ar, symbolizes rth coefficient of the DFT and Xk symbolizes kth sample of a time arrays
composed of N sample. Xk’s may be complex numbers while Ar’s are always complex
numbers. The formation of the formula that is consistent with the notation given in the
equation 3 is mostly shown with the formula given in the equation 4:
∑ ( X k )W rk
N −1
Ar = r = 0,..., N − 1 (4)
k =0
W = exp( −2π j / N ) (5)
Since Xk’s are commonly the values of discrete time points of a function, r arrays are
sometimes referred as the frequency of DFT. DFT, too, is termed as “discrete time, finite
range Fourier transform”.
If we take a look at the inverse transform of equation 4:
Xı = (1 / N ) ∑ Ar W − rl l = 0,..., N − 1
N −1
(6)
r =0
This relation (equation 6) is termed as the inverse of DFT (Cochran et al., 1967).
In the figures between 3 and 7 below, the flow diagram of DFT algorithm that was taken as a
basis in the programming of the pocket computer was given. The main module seen in
figure 3 calls for four modules, which are used in the calculation process.
The next module, shown in figure 4, sets up the twiddle factor arrays, that is the
computation and storage of the sample values of cosines and sinus over one cycle, with
argument starting from zero with increment of 2П/N. This module is named as twid_fac. i
variable, here, is used as loop counter. On the condition that i variable reaches N that is the
data size, the loop is terminated. The tfc and tfs arrays whose sizes are equal to N are used to
hold the sample values of sinus and cosines respectively.
The next module seen in figure 5 reads the real and imaginary parts of the complex input
data, and in turn transfers them into N dimensioned xr and xi arrays. It is assumed that the
real parts of the input data are stored before the imaginary parts in the input file. If the data
is real, the data in the xi array is initialized to 0 and the data in the xr array is read. This
module is called in-put.
www.intechopen.com
Fig. 3. The Main Module for Direct Implementation of the DFT
Fig. 4. The Twiddle Factor Module
www.intechopen.com
Fig. 5. The Input Module

The next module that is dir_dft module is given in figure 6. dir_dft module is used in the
calculation of DFT coefficients. The calculation has been achieved via two nested loops.
While the outer loop controls the frequency index, the inner loop controls the access of the
data values. In each outer loop’s iteration only one coefficient is calculated. The real and
imaginary parts of the coefficients are stored, respectively, in arrays XR and XI, each of size
N. The access of correct twiddle factor values is carried out using the mod function. Inside
the inner loop, each coefficient is computed according to the DFT definition.
Fig. 6. The DFT Module
www.intechopen.com
The last module, shown in Figure 7, prints the real and imaginary parts of the coefficients,
respectively, from the arrays XR and XI, one coefficient in each iteration. This module is
called out-put (Sundararajan, 2001).
Fig. 7. The Output Module

DFT calculations are used in many scientific and engineering applications (Winograd, 1976).
For the effective and advantageous use of DFT, some of its basic properties need to be
learnt. It is beneficial to talk about its features shortly. The first characteristic of DFT is the
linearity feature. DFT is a linear transform. Another feature of its is the symmetry feature.
DFT values that are made of real values and equal to a periodical array are complex and
periodical. One more feature is the principle of similarity concerning the selectivity of time
and frequency. The indefinity principle of DFT is connected with the terms in the time and
frequency domain of DFT. It is equal to the well-known indefiniteness principle in physics.
This term is not the result of physical properties, but it is the result of a chief mathematical
formulation. The final feature is the conditions of link-equivalence between DFT and Fourier
Transform. Since DFT is the approximate of the continuous Fourier transform, researchers
are concerned with it. The validity of this approximation is certainly based on the related
wave form (Kayran & Ekşioğlu, 2004).
4. DFT application on pocket computer

In this study, the frequency analysis of the heart sounds taken from the patient via electronic
stethoscope was actualized, and on the pocket computer it was displayed both in the time
and frequency domain. Just as the system was completely developed mobile, it was aimed
that the physicians may use it in the course of clinical examination. The physician using the
system in the course of clinical examination will be able to make a more accurate and fast
diagnose through listening to the heart sounds and following the heart sounds both in the
time and frequency domain on the device.
4.1 Digitization of the sound

In order to play and store the heart sounds on the computer media, we need to have some
kind of file formats. Because of these formats, multimedia file can be listened and stored on
the computer. The wav (Waveform Audio Format) format to be used in our study is a kind
of sound file. It is a commonly used sound file format. Unlike the other sound formats, the
sound in the wav file format cannot be stored by compressing; it can only be stored through
digitizing. As this file format is not compressed, it holds a lot of space. On the other hand,
the sound quality is quite good. In this study, the wav sound format has been selected as it
saves the digitized sound with its original state before compressing.
www.intechopen.com
One wav file is in general consisted of two parts. In the first part, there is general information
concerning the data. The second part is the part where the original data starts.
In figure 8, the structure of a wav file format in general has been given (a. Güraksın, 2009).
Field Name Field Size in Bytes

Chunk ID 4 The format of concers
here is “WAVE” which
Chunk Size 4 requires two sub-
chunks: “fmt” and
Format 4 “data
Subchunk1 ID 4
Subchunk1 Size 4
Audio Format 2
Num Channels 2 Describes the format
of the sound
information in the
Sample Rate 4 data sub-chunk
Byte Rate 4
Block Align 2
Bits Per Sample 2
Subchunk2 ID 4
Indicates the size of the
Subchunk2 Size 4 sound information ad
contains the raw sound
data
Data Subchunk2 Size
Fig. 8. The Canonical wav File Format

The heart sounds taken from the patient are digitized through a piece of code written
accordingly to the mentioned format. During the digitization process, the digital data that
are in the hexadecimal basis are obtained by reading wav format and are then converted
into decimal basis. Thanks to the obtained digital format of the sound, it is made possible to
draw a graphic of the sound on the pocket computer and DFT can be applied to the sound.
With the help of a software that operates on the pocket computer, the digital data obtained
from the wav sound format are turned into sound graphic in another code, and again with
the help of another programming code it is subjected to DFT.
4.2 Raw data obtainment

The sounds used in this study were taken by using Littmann 4100 model electronic
stethoscope from Afyon Kocatepe University with the 07.AFMYO.01 numbered scientific
research project (a. Güraksın, 2009). With model 4100 Littmann electronic stethoscope, it is
possible to record 6 different heart sounds. Thanks to this, the sounds taken continuously
from six patients are stored within the stethoscope itself. In addition, the saved sounds will
be able to be transferred to pocket computer with the help of infra-red technology within the
stethoscope. The sounds recorded with Littmann 4100 model electronic stethoscope are
www.intechopen.com
stored in e4k format. This format was converted into wav format on the pocket computer via
a program given by Littmann. After converting the heart sounds into wav format, we pass to
the next phase of signal processing. By and large, there are some stethoscopes in the market
that are able to convert the stored sounds directly into wav format. By using such a kind of
stethoscope, there will be no need to use a second transform. In order for the sounds taken
by stethoscope to be processed, HP iPAQ hx2000 pocket computer that allows for infrared
transaction was used.
4.3 Drawing DFT graphics of the data of heart sounds on pocket computer
The aim of this study is to get the frequency domain graphic that was obtained by DFT and
the time domain graphic of the heart sounds taken from the patients drawn on the pocket
computer. The software developed for this purpose was prepared through the use of
programming language of C# in Microsoft Visual Studio 2005 media.
As you can see in figure 9, firstly the sound was taken from the patient by using the
Littmann 4100 model electronic stethoscope. Next, the sound to be analysed in a program
written in Visual Studio 2005 media was digitized, and it was subjected to DFT method. In
the final stage, the graphics of the processed heart sounds both in the time domain and
frequency domain on the pocket computer were drawn. The user interfaces of the software
developed on the pocket computer were shown below in figures between 10 and 13.
Fig. 9. Flow Diagram of the System

On the introduction screen of the software produced with C# programming language in
Microsoft Visual Studio 2005 medium, there appears a selection screen shown in figure 10
on which you could select one of the heart sounds that was at first taken via Littmann 4100
model electronic stethoscope and converted into wav format on the pocket computer.
On this screen, the heart sound that belongs to a patient is selected, and the selected heart
sound is prepared with the algorithm adapted to the pocket computer from wav format to
the signal processing phase. After that, it moves to graphic interface. On the graphic
interface, after selecting the button named as “Graph”, the software on the computer
subjects the sound data that was digitized beforehand to DFT. After this process is
completed, the data obtained from the DFT is sent into another array. As the signal
processing phase is finished, the graphic of the processed heart sound in the time domain is
drawn on the upper screen of the pocket computer (the graphic plane writes as “SND” on
www.intechopen.com
Fig. 10. The selection of patient’s heart sound data to be examined

the left), and the frequency spectrum obtained as a result of DFT is drawn on the lower
screen of the pocket computer (the graphic plane writes as “DFT” on the left). This process
almost takes between 5 and 10 seconds. The screen display obtained by drawing on the
pocket computer of the time domain graphic and frequency spectrum that belongs to a
normal heart sound has been given in figure 11.
Fig. 11. The screen display with drawing of time domain graphic and frequency spectrum of
heart sounds taken from a normal heart
www.intechopen.com
In figure 12 and 13, the screenshots of time domain graphics and frequency spectrum that
belong to the heart sounds in which in turn Pulmonary Stenosis and Mitral Stenosis
disorders were detected were shown on the pocket computer.
Fig. 12. The screenshot with drawing of a time domain graphic and frequency spectrum of a
heart sound that owns Pulmonary Stenosis Heart Disorder
Fig. 13. The screenshot with drawing of a time domain graphic and frequency spectrum of a
heart sound that owns Mitral Stenosis Heart Disorder
www.intechopen.com
When there is a comparison between the frequency spectrums of normal heart sound and
the heart sounds that have mitral and pulmonary stenosis heart disorders, it is noticed that
the heart sounds taken from the normal heart possess less frequency constituents. (b.
Güraksın et al., 2009). Because a healthy heart produces a periodical sound while a heart
that has any kind of disorder produces some different sounds other than S1 and S2 sounds.
Thus, these sounds include noise. The irregular and turbulent blood flows that cause all
these sounds lead to inclosure of high frequency constituents in heart sounds.
5. Conclusion
In this actualized study, the heart sounds gained with the use of electronic stethoscope were
digitized and then subjected to DFT. Finally, the graphic in the time domain and frequency
spectrum that belong to the heart sounds was obtained on the pocket computer. Thus,
frequency spectrum of normal and abnormal heart sounds was gained via DFT. As a result,
the physicians were prepared with more data in the course of auscultation. By providing the
physicians with alternative methods other than listening, it was aimed to help them make a
more accurate and faster diagnosis.
The basic purpose in this study is not to diagnose the heart sounds directly. Instead, it was
aimed to form a substructure for the prediction methods that may be used in the diagnosis
phase. This formed structure sets a substructure for the artificial intelligence programs such
as neural network, support vector machines, neuro-fuzzy, which are used for classification.
(c. Güraksın et al., 2009). The obtained DFT data can simply be made use of in the
classification mechanisms.
Acknowledgement The authors wish to acknowledge the Afyon Kocatepe University
Scientific Research Council with the Project number “07.AFMYO.01” for the collection of the
heart sounds used in this study.
6. References
Abdel-Alim, O. ; Hamdy, N. & El-Hanjouri, M. A. (2002), Heart Diseases Diagnosis Using
Heart Sounds, Proceedings of Radio Science Nineteenth National Conference, pp. 634–
640, Nov., 2002.
Ahlström, C. (2006), Processing of the Phonocardiographic Signal-Methods for the Intelligent
Stethoscope, Thesis, Linköping University, Institute of Technology, Linköping,
Sweden.
Barschdorff, D. ; Ester, S. ; Dorsel, T. & Most, E. (1990), Neural network based multi sensor
heart sound analyses, Proceedings of Computers in Cardiology, pp.303–306, USA, Sep.,
1990, Chicago.
Cochran, W. T.; Cooley, J. W.; Favin, D. L.; Helms, H. D.; Kaenel, R. A.; Lang, W. W.; Maling,
G. C.; Nelson, D. E. ; Rader, C. M. & Welch, P. D. (1967), What is the Fast Fourier
Transform?, Transactions on Audio and Electroacoustics, Vol. 15, Oct., 1967, pp.45–55.
Debbal, S. M. & Bereksi-Reguig, F. (2007), Time-Frequency Analysis of the First and the
Second Heartbeat Sounds, Applied Mathematics and Computation, Vol. 184, Jan, 2007,
pp. 1041–1052.
www.intechopen.com
El-Segaier, M. ; Lilja, O. ; Lukkarinen, S. ; Sörnmo, L. ; Sepponen, R. & Pesonen, E. (2005),

Computer-Based Detection and Analyses of Heart Sound and Murmur, Annals of
Biomedical Engineering, Vol. 33, No. 7, Jul., 2005, pp. 937–942.
Folland, R. ; Hines, E. L. ; Boilot, P. & Morgan, D. (2002), Classifying Coronary Dysfunction
Using Neural Networks Through Cardiovascular Auscultation, Med. Biol. Eng.
Comput., Vol. 40, 2002, pp. 339–343.
Güraksın G.E. (2009), Classification of the Heart Sounds via Artificial Neural Network, Ms Thesis,
Institute of Natural and Applied Sciences, Afyon Kocatepe University,
Afyonkarahisar, Turkey, 2009.
Güraksın G.E.; Ergün U. & Deperlioğlu Ö. (2009), Performing Discrete Fourier Transform Of
The Heart Sounds On The Pocket Computer, 14th National Biomedical Engineering
Meeting, İzmir, Turkey, 2009, pp. 124-127.
Güraksın G.E.; Ergün U. & Deperlioğlu Ö. (2009), Classification of the Heart Sounds via
Artificial Neural Network, International Symposium on Innovations in Intelligent
Systems and Applications, Trabzon, Turkey, 2009, pp. 507-511.
Hayes, M. H. (1999), Digital Signal Processing, The McGraw-Hill Companies Inc., USA.
Jiang, Z. & Choi, S. (2006), A cardiac sound characteristic waveform method for in home
heart disorder monitoring with electric stethoscope, Expert Systems with
Applications, Vol. 31, 2006, Aug., pp. 286–298.
Kandaswamy, A. ; Kumar, C. ; Ramanathan, R. ; Jayaraman, S. & Malmurugan, G. (2004),
Neural classification of lung sounds using wavelet coefficients, Computers in Biology
and Medicine, Vol. 34, Jul., 2004, pp. 523–537.
Kayran A. H. & Ekşioğlu E. M. (2004), Bilgisayar Uygulamalarıyla Sayısal İşaret İşleme, Birsen
Publishing House, 1. Edition, İstanbul, Turkey.
Kemaloğlu, S. & Kara, S. (2002), Instrumentation For Synchronous Acquiring Of ECG
Signals Together Heart Sounds, Journal of Erciyes University Institute of Science and
Technology, Vol. 18, (1-2), pp. 28–33.
Leung, T. S.; White, P. R. ; Collis, W. B. ; Brown, E. & Salmon, A. P. (2000), Classification of
Heart Sounds Using Time-Frequency Method and Artificial Neural Network, 22nd
Annual International Conference of the IEEE Engineering in Medicine and Biology
Society, Vol. 2, pp. 988–991.
Ölmez, T. & Dokur, Z. (2003), Classification of Heart Sounds Using an Artificial Neural
Network, Pattern Recognition Letters, Vol. 24, pp. 617–629.
Say, Ö. (2002), Analysis of Heart Sounds and Classification by Using Artificial Neural Networks,
Ms Thesis, İstanbul Technical University, Institute of Natural and Applied Sciences,
İstanbul, Turkey, 2002.
Sharif, Z. ; Zainal, M. S. ; Sha’ameri, A. Z. & Salleh, S. H. S. (2000), Analysis and
Classification of Heart Sounds and Murmurs Based on the Instantaneous Energy
and Frequency Estimations, Proceedings of TENCON, Vol. 2, pp. 130–134, 2000.
Sinha, R. K. ; Aggarwal, Y. & Das, B. N. (2007), Backpropagation Artificial Neural Network
Classifier to Detect Changes in Heart Sound due to Mitral Valve Regurgitation,
Journal of Medical Systems, Vol. 31, Jun., 2007, pp. 205–209.
Smith, S.W. (1997), The Scientist and Engineer’s Guide to Digital Signal Processing, California
Technical Publishing, 1. Edition, San Diego, USA.
www.intechopen.com
Sundararajan, D. (2001), The discrete fourier transform: theory, algorithms and applications,
World Scientific Publishing Co. Pte. Ltd., USA.
Swanson, D. (2000), Signal Processing for Intelligent Sensor System, Marcel Deker, 1.Edition,
NY, USA.
Winograd, S. (1976), On Computing the Discrete Fourier Transform, Proceedings of the
National Academy of Sciences, Vol. 73, No. 4, Apr., 1976, pp. 1005-1006.
www.intechopen.com
10
Towards a Reconfigurable FFT : Application

to Digital Communication Systems
Florent Camarda,Jean-Christophe Prevolet and Fabienne Nouvel
IETR/INSA Rennes
France
1. Introduction
Since the beginning of humanity, people have always wanted to communicate with each other.
Today, this objective remains the same although a huge technological step has been taken since
the first communication link. More recently, with the development of electronics systems, the
number of interconnected devices has considerably increased and the quest of performances
becomes more and more a reality.
In this context, digital communication systems have provided an efficient alternative to
classical analog devices. For example, in the telecommunication domain, as analog television
was a great technological progress less than a century ago, 3D HD television sets are currently
being commercialized and will progressively replace the old devices.
The proliferation and variety of such digital devices have also led to the elaboration of
multiple standards destined to ease the exchange of information between the different
devices. Unfortunately, there are still too many standards to improve the portability and
interconnections between different systems. The existence of such an amount of various
standards is related to economical and technological constraints but also to political and
strategical reasons which makes the interoperability between standards very difficult. This
generally imposes the engineers to design as many circuits as standards which seems very
inefficient and may be seen as a waste of time and money. This solution is even worse as a
particular standard does not differ completely from another one. There are generally a lot of
similarities between standards in terms of proposed algorithms and functions. It is sometimes
sufficient to modify some parameters of a specific algorithm to change from one standard to
another.
For example, in current communication systems, many applications referring to several
standards make use of Orthogonal Frequency Division Multiplexing (OFDM) modulations
(Weinstein & Ebert, 1971) such as Digital Terrestrial Television Broadcasting (DTTB), Digital
Audio Broadcasting (DAB), Wireless Local or Personal Area Network (WLAN-WPAN),
HomePlug 1.0 and HomePlug AV, HyperLAN, 802.11 standards, Digital Subscriber Line
standards (xDSL), 3GPP, Long Term Evolution (LTE), etc.
The OFDM principle is mainly based on the use of the Fast Fourier Transform (FFT) and its
Inverse (IFFT) (P. Duhamel & M. Vetterli, 1990). However, the sizes of FFT employed in all
www.intechopen.com
these standards are various. DVB standards use 2048, 4096, 8192, 16384 or 32768 points FFT,
Homeplug standards require 128 or 3072 points FFT, the chinese DTV standard employs a
3780 points FFT. Moreover, some of these communication systems need to implement several
FFT sizes within a particular standard. It is the case for SC-FDMA or for all Time Domain
Synchronous OFDM (TDS-OFDM) systems (J. Wanget al., 2003), in which a second FFT size is
often needed for synchronization.
In order to develop a multi-standards receiver, the solution which is usually retained by
manufacturers consists in implementing various Intellectual Properties (IPs) blocks, each
associated to a unique function and a unique standard. This solution is often under-optimized
in the sense that it does not allow the reuse of components within the chip and thus constitutes
a waste of hardware resources.
Another approach is to design a generic and unique system capable of dealing with the
specifications of different standards. This circuit has to be designed in such a way that it
must be able to adapt itself to a particular standard or a dedicated application.
This concept, known as dynamic reconfigurability, provides not only a high flexibility but also
permits to considerably reduce the time to market and the design effort.
Dynamic reconfigurability is a powerful concept since it enables a circuit to change its
functionality on the fly according to the users’ needs or the environmental conditions.
The dynamic reconfiguration may be performed at different granularities depending on
the technology. For example, some circuits allow to be reconfigured at a gate level by
modifying the contents of a logic equation. This is typically the case of FPGAs circuits (Field
Programmable Gate Array) which are widespread circuits that constitute a good alternative
to ASICs (Application Specific Integrated Circuits). Other circuits may be reconfigured at a
higher level of granularity, for example at a functional level. In this case, a specific operator
(or function) is modified to change the behavior of an application.
In general, the level of performances depends on the granularity of reconfiguration. If the
granularity is very fine, the flexibility is high but the timing performances are very low since
a lot of resources have to be reconfigured. As a consequence, the time which is necessary to
reconfigure the system constitutes a real issue and limits the use of such circuits in real time
embeddded systems where timing constraints are generally tight.
Nevertheless, some circuits make intensive use of reconfiguration at a high level of granularity
and permit to obtain not only high performances but also flexibility. This is generally
performed by reconfiguring specific operators or group of operators in order to change the
design functionality.
In this article, we present a new system capable of dealing with multi communication
standards. The basis of this work has consisted in developing a unique FFT operator which
is capable of being reconfigured to adapt to the different standards’ specification. In our case,
the FFT operator should be able to compute FFTs of size 2K, 4K, 8K and 3780 points.
In section 2, the nature of the FFT operator will be presented as well as the different
possibilities for its implementation. Section 3 will describe the proposed reconfigurable
architecture and obtained performances after implementation. An optimization is presented
in section 4 with its performances results.
www.intechopen.com
Towards a Reconfigurable FFT : Application to Digital Communication Systems 187
2. FFT Algorithms
The Fourier transform is a very useful operator for image or signal processing. Thus it has
been extensively studied and the litterature about this subject is very rich. The Discrete Fourier
Transform (DFT) is used for the digital signal processing and its expression is given in (1)
∞ nk
X (k) = ∑ x (n )e j2π N (1)
n=− ∞
It appears obvious that this expression can not be computed in a finite time due to the infinite
bounds. From that, the usully computed expression is the N-points Fast Fourier Transform
given in (2)
N −1 nk
X (k) = ∑ x (n )e j2π N (2)
n =0
The expression of the FFT is bounded and computable with a finite algorithmic complexity.
This complexity is expressed as an order of multiplications and additions. Computing
a N-points FFT without any simplification requires an algorithmic complexity of O( N 2 )
multiplications and O( N 2 ) where O denotes the "order of" multiplications and additions.
Note that the real number of additions is N ( N − 1) which is O( N 2 ). This reduction of
complexity is however not sufficient for the large FFT sizes that are used in many digital
communications standards.
A great reduction of this complexity can be achieved by using an efficient algorithm. It exists
many FFT algorithms through the litterature like the Radix algorithm (J. W. Cooley & J. W.
Tukey, 1965), the Prime Factor Algorithm (PFA) also called Good Thomas Algorithm (I. J.
Good, 1958). We also can cite the algorithms of Rader-Brenner (C. M. Rader, 1968), Bruun’s
(G. Bruun, 1978) or the Winograd Fourier Transform Algorithm (WFTA) (S. Winograd, 1978).
The most known and used of them is unmistakably the Radix algorithm. However other
algorithms like the WFTA can be very convenient for specific FFT sizes. In this section, we
focus on the development of the Radix algorithm. The Mixed-Radix algorithm will then be
introduced. We also present two advantages of the WFTA and propose a way to combine the
Radix and WFTA algorithms.
2.1 The radix algorithm

In 1965 Cooley and Tukey proposed a new way to compute the FFT for sizes that are power of
2, this is the birth of the Radix algorithms. The basic principle is based on the decomposition
of a FFT of size N in two FFTs of size N2 .
2.1.1 Decimation in time / Decimation in frequency

This decomposition can be achieved by two means which are called Decimation In Time (DIT)
and Decimation In Frequency (DIF) depending on the regrouped terms. Equation (3) present
the decomposition for the DIT while equations (4) and (5) are for the DIF. Thanks to this the
2
algorithmic complexity becomes reduced to O( N
2 ) since N = 2.
www.intechopen.com
N −1 nk
X (k) = ∑ x (n )e j2π N
n =0

k N
DFT N x (2n ′ ) + e j2π N DFTN/2 x (2n ′ + 1) with n ′ ∈

= 0, ..., −1 (3)
2 2
This decomposition is called DIT because the N time samples are reordered in two groups of
N/2 samples. One group contains the even indexed samples, while the second one contains
the odd indexed samples. FFT of size N/2 are then computed for each group. The frequency
samples are computed in order.
N −1 n2k ′
X (2k′ ) = ∑ x (n )e j2π N
n =0

N N
= DFT N x (n ) + x (n + ) with k′ ∈ 0, ..., − 1 (4)
2 2 2
N −1 n (2k ′ + 1 )
X (2k′ + 1) = ∑ x (n )e j2π N
n =0

N n N
= DFT N x (n ) − x (n + ) e j2π N with k′ ∈ 0, ..., − 1
2 2 2
This decomposition is called DIF because the even indexed frequency samples are computed
by the first N/2-points FFT while the odd indexed samples are computed by the second one.
The time samples are not reordered before computation. In order to simplify these equations
x
all terms in e j2π N are usually expressed as WN
x.
x(0) + X(0)
N/2-point WN0
FFT
N/2 N/2 N/2
Even index
x(N − 2) + X( N2 − 1)
N
−1
WN2
x(1) + X( N2 )
N
N/2-point WN 2
FFT
N/2 N/2 N/2
Odd index
x(N − 1) + X(N )
WNN −1
Fig. 1. Radix-2 Decimation In Time
www.intechopen.com
Figure (1) represents the computation of the previous DIT. With this structure it is still
necessary to perform N multiplications by the twiddle factors WN k to compute the N FFT
outputs. However it is possible to use a property of the twiddle factors expressed in (5), to
remove half of these multiplications, by replacing some additions by substractions.
x + N2 x
WN = − WN (5)
Then the opimized DIT decomposition will be computed as presented in figure (2)
x(0) + X(0)
N/2-point
FFT
N/2 N/2 N/2
Even index
x(N − 2) + X( N2 − 1)
x(1) − X( N2 )
N/2-point WN0
FFT
N/2 N/2 N/2
Odd index
x(N − 1) − X(N )
N
−1
WN2
Fig. 2. Simplified Radix-2 Decimation In Time
Figure (3) presents the computation scheme for the DIF decomposition. In this last one, it
appears that the N time samples are precomputed in order while the frequency samples
are reordered in an even indexed group and an odd indexed group. This decomposition
also requires N 2 multiplications by the twiddle factors and use the same Radix-2 butterfly
in pre-computation that was used for the post computation of the DIT computation.
For these last two schemes, the algorithmic complexity is again reduced of N2 multiplications
2
but is still O( N2 ). However, if the FFT size is power of 2, this decomposition can be performed
until requiring O( N 2 log2 ( N )) 2-points FFT butterflies. Because the computation of a 2-points
FFT, represented figure (4), does not involve any multiplication, such decomposition requires
only multiplications by the twiddle factors. This leads to an algorithmic complexity of
O( N2 log2 ( N )).
Figure 5 shows the result of such a decomposition for a 8-points FFT with both decimations.
The first thing that we can observe on this picture is that many of N 2 log2 ( N ) twiddle factors
are WN 0 = 1 which allows a reduction of the algorithmic complexity. A second important
point is the high regularity in the data flow mapping. The data reordering is also very regular
and can be determined by the "bit reversed" algorithm.
While all this study has been led for the Radix-2 algorithm, it is possible to generalize it to
other Radix, and to the Radix-4 in particular.
www.intechopen.com
x(0) + X(0)
N/2-point
FFT
N/2 N/2
Even index
x( N2 − 1) + X(N − 2)
x( N2 ) − X(1)
WN0 N/2-point
FFT
N/2 N/2
Odd index
x(N − 1) − X(N − 1)
N
−1
WN2
Fig. 3. Radix-2 Decimation In Frequency

(0) + (0) = (0) + (1) (0) (0)
(1) ¡ (1) = (0) ¡ (1) (1) (1)
Fig. 4. Radix-2 Butterfly

x(0) X(0) x(0) X(0)
x(4) X(1) x(1) X(4)
W40 W40
x(2) X(2) x(2) X(2)
W41 W41
x(6) X(3) x(3) X(6)
W80 W80
x(1) X(4) x(4) X(1)
W81 W81
x(5) X(5) x(5) X(5)
W41 W82 W82 W40

x(3) X(6) x(6) X(3)
W40 W83 W83 W41

x(7) X(7) x(7) X(7)
(a) DIT (b) DIF
Fig. 5. 8-points FFT by using Radix-2 algorithm with DIT or DIF
2.1.2 Radix-R algorithm

As the Radix-2 butterfly computes a FFT of size 2, the Radix-4 butterfly does the same for FFTs
of size 4. The four outputs of this butterfly only need trivial multiplications by {−1, 1, − j, j }
as presented in the equation (6).
X (0) = x (0) + x (1) + x (2) + x (3)

X (1) = x (0) + jx (1) − x (2) − jx (3)
X (2) = x (0) − x (1) + x (2) − x (3) (6)
X (3) = x (0) − jx (1) − x (2) + jx (3)
www.intechopen.com
This butterfly however presents a gain regarding the algorithmic complexity when it is used
instead of the Radix-2 by reducing the necessary number of stages in the decomposition. The
major drawback of using this algorithm is that it is suitable for FFT sizes that are power of 4,
which reduces considerably the field of applications.
As with the Radix-2 or the Radix-4 algorithm, it is possible to use the Radix-R algorithm for
other R values. Nevertheless, the Radix-R butterfly will require no trivial multiplications
into the butterfly and thus increase its algorithmic complexity. The equation 7 shows that
Radix-3 butterfly computation requires 4 complex multiplications by coefficients around the
unit circle. Pay attention not to confuse these coefficients with the twiddle factors needed
between the stages.
X (0) = x (0) + x (1) + x (2)

X (1) = x (0) + W31 .x (1) + W32 .x (2) (7)
X (2) = x (0) + W32 .x (1) + W31 .x (2)
Except for R multiple of 2, where it may be decomposed by using R = 2 or R = 4, a Radix-R

butterfly requires ( R − 1)2 muliplications and R( R − 1) additions. Thus the complexity
becomes O( R2 logR ( N )). This is the reason why sizes which are power of two are promoted
in most of digital communication standards.
2.2 Mixed-radix algorithm

The Radix-2 algorithm proposed by Cooley and Tukey can be derived to provide the
mixed-Radix algorithm (R. C. Singleton, 1969) (G. L. DeMuth, 1989). This last one employs
the combination of different Radix-R algorithms allowing the computation of more FFT sizes.
For example, to compute a 12-points FFT, the size N can be decomposed in N = 2 × 2 × 3.
In this configuration, the FFT will be computed by a first and a second stage of six Radix-2
butterflies and a third stage of four Radix-3 butterflies. Nevertheless this FFT can also be
computed by other decompositions such as N = 2 × 3 × 2, N = 3 × 2 × 2, N = 3 × 4 or
N = 4 × 3. Due to the reduced number of computation stages and to the optimal complexity
of the Radix-4 butterfly, the last two decompositions will require the lower complexity.
To summarize, the complexity of a Radix or Mixed-Radix algorithms will depend on the
complexity of the Radix-R butterflies that are used and the number of twiddle factors
necessary between the stages.
2.3 Advantages of the Winograd fourier transform algorithm

In 1978, S. Winograd presented his works on the computation of the DFT by using cyclic
convolution and introduced the Winograd Fourier Transform Algorithm which reduces the
algorithmic complexity of the FFT. This reduction is achieved thanks to two points.
The first gain comes from the computation of butterfly of size R, which may be computed with
a complexity of O( R) instead of O( R2 ). The algorithms for R ∈ {2, 3, 4, 5, 7, 8, 9} are available
in (S. Winograd, 1978).
The second optimization is performed thanks to an ingenious mapping of the data flow. This
mapping does not implement any twiddle factors between stages. However, this mapping
can only be realized for FFT sizes which can be decomposed in several terms Ri that must be
mutually primes. This is the major constraint of this type of optimization.
www.intechopen.com
For a FFT of size N that can be decomposed in N = R1 × R2 which are mutually primes, the
mapping will be determined by the following rules.
First the input data vector [ x (0)...x ( N − 1)] have to be reordered in a two dimensionnal matrix
of size ( R1 , R2 ). The position (rn , cn ) of the sample x (n ) in this matrix is given by the relation
(8).
rn = n mod R1
cn = n mod R2 (8)
Then the computation of the Winograd butterfly R1 will be computed over the R2 groups of
data which will be taken column-wide and similarly for the R2 butterfly.
If the FFT size has to be decomposed in I > 2 terms (still mutually primes), this mapping can
be applied with a two dimentionnal matrix of size ( R1 , ∏iI=2 Ri ). Thus the computation of the
∏iI=2 Ri = RN1 will be performed using the same algorithm with the dimensions ( R2 , ∏iI=3 Ri )
for the R1 groups generated by the first grouping.
2.4 Combined radix and WFTA algorithms

As the decomposition of the FFT in small-N FFTs was explained previously, it is possible to
regroup and compute these small-N FFT as desired. So it is possible to compute a part of the
FFT using the WFTA algorithm and the other part using the Mixed-Radix algorithm. However
the decomposition has to be properly performed in order to take advantage of the properties
of these two algorithms.
As the WFTA has better performances than the Radix algorithm, it is preferable to first regroup
the mutally primes factors and compute a small FFT with the WFTA. Then the other terms will
require a Mixed-Radix mapping. However the first advantage of the WFTA which reduce the
complexity of the butterflies have to be use for Radix-R butterflies when R ∈ / {2, 4}1 .
Another optimization can be achieved by an efficient organization of the Radix/WFTA FFT.
The number of different twiddle factors NbW that will have to be stored in memory for the FFT
computation can be reduced if the different stages are well organized. This number of different
i0
coefficients is O( ∏ ( Ri ) for the stage i0 , i0 = 1 with the DIT computation, or for the I − i0 ,
i =1
0 = 0.
i0 = 1 for the DIF. For i0 = 1 all the coefficients are WN
3. Proposed architecture
3.1 Presentation
In order to design the most flexible architecture capable of implementing the FFT, we
decomposed the algorithm in several WFTA or Radix blocks depending on the considered N
size. For example, a 64-points FFT may be decomposed in N = 4 × 4 × 4 computation stages,
each stage being configured using the Radix-4 butterfly. In the case of a 3780-points FFT, it may
be decomposed in N = 3 × 3 × 3 × 4 × 5 × 7 which leads to implement the WFTA-3, Radix-4,
WFTA-5 and WFTA-7 butterflies (Z.-X. et al., 2002).
Such decompositions make it possible to consider a reconfigurable WFTA/Radix pipelined
architecture that allows to compute several sizes of FFT by reconfiguring each stage’s butterfly.
1 in fact the WFTA butterfly complexity is the same as the Radix butterfly for R = 2 or R = 4
www.intechopen.com
In this section, we focus on the presentation of an architecure dedicated to a multiple Digital

Terrestrial Television Broadcasting (DTTB) demodulation. The concerned sandards are DVB-T
(ETSI, 2004), ISDB-T (ARIB, 2001) and DTMB (Chinese National Standard, 2006) for which the
possible FFT sizes are 2048 (2K), 4096 (4K), 8192 (8K) and 3780 points. Nevertheless, it is
possible to adapt the architecture to other FFT sizes. Table (1) presents a way to decompose
these sizes by implenting the different Radix or WFTA butterflies. The presence of a Radix-8
butterfly is necessary to compute the 8192-points FFT in six stages, but another stage of
computation can be added in order to split this butterfly in two by using Radix-4 and Radix-2
butterflies.
As shown in section 2.4, the number of different twiddle factors coefficients that will have
to be computed depends on the placement of the different butterflies. To provide the output
samples in order, a DIT computation is considered. With such a decomposition, the quantity
of required twiddle factors will be minimized by placing the higher radix butterflies at the last
stages of computation. Moreover, as the second advantage of the WFTA can be employed in
the 3780 decomposition for the 3 × 4 × 5 × 7 = 420 part, all these 420-points FFT computations
have to be processed in the last stages.
Butterflies Configuration
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Stage 6
3780 WFTA-3 WFTA-3 WFTA-3 Radix-4 WFTA-5 WFTA-7
2048 Radix-2 Radix-4 Radix-4 Radix-4 Radix-4 Radix-4
Table 1. A possible decomposition of the concerned FFT sizes
Thus, such a decomposition in six stages of computation may be performed by the 6-stages
reconfigurable architecture presented in figure (6).
Each stage is built on a reconfigurable WFTA/Radix module implementing the corresponding
butterfly. The module may be reconfigured depending on the chosen FFT size. The butterfly
input and output data are stored in the symbol memory. The address decoders select the
memory address to read the inputs and write the results. A twiddle block is also implemented
at each stage of the computation (except for the last one). This block is located between
the butterfly module and the symbol memory and performs the multiplications by the
appropriate twiddle factors. The input and output buffers are used to respectively store the
time domain and frequency domain OFDM samples. Finally, the “memory data and address”
multiplexers switch the memories between stages of computation.
3.2 Operating description

The N complex samples corresponding to the Sn OFDM time domain symbol are stored in
the input buffer. During this storage, the WFTA/Radix module of the first stage processes
its computation according to the required configuration (Radix-2, WFTA3 or Radix-4). This
process is performed on the N previously stored samples that constitute the Sn−1 OFDM
symbol.
Each subsequent module performs the same operation in parallel until the WFTA/Radix
module in the stage 6, that operates on the Sn−6 OFDM symbol. While these computations are
performed, the frequency domain samples corresponding to the FFT of the Sn−7 are available
at the output buffer.
www.intechopen.com
Address Address Address Address Address Address Address Address

Decoder Decoder Decoder Decoder Decoder Decoder Decoder Decoder
Input Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Stage 6 Output
R W R W R W R W R W R W R W R W
Dual Port Memories Address Multiplexor
Input Symbol Symbol Symbol Symbol Output

Symbol Symbol
Buffer Memory Memory Memory Memory Memory Memory Buffer
Sn Sn−1 Sn−2 Sn−3 Sn−4 Sn−5 Sn−6 Sn−7
Symbol Symbol Symbol Symbol Symbol Symbol Symbol Symbol
Dual Port Memories Data Multiplexor
Data Data
Twiddle Twiddle Twiddle Twiddle Twiddle
Input utput
Block Block Block Block Block
Radix/WFTA Radix/WFTA Radix/WFTA Radix/WFTA Radix/WFA Radix/WFA

econfigurable econfigurable econfigurable econfigurable econfigurable econfigurable

Control Control Control Control Control Control
External Control
Reconfigurable Synchronisation
Fig. 6. 6 stages reconfigurable architecture for FFT
As soon as the Sn symbol is completely stored, the different computation stages must be
completed. Memories are then circularly switched i.e. the previously input buffer becomes the
symbol memory 1, the symbol memory i becomes the symbol memory i + 1 and the output
buffer becomes the new input buffer. This permutation is performed by the two memory
multiplexers, one for the data bus, the other for the address bus.
When the algorithm requires the twiddle factors multiplications, these operations are
computed by the twiddle block at the output of the butterfly, before being stored in the symbol
memory.
3.3 Presentation of the Processing Blocks

3.3.1 WFTA/radix module
Each WFTA/Radix module is composed of three stages. A first stage aims to compute
additions, a second performs the multiplications by coefficients and the last adder stage
generates the output results. The figure (7) presents the architecture of such a module.
First, the set of butterfly inputs are stored in the butterfly input register. As soon as possible,
the adder/substractor starts to process two values. These values are either the time domain
samples stored in the register or a previous result which comes from a reconfigurable delay
block. The second stage performs a multiplication of the results coming from the first stage by
the coefficients which are stored in the coefficient memory. As the coefficients are either real or
www.intechopen.com
Data from Data toward
Symbol memory Symbol memory
econfigurable Delay econfigurable Delay
Butterfly
Datas Datas Datas
Input MUX
+/- egister
MUX
+/-
egister Formating Formating Formating
econfigurable Delay Butterfly coefficient memory econfigurable Delay
Fig. 7. Architecture of a WFTA/Radix module
pure imaginary, only two real multipliers are necessary to implement the complex multiplier.
Since there are only few coefficients, their resource utilization is negligible compared to the
resources required by symbol memories. Finally, the output adder/substractor part operates
similarly to the first one and provides the butterfly outputs.
3.3.2 Twiddle blocks

As no twiddle factors are needed at the output of the last stage, the twiddle block does not
appear on the figure 6. The architecture of the twiddle blocks is depicted in figure (8). This
block just performs the complex multiplications by the twiddle factors stored in the associated
memory. When multiplications are not necessary, a multiplication by WN 0 = 1 is computed to
maintain a constant propagation delay through the architecture.

Such a multiplication can be implemented by either four real multipliers and four real adders
or three real multipliers and five real adders.
Data toward the symbol memory
Data Formating
Twiddle factors
Memory
Data from the butterfly module

Fig. 8. Twiddle block
www.intechopen.com
3.3.3 Symbol memories, address decoders

The symbol memories have to be connected to the right stage depending on the computation
step. Then, during the computation stage, the memory will send data to the butterfly while
recording data from the twiddle block. The time domain samples are overwritten by the
frequency domain samples. Nevertheless, the read and write addresses are always different
at the same time, due to the butterfly latency. For that purpose, a Simple Dual Port Random
Acces Memory (SDP-RAM) have been implemented. The size of this Random Access Memory
(RAM) has been defined according to the maximum number of points to be computed in the
design (8192 samples). Each complex sample is coded into 32 bits (16 bits for the real part
and 16 bits for the imaginary part). Thus, the size of the symbol memory has been set to
2 × 16 × 8192 bits which corresponds to 32kB.
The address decoders compute the right sequence of memory address that have to be read
and written. Thus an address decoder block is dedicated to a stage of computation. As the
size of the FFT can be reconfigured, these blocks also have to be reconfigured.
3.4 Performances of the 6-Stages architecture

The complete architecture has been simulated and implemented on an Altera stratix
EP3SE50F484C2. This FPGA has been envisaged as a prototyping circuit for future
implementation in an Application Specific Integrated Circuit (ASIC). The considered FPGA
exhibits a lot of logical resources as well as dedicated blocks that are optimized for MAC
(Multiplications/Accumulations) operations and memory. Thus, it is perfectly suited for the
considered types of applications.
3.4.1 Resources usage

Implementation results for the 6-stages architecture are provided in Table (2). In this table, the
implemented architecture is compared to commercial IPs provided by Altera (Altera, 2009) in
terms of resource usage.
Logic Cells Registers Memory bits DSP blocks
Available on Stratix III 38,000 38,000 5,455,872 384
6-stages architecture 4,077 2,544 3,769,080 44
Altera 2K IP 4,138 6,943 208,329 40
Altera 4K IP 4,557 7,530 425,563 40
Altera 8K IP 5,270 8,670 884,785 48
Table 2. Resources Utilization for the 1-Stage and the 6-Stages Architecture After
Implementation on Stratix III
It seems important to notice that the proposed architecture requires less logic cells, registers
and DSP block than its counterparts. This is especially true if we consider that this architecture
is also able to compute different sizes of FFT (2K, 3780, 4K, 8K).
Another interesting point is that the memory resource is the most consuming in this
architecture. This is mainly due to the parallelization of several stages in the architecture.
Thus the equivalent of six OFDM symbols must be stored in memory.
Globally, implementation results demonstrate the feasibility of implementing such
architectures on reconfigurable circuits without consuming too many resources. This makes
www.intechopen.com
it possible to envisage the implementation of this architecture onto ASICs which generally
present higher performances and exhibit more logical resources.
3.4.2 Time performances

The time that is required to store a complete symbol into the input buffer RAM is the OFDM
symbol duration TOFDM . This duration depends on the sampling frequency, the size of the
IFFT and the guard interval duration, used for the transmission. The minimun TOFDM and
associated maximum OFDM throughput for each standard and FFT sizes are provided in
Table (3).
DVB ISDB TDMB
Duration (µs) - - 555
3780
Throughput (symb/s) - - 1802
Duration (µs) 231 259 -
2048
Throughput (symb/s) 4329 3861 -
4096
8192
Table 3. Minimum OFDM symbol duration with Guard Interval and corresponding
throughput
The architecture must, at least, respects these performances to be suitable in the OFDM
receiver. Table (4) describes the number of clock cycles required to compute a complete stage
of the FFT. According to this table, it may be seen that the WFTA-7 stage demands the most
important computation time per sample.
FFT size 3780 2K 4K 8K
WFTA7 11351 - - -
WFTA5 7570 - - -
WFTA3 3787 - - -
Radix-8 - - - 16398
Radix-4 3787 2055 4103 8199
Radix-2 - 2055 - -
Table 4. Module Execution Time in Number of Clock Cycles.
In order to provide a comparison between the proposed architecture and existing solutions,
several implementation tests have been led. The obtained results are presented in Table (5)
which summarizes the symbol rate as well as the computation latencies for the proposed
architecture according to different configurations. The comparison has been performed with
different IPs of Altera implementing the corresponding sizes of DFT. The maximum rate
that may be achieved is conditioned by the WFTA/Radix Module 6 which exhibits the most
important complexity.
Regarding Table (5), for each FFT size, the 6-stages architecture delivers a similar symbol rate
than the commercial IPs. Note that a significant difference remains for the 8192 points FFT.
This is due to the internal structure of our proposed architecture that only consists of 6 stages
www.intechopen.com
Proposed 6-stages
Altera IP
Architecture
Symbol rate 2K 48,661 48,828
@ 100MHz 4K 24,372 24,414
8K 6,098 12,207
(OFDM symbols / s)
3780 8,809 NC
2K 10,282 6,144
Latency
4K 20,522 12,288
8K 49,201 24,576
(clock cycles)
3780 30,289 NC
Table 5. Symbol Rate and Latency According to Different Configurations
of computations. In order to compute a 8K FFT, it is then necessary to use a Radix-8 Module

which requires much more computing cycles compared to Radix-4 or Radix-2 Modules.
Concerning the latency, an increase has to be deplored in the proposed architecture. This is
mainly due to the time that is necessary to store a complete symbol into an intermediate buffer
between two consecutive stages of computation. Nevertheless, the drawback of latency may
be compensated by the high symbol rate which is available at the output of the design.
4. Optimization for the studied applications

The previous architecture is optimized for a high symbol rate. However, it is possible to
modify it by reducing the number of stages from six to one as described in figure (9).
4.1 Operating description

According to the figure (9), during the storage of a current symbol in the input buffer, the
unique WFTA/Radix Module will be reconfigured to compute the different stages of the
whole FFT. The connected symbol memory is employed for the computation of all stages.
At the end of the computation, the input buffer becomes the new symbol memory, the symbol
memory becomes the output buffer with the FFT transform results and the previous output
buffer becomes the new input buffer which will store the new incoming OFDM symbol.
This operating mode allows a flexible number of computation stages. Thus the 8192-points
FFT may be computed by using only Radix-4 and Radix-2 algorithms in seven stages of
computation (8192 = 4 × 4 × 4 × 4 × 4 × 4 × 2).
The used module can be implemented in 6 possible configurations: Radix-8, WFTA7, WFTA5,
Radix-4, WFTA3 and Radix-2. All combinations of these possibilities can be envisaged to
compute several FFT sizes. The number of stages is related to the number of terms in the
FFT size decomposition. Nevertheless, an increase of this number implies a reduction of the
symbol rate.
Furthermore, for particular sizes of FFT which are neither power-of-2 nor adapted to the
WFTA mapping algorithm, new Twiddle factors have to be calculated and stored in the
twiddle block memory. In order to take into account all possible configurations, it is possible
to implement an external memory which contains all these coefficients and load them in the
Twiddle memory block according to the chosen FFT size.
www.intechopen.com
Address Address Address

Decoder Decoder Decoder
Input Stage 1 Output
R W R W R W
Dual Port Memory Address Multiplexor
Input Output
Symbol
Buffer
Memory Buffer
Dual Port Memory Data Multiplexor
Data Twiddle Data
Block
Input utput
Radix/WFTA
All stages
econfigurable
Control
Stage 1
External Control
Reconfigurable Synchronisation
Fig. 9. 1-stage reconfigurable architecture for FFT
4.2 Reconfiguration of the architecture

As the reconfiguration of the WFTA/Radix Modules for the 6-stages architecture occurs only
when the FFT size is modified, the reconfiguration is completely static. On the other hand,
the Butterfly module of the 1-stage architecture must be reconfigured during runtime. This
reconfiguration is possible almost instantaneously. In fact, only the control of the resources
has to be reconfigured. Thus it is possible to reconfigure it by using a multiplexed control.
www.intechopen.com
4.3 Resources Usage for the 1-Stage architecture

The 1-stage architecture has been implemented into the same FPGA target. The corresponding
resources usage is summarized in Table (6). In this configuration, the obtained results are more
competitive than Altera IPs, especially for memory usage. This may easily be seen when
comparing the number of memory bits in the 8K mode. Moreover, it is clear that logical
resources are reduced significantly.
Logic Cells Registers Memory bits DSP blocks
Available on Stratix III 38,000 38,000 5,455,872 384
1-stage architecture 811 548 622,896 8
Altera 2K IP 4,138 6,943 208,329 40
Altera 4K IP 4,557 7,530 425,563 40
Altera 8K IP 5,270 8,670 884,785 48
Table 6. Resources Utilization for the 1-Stage and the 6-Stages Architecture After
Implementation on Stratix III
4.4 Timing considerations for the 1-Stage architecture

The execution times and latency that are depicted in Tables (4) and (5) are given for the 6-stages
architecture but are still available for the 1-stage architecture. On the other hand, a comparison
between the symbol rate performances achieved by the 1-stage architecture and the required
symbol rate imposed by the DTTB standards are provided in Table (7).
FFT 1-stage architecture
DVB-T/H ISDB-T TDMB FLO
Size @ 100MHz
2048 8110 4329 3861 - -
4096 4062 2164 1926 - 1200
8192 1742 1082 962 - -
3780 2935 - - 1801 -
Table 7. OFDM Symbol Rate Comparison Between Studied Standards and Proposed
Architecture
According to these results, we may conclude that the optimized structure is still suitable for
the FFT computation of the considered standards. However a compomize is always possible
by combining the two methods seen in sections 3 and 4
5. Conclusion
In this chapter, we have presented two algorithms that may be combined to compute a FFT.
Depending on the size of this transform, some advantages can be exploited by taking into
account a meticulous organization in the algorithms combination. Thus, an optimal reduction
of the algorithmic complexity will imply an optimal use of the hardware resources. Moreover,
an architecture has been proposed for the computation of these algorithms. This architecture is
able to deal with a large amount of FFT sizes, decomposable in product terms that are 2,3,4,5,7
or 8. A growth either of the largest FFT size or of the number of reconfigured sizes imply the
use of more memory resources which is the most delicate point of the architecture.
www.intechopen.com
A major advantage of this architecture is its possibility to be dynamically reconfigured

from one to another FFT size. This allows to reuse the same circuit to compute several
FFT sizes within the same modulation/demodulation standard, as it is necessary with the
DTMB standard or the SC-FDMA modulation. As explained in section 4, this architecture
provides a high flexibility that permits to achieve the best compromize between hardware
resources and computation throughput. For prototyping purposes, this architecture has been
successfully simulated and implemented in a FPGA. Nevertheless, the architecture targets an
implementation in ASIC circuits whose technology exhibits higher performances than FPGA.
Furthermore, the field of application is wide for this architecture. As mainly expressed, the
convergence of multiple standards in a same device may be performed without implementing
several modulation or demodulation circuits. Another example may be to envisage an
implementation of OFDM/MIMO (Multiple Input Multiple Output) systems since the
number of FFTs directly varies with the number of antennas in this type of applications.
Finally, these concepts may also be used in a software radio context since it deals with
reconfiguration of resources according to the channel or to environmental changes.
6. References
Weinstein, S. B. & Ebert, P. M. (1971). Data transmission by frequency-division multiplexing
using the discrete fourier transform, In : IEEE Transactions on Communication
Technology, vol 19, pp. 628-634
J. Wang, Z. Yang, C. Pan, M. Han & L. Yang (2003). A combined code acquisition and symbol
timing recovery method for TDS-OFDM , In : Broadcasting , IEEE Transactions on, vol.
49, pp. 304-308
I. J. Good (1958). The interaction algorithm and practical Fourier analysis , In : J. R. Statist.
Soc., ser. B, vol.20, pp. 361-372
P. Duhamel & M. Vetterli (1990). Fast Fourier transforms: a tutorial review and a state of the
art, In : Signal Processing, vol. 19, pp. 259-299
J. W. Cooley & J. W. Tukey (1965). An algorithm for the machine calculation of complex fourier
series , In : Proc. Math. Comp., pp. 297-301
S. Winograd (1978). On computing the Discrete Fourier Transform, In : Proc. Math. Comp., pp.
175-199
C. M. Rader (1968). Discrete Fourier transforms when the number of data samples is prime, In
: Proc. IEEE, vol. 56, pp. 1107-1108
G. Bruun (1978). z-Transform DFT filters and FFTs, In : IEEE Trans. on Acoustics, Speech and
Signal Processing (ASSP), vol. 26, pp. 56-63
R. C. Singleton (1969). An algorithm for computing the mixed-radix fast fourier transform, In
: IEEE Transactions on audio and electroacoustics, vol 17, NO.2, pp. 93-103
G. L. Demuth (1989). Algorithms for defining mixed radix FFT flow graphs, In : IEEE
Transactions on acoustics, speech and signal processing, vol 37, NO.9, pp. 1349-1358
ETSI (2004). Digital Video Broadcasting-Terrestrial (DVB-T); Framing structure, channel
coding and modulation for DTV, In : ETSI standard, Nov. 2004
ARIB (2001). Integrated Services Digital Broadcasting-Terrestrial (ISDB-T); specification on
channel coding, framing structure and modulation, In : ARIB standard, May 2001
www.intechopen.com
Chinese National Standard (2006). Framing Structure, Channel Coding and Modulation for
Digital Television Terrestrial Broadcasting System, In : Chinese National Standard GB
20600-2006
Z.-X. Yang, Y.-P. Hu, C.-Y. Pan & L. Yang (2002). Design of a 3780-point IFFT processor for
TDS-OFDM, In : Proc. IEEE, vol. 48, pp. 57-61
Altera (2009). In : FFT mega core function user guide
www.intechopen.com
11
Fourier Transform Based Transmission

Systems for Broadband Wireless
Communications
Mingqi Li, Yun Rui and Zhiyong Bu
Shanghai Institute of Microsystem and Information Technology, & Shanghai Advanced
Research Institute, Chinese Academy of Sciences, CAS
China
1. Introduction
In recent years, Fourier Transform (FT), as an effective signal processing technology, is more
and more popularly applied to wireless communications. By the FT technologies, it can not
only reduce the implementation complexity of traditional transmission systems, but also
bring in some new features, thus constructing new transmission systems. The current main-
stream transmission schemes utilizing DFT technologies include Orthogonal Frequency-
Division Multiplexing (OFDM) [1], Discrete Fourier Transform Spread Orthogonal
Frequency Division Multiplexing (DFT-S-OFDM) [2] and Filter bank modulation [3]. For
OFDM systems, by an IDFT at the transmitter, the whole frequency-selective wideband
channel is divided into several flat narrow band sub-channels, which is benefit to overcome
the effects of multi-path in wireless channels. For DFT-S-OFDM system, the uplink
transmission scheme for 3GPP-Long Term Evolution (3GPP-LTE) standard, besides the
IDFT served the same function as in the OFDM systems, additional DFT processing is
performed to the transmitted constellation symbols before OFDM modulation. In this way,
the whole modulation method can be viewed as a DFT-based interpolation processing, and
the modulated signals can be regarded as single carrier signals with low Peak-to-Average
Power Ratio (PAPR) property. For filter-bank systems, the FT can be used both to reduce
implementation complexity and to construct the cyclic prefix (CP) based block transmission
scheme, which has merits of both the filter-bank systems with robustness against to multiple
access interference (MAI) and the CP based block transmission systems with simple
frequency equalization.
The chapter is organized as follows. Firstly, the implementation structure of OFDM
transmitter, time-frequency description of OFDM signals, and effects of timing- and
frequency-offset and channel multi-path are discussed detailed. Then, we present DFT-S-
OFDM system model, describe time-frequency property of DFT-S-OFDM signals, analyze
the effects of carrier frequency-offset (CFO) quantitatively and compare the SIR and PAPR
performances with that of OFDM systems. Next, a DFT spread Generalized Multi-Carrier
(DFT-S-GMC) system is presented. The time-frequency properties of DFT-S-GMC signal, the
DFT-based implementation method and the receive SINR are addressed. Finally,
conclusions are collected.
www.intechopen.com
2. OFDM transmission systems

OFDM plays a significant role in modern broadband communication systems. The
wireline high-speed access technology, i.e. Asymmetric Digital Subscriber Line (ADSL),
was the first widely used application for the FT-based OFDM system. Several other
wireless standards, such as the IEEE 802.11a Wireless Local Area Network (WLAN) and
IEEE 802.16 (WiMAX) series, have adopted OFDM as a key transmission technology. IEEE
802.20 working group on mobile broadband wireless access uses OFDM as the wireless
high speed transmission technology. In the area of cellular mobile communications,
OFDM was also adopted as a basic downlink transmission scheme of 3GPP-Long-Term
Evolution (3GPP-LTE) standard and the incoming 3GPP-LTE-Advanced standard [6].
OFDM is also widely applied in the areas of audio and video broadcasting. Digital Audio
Broadcasting (DAB), initiated as a European research project in 1980s, adopts coded
OFDM as the transmission technology. DVB-T based on OFDM in an 8 MHz channel is
now a popular technology for terrestrial video broadcast in the world. An additional new
application area of OFDM is in Ultra-Wideband (UWB) personal area networks [4].
2.1 OFDM system model
specific K data symbols are {ak } , 0 ≤ k ≤ K − 1 . After the OFDM modulation, the transmit
Figure 1 illustrates the principle structure of OFDM transmitter. Assume that the user-
signals can be expressed as
sn =
1 K −1
N k =0
( )
∑ ak exp j 2π n ( k0 + k ) / N ,
(1)
n = − N g ,..., −1,0,1,..., N − 1
where N g is the CP length. To simplify the system model, the localized sub-carrier
⎧ ak , k ' = k0 + k
mapping is applied, i.e., bk ' = ⎨ , k ' = 0,..., N − 1 , k0 is the user-specific sub-
⎩ 0 otherwise
carrier allocation offset.
0 b0
a0
b1
a1 Sub- CP sn
N-point
. carrier .
.. . Padding
Mapping . IDFT
aK −1
0 bN −1
Fig. 1. OFDM transmitter
www.intechopen.com
Fourier Transform Based Transmission Systems for Broadband Wireless Communications 205
2.2 Time-frequency properties of OFDM signal

2.2.1 Time-frequency description of OFDM signal
Fig. 2 describes the time-frequency property of OFDM signal. According to the
implementation principle, OFDM is a block-based transmission scheme. As a result, each
OFDM symbol has a rectangle waveform in time-domain. The rectangle waveform can be
expressed as
⎧⎪1 , |t|< T / 2
p (t ) = ⎨
⎪⎩0
(2)
, otherwise
function shape in frequency-domain. Namely, the Fourier transform of p ( t )

By the properties of the Fourier transform, the spectrum of each sub-carrier has a sinc-
is P ( f ) = sinc ( πTf ) πf . As it is shown, the spectrum function P ( f ) has following properties
⎧⎪1 , f = 0
P( f ) = ⎨
⎪⎩0 , f = n / T , ( n = ±1, 2,...)
(3)
Equation (3) shows that P ( f ) , P ( f − Δf ) ,...., P ( f − nΔf ) are orthogonal to each other, and
Δf = 1 T . This means although the spectrum of sub-carriers are overlapped each other, the
orthogonality among each sub-carriers can be maintained when the sub-carrier spacing is
set to be Δf .
Fig. 2. The time-frequency property of OFDM signal
2.2.2 Cyclic prefix and frequency-domain equalization

When the transmitted signal in time domain is sn and the discrete Channel Impulse
Response (CIR) is { hi } Ki =−01 ，the received signal can be given as
∑ hi sn −i + wn
K −1
rn = (4)
i =0
where wn is the Additive White Gaussian Noise (AWGN).

According to the DFT theory, the frequency-domain multiplication is equivalent to the time-
domain circular-convolution. When the signal sn , sn = sn , n = 0,..., N − 1 , is passed through
www.intechopen.com
the channel, only linear-convolution with the CIR occurs. To mimic circular-convolution, we
can copy a partial of samples in the tail of sn , and pad to the head of sn to construct a CP.
Usually the length of CP is kept greater or at least equal to the length of the channel delay
spread, i.e., L . Consequently, by performing DFT to the time-domain signal rn , we can get
the frequency-domain received signal in the k -th sub-carrier
Rk = bk H k + Wk , k = 0,..., N − 1 (5)
where H k is channel frequency response and H k = ∑ i = 0 hi e j 2 πik /N . Wk is the AWGN in

L −1
frequency domain. bk is the transmitted modulated symbol in the k -th sub-carrier as

described in Equation (1).
To recover the transmitted symbol, we can apply single-point frequency-domain
equalization to each sub-carrier. For example, using ZF equalization, the desired signal can
be denoted as
bˆk = k , k = 0,..., N − 1
R
(6)
Hk
2.2.3 Effects of time and frequency offset on demodulated signal

a. Effect of timing offset
Symbol timing is one of the key factors in OFDM synchronization, which decides the
accurate selection of FFT window starting position for OFDM demodulation. As shown in
Fig. 3, the ideal synchronization position of symbol timing is the first sample of the
transmitted signal removing the CP. If the FFT window is opened at the permitted zone, the
received signal will produce no Inter-Symbol Interference (ISI), only inducing the common
phase rotation on demodulated symbols. While the initial location is selected outside the
permitted area, it inevitably will produce the ISI, thus resulting in the Inter-sub-carrier
Interference (ICI).
Assuming over the AWGN channel, if the symbol timing position is captured in the
permitted zone, then the output demodulation signal on the k ' -th sub-carrier of the m -th
OFDM symbol is expressed as
aˆ m , k ′ = am , k ′ exp ( − j 2πξ k′ / N ) , k′ = 0,… , N − 1 (7)
where ξ is the timing offset. However, if the symbol timing position is outside the
permitted zone, for example delay ξ samples, then the output demodulation signal on the
k ' -th sub-carrier of the m -th OFDM symbol can be denoted as
am , k ' exp ( j 2πξ k '/ N )

N −ξ
aˆ m , k ' =
(
∑ ∑ am , k ' exp j 2π ( n + ξ ) / N )
N
1 N − 1 −ξ K − 1
+ (8)
( )
N n=0 k =0 ; k = k '
∑ ∑ am + 1, k exp j 2π ( n − N − N g + ξ ) k / N exp ( − j 2π nk '/ N )

1 N −1 K −1
+
N n = N −ξ k = 0
www.intechopen.com
where the second item on the right side is for ICI, and the third part is for ISI. As shown in
Fig. 4, the constellations of the demodulated signals are divergent besides the phase
rotation.
Fig. 3. OFDM symbol timing
Fig. 4. The constellation of demodulated signal with timing offset

b. Effect of carrier frequency offset
In OFDM system, the existed CFO will lead to frequency shift of the received signal. If the
offset of the sub-carrier frequency is integral multiple of sub-carrier spacing, the
orthogonality among sub-carriers is still maintained, just with a shift relative to the sub-
carrier for the data symbols. However, if the frequency offset is a fractional sub-carrier
spacing, the ICI will be introduced. For OFDM systems composed of a large number of sub-
carriers, sub-carrier bandwidth is relatively much smaller compared with the channel
www.intechopen.com
bandwidth. Therefore, the small amount of frequency offset will result in substantial BER
Assuming the normalized fractional CFO is ε , and then the demodulated signal can be
performance degradation.
expressed as
ak' = α ak' exp ( jπε ( N − 1 ) / N ) + ∑ ∑ ak' exp ( j 2π n ( k − k'+ ε ) / N )

N −1 K
(9)
n = 0 k = 0 ;k ≠ k'
where α is the attenuation experienced by all sub-carriers, and
sinc ( ε )
α=
sinc ( ε / N )
(10)
The second part on the right side of Equation (9) is an ICI item. Fig. 5 shows the
demodulated signal constellation under the AWGN channel with ε = 0.05. As shown in the
figure, the CFO on one hand results in the whole constellation phase rotation, and on the
other hand, due to the impact of ICI, a divergent phenomenon is generated among
constellation points.
Fig. 5. The constellation of demodulated signal with CFO = 0.05
2.3 Performances analysis and numerical results

a. SIR effects caused by the carrier frequency-offset
According to Equation (9), the SIR of k ' -th sub-carrier is given as
α
SIRk'O =
2
E ⎡⎢ ICI k' ⎤⎥
(11)
⎣ ⎦
2
www.intechopen.com
where ICI k' is the inter-sub-carrier interference on the k ' -th sub-carrier, and its variance can
be expressed as[15]
K −1 ⎛ sin c ( k − k '+ ε ) ⎞
E ⎡⎢ ICI k ' ⎤⎥ = ∑ ⎜
( )
⎟
2
⎣ ⎦ k = 0 ⎜ sin c ( k − k '+ ε ) / N ⎟
2
(12)
k≠k'⎝ ⎠
b. Uncoded BER Performance of OFDM

In order to evaluate the error probability, without loss of generality, we focus on the signal
received on the first sub-carrier. Moreover, let us now consider the QPSK modulation. By
the conditional SINR, the approximate BER becomes [14]
sin 2 πε π πε ( M − 1) sin 2 πε π πε ( M − 1)
+ +
2 πε πε
cos 2 ( ) sin 2 ( )
4 M 4 M
BER ≈ )+
2 2 2
M sin M sin
sin 2 πε sin 2 πε
1 M 1 M
Q( Q( ) (13)
(1 − )+ (1 − )+
2 πε 2 πε
2 1 1 2 1 1
2 2
M sin SNR 2 2
M sin SNR
M M
where M is the modulation order, and SNR is the average received signal to noise ratio. Fig.
6 shows simulation results of the impacts of the CFO on the BER performance of OFDM
system with QPSK modulation over the AWGN channel.
Fig. 6. BER of OFDM system with QPSK modulation under the impacts of the CFO over the
AWGN channel
www.intechopen.com
3. DFT-S-OFDM transmission systems

In order to meet the emerged new requirements, many powerful and implementation-
efficient transmission schemes have been proposed for the standardization of the latest and
future communication systems. For the downlink transmission, the OFDM scheme has been
widely accepted due to its high spectral efficiency and flexible resource allocation. For the
uplink transmission, however, the power efficiency is particularly critical for mobile
terminals with the restriction on the transmission power and power consumption.
Therefore, the PAPR performance becomes one of the most important criterions in selecting
the transmission scheme for the uplink. From this point of view, the single-carrier based
frequency division multiple access (SC-FDMA) scheme is favoured by future wideband
wireless communications. In fact, one kind of SC-FDMA schemes, i.e., DFT-S-OFDM, is
accepted as the uplink basic transmission scheme by the 3GPP-LTE standard and incoming
3GPP-LTE-Advanced standard [5][6]. In order to reduce the PAPR, the DFT-S-OFDM
scheme utilizes the DFT spreading processing on the transmitted constellation symbols
before OFDM modulation. By this way, the modulation method can be view as a DFT-based
interpolation processing, and the modulated signals can be regarded as single carrier
signals.
3.1 DFT-S-OFDM system model
data symbols are {am } , 0 ≤ m ≤ K − 1 . After the DFT based spreading, the output signals can
The structure of DFT-S-OFDM transmitter is shown in Fig. 7. Assume the user-specific K
be expressed as
∑ am exp ( − j 2π mk / K ) ,
K −1
xk = 0≤ k ≤K −1
1
(14)
K m=0
Then according to the localized allocation pattern, the signal is converted into a time-
domain signal by N -point IDFT processing. N is greater than K . After the CP padding, the
time-domain signal, i.e., the SC-FDMA symbol, can be written as
∑ xk exp ( j 2π n ( k0 + k ) / N ) ,
K −1
sn = n = − N g ,..., −1,0,1,..., N − 1
1
(15)
N k =0
where k0 is the user-specific sub-carrier offset and N g is the CP length.
0 b0
a0 x0
a1 b1
x1 Sub- CP s (n )
N-point
. K-point . carrier .
.. DFT .. .. Padding
IDFT
Mapping
aK −1 xK −1
0 bN −1
Fig. 7. DFT-S-OFDM transmitter
www.intechopen.com
3.2 Time-frequency properties of DFT-S-OFDM signals

a. Time-frequency description of DFT-S-OFDM signal
Figure 8 describes the time-frequency property of DFT-S-OFDM signal. For DFT-S-OFDM
signals, the time-domain waveform can be viewed as a DFT-based interpolation of
transmitted constellation symbols. Therefore, the energy distribution within one DFT-S-
OFDM symbol keeps the same transmission order of the constellation symbols in the time-
domain. In the frequency-domain, due to the DFT based spread spectrum processing, the
spectrum of each transmitted constellation symbol is distributed on all occupied sub-
carriers, i.e., each sub-carrier contains only a part of spectrum component of the transmitted
constellation symbol, which is substantially different form OFDM signals.
Fig. 8. The time-frequency property of DFT-S-OFDM signal

b. Effects of time and frequency offset on demodulated signal
DFT-S-OFDM systems are subjected to the CFO and TO as well. Although the DFT-S-OFDM
scheme is based on the OFDM technology, the effects of CFO on two systems are very
different, because the DFT-S-OFDM systems can be viewed as transmitting symbols in the
time-domain, whereas the OFDM systems are usually regarded as transmitting symbols in
the frequency domain.
1. Effect of time offset
First, if the timing point is allocated inside the permitted zone, similar as the analysis for
OFDM system, the received signal in the frequency domain for DFT-S-OFDM can be
denoted as
xˆ k = x k exp ( − j 2πξ k / N ) , k = 0,1,..., K − 1 (16)
While after the K-point IDFT despreading, then the demodulated symbol can be given as
∑ xˆ k exp ( j 2π m ' k / K ) ,
K −1
aˆ m ' = m ' = 0,1,..., K − 1
1
(17)
K k =0
From the above expression, we can see that although the timing point is allocated inside the
permitted zone, it not only induce the common phase rotation, but also brings the ICI on the
www.intechopen.com
demodulated symbols, which is shown in Fig. 9. This character is different from OFDM
system. Furthermore, if the timing point is outside the permitted zone, it will cause both the
ISI and ICI, which is similar with OFDM.
Scatter plot
0.5
Quadrature
-0.5
-1
-1 -0.5 0 0.5 1
In-Phase
Fig. 9. The constellation of demodulated signal with timing offset

2. Effect of CFO
Considering CFO effects, following CP removing, the received N samples of SC-FDMA
symbol for demodulation are
∑ H k + k xk exp ( j 2π n ( k0 + k + ε ) )
K −1
rn = N + wn , n = 0,1,..., N − 1
1
(18)
k =0
0
N
where wn is the complex-valued AWGN on the n -th time-domain sample. H k is the
H k = ∑ i = 0 hi ⋅ exp ( − j 2π kτ i / N ) . By DFT and single sub-carrier equalization, the output is

channel frequency response (CFR) at the k -th sub-carrier, and
L −1
∑ ∑ Wk + k H k + k xk exp ( j 2π n ( k + k0 + ε − k ') / N ),
N −1 K −1
yk ' = k ' = 0,1,..., N − 1
1
(19)
n =0 k =0
0 0
N
∑ am exp ( − j 2π mk / K ) , 0 ≤ k ≤ K − 1 .
1 K −1
where x k =
K m=0
Then, after the sub-carrier demapping, generally, the K elements extracted from the N -
sample output of DFT are processed by a K -point IDFT, and yields the estimated symbols
www.intechopen.com
∑ y k '+ k exp ( j 2π m ' k '/ K ) , m ' = 0,1,..., K − 1

K −1
aˆ m ' =
1
(20)
k '= 0
0
K
Because the energy of ICI inducing by CFO is distributed in all the sub-carriers, as shown in
Fig.10, the CFO brings about not only the ICI and the linear phase rotation, but also the ISI
and the attenuation on the demodulated symbols.
Scatter plot
0.5
Quadrature
-0.5
-1
-1 -0.5 0 0.5 1
In-Phase
Fig. 10. The constellation of demodulated signal with CFO = 0.05

a. SIR effects caused by the carrier frequency-offset
For DFT-S-OFDM systems with CFO and over a flat fading channel, the demodulated
symbol can be given as [10]
aˆ m ' = am 'α m 'exp ( j 2π m 'ε / K ) + ISI m ' + ICI m ' (21)
where α m ' , ISI m ' and ICI m ' are the attenuation term, inter-symbol interference and inter-
sub-carrier interference on the m ' -th demodulated symbol respectively.
αm' = ∑ ∑ exp ( − j 2π ( k − k '+ k0 + ε ) m '/ K )

1 N −1 K −1
sin c ( k − k '+ k0 + ε )
K k '= 0 k = 0
⋅exp ( jπ ( k − k '+ k0 + ε )( N − 1 ) / N )
sin c ( ( k − k '+ k0 + ε ) / N )
(22)
www.intechopen.com
∑ ∑ ∑ am exp ( − j 2π ( mk − m ' k '+ m ' k0 ) / K )

1 N −1 K −1 K −1
ISI m ' =
K k '= 0 k = 0 m = 0
sin c ( k − k '+ k0 + ε )
m≠m'
(
⋅exp jπ ( k − k '+ k0 + ε )( N − 1 ) / N ) sin c
(23)
( ( k − k '+ k 0 + ε ) /N )
∑ ∑ ∑ ∑ am exp ( − j 2π mk / K ) exp ( j 2π n ( k + k0 ) / N )
N −1 N −1 K −1 K −1
ICI m ' =
1
(
⋅ exp ( j 2πε n / N ) exp ( − j 2π nk '/ N ) exp j 2π m ' ( k '− k0 ) / K )
NK k '= K n = 0 k = 0 m = 0 (24)
The SIR of the m ' -th demodulated symbol is
αm'
SIRm( ') =
2
E ⎡⎢ ISI m ' + ICI m ' ⎤⎥

2
(25)
⎣ ⎦
2
60
DFT-S-OFDM ε = 0.01
55 DFT-S-OFDM ε = 0.1
OFDM ε = 0.01
OFDM ε = 0.1
50
45
SIR(2) /SIRO(dB)
40
35
30
25
20
15
10
0 5 10 15
Symbol index (m') /Sub-carrier index (k')
Fig. 11. SIR comparison between DFT-S-OFDM and OFDM system with CFO
As shown in the figure, except the first symbol, all other demodulated symbols of DFT-S-
OFDM system have a much higher SIR than that of OFDM system under the same CFO
condition.
b. Uncoded BER Performance of CFO Effect [16]
We first derive an exact closed-form BER expression for the DFT-S-OFDM system without
channel coding. As we know, an arbitrary rectangular QAM can be viewed as two
independent pulse amplitude modulation (PAM), i.e., I -ary and J -ary PAM’s, through two
www.intechopen.com
quadrature branches. As a result, the average bit probability of the detected symbol am ', j in
the presence of CFO can be obtained by averaging the bit error probabilities from [7]
Pmj ′ = ( ∑ PI ( k ) + ∑ PJ (l ))
log 2 I log 2 J
1
log 2 ( I ⋅ J ) k = 1
(26)
l =1
where
⎡ k −1 ⎢ i ⋅ 2 k −1 1 ⎥ ⎤
⎢ i ⋅ 2 k −1 ⎥
∑
−k ⎢ ⎥
1 (1 − 2 ) I − 1
PI ( k ) = ( −1)⎣ ⎢2 − ⎢ + ⎥⎥
⎢ I ⎦⎥
I i =0 ⎢⎣ ⎣ I 2 ⎦ ⎥⎦
⎛ 3log 2 ( I ⋅ J )SINRmj ' ⎞
(27)
⋅ erfc ⎜ (2 i + 1) ⎟
⎜ ( I 2 + J 2 − 2)M ⎟
⎝ ⎠
where M is the number of bits for the special modulated symbol, and SINRmj ' denotes the
SINR per modulated symbol, which can be obtained from the above subsection. In addition,
erfc(⋅) is the complementary error function, and ⎢⎣ x ⎥⎦ denotes the largest integer to x .
Similarly, the PJ (l ) can be denoted as the above. Note that for I = 2 and J = 1 , equation (26)
reduces to the BER of a BPSK signal.
-1
10
-2
10
BER
-3
10
theoretical,no CFO,DFT-S-OFDM
simulation, no CFO, DFT-S-OFDM
theoretical,CFO=0.01,DFT-S-OFDM
simulation,CFO=0.01,DFT-S-OFDM
theoretical,CFO=0.1,DFT-S-OFDM
-4 simulation,CFO=0.1,DFT-S-OFDM
10
2 3 4 5 6 7 8
SNR(dB)
Fig. 12. BER comparison of the CFO effect for uncoded DFT-S-OFDM systems
www.intechopen.com
Therefore, based on the derived SINR for the demodulated symbol in the above section, the
closed-form BER expression in the presence of CFO for DFT-S-OFDM system can be given
as
∑ ∑P
1 N −1 D−1 j
=
Q j = 0 m′ = 0 m′
uncoded
PDFT-S-OFDM (28)
Then, using the average BER of the DFT-S-OFDM system, we can compute the effective
SINR by the mapping function as shown in (29). For the individual modulation, such as
BPSK, the effective SINR can be denoted as
uncoded
SINReff = [ erfc −1 ( PDFT-S-OFDM
uncoded
)]2 (29)
As a result, the SNR degradation for DFT-S-OFDM system can be presented as
DDFT-S-OFDM = 10 log10
σ ⋅ SINReffuncoded
Es
2
(30)
As can be seen from the figure, the theoretical result using the exact expression agrees well
with the simulation results. This clearly shows that the exact expression for calculating the
effective SINR in equation (29) can be used in order to assess the effect of the carrier
frequency offset accurately.
c. Turbo coded BER Performance of CFO Effect
In this section, we will focus on the BER performance in the presence of CFO for the coded
DFT-S-OFDM system. Due to the mathematical complexity of the iterative turbo decoding
algorithm, the analytical derivation of the BER of turbo codes is not available. To simplify
the analysis, the BER of coded system in AWGN channels can be approximated by an
expression of the form [8]
γ
− S − OFDM = exp( −
β
Coded
PDFT ) (31)
where γ is the received effective SINR. Parameters β is mode-dependent, and can be

obtained by fitting the curves to the exact simulated BER. As a result, the exponential
effective SINR mapping (EESM) method is considered to incorporate the SINR of all the
detected symbols. The formulation of the incorporated SINR can be expressed as [9]
⎛ 1 N −1 D−1 ⎪⎧ SINRm ' ⎪⎫ ⎞

= − β ln ⎜ ∑ ∑ exp ⎨− ⎬⎟
j
⎜ D j = 0 m′ = 0 β ⎭⎪ ⎟⎠
⎩⎪
coded
⎝
SINReff (32)
Similar to (30), the SNR degradation can be also obtained.

Fig.13 shows the BER performance in DFT-S-OFDM systems for turbo coded modulation,
where the QPSK modulation and 1/2 coding rate is assumed. In addition, the EESM method
presence of CFO, and β is achieved through the simulation. As we can see from the figure,
in (32) is used to incorporate the SINR of all the detected symbols, which is different in the
the BER mapping curve for CFO ε = 0.01 and ε = 0.1 is very close to that without CFO over
the equivalent SINR.
www.intechopen.com
-1
10
CFO=0,simulation
CFO=0.01,EESM
CFO=0.1,EESM
-2
10
BER
-3
10
-4
10
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
Equivalent SNR(dB)
Fig. 13. BER performance with and without the CFO effect
d. Post-processing SINR of DFT-S-OFDM [11]
For DFT-S-OFDM system, the transmit signal vector without CP can be given as
S = FNHTNm, M FM D (33)
where TNm, M is the mapping matrix for sub-carrier assignment, FM is the M point FFT matrix
and FNH is the N point IFFT matrix, D = ⎡⎣ d1 dM ⎤⎦ is the data vector.
T
Then, at the receiver, the vector of detection metric after FDE is then given as
D = FMHTNm, M H H H W ( HTNm, M FM D + Z )
= FMH H W , M FM D + FMH H W , M ZM
m m
(34)
where W = diag{ω1 , ω2 , , ωN } is the frequency domain equalizer, and
m
HW {
, M = diag H m ωm , H m + 1 ωm + 1 ,… , H m + ( M − 1) ωm + ( M − 1)
2 2 2
} (35)
m
HW {
, M = diag H mωm , H m + 1ωm + 1 ,… , H m + ( M − 1)ωm + ( M − 1)
* * *
} (36)
Hence, the SINR for DFT-S-OFDM with FDE is given as
www.intechopen.com
∑ H m + k ωm + k
1 M −1
2
2
SINR =
M k =1
∑ ∑ ∑ H m + k ωm + k
(37)
1 M −1 1 M −1 1 M −1
H m + k ωm + k σ 2 + H m + k ωm + k −
2
2 2 4 2 2
M k =1 M k =1 M k =1
Furthermore, for ZF and MMES equalizer, the SINR expression can be simplified
respectively as
⎡ 1 M −1 σ 2 ⎤
−1
SINR = ⎢ ∑ ⎥
⎢ M k =0 Hm+ k ⎥
(38)
⎣ ⎦
2
and
⎡ ⎤
−1
⎢ ⎥
⎢ ⎥
SINR = ⎢ − 1⎥
1
⎢ 1 M −1 Hm+ k ⎥
⎢M ∑
2
(39)
⎥
⎣⎢ = + σ ⎥⎦
2
m+ k
2
k 0 H
4. DFT-S-GMC transmission systems

4.1 DFT-S-GMC system model
0 bd (0 )
ad (0 ) Ad (0 )
bd (1)
ad (1) Ad (1) Buffering
Subband M-band g d (n ) & s (n ) CP
. K-point . . Tx
.. DFT .. .. Circulated
Padding
Mapping IFBT Shift signal
Cumulating
ad (K − 1) Ad (K − 1)
0 bd (M − 1)
Fig. 14. DFT-S-GMC transmitter
input parallel modulated constellation symbol sequence is ad ( k ) , 0 ≤ k ≤ K − 1 and { }

The structure of the DFT-S-GMC transmitter is illustrated in Fig. 14 [12]. Assume that the
0 ≤ d ≤ D − 1 . Note that K is the number of the user-specific occupied sub-bands, and D is

the number of inverse filter-bank transform (IFBT) symbols transmitted during each DFT-S-
GMC symbol.
The input data sequence is passed through K-point DFT for spectrum spreading, yields the
output signal
Ad ( k ') = ∑ ad ( k ) exp ( − j 2π kk '/ K ) ,

K −1
0 ≤ k' ≤ K − 1
1
(40)
K k =0
Then, the signals are sent to the sub-band mapping module, yields
www.intechopen.com
⎪⎧ A ( k ') , m = k0 + k ' R
bd ( m ) = ⎨ d , 0≤m≤ M −1
⎪⎩ 0, m ≠ k0 + k ' R
(41)
where M is the total number of sub-bands, k0 is the user-specific sub-band offset, R is the
repetition factor. For DFT-S-GMC system, both distributed and localized mapping policy
could be supported, which is corresponding to R greater than or equal to one respectively.
After sub-bands mapping, an IFBT is performed on the data sequence, yields the output
signal, i.e., the IFBT symbol, as
gd ( n ) = ∑ bd ( m ) exp ( j 2π mn / M ) f p ( n ) ,
M −1
m=0 (42)
0≤n≤L−1
where f p ( n ) is the impulse response of the prototype filter for the filter bank, and can be set
energy, i.e. ∑ n = 0 f p ( n ) = 1 and a length L which is integer times of M .

to a Square Root Raised Cosine (SRRC) function. The prototype filter is with a normalized
L −1 2
Then, each IFBT symbol is zero-padded to form a Q -sample data block
⎪⎧ g ( n ) , 0 ≤ n ≤ L − 1
gd ( n ) = ⎨ d
⎪⎩ 0, L≤n≤Q−1
(43)
where Q = D × N .
Next, D consecutive Q-sample data blocks will be processed by buffering and cyclic-shift
cumulating with shift interval N , and the output can be described as
s (n) = ∑ gd ( ( n − dN ) )Q ,
D− 1
0≤n≤Q−1 (44)
d =0
where ( ( ⋅) )
Q
denotes modulus operation.
4.2 Time-frequency properties of DFT-S-GMC signals

From equation (44), the DFT-S-GMC symbol is formed by cyclically accumulating several
IFBT symbols in time domain, and each IFBT symbol has a SRRC waveform. Therefore, as
shown in Fig. 15, the spectrum of each sub-band has a Raised Cosine function shape in
frequency-domain. Moreover, the sub-band spacing is specially designed and differs from
any conventional filter-bank systems in that a certain guard band is inserted between
neighbouring sub-bands. By adding some guard band between the sub-bands, the near-
orthogonality between neighbouring sub-bands is guaranteed, which further simplifies the
detection algorithm greatly. Meanwhile, interferences among successive symbols for each
sub-band could be easily mitigated in the receiver because of the shift orthogonality of
prototype filter and the narrow-band single carrier transmission. Similar to DFT-S-OFDM
systems, DFT-S-GMC systems also exploit DFT based spreading among sub-bands.
Therefore, each sub-band contains only a part of spectrum component of transmitted
constellation symbols, and transmitted signal over all occupied sub-bands can be viewed as
single-carrier signal as a whole.
www.intechopen.com
Channel bandwidth
IFBT symbols
-3dB sub-band
Virtual sub-bands Effective sub -bands bandwidth
...
... ... ...
Frequency
Sub-band spacing
Virtual sub-band
FBT/IFBT
...
CP
Time
Fig. 15. The time-frequency property of DFT-S-GMC signal
4.3 Frequency-domain implementation structure

a. The frequency-domain equivalent implementation of the DFT-S-GMC transmitter
As described in equation (44), it can be seen that transmitted symbols are multiplexed
within each DFT-S-GMC symbol by both time and frequency dimensions. Taking Q-point
FFT on the DFT-S-GMC modulation signal, the output signal can be expressed as
S(q) = ∑ ∑ g((t − nN )) ( n )exp ( − j2π qt / Q )

Q −1 D−1
1
(45)
Q t =0 n=0 Q
Since
∑ g((t −nN )) ( n )exp ( − j2π qt / Q )

Q−1
1
Q t =0
∑ gt ( n )exp ( − j2π qt / Q ) exp ( − j2π qn / D )

Q
Q −1
(46)
=
1
Q t =0
∑ gt ( n )exp ( − j2π qt / Q )
Q−1
1
Q t =0
∑ bm ( n ) ∑ f p ( t ) exp ( j 2π mt / M ) exp ( − j2π qt / Q )

M −1 Q −1
=
1
(47)
m=0 Q t =0
∑ bm ( n )Fm ( q )
M −1
=
m=0
where
Fm ( q ) = ∑ f p ( t ) exp ( j 2π mt / M ) exp ( − j2π qt / Q )

Q −1
1
Q t =0 (48)
, 0 ≤ q ≤ Q − 1; 0 ≤ m ≤ M − 1
www.intechopen.com
S(q) = ∑ Fm ( q ) ∑ bm ( n ) exp ( − j2π qn / D)

M −1 D−1
(49)
m=0 n=0
In fact, Fm ( q ) is the frequency response of the prototype filter for m -th sub-band.
Therefore, the DFT-S-GMC modulation signal can be expressed as
s (t ) = ∑ S ( q )exp ( j2π qt / Q )
Q −1
1
Q q =0
∑ ∑ Fm ( q ) ∑ bm ( n ) exp ( − j2π qn / D)exp ( j2π qt / Q )

Q −1 M −1 D−1
=
1
Q q =0 m=0 n=0 (50)
Cyclic extended D-point DFT
Weighting in frequency-domain
multi-sub-bands summation
Q-point IDFT
From equation (50), the frequency-domain implementation structure of DFT-S-GMC

transmitter can be given by Fig. 16.
{F0 (q)}
a0 (n)
D-point Cyclic
S/P
a1 (n) DFT extending
add P
DFT
spreading
Subband
mapping
S/P
D-point Cyclic
{FM −1 (q)}
∑ Q-Point
IDFT
cyclic
prefix
/
S
DFT extending
aK −1 (n)
Fig. 16. Frequency-domain implementation structure of DFT-S-GMC transmitter

b. The simplified frequency-domain equivalent implementation of the DFT-S-GMC
transceiver
From equation (49), the frequency response of the prototype filter for each sub-band has the
energy over all Q tones. However, in fact, the most energy of the frequency response is just
proportional of the total energy of Fm ( q ) , Fm ( q ) can be simplified as

over several tones. Therefore, by selecting a set of values with energy greater then a
⎪Fm ( q ) , ∑ Fm ( q ) > ξ ∑ Fm ( q ) , q ∈ Ω
⎧ Q−1
Fm ( q ) = ⎨
2 2
q∈Ω q =0
⎪ 0,
(51)
⎩ otherwise
With proper designed prototype filter, the frequency response of the prototype filter for
each sub-band can be simplified such that the number of total selected values for all sub-
bands is equal to the number of total tones Q. By this way, the signals for each sub-band can
be even mapped to the tones directly and exclusively, and the simplified implementation
structure is shown in Fig.17.
As shown in Fig. 17, by tone mapping, rather than summation processing, the
implementation complexity of DFT-S-GMC transmitter can be significantly reduced, and the
performance loss is very limited as illustrated by the following simulation results.
www.intechopen.com
{Fɶ (q)}
0
a0 (n)
D-point Cyclic
S/P
a1 (n) DFT extending
Tone add P
DFT
spreading
Subband
mapping
{Fɶ
M −1 }
(q) Mapp-
Q-Point
IDFT
cyclic /
ing prefix S
D-point Cyclic
S/P
DFT extending
aK −1 (n)
Fig. 17. Simplified implementation structure of DFT-S-GMC transmitter

a. Post-processing SINR with frequency-domain equalization
wd (0 )
wd (1)
L- aˆd (0 )
CP sample
Rx r (n ) r ' (n ) Subband
K- aˆd (1)
Remov-
SC-
Cyclic-
M-band . .. point .
FDE FBT .. Demapp- ..
Signal ing shift
ing . IDFT
Extract-
ing
wd (M − 1) aˆd (K − 1)
Fig. 18. DFT-S-GMC receiver

At the receiver side, shown in Fig. 18, after removing the CP, and going through the Q-point
SC-FDE, the output signal vector can be given as [13]
r = FQH H H WHFQ ΩQ , L ϒ L Γ L , M FMHTM , K FK α + FQH H H WFQ z (52)
Then, the post-processing SINR can be given by
SINR =
σ + σ ISI + σ ITSI
Es
2 2 2
n
∑ Hk
1 K −1
2
(53)
=
K k =0
σ2
∑Hk + K ∑ ∑ H k + ∑ ∑ Hd ;k
K −1 K −1
1 K −1 1 D−1 K −1
Hk −
2
1 2 2
K k =0 k =0 K k =0 K d =1 k =0
where σ n2 is variance of AWGN on demodulated symbols. σ ISI

2
is the variance of the inter-
symbol-interference (ISI) within the d -th IFBT symbol, and can be expressed as
σ ISI = ∑ hk = ∑ H k − ∑ . σ ITSI
K −1
1 K −1 1 K −1
2
2 2
2 2
Hk is the variance of the inter-IFBT symbol
k =1 K k =0 K k =0
interference (ITSI) on the demodulated symbols, and can be expressed as
= ∑ ∑ Hd ;k .
1 D−1 K −1
σ ITSI
2
2
K d =1 k =0
www.intechopen.com
25
20
Post-processing SINR (dB) 15
10
5 Analysis (MMSE, 8 sub-bands)

Simulation (MMSE, 8 sub-bands)
Analysis (MMSE, 1 sub-band)
0 Simulation (MMSE, 1 sub-band)
Analysis (ZF, 8 sub-bands)
-5 Simulation (ZF, 8 sub-bands)
Analysis (ZF, 1 sub-band)
Simulation (ZF, 1 sub-band)
-10
0 5 10 15 20 25 30
Average received SNR (dB)
Fig. 19. The post-processing SINR of DFT-S-GMC with SC-FDE

Both the theoretical and simulated post-processing SINR are shown in Fig.19 for the DFT-S-
GMC receiver with MMSE and ZF SC-FDE respectively. Over the same channel condition,
with MMSE equalization, DFT-S-GMC receiver achieves higher SINR than that with ZF
equalization in the low SNR range and with wider band transmission, due to the noise
enhancement effects of ZF SC-FDE.
b. Performance of frequency-domain implemented DFT-S-GMC transceiver
System Parameters Value

Carrier frequency (GHz) 2
Carrier Bandwidth (MHz) 5
Sampling frequency (MHz) 5.6
# of total sub-bands (M) 28
# of useful sub-bands 24
Upsampling rate (N) 32
# of IFBT symbols in each data block (D) 16
Sub-band BW (kHz) 200
Sub-band 3dB-BW (kHz) 175
Occupied BW (MHz) 4.8
FFT size for FDE (Q) 512
Prototype filter type SRRC
Simulation parameters
Channel model PB (3km/h)
Channel coding(coding rate) Turbo (1/2)
Modulation QPSK
Equalization MMSE
# of Tx/Rx antennas 1/1
Table 1. Simulation specification
www.intechopen.com
The simulation specification is shown in Table 1. As shown in the Fig.20, the BER
performance of frequency-domain implemented DFT-S-GMC transceiver is almost the same
as that of time-domain implemented DFT-S-GMC transceiver.
Fig. 20. BER performance comparison of FD transceiver with TD transceiver

As presented in the Table 2, the CM performance of frequency-domain implemented DFT-S-
GMC transmitter is very close to that of time-domain implemented DFT-S-GMC transmitter.
Moreover, the CM of DFT-S-GMC is smaller 1.7 and 1.1dB than that of OFDM for QPSK and
16QAM modulation respectively.
Systems
Frequency-domain Time-domain
OFDM
DFT-S-GMC DFT-S-GMC
Used sub-carriers
QPSK 16QAM QPSK 16QAM QPSK 16QAM
/sub-band(s)
16 / 1 0.5 1.5 0.5 1.5 3.2 3.2
32 / 2 1.4 2.0 1.3 2.0 3.3 3.3
64 / 4 1.6 2.2 1.6 2.2 3.4 3.4
256 / 16 1.7 2.3 1.7 2.3 3.4 3.4
Table 2. Cubic Metric (dB) comparison of frequency- and time-domain implemented DFT-S-
GMC system with OFDM system
# of used Computation complexity Percentage of

sub-band(s) (# of real multiplications) reduction
TD transmitter/ FD transmitter/
receiver receiver
28 18800/58688 15184/31008 19％/47％
1 18432/55404 9416/18868 49％/66％
Table 3. Complexity comparison of frequency- and time-domain implemented DFT-S-GMC
transceiver
www.intechopen.com
As shown in the Table 3, with frequency-domain implementation structure, the computation

complexity of DFT-S-GMC transceiver with equalization can be reduced significantly,
compared with that with time-domain implementation structure. For 28 and 1 sub-band(s)
transmission, the computational complexity can be reduced about 47% to 66%.
5. Conclusion
In this chapter, the principle, implementation structure, time-frequency property of three
Fourier Transform-based transmission systems, namely OFDM, DFT-S-OFDM and DFT-S-
GMC, are presented for broadband wireless communications. For OFDM systems, the
spectrum of each sub-carrier has a sinc-function shape, spectrums of all sub-carriers are
independent each other which cause high PAPR of transmitted signal; For DFT-S-OFDM
systems, each sub-carrier contains only a part of spectrum component of transmitted
constellation symbols, and the time-domain waveform can be viewed as a DFT-based
interpolation of transmitted constellation symbols, which bring in lower PAPR of
transmitted signal; For DFT-S-GMC systems, each DFT-S-GMC symbol is formed by
cyclically accumulating IFBT symbols with SRRC waveform in the time domain, hence, the
spectrum of each sub-band has a Raised Cosine function shape, and due to DFT based
spreading among sub-bands , the transmitted signal over all occupied sub-bands can be
viewed as single-carrier signal as a whole. Moreover, the effects of time and frequency offset
on OFDM and DFT-S-OFDM systems are analyzed quantitatively. Theoretical analysis and
simulation results show that except the first symbol, all other demodulated symbols of DFT-
S-OFDM system have a better SIR than that of OFDM system under the same CFO
condition. Furthermore, the post-processing SINR of DFT-S-OFDM and DFT-S-GMC are
addressed for different equalizer. The closed-from expressions of SINR are presented and
verified by the simulation results.
6. References
[1] S,B ; Weinstein and Ebert, P.( 1971). Data transmission by frequency-division
multiplexing using the discrete Fourier transform. IEEE Transactions on
Communications, Vol. 19, No.5, (Oct. 1971) 628-634, ISSN: 0018-9332
[2] Myung, H.; Lim, J.; and Goodman, D.(2006). Single Carrier FDMA for Uplink Wireless
Transmission. IEEE Vehicular Technology Magazine, Vol. 1, No. 3, (Sep. 2006) 30-38,
ISSN: 1556-6072
[3] Cherubini, G.; Eleftheriou, E.; Cioffi, J.; Oker, S.(2000). Filter bank modulation techniques
for very high speed digital subscriber lines. IEEE Communications Magazine, Vol. 38 ,
No. 5, (May 2000) 98-104, ISSN: 0163-6804
[4] Qiu, RC.; Liu, H.; Shen, X. (2005). Ultra-Wideband For Multiple Access Communications.
IEEE Communications Magazine, Vol. 43, No.2, (February 2005) 80-87, ISSN: 0163-
6804
[5] 3GPP TS 36.211 v8.7.0. Evolved Universal Terrestrial Radio Access (E-UTRA); Physical
Channels and Modulation(Release 8)，May 2009.
[6] 3GPP TR 36.913 V8.0.1 2009-03 3rd Generation Partnership Project, Technical
Specification Group Radio Access Network, Requirements for further
advancements for Evolved Universal Terrestrial Radio Access (E-UTRA) (LTE-A)
(Release 8).
www.intechopen.com
[7] Yoon, D. and Cho, K.(2002) . On the general BER expression of one- and two dimensional
amplitude modulations. IEEE Transactions on Communications, Vol. 50, No.7, (July
2002) 1074–1080, ISSN: 0018-9332
[8]Liu, Q.; Zhou,S.; and Giannakis, G.(2004). Cross-layer combining of adaptive modulation
and coding with truncated ARQ over wireless links. IEEE Transactions on Wireless
Communication., Vol. 3, No.5, (sept. 2004) 1746-755, ISSN: 1536-1276
[9] 3GPP TSG-RAN1 WG1 \#36, R1-040189, New results on realistic OFDM interference,
Malaga, Spain, Feb. 16-20, 2004.
[10] M, Li., Y, Rui. (2009). Analysis of CFO effects on and phase compensation method for
SC-FDMA systems, Science in China Series F: Information Sciences, Vol.52, No. 12,
(Dec. 2009) 2397-2405, ISSN: 1674-733X
[11] 3GPP TSG-RAN1 WG1 \#42, R1-050719, Simulation Methodology for EUTRA UL:
IFDMA and DFT Spread-OFDMA, London, U.K., Aug. 29 – Sept. 2, 2005.
[12] Zhang, X. ; Li, M.; Hu, H.; Wang, H.; Zhou, B.; You, X.(2006). DFT Spread Generalized
Multi-Carrier Scheme for Broadband Mobile, IEEE PIMRC Communications. pp. 1-5,
ISBN: 1-4244-0329-4, Helsinki , Sept. 2006.
[13] Li, M. ; Zhang, X.(2009). Performance analysis of DFT spread generalized multi-carrier
Systems, Science in China Series F: Information Sciences, Vol.52, No.12, (Dec. 2009)
2385-2396, ISSN: 1674-733X
[14] Liu, P.; Bar-Ness, Y.(2006). Comparing the effect of Carrier Frequency Offset on OFDM
and Single-Carrier Block Transmission in AWGN Channels, in Proc IEEE
GLOBECOM 2006, pp. 1-5, ISBN: 1930-529X, San Francisco, Nov. 2006.
[15] Moose, Paul H.(1994). A technique for orthogonal frequency division multiplexing
frequency offset correction, IEEE Transactions on communications, Vol.42, No.10,
(Oct. 1994) 2908-2914, ISSN: 0090-6778
[16] Rui, Y.; Hu, H.; Li, M.; et al.(2009). Comparing Effects of Carrier Frequency Offset on
Generalized Multi-carrier and OFDM Systems, in Proc IEEE ICC 2009, pp. 1-6, ISBN:
1938-1883, Dresden, Jun. 2009
www.intechopen.com
12
Applications of Fiber Optic Coupled-Grazing

Angle Probe Reflection-Absorption
FTIR Spectroscopy
Oliva M. Primera-Pedrozo, Leonardo C. Pacheco-Londoño and
Samuel P. Hernandez-Rivera
ALERT DHS Center of Excellence for Explosives
Center for Chemical Sensors Development
Department of Chemistry
University of Puerto Rico-Mayagüez
PO Box 9000, Mayagüez, PR, 00681
Puerto Rico
1. Introduction
The recent increase in attention of detection of chemical threats, explosives and narcotics has
led to the development of instruments and sensors that can be effective in a variety of
operating environments. Various approaches can be used for in situ analysis of explosives,
including the widely used technique, Ion Mobility Spectrometry (IMS). The major
advantages of IMS are its sensitivity in the picogram range, its continuous real time
monitoring capability, its reasonable price due to instrumental simplicity and the ease of
automation (Salleras, 1995). A main disadvantage of IMS is its limited linear range and that
it cannot be used for quantitative analysis (Salleras, 1995). It is relatively easy to overload an
IMS and, therefore, sample size must be controlled with care (Brambilla, 1997). Another
weakness is the response variation that occurs with different background gas compositions
and with different sample compositions (Salleras, 1995). However, spectroscopic techniques
have the potential to afford the best selectivity for explosives. The infrared spectra of
molecules can provide an information-rich fingerprint that allows for near unambiguous
identification. A few years ago, direct detection by infrared absorption spectroscopy was not
possible because of the limited sensitivity of this method. Fourier Transform Infrared
Reflection Absorption Spectroscopy (IRRAS), operating at the grazing-angle, is the most
sensitive optical absorption technique available for measuring low concentrations of
chemical compounds adhered to reflective surfaces such as metals (Griffiths, 1986). The
disadvantage of conventional spectroscopic techniques for applications such as explosives
detection is that the test materials must be placed physically within the spectrometer’s
sample compartment for measurement. FT-IRRAS combined with grazing angle probe
(GAP) can now be used outside the boundaries of the sample compartment. Fiber-optic
cables (FOCs) that transmit in the mid-IR (MIR) range have made it possible to develop a
range of spectroscopic probes for in situ analysis (Melling, 2001; Melling, 2002; Mehta, 2003;
Bacci, 2001). Thus, FTIR spectroscopy can now be effectively used outside the confinement
www.intechopen.com
of the sample compartment, making it available for field work (Mizaikoff, 2002). The
attractive features of this technique include portability, simple design and rugged design,
high sensitivity and short analysis time. These features lead to potential uses of Mid FTIR
for airport screening and within the military.
Sample preparation methodology constitutes a critical a step in the detection scheme pursed
because a uniformly covered, thin film is needed for preparation of the standard samples on
which precision relies on. Several approaches have been used in the lab for sample
preparation including: standard preparation using an airbrush aerosol spray, sample
the smearing method 20 μL aliquots of standard solutions were deposited over the plate
smearing and direct transfer from solutions using micropipettes (Primera-Pedrozo, 2004). In
then the solution is spread using a Teflon sheet. Smearing transfer method has led to detect
TATP over stainless steel surface. This transfer method let to detect and quantify TATP for
first time on metallic surfaces. Despite the fact of tendency towards sublimation of TATP, a
limiting value of 8 µg/cm2 could be detected. Samples ranging from micrograms/cm2 to
nanograms/cm2 of 2,6-dinitrotuelene (DNT), 2,4,6-trinitrotoluene (TNT), pentaerythritol
tetranitrate (PETN), nitroglycerine (NG) and triacetone triperoxide (TATP) have been
detected using this new method of deposition. A smearing deposition was used for
depositing the target explosives over substrates to be used as standards. The sample transfer
method gave good sample distribution, reduced sample loss on transfer and was easy to
manipulate giving good reproducible distributions (Primera-Pedrozo, 2004; Primera-
Pedrozo, 2009; Primera-Pedrozo, 2010; Pacheco-Londoño, 2010). Although, the smearing
technique has given good results for explosives detection, many samples are needed for
sample preparation transfer method because it depends on human error since the sample is
placed using a piece of Teflon sheet and is distributed with the hands on the surface and
sometimes good distribution is not found. In this case other samples have to be prepared.
Another disadvantage of this transfer method is solvent interference since various solvents
are adhered to the stainless steel plated producing poor distribution of the material over the
surfaces. The use of slow evaporating solvents makes the sample preparation more
complicated. For these reasons the development of an automatic method for explosives
transfers on the surfaces must be devised. Thermal ink jet (TIJ) was selected as transfer
technique to avoid human errors during preparation of standards and to decrease the time
for sample preparation. In thermal ink jet a thin film resistor superheats less that 0.5% of the
fluid in the chamber to form a gas bubble. This bubble rapidly expands (less than ten
microseconds) and forces a drop to be ejected through an orifice (Beeson, 1998).
When comparing the inkjet based method of sample transfer to the smearing method used
in previous works (Primera-Pedrozo, 2004), the former has notable advantages. The loading
concentration of the sample on the surface can be controlled by varying parameters such as:
number of passes, dispensing frequency, applied energy, and pen architecture. Precise
delivery of the number of droplets with known volume and concentration controls the mass
deposited. Also only one solution needs to be used, avoiding dilutions that can increase the
analytical errors.
2. Description of methodology
FOC-GAP FTIR spectroscopy has made possible to develop new methods for detection of
traces of chemical compounds on surfaces. Thermal inkjet (TIJ) technology is able to deposit
very small amounts of chemical compounds, including energetic materials, in a specific
www.intechopen.com
Applications of Fiber Optic Coupled-Grazing Angle Probe
Reflection-Absorption FTIR Spectroscopy 229
location on a surface (Primera-Pedrozo, 2005). Aliquots of TNT and RDX solutions were
deposited on stainless steel film. A thin coating of the explosives can be produced by
controlling the concentration of TNT or RDX, the number of drops dispensed and the
distribution of drops on the surface. A Vector 22, a Bruker Optics FTIR fiber coupled to a
Remspec Corp. Grazing Angle Probe head was used for the experiments. The spectra were
recorded at 4 cm-1 resolution and 50 scans. The results of the experiments gave intense
absorption bands in the fingerprint region of the infrared spectra that were used for
quantification. Chemometrics routines were applied for enhancing quantitative analysis.
The sample analysis setup is schematically presented in Fig. 1. A Remspec mid-IR grazing
angle probe was used to collect the spectra. The grazing-angle head uses carefully aligned
mirrors to deliver the mid-IR beam to the sample surface at the grazing angle
(approximately 80° from normal), to collect the reflected beam, and to return it to a mid-IR
detector (in this case, external liquid nitrogen cooled MCT detector). The signal is delivered
from the spectrometer to the head by IR transmitting fiber optic cables. The grazing angle
accessory is connected to the external beam port of the Bruker Vector 22 spectrometer by a
1.5 m, 19–fiber chalcogenide glass optical bundle in the As-Se-Te system, which transmits
throughout the mid–IR with the exception of a strong H-Se absorbance band at 2200 cm-1.
The IR footprint produced by the grazing angle probe is elliptical with the intensity
decaying from the middle towards the edges. The specially configured head illuminates a
large spot on the sample surface. The spot is an ellipse 1 inch by six inches that is defined by
a Gaussian distribution with a center spot about 1/8 inch by an inch. The electric signal
from the MCT is delivered to the FTIR using an amplifier.
Fiber Optic Detector

Cables
Amplifier
Sample
Vector 22 PC
Gold
Opus
Mirrors
Fig. 1. Experimental setup for Fiber Optic Coupled-Grazing Angle Probe FTIR
3. Evaluation of samples and standards

FOP-GAP FTIR absorption-reflection spectroscopy may be successfully implemented in
assessing sample loading distribution of solids deposited on substrates (Primera-Pedrozo,
2008; Primera-Pedrozo, 2009). Two methods were used to prepare the samples by depositing
the analytes onto the test surfaces: sample smearing and thermal inkjet transfer. Stainless
steel (SS, non-magnetic, type 316) metal sheets with an effective or area for coverage of 46.3
cm2 (3 cm x 15.4 cm) were used. The plates were cleaned with HPLC grade methanol and
www.intechopen.com
air-dried at room temperature before the experiments. Aliquots of 20 μL of standard

solutions were placed at one side of the SS plate. A Teflon sheet was inclined towards the
right or left and the smearing was done quickly. Fig. 2-a illustrates how the smearing is
done.
(a) (b)
Fig. 2. Methods used for transferring a solid sample onto a substrate: (a) TIJ; (b) sample
smearing
This method is rapid and easily executed without specialized equipment. The amount is
readily controlled and can be calculated without the need for an independent analysis. Once
the solvent had evaporated, the spectrum of the sample was collected immediately.
Solutions were dispensed using an ImTech Imaging System model I-Jet 312S, (ImTech, OR,
USA) equipped with a HP 51645A inkjet cartridge, illustrated in Fig. 2-b. Aliquots of 10 mL
were placed into the inkjet cartridge and the backpressure was set to 3 inches of water using
an external backpressure controller. The solutions were dispensed over stainless plates at
zero dot spacing (space between drops using HP ink) using a printing resolution of 600 dots
per inch (dpi). Once the solvent had evaporated, the spectrum of the sample was collected.
4. Cleaning validation of pharmaceutical batch reactors

FOC-GAP FTIR spectroscopy can be used in cleaning validation applications for active
pharmaceutical agents on metallic surfaces (Mirza, 1999); Mehta, 2002; Primera-Pedrozo,
2005-b; Fierro-Mercado, 2010). A method based on smearing a known amount of the sample
in solution was used for preparing samples and standards to develop cleaning validation
methodologies using IR spectroscopy. The samples were deposited on the surface using
smearing transfer method. Using this method ibuprofen was detected on stainless steel
plates, a common material on pharmaceuticals reactors. This new technology combining to
smearing can decrease the consuming time in cleaning validation process, being
advantageous in an in process laboratory. Detection limit for this compound was 0.5 μg/cm2
concentrations in the range of 0.1 – 20 μg/cm2 FT-IR spectra were collected from each plate
loading concentration. Grazing angle spectra of samples were collected for surface
using the grazing angle probe. The spectra are shown in Figure 3. Fingerprint signals
intensities of the spectra decrease with lower ibuprofen loading concentrations. This
amplified region was used for the chemometrics calculations (Beebe, 1998). The most intense
band of ibuprofen in the region of 1760-1650 cm-1 was used for peak area and peak height
calibration curve generation. This band is assigned to C=O stretch (Griffith, 1986; Lin-Vien,
1991).
www.intechopen.com
Absorbance
1000 1150 1300 1450 1600 1750
Wavenumbers / cm-1
Fig. 3. Grazing angle FTIR spectra of ibuprofen on stainless steel for various surface loadings
For quantification studies, two types of calibration curves were generated using two
methods: measurement of the absorbance peak heights and integration of areas spectral
region within the 1760 to 1650 cm-1 range. Fig. 3 shows the calibration curves of the
absorbance peak heights. Results for peak areas are not shown. These plots exhibit a high
degree of linear correlation. The calibration curve graph using height peaks shows better in
linearity. However, the errors are higher for the calibration curve using peak areas. The
calibration graphs using peak areas represents a better choice for quantitative analysis when
compared to peak height analysis (Lavine, 2002; Kramer, 1998).
Peak height (1760-1650 cm-1)
y = 0.0028x - 0.0025
R2 = 0.9949
0 3 6 9 12 15 18
Loading concentration / ug/cm2
Fig. 4. Calibration curve of ibuprofen using peak height analysis

The calibration model was built using the Quant 2 package, an add-on software package to
the OPUS™ (Bruker Optics) data acquisition and analysis software. In this study, the model
parameters were optimized in the spectral region 1770 – 1016 cm-1 and 3104 – 2750 cm-1. No
spectral data preprocessing was done. The resulting model was cross-validated using the
‘’leave one out’’ method in which each spectrum is omitted in turn from the training set and
then tested against the model built with the remaining spectra. The results are illustrated
www.intechopen.com
graphically in Fig. 5. The root mean square error of the cross validation was 0.401, and R2
was 0.9952. Clearly, low levels of ibuprofen can be detected and measured on a metal
surface with quantitative results.
10
Predicted concentration / µg/cm2 9 y = 0.9947x - 0.0048

8 R2 = 0.9952
7
6
5
4
3
2
1
0
0 1 2 3 4 5 6 7 8 9 10
True concentration / µg/cm2
Fig. 5. Cross validation for Ibuprofen quantification on metallic surfaces

Discriminant analysis (Huberty, 1994) was also performed to classify ibuprofen loading
concentration in two groups (Fig. 6). The first one corresponds to concentrations < 2 μg/cm2
and the second one to concentrations > 2 μg/cm2. Peak areas of signals in the range of 1273 –
1978 cm-1 were used for the discrimination. This model was generated using 10 PLS and
submitting the data to a pre-processing of straight line subtraction. Results show that
discriminant analysis can be used to classify ibuprofen samples according to their surface
concentration on the metal surfaces. The minimum amount of this API on the reactor after
cleaning must be considered for future works. This will allow having a real model of
discrimination.
conc < 2ug/cm2 conc > 2ug/cm2

2.5
2
Discrimination
1.5
0.5
0
0.6 1.1 1.6 2.1 2.6
Discriminant function
Fig. 6. Discrimination study for ibuprofen: samples were separated according to surface
loadings
www.intechopen.com
30 27.12
Variation coefficient (%)

25
20 17.71
15
11.31
10
6.52 6.36
4.96
5
0
0.5 1 2 5 10 20
Surface loading (μg/cm 2)
Fig. 7. Variation coefficient with surface loadings for detection and quantification limit
calculations
The variation coefficient (Fig. 7) increases at low surface loadings. For 0.5 μg/cm2 loading
concentration, the variation was higher than the others. This indicates that at this surface
loading, the analytical response gets its minimum value and can be confused with the noise.
This value can be considered as the detection limit. In order to verify this, an ANOVA test
was performed for surface concentrations of 0.5, 1 y 2 μg/cm2, and for a 95.0% confidence
level there was a statistically significant difference between these values. However at 99.0%
confidence level, there was no difference for samples of 0.5 and 1 μg/cm2 (Statgraphics Plus
for Windows™, 1999).
5. Detection of explosives
FOC-GAP-FTIR spectroscopy is suitable for development of methods for detection of traces
of explosives on surfaces. A smearing transfer method can be used for depositing the target
explosives on the substrates to be used as standards and samples. The sample transfer
method is appropriate to compare with other methods of sample preparation due the fact
that a mass balance is not needed in order to know the amount of the sample on the surface.
Besides that, many plates were prepared, good reproducible distributions were found (the
analyte was distributed almost homogeneous on the surface). Samples ranging from
micrograms/cm2 to nanograms/cm2 of 2,6-dinitrotuelene (DNT), 2,4,6-trinitrotoluene
(TNT), pentaerythritol tetranitrate (PETN), nitroglycerine (NG) and triacetone triperoxide
(TATP) were deposited as on stainless steel surface. Methanol, acetone and acetonitrile were
used as transfer solvents. The IR reflectance spectra were recorded at 4 cm-1 resolution and
50 scans. The results of the experiments gave intense absorption bands in the fingerprint
region of the IR spectra that were used to calculate the detection limit for each of the target
explosives. The nitro band can be used for explosives detection since it acts as a vibrational
signature of several classes of explosives: nitroaromatic, nitroaliphatic, nitramines and
nitrate esters. Figs. 8 and 9 show the prominent signal of nitro explosives deposited on
stainless steel surfaces. Only one signal in the range 1200 – 1400 cm-1 was significant for
quantitative and qualitative analysis. This band can be attributed NO2 stretching vibration.
Nitro stretching vibration of PETN and NG appears in the 1250 – 1320 cm-1 region. For
nitroaromatic explosives such as TNT and 2,6-DNT the band appears at 1320 – 1360 cm-1 [42,
43].
www.intechopen.com
This difference can be explained in terms of the fact that the group NO2 in PETN and NG is
attached to an oxygen atom. However, in DNT and TNT, this nitro group is directly
attached to the aromatic ring. The high electronegativity of the oxygen atom in PETN and
NG attracts electron density from the nitro group leading to a lowering of the oscillator
strength and causing a shift to lower frequencies. This effect is lower or not present in the
aromatic ring for TNT and DNT.
a 8.5 µg/cm²
4.4 µg/cm²
0.3 µg/cm²
Absorbance
1100 1175 1250 1325 1400 1475
Wavenumbers / cm-1
1.6 µg/cm²
b
0.8 µg/cm²
0.4 µg/cm²
Absorbance
1100 1175 1250 1325 1400 1475
Wavenumbers / cm-1
Fig. 8. Grazing angle spectra of nitroexplosives: a. PETN; b. NG

Fig. 10 includes the calculated classical detection limits for some nitroexplosives. For surface
loadings near the detection limit only the NO2 signal can be observed without aid of
software. The detection limit varies according to macro properties. Properties such vapor
pressure, physical adsorption, sublimation rate and surface-adsorbate thermodynamics can
influence the detection limit. A close relation between vapor pressure and limit detection is
shown for nitro explosives. Table 1 shows the values for the vapor pressures near room
temperature of the explosives studied. The amount of explosive on surface of stainless steel
www.intechopen.com
and the residence time depends on this property, because at this loading concentration the
detection limit of 8 μg/cm2. Some macro properties also can affect sublimation at room
explosive goes to vapor phase fast. This phenomenon is remarkably observed in TATP with
temperature and TATP detection limit.

Fig. 11 illustrates that several TATP bands can be used for its detection with accuracy. It
shows the prominent presence of the most intense bands of TATP in the IR fingerprint
region. The band at 1205 cm-1 belongs to the C-O stretch, 1365 cm-1 is a deformation of CH3
group and the band at 1471 cm-1 is an asymmetric deformation of CH3 group. The
spectroscopic window used for TATP detection was spectral range of 1320-1407 cm-1.
Forward selection analysis of variable significance affirms that the significant peaks were
contained within the spectral range of 1330-1407 cm-1 (Beeson, 1998; Demuth, 1998). The best
discriminant model was done using peak areas. It was selected based on statistical
significance and the percent of cases correctly classified. The statistical significance value (p-
value) was p < 0.0001 and the percent of cases correctly classified was 90.6%.
a
Absorbance
10µg/cm²
5 µg/cm²
2.5 µg/cm²
1100 1175 1250 1325 1400 1475

Wavenumbers / cm-1
b
Absorbance
2.5 µg/cm²
1.25 µg/cm²
0.3 µg/cm²
1100 1175 1250 1325 1400 1475

Wavenumbers / cm-1
Fig. 9. Grazing angle spectra of nitroexplosives: a. 2,6-DNT; b. TNT
www.intechopen.com
2.5
2.5
1.5
µg/cm² 1
0.4
0.5 0.3 0.3
0
DNT > NG > TNT > PETN
Limit of detection and deceasing vapor presure trend
Fig. 10. Detection limits for selected nitroexplosives and correlation with decreasing vapor
pressure of the energetic compound
Explosive Vapor pressure(mm Hg)

2,6-dinitrotoluene 5.67 x 10-4 at 25 °C
2,4,6-trinitrotoluene 1.99 x 10-4 at 20 °C
nitroglycerin 2.0 x 10-4 at 25 °C
pentaerythritol tetranitrate 1.035 x10-10 at 25°C
TATP 5.25 x 10-2 at 25 °C
Table 1. Vapor pressure of nitroexplosives and TATP

It is important to emphasize that measuring surface concentrations using the peak area
method is conceptually simple and easy to use, but it has limitations. The method is
univariate (the concentration is determined with a single spectral peak) and depends on a
linear correlation between the concentration and the spectral response. The results can,
therefore, be undermined by perturbations such as fluctuations caused by detector noise,
temperature variations, or molecular interactions. Statistically based, multivariate
calibrations use spectral features over a wide range. Information from a calibration spectral
set (a training set) was compared to independently determined concentration data using
partial least squares (PLS) regression. This method is based on the assumption that
systematic variations in the spectra are a consequence of concentration changes. A
calibration model for analysis of 2,6-DNT was built using the Quant 2 package, an add-on
software package to the OPUSTM data acquisition and analysis software (Bruker Optics). The
best spectral region was in the range of 1702 – 1269 cm-1. This range was used for model
generation. No spectral data preprocessing was applied to the spectra. The results are
illustrated graphically in Fig. 12. The root mean square error of the cross validation
(RMSCV) was 0.957, and % R2 was 97.75. This calibration was used in order to predict
μg/cm2, 3.58 and 7.35 μg/cm2 were detected, respectively. Clearly, low levels of explosives
unknown loading concentrations. For a deposited loading concentration 3.78 and 7.56
can be detected and measured on a metal surface with good results. So, chemometrics easily
leads to a powerful technique for surface contamination detection and measurement.
Moreover, classical detection limits do not apply any longer. Thus the reported values are
on the conservative side.
www.intechopen.com
0.16
100 µg/cm² 40 µg/cm² 20 µg/cm²
0.14
0.12
0.1
Absorbance
0.08
0.06
0.04
0.02
0
1100 1150 1200 1250 1300 1350 1400 1450 1500
-1
Wavenumbers (cm )
Fig. 11. GAP spectra of TATP at three surface loadings: 20, 40 and 100 μg/cm2
25
20
Predicted
15
10
0
0 3 6 9 12 15 18 21
True
Fig. 12. Partial least squares regression cross validation plot for 2,6-DNT
100
80
R %
60
2
40
20
0 2 4 6 8 10 12
Numbe r of PLS
Fig. 13. Variation of % R2 with the number of PLS carried out
www.intechopen.com
The importance of applying PLS that it is used to design and build robust calibration models
for quantitative analysis. PLS regression is a quantitative spectral decomposition technique
that is closely related to principal component (PC) regression. It uses the concentration
information during the decomposition process. This causes spectra containing higher
constituent concentrations to be weighted more heavily than those with low concentrations.
The main idea of PLS is to get as much concentration information as possible into the first
few loading vectors or number of PLS (Kramer, 1998; Otto, 1999).
Fig. 13 shows how the calibration model improves with the addition of PLS. Five PLS
executions were necessary to build a good calibration model. This indicates that the
relationship that exists between the loading concentration and the spectral absorbance in
this technique is multidimensional. The robustness of model calibration was evaluated
using internal jackknifing validation [44]. Model with lower PLS than 6 was not capable to
predict new data with good precision. Fig. 14 shows the appearance of TNT deposits under
high magnification of a white light microscope. At these loading concentrations almost all
the stainless steel surface was covered by crystals. Using this transfer method, positive and
inverted bands were observed (Fig. 15). For loading concentrations > 8 μg/cm2, only
positive bands were observed. This fact can be attributed some changes in the refraction
index of TNT.
a. 20 μg/cm2 50 x b. 40 μg/cm2 100 x

Fig. 14. Optical images for TNT deposited on stainless steel using smearing method
Sample Detection Limit

Substance Solvent
Preparation Method (μg/cm2)
TNT Smearing dichloromethane 0.3
2,6-DNT Smearing acetone 0.3
PETN Smearing methanol 0.3
NG Smearing acetonitrile 0.4
TATP Smearing dichloromethane 8.0
Table 2. Detection limits of Explosives using FT-RAIRS on stainless steel

The calibration model was built using the Quant 2 package, an add-on software package to
the OPUSTM (Bruker Optics) data acquisition and analysis software. In this study, the model
parameters were optimized in the spectral region 1668-1045 cm-1. No spectral data
preprocessing was done. The resulting model was cross-validated using the ‘’leave one out’’
www.intechopen.com
method in which each spectrum is omitted in turn from the training set and then tested
against the model built with the remaining spectra. The results are illustrated graphically in
Fig. 16. The root mean square error of the cross validation was 0.918, and R2 was 0.9858.
TNT can be detected and quantified on metallic surfaces and low concentrations as 500
ng/cm2. Table 2 shows a summary of the detection limits of the explosives using smearing
transfer method. The detection limit depends on the vapor pressure.
0.5 ug/cm2 1 ug/cm2 8 ug/cm2 20 ug/cm2 40 ug/cm2

Absorbance
1000 1100 1200 1300 1400 1500 1600 1700 1800 1900
Wavenumbers (cm-1)
Fig. 15. FT absorption-reflection IR spectra of TNT on stainless steel using smearing transfer
method
concentration μg/cm2
1.8
Predicted surface
y = 1.005x - 0.0019
1.3 R² = 0.9858
0.8
0.3
-0.2
True surface concentration / μg/cm2

0 0.25 0.5 0.75 1 1.25 1.5 1.75
Fig. 16. Leave one out cross validation for TNT surface concentration deposited on stainless
steel
6. Sublimation studies of TATP deposited on metal surfaces

The rapid sublimation of the TATP under normal atmospheric conditions presents a
significant challenge in applying the fiber-optic grazing-angle method to surface detection
www.intechopen.com
of the compound. Even under controlled laboratory conditions, where spectroscopy was
carried out on freshly prepared samples, it proved impossible to develop PLS1 calibrations
that met a high standard of robustness or usability. The rapid change in the surface
concentration of TATP made it impractical to collect more than one spectrum from each
sample and this further limited the possibilities of building a statistically useful data set. The
samples were deposited on the surface using a smearing method. To carry out the
experiments, TATP was synthesized in the laboratory. For the calibration curves TATP was
dissolved in dichloromethane. A solution with an initial concentration of 0.23 g/mL was
prepared and then dilutions were made until obtain 0.23 g/mL. The resulting average
surface concentrations of TATP ranged from 8 to 200 μg/cm2. Since dichloromethane
evaporates very fast (Boiling point = 39.8 °C), a thin sample film was observed after
smearing. Once the solvent had evaporated, the spectrum of the sample was collected
immediately to minimize the impact on the calibration of rapid TATP sublimation. The data
other experiments done with stainless steel plates coated with 25-100 μg/cm2 TATP, spectra
was analyzed using chemometrics routines; in particular multivariate PLS was used. In
were recorded every 27 seconds at 20-30 °C and the sublimation behavior at the studied
temperatures was observed.
The readiness with which TATP sublimates (vapor pressure at room temperature = 7 Pa)
under normal atmospheric conditions complicates the task of calibrating the detection of the
compound on surfaces, as the surface concentration of TATP decreases over the timescale of
the experiment. To explore this effect, experiments were performed with stainless steel
plates initially given a nominal loading of 100, 80, 50, 20, and 10 μg/cm2 TATP. Immediately
after deposition of the TATP, the probe head was positioned on the surface, and spectra
were collected every 27 seconds. Fig. 17 shows successive spectra taken at 27 s intervals
from an initial loading of 100 μg/cm2 TATP.
16
14
12
10
0
0 2 4 6 8 10 12
Fig. 17. Peak areas for the grazing-angle mid-IR spectra in the range from 100-1300 cm-1, for
initial loadings of 100, 80, 50, 20, and 10 μg/cm2
A peak-fitting model of the spectral region from approximately 1300-1100 cm-1 was fitted
against each of the spectra in turn to give the total peak area for the selected region. Fig. 18
www.intechopen.com
shows the result for samples of stainless steel substrates initially loaded with 10-100 μg/cm2,
represented as a graph of peak areas versus time. The amount of TATP detected on the
surface drops below the apparent detection limit of the technique within about 9 min.
Standard were prepared as described in the Experimental Section. Grazing angle FT-IR
spectra of freshly prepared samples were collected for a series of different surface
concentrations, as shown in Table 1. When a PLS1 model was built from all of the 79 spectra
listed, using the spectral region from 1066-1506 cm-1 and no spectral preprocessing, it
proved impossible to develop a model that met a reasonable standard (the maximum
obtainable value for R2 was about 0.75). In this study, the model parameters were optimized
in the spectral region 1498 – 1113 cm-1. No spectral data preprocessing was done. When data
used for the model was limited to loadings below 40 μg/cm2, it was possible to build a
calibration with R2 = 0.869, and root mean square error of cross validation (RMSECV) = 3.69
(obtained from a leave-one-out cross validation); the results are shown graphically in Fig.
18. The graph shows the degree of scattering in the data. While some of this scattering may
be attributable to variations in the amount of TATP deposited on each coupon, sublimation
of the TATP during the experiment is another likely contributing factor. Given these
limitations, the quality of the calibration that has been developed is surprisingly good and it
is clearly quite possible to detect microgram quantities of TATP on metal surfaces using
grazing angle FTIR methods. This is in agreement with previous results for a range of
organic compounds on metal and glass surfaces (White, 1992; Lin-Vien, 1991).
40
True concentration / µg cm-2
35 y = 0.8788 x + 2.8874
R2 = 0.8686
30
25
20
15
10
0
4 9 14 19 24 29 34 39
Predicted concentration / µg cm-2
Fig. 18. “Leave-one-out cross” validation of predicted values vs. true values for TATP on
stainless steel substrates. All data are shown illustrating point scattering at each value
measured
TATP quantification was done using the calibration generated by chemometrics. These
results are shown in Table 3. Clearly, low levels of TATP can be detected and measured on a
the loadings of 40 μg/cm2 and above were unsuccessful may be attributable to the change in
metal surface with quantitative results (Table 4). Attempts to build a separate calibration for
the nature of the surface coating at high loadings, from a thin film capable of generating a
double-pass transmission spectrum to a bulk material generating diffuse surface reflectance
www.intechopen.com
independent of the thickness (Blaudez, 1998; Bradshaw, 1988; Hayden, 1987). Calibration
curve is shown in Fig. 19.
A true "transflectance" experiment is possible only when the layer of organic material is
“very” thin, so that the IR radiation can pass all the way through it to be reflected from the
substrate (e.g. the metal). If the coating is thicker than about 1 or 2 micrometers, the IR
radiation does not reach the substrate and is reflected from the top surface of the organic
material (as if it were a "bulk" sample of the organic material) to give diffuse reflectance
(Kaihara, 2001). This changes the spectrum that is obtained from the organic material from
the "transflectance" spectrum to a diffuse reflectance spectrum. The diffuse reflectance
spectrum is not affected by the thickness of the coating, since it does not come from the
whole coating but only from the top layer, and so it does not contribute to a useful
calibration.
TATP Loading (μg/cm2) Number of Spectra Included in Calibration
5 1 No
8 3 Yes
10 4 Yes
12 4 Yes
15 4 Yes
18 3 Yes
20 7 Yes
22 5 Yes
25 3 Yes
30 9 Yes
35 7 Yes
38 7 Yes
40 8 No
50 2 No
60 2 No
80 2 No
90 2 No
150 1 No
200 1 No
Table 3. List of TATP spectra collected for calibration
www.intechopen.com
Deposited Detected by grazing angle Difference

(μg/cm2) (μg/cm2)
14.96 13.8 1.16
17.96 16.5 1.46
24.94 25.3 -0.36
34.91 32.6 2.36
Table 4. Quantification of TATP on metal plates by grazing angle Fiber optic FT-IR
Discriminant analysis was also performed to classify the TATP loading concentration in two
groups (Fig. 20). The first one corresponds to concentrations lower than 25 μg/cm2 and the
second one to concentrations higher than 25 μg/cm2. Peak areas of signals in the range of
1330-1407 cm-1 and 1407-1503 cm-1 were used for the discrimination. Forward selection
analysis of variable significance affirms that the significant peaks were contained within the
spectral range of 1330-1407 cm-1. The best discriminant model was selected based on
statistical significance and the percentage of cases correctly classified. The percentage of
cases correctly classified was 90.6% and the significance statistical p-value was < 0.0001.
1.5
Peak Areas / a.u.
y = 0.0369x + 0.0287
R² = 0.9376
0.5
0
0 25 50 75 100 125 150 175 200
Loading / µg cm-2
Fig. 19. Calibration curve for TATP on SS surface using peak areas (1330-1407 cm-1)
21. For temperatures lower than 20 °C TATP sublimation rate is significantly low. Beyond
TATP sublimation behavior depends of the temperature as is expected. This is shown in Fig.
this point the sublimation rate starts to increase very fast. The rate of sublimation for TATP
was calculated in units of peak-area/second for the slope of the curve, taking into account
the linear range of the graphs (Fig. 21-a and -b). Since the surface concentration is well
approximated by the measured peak area in this range because these are proportional (Fig. ,
the mass transferred to the vapor phase is reasonably well estimated by the decrease in peak
www.intechopen.com
areas in the MIR. This range was selected because here the rate decays more rapidly. These
temperature in the range of 20-30 °C is -0.0013 peak area-s-1 °C-1. This is equivalent to ~ -0.81
results can be confirmed in Fig. 19. The estimated value of the rate sublimation with
μg cm-2 s-1 °C-1 in the range of 20-30 °C. These experiments confirm the fact that TATP
sublimates very fast in the absence of vapor pressure reducing and stabilizing agents.
Leave in < 25 μg/cm²

Discrimination
Leave in > 25 μg/cm²
1
Leave out < 25 μg/cm²
Leave out > 25 μg/cm²
0
0 0.5 1 1.5 2 2.5
Peak areas
Fig. 20. Plot of peak area selected to construct the discrimination for grouping 67
concentrations
1.8
20 C 25ug/cm2
1.6 2 26 C 25ug/cm2
20 C 50ug/cm2
1.4 26 C 50 ug/cm2
20 C 100 ug/cm2 1.6
1.2 26 C 100ug/cm2
Peak areas / a.u.
Peak areas / a.u.
1 1.2
0.8
0.6 a 0.8 b
0.4
0.4
0.2
0
0
0 100 200 300 400 500 600 700 800 900
0 100 200 300 400 500 600
Time / s Time / s
Fig. 21. Plots of peak areas versus time (second) for 25, 50 and 100 μg/cm2 at: a: 20 ºC; b: 26 ºC
www.intechopen.com
-0.0005
30 C 100ug/cm2
Peak areas / a.u.
Slope / Peak Areas/s

y = -0.0013x + 0.0263
-0.0045 R² = 0.9998
30 C 25ug/cm2
-0.0085
30 C 50ug/cm2
a -0.0125 b
-0.0165
0 100 200 300 400 500 600 20 21 22 23 24 25 26 27 28 29 30
Time / s Temperature / °C
Fig. 22. a: Plot of peak areas versus time (s) for 25, 50 and 100 μg/cm2 at 30 ºC; b:
sublimation rate for different temperatures
7. Characterization studies of TNT deposited on metal surfaces by TIJ

technology
Although smearing transfer method gave almost a homogenous distribution of the
explosives on surfaces, the method of mass transfer required the preparation of many
samples. This was due to the fact that the sample transfer method is prone to uncontrolled
operator errors. To circumvent the problem, Thermal inkjet technology (TIJ) was used for
the experiments. TNT was selected as an explosive for dispensing on TIJ.
Thermal Inkjet Technology is able to deposit very small amounts of chemical compounds,
including energetic materials, in a specific location on a surface. Aliquots of TNT solutions
were deposited on stainless steel film. A coating of TNT can be produced by controlling the
concentration of TNT, the number of drops dispensed and the distribution of drops over the
surface. The loading concentration of the sample on the surface can be controlled by varying
parameters such as: number of passes, dispensing frequency, applied energy and pen
architecture. Precise delivery of known number of droplets with known mass and
concentration are known. Also only one solution can be used, avoiding dilutions that can
increase the analytical errors.
The precise amount of TNT on the surface can be known in different ways: one can be
known if the drop weight of the solutions is known and the other using an alternative
method for quantification (GC, UV or HPLC). In our study, the stainless steel sheets loaded
with TNT samples were rinsed with 25-100 mL of acetonitrile to remove the entire TNT that
was delivered on the surface. Then solutions were transferred a volumetric flask and filled
to mark with solvent. Gas chromatography was used in our experiments as an alternative
method in order to determine the surface loading concentration. The analysis was carried
out using an Agilent Technologies 6890N, Network GC System with A micro cell 63Ni
Electron Capture Detector (µECD). For GC separation, a capillary column was used: RTX-5
(cross bonded 5% diphenil-95% diethyl polysiloxane) 15 m x 0.25 mm ID x 0.25 μm df,
Restek Corp, Bellefonte, PA. The GC oven was held at 80 °C for 1 min and then programmed
at 10 °C/min to 180 °C, followed by a 30 °C/min ramp to 300 °C. The temperature at the
injection port was 250 °C.
www.intechopen.com
Calibration curves for GC-µECD were prepared with 1000 ppm standard solutions of TNT
obtained from Restek Corp. Stock solutions of concentrations: 1, 0.5, 0.1, 0.05, 0.01 ppm
diluted in HPLC grade acetonitrile (Sigma-Aldrich Chemical Co., Milwaukee) were
prepared. All solutions were injected directly using glass syringes (Hamilton Series 7101)
into the injection port, this analysis were carried out with three replicates for each
concentration. Using TIJ method it was possible to generate standards samples with more
uniform coverage, and one advantage is the fact that the surface loading concentration could
be varied by changing the numbers of passes delivered to the sample without the need for
serial dilutions. Fig. 23 shows the linear correlation of the variation of the number of passes
for a 5292 ppm solution. A direct relationship is observed.
4
Loading concentration (ug/cm2)
y = 0.765x - 0.281
3 R² = 0.9893
0
1 2 3 4 5
Number of TIJ passes
Fig. 23. Variation of the number of passes delivered to the surface for a TNT 5292 ppm
solution
For visualization of substrates after sample deposition, optical microscopy images were
captured using Leica microscope (model LS). These images can be seen in Fig. 24. The
images clearly reveal that TNT is collecting on the stainless steel surface as droplets and
increases. At 1.25 μg/cm2 loading concentration, the surface is practically covered by

crystals. As the number of passes increases on the surface, the size of the droplets
crystals.
The way how the sample is distributed on the surface plays an important role in the task of
calibrating the detection of the TNT or other compound on surfaces. To explore this effect,
experiments were performed by positioning the probe head on different parts on the
surface, and spectra were collected. Fig. 25 shows how the analytical response varies with
the loading concentration. A peak-fitting model of the spectral region from approximately
1583-1396 and 1418-1220 cm-1 was fitted against each of the spectra in turn to give the total
peak area for the selected region. This confirmed that although the coating is not
homogenous a pattern with TIJ is generated, giving a few errors.
www.intechopen.com
a. TNT 0.39 μg/cm2 50 μm b. TNT 0.39 μg/cm2 50 μm
c. TNT 1.25 μg/cm2 50 μm d. TNT 1.25 μg/cm2 50 μm

Fig. 24. White light images of TNT deposits on stainless steel substrates using TIJ for
deposition
The asymmetric and asymmetric vibrational stretches of the nitro (NO2) group can be used
for explosives detection since they act as vibrational signatures of several classes of
explosives: nitroaromatic (TNT, DTNT), nitroaliphatic (CH3NO2), nitramines (RDX, HMX)
and nitrate esters (nitroglycerine, PETN). Fig. 26 shows the prominent signals of TNT
deposited on stainless steel surfaces. All nitro signals were significant for quantitative and
qualitative analysis. Nitro symmetric stretching vibration of TNT band appears at 1320 –
1360 cm-1 and nitro asymmetric stretching vibration is typically located in the wavenumber
downwards is finally obtained for 1.73 μg/cm2 and 3.58 μg/cm2 loading concentration (Fig.
range: 1477 – 1600 cm-1. A completely inverted spectrum with all bands pointing
loadings and for high surface concentrations (3.58 μg/cm2 shown in trace of Fig. 27-A), the
26). Fig. 27 shows TNT IRRAS spectra of two types of deposits. For very low surface
typical downward looking percent transmission profiles are observed. However, IRRAS
spectroscopic features of thins films on surfaces frequently show strongly variations of the
observed when the loading concentration of TNT was bellow 1.73 μg/cm2 (Fig. 27-B).
relative intensities and wavelength shifts as the surface loadings changes. These effects were
Significant wavenumber shifts (> 10 cm-1) were found for the nitro vibrations bands. An
inverted upward looking profile similar to an absorption spectrum of the bands is clearly
seen in lower trace in Fig.
www.intechopen.com
1583.1-1396.4 cm-1 1417.7-1220.3 cm-1
2.50
2.00
Peak areas
1.50
1.00
0.50
0.00
c0.27 c0.39 c0.49
2
Loading concentration (ug/cm )
Fig. 25. Variation of the distribution of the TNT deposited on stainless steel using TIJ with
the analytical response
The reflected radiation depends of the nature of the film as well as the incident angle of
the IR beam reaching the surface. The properties of the reflected IR light depend critically
on localized optical characteristics of the film such as the local index of refraction. Sample
heterogeneities give rise to local changes in the refractive index across the sample and
affect the reflectance spectra. Typical Infrared Reflection Absorption Spectroscopy
(IRRAS) spectra are reported in terms the measured absorbance defined as (Bradshaw,
1988; Hayden, 1987):
A = Log (Rr/Rs) (1)

where: Rr is the reflectance of the of the plate (substrate) and Rs is the reflectance of the
(0.27- 0.47 μg/cm2). Fig. 26-B shows corresponding RAIS spectra for 1.73 and 3.58 μg/cm2
sample-substrate [22]. Fig. 26-A shows “normal” (vertical up) for low surface loadings
wavenumber range for Rr and Rs of 0.27 μg/cm2 and 3.58 μg/cm2. The metallic substrate
illustrating downward pointing peaks. Figs. 27-A and B contains expanded views in the
reflects more than the sample (TNT), it means that Rr > Rs and a normal peak is found.
This can be observed in Fig. 27-B. The inverted peak is found in 1561 because the
difference between Rr and Rs in this frequency value is lower than in other values around
this one in the frequency range 1548-1565 cm-1. This means that the sample is reflecting
almost at the same level as the metal substrate. This fact can be explained based on
changes in the refractive at the film-metallic surface interphase or by the way that TNT
layers are packed on the surface. Another reason can be argued is the fact that these
inverted bands are found when most of the coverage of the surface is TNT crystals and
not amorphous TNT droplets.
It is very important to emphasize that measuring surface concentrations using the peak
area method is conceptually simple and easy to use, but it has some limitations. The
method is univariate: the concentration is determined with a single spectral peak and it
www.intechopen.com
depends on a linear correlation between the concentration and the spectral response. The
results can, therefore, be undermined by perturbations such as fluctuations caused by
detector noise, temperature variations, or molecular interactions. Statistically based,
multivariate calibrations use spectral features over a wider range. Information from a
calibration spectral set (a training set) was compared to independently determined
concentration data using partial least squares regression (PLS). The method is based on
the assumption that systematic variations in the spectra are a consequence of
concentration changes.
0.27ug/cm2 0.39 ug/cm2 0.47 ug/cm2

a
Absorbance
1000 1200 1400 1600 1800 2000

Wavenumbers / cm-1
1.73 ug/cm2
Absorbance
3.58 ug/cm2
1000 1200 1400 1600 1800 2000

Wavenumbers / cm-1
Fig. 26. Grazing angle spectra of TNT: a. Positive bands (normal). b. Inverted bands
www.intechopen.com
0.27ug/cm2 3.58 ug/cm2 0.13

3.58 ug/cm2 spectra 0.27 ug/cm2 spectra
Rr Background
3.58 ug/cm2 Rs
0 .5
a 0.11
0.45
0 .4
0.35
0 .3
0.25 1560
0 .2
0 . 15
c
15 4 8 15 5 0 15 5 2 15 5 4 15 5 6 15 5 8 15 6 0 15 6 2 15 6 4
0.09 A
Absorbance
1365.4 0.07
Absorbance
1562.1 Rr Background
0.27 ug/cm2 Rs 1562.1
0.47 1550
0.05 0.42
0.37
0.32
0.27
1353.8 1550.5 0.22
0.03 15 4 1 15 4 3 15 4 5 15 4 7 15 4 9 15 5 1 15 5 3 15 5 5 15 5 7
B 1550.5
b
0.01
-0.01
1000 1200 1400 1600 1800 2000
1500 1505 1510 1515 1520 1525 1530 1535 1540 1545 1550 1555 1560 1565 1570 1575 1580
Wavenumbers (cm-1) Wavenumbers (cm-1)
Fig. 27. IRRAS symmetric and asymmetric Nitro stretching vibrations of TNT a: normal
(downward pointing) %T profile; b: inverted (upward pointing) %T profile; c: behavior of
Rs and Rr for the peak formation
4.0
y = 0.9694x + 0.0219
3.5 R2 = 0.9959
Predicted loading / µg cm-2
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0 0.5 1 1.5 2 2.5 3 3.5 4
True loading/ μg cm-2
Fig. 28. Graph of averaged predicted surface loading values versus true values obtained by
the “leave-one-out” cross validation of a Quant2 calibration for TNT on steel surfaces
Grazing angle FTIR spectra of prepared samples were collected for a series of different
surface concentrations. When a PLS1 model was built from all of the 84 spectra listed, using
the spectral region from 1028-1713 cm-1 and no spectral preprocessing, it was possible to
build a calibration with R2 = 0.9959, and root mean square error of cross validation
www.intechopen.com
(RMSECV) = 0.258 (obtained from a “leave-one-out” cross validation); the results are shown
graphically in Fig. 28. According to these results TIJ-IRRAS combined with PLS can be used
for TNT detection and quantification on metallic surfaces.
8. Conclusion
Fiber optic coupled FTIR spectroscopy has been shown to provide a useful technique for
ingredients and excipients on metallic surfaces. Very low detection limits (∼ 0.3 μg/cm2)
methods development for in situ detection explosives and active pharmaceutical
were found for nitroexplosives and for organic peroxides. The present study was limited
to highly reflective metallic surfaces such as stainless steel. However, since it is based on a
grazing angle probe that is sensitive to adsorbates deposited on substrates regardless of
the nature of the surface, and at least to first order, the methodology should be applicable
for trace detection of explosives on other types of surfaces. The coupling of data
acquisition to powerful non linear, multivariate analysis based on chemometrics routines
such as partial least squares resulted in an even more robust analytical methodology.
Explosives detection for this technique as applied to the target molecules used was found
to be limited by the residence time of the substance on the surface. At low loading
concentrations, high vapor pressure explosives escape to the vapor phase by sublimation
high vapor pressure organic peroxide, was 8 μg/cm2. For nitroexplosives such as 2,6-
even faster, limiting the low limit of detection achievable. The detection limit for TATP,
DNT, NG TNT, and PETN LODs found were 2.5, 0.4, 0.3, and 0.3 μg/cm2 respectively. The
results are in good agreement with a decreasing low limit of detection when arranged in
the order of decreasing vapor pressure.
In this work we present the first method for detection and quantification of TATP on
metallic surfaces. The method entails the coupling of optical fibers operating in the mid
infrared (MIR) to a high performance interferometer. A grazing angle head coupled to
liquid nitrogen cooled, MCT detector was used to excite and collect IR absorption spectra
from stainless steel surfaces loaded with the organic peroxide. A smearing technique was
used to prepare standards for surface loading of the explosive.
The novel aspects of the methodology involve in situ, remote sensed, fiber optic coupling of
a grazing angle probe to a high sensitivity FTIR interferometer. It must be noted that the
sample must to be measured immediately after the smearing to avoid sublimation of TATP
from the metal surface. The robustness of the new methodology relies on utilizing powerful
Chemometrics routines and Discriminant Analysis for statistical enhancement of data.
Discriminant Analysis can be used to quantify TATP by classifying loading concentration in
two groups: the first one corresponds to concentrations lower than 25 μg/cm2 and the
second one to concentrations higher than 25 μg/cm2.
This work describes the first demonstration of a method for direct, in-situ detection and
quantification of TATP on metallic surfaces. The rapid sublimation of TATP from the test
surface has been demonstrated by tracking the decreasing intensities of the IR spectrum of
TATP; the signal falls below the detection limit after about 9 minutes, from an initial value
of 100 μg/cm2. This phenomenon limits the extent to which the surface concentration of
TATP can be calibrated. However, it has been possible to show that TATP on metallic
surfaces can be quantitatively detected in a matter of seconds using the grazing-angle FTIR
approach.
www.intechopen.com
The first method for detection and quantification of TNT on metallic surfaces using thermal
inkjet technology as a transfer method was discussed. Spectroscopic characterization and
thin layer deposits quantification was achieved using the powerful Grazing Angle Probe
Fiber Coupled-FTIR developed for surface analysis. Chemometrics routines were applied as
data enhancers in order to accomplish fully the difficult task proposed. Inkjet printing of
explosives demonstrated to have the following important characteristics: precision in
sample deposit, drop delivery with non-contact fluid transfer and high reproducibility. The
methodology could be applied for development of standards in Trace Explosive Reference
materials on surfaces.
Sample transfer methods coupled to RAIRS promises to be an excellent support means
when used as sensor for explosives detection on surfaces. A grazing angle head coupled to
liquid nitrogen cooled, MCT detector was used to excite and collect IR absorption spectra
from stainless steel surfaces loaded with the explosive. This case study suggests that MIR
reflectance spectroscopy using a fiber-optic probe with grazing angle head is a viable
method for detecting and measuring low (µg/cm2) quantities of organic contaminants on
metallic surfaces. Adding chemometrics algorithms, which are developed and automated
easily, leads to powerful techniques for surface contamination detection.
9. Acknowledgments
Parts of the work presented in this contribution were supported by the U.S. Department of
Defense, University Research Initiative Multidisciplinary University Research Initiative
(URI)-MURI Program, under grant number DAAD19-02-1-0257. The authors also
acknowledge contributions from Mr. Aaron LaPointe from Night Vision and Electronic
Sensors Directorate, Fort Belvoir, VA, Department of Defense, Dr. Jennifer Becker MURI
Program Manager, Army Research Office, DoD and Dr. Stephen J. Lee Chief Scientist,
Science and Technology, Office of the Director, Army Research Office/Army Research
Laboratory, DoD.
Support from the U.S. Department of Homeland Security under Award Number 2008-ST-
061-ED0001 is also acknowledged. However, the views and conclusions contained in this
document are those of the authors and should not be interpreted as necessarily representing
the official policies, either expressed or implied, of the U.S. Department of Homeland
Security.
Thanks are due to Dr. Luis. F De La Torre-Quintana for his assistance with statistical
analysis and coupling with chemometrics routines and for his support and collaboration in
design and preparation some figures.
10. References
Akhavan, J. (2004). The Chemistry of Explosives. 2nd ed. T J International Ltd, Padstow,
Cornwall, UK,
Bacci, M.; Fabbri, M.; Picollo, M.; Porcinai, S. (2001). Non invasive fibre optic Fourier
Transform-infrared reflectance spectroscopy on painted layers identification of
materials by means of principal component analysis and Mahalanobis distance.
Anal. Chim. Acta. 446, 15-21.
www.intechopen.com
Beebe, K.R.; Pell, R.J.; Seasholtz, M.B. (1998). Chemometrics: A Practical Guide; John Wiley &
Sons, Inc. New York, NY.
Beeson, R. and Skip, R.D., (1998). Thermal ink jet technology –review and outlook, Rung
Research Laboratory, Inkjet Business Unit, Hewlett-Packard Company, Corvallis
Oregon, US 3rd International Conference on Imaging Science. Chongqing, China.
Bellamy, A.J. (1999). Triacetone triperoxide: its chemical destruction. J. Forensic Sci. 44, 603-
608.
Blaudez, D.; Buffeteau, T.; Desbat, B.; Fournier, P.; Ritcey, A.; Pezolet, M. (1998). Infrared
refection – absorption spectroscopy of thin organics films on nonmetallic
substrates: optimal angle of incidence. J. Phys. Chem. B. 102, 99-105.
Bradshaw, A.M.; Schweitzer, E.; R.J.H. Clark.; Hester, R.E. (1988). Infrared reflection absorption
spectroscopy of adsorbed molecules, in spectroscopy of surfaces. John Wiley & Sons, New
York, NY.
Brambilla, G., Loizzo, A., Fontana, L., Strozzi, M., Guarino, A. and Soprano, V. (1997). J Am
Med Assoc., 278, 635.
Demuth, H.; Beale, M. (1998). Neural Network Toolbox for use with MATLAB User’s Guide
Version 3.0; 5th ed.; The MathWorks, Inc., Natick, MA.
Diem, M. (1993). Introduction to Modern Vibrational Spectroscopy, John Wiley, New York, NY.
Evans, H.K.; Tulleners, F.A.J.; Sanchez, B.L.; Rasmussen, C.A. An Unusual Explosive,
Triacetonetriperoxide (TATP). J. Forensic Sci. 1986, 31, 1119-1125 .
Fierro-Mercado, P.M., Primera-Pedrozo, O.M., Hornedo, A., Hernández-Rivera, S.P., An In
situ FTIR Fiber Optic Method for the Detection of Active Pharmaceutical
Ingredients and Excipients on Metallic Substrates, in “Fourier Transform Infrared
Spectroscopy: Developments, Techniques and Applications”, Rees, O.J., ed.,
Chemical Engineering Methods and Technology Series, Nova Science Publishers,
Inc. Hauppauge, NY, third quarter 2010, ISBN: 978-1-61668-835-6.
Griffiths, P.R. and De Haseth, J.A. (1986). Fourier- Transform Infrared Spectrometry. John
Wiley, New York, NY.
Hamilton, M.L.; Perston, B.B.; Harland, P.W.; Williamson, B.E.; Thomson, M.A.; Melling, P.J.
(2005). Org. Process Res. Devel. 9, 337−343.
Helmut, G.; Hans-Ulrich, G. (2002). IR Spectroscopy: An introduction. Wiley-VCH Federal
Republic of Germany.
Hollins, P. (1992). The influence of surface defects on the infrared spectra of adsorbed
species. Surf. Sci. Rep. 16, 51-94
Huberty, C. J. (1994). Applied Discriminant Analysis. Wiley-Interscience, NJ.
Hayden, B.E.; Yates, J.T.; Madey, T.E. (1987). Reflection Absorption infrared spectroscopy, in
Vibrational Spectroscopy of molecules on surfaces. Vol. I. John Wiley & Sons, New York,
N.Y.
http://chesapeake.towson.edu/data/all_electro.asp
http://toxnet.nlm.nih.gov/cgi-bin/sis/htmlgen?HSDB.htm
http://www.wcer.wisc.edu/step/ep301/Spr2000/esselman/IBUb.html
Kaihara, M.; Sato, Y.; Sato, T.; Takahashi, T.; Takahashi, N. (2001). Chemometrics for FT-IR
reflectance spectroscopy. The distinction between the specular and diffuse
reflection. Analytical Sciences. 17, 1701-1704.
www.intechopen.com
Katona, Z.; Vincze, L.; Vegh, Z.; Trompler, A.; Ferenczi-Fodor, K. (2000). Cleaning validation
procedure eased by using over pressured layer chromatographic. J. Pharmaceutical
and Biomedical Analysis. 22, 349-353.
Kevin, J, Kochlowski.; Arbuckle-Keil, G. (2000). Infrared spectroscopy. Anal Chem. 72, 71-
79R.
Kramer, R., (1998). Chemometric Techniques for Quantitative Analysis, Marcel Dekker, New
York, NY.
Kurth, D. (1998). Analysis of artifacts in infrared spectroscopy of thin organic films on
metallic substrates. Langmuir. 14, 6987-6991.
Lavine, B., and Workman, J. Jr. (2002). Chemometrics. Anal Chem. 74, 2763-2770.
Lin-Vien, D.; Colthup, N. B.; Fateley, W. G. and Grasselli, J. G. (1991).The Handbook of Infrared
and Raman Characteristic Frequencies of Organic Molecules; Academic Press, San
Diego, CA.
Mehta, N.K., Goenaga-Polo, J.E., Hernández-Rivera, S.P., and Hernández, D., Thomson,
M.A. and Melling, P.J. (2002). Development of an In-Situ Spectroscopic Method for
Cleaning Validation Using Mid-IR Fiber Optics. Bio Pharm. 15, 36-42.
Mehta, N.; Goenaga, J.; Hernandez-Rivera, S.P.; Hernandez, D.; Thomson, M.; Melling, P.
(2002). Development of an in situ spectroscopic method for cleaning validation
using mid-IR fiber optics. Spectroscopy. 18 (4),14-19.
Melling, P.J. and Shelley, P., (2001). Spectroscopic Accessory for Examining Films and
Coatings on Solid Surfaces. U.S Patent 6,3,10,348, US Patent and Trademark Office,
Washington, DC.
Melling, P, and Thomson, M. (2002). Fiber optic probes for mid-infrared spectrometry. John
Wiley & Sons Ltd., UK.
Mirza, T.; Lunn, M.; Keeley, F.; George, R.; Bodenmiller, J. (1999).Cleaning level acceptance
criteria and high pressure liquid chromatography procedure for the assay of
Meclizine Hydrochloride residue in swabs collected from pharmaceutical
manufacturing equipment surfaces. Journal of Pharmaceutical and Biomedical Analysis.
19, 747-756.
Mizaikoff, B. (2002). Sensory Systems Based on Mid-infrared Transparent Fibers. In
Handbook of Vibrational Spectroscopy, Vol. 2, Eds. Chalmers, J.M. and Griffiths, P.R.
John Wiley & Sons, Chichester, UK.
Nozal, M.; Bernal, J.; Toribio, L.; Martin, M.; Diez, F. (2001). Validation of a liquid
chromatographic method for the determination of ranitidine hydrocloridine
residues on surfaces in the manufacture of pharmaceuticals. J. Chromatogr. A. 919,
87-93.
Otto, M. (1999). Chemometrics. Statistics and Computer Application in Analytical Chemistry.
Wiley-VCH. Federal Republic of Germany.
Pacheco-Londono, L., Santiago, A., Pujols, J., Primera-Pedrozo, O.M., Mattei, A., Ortiz, W.,
Ruiz, O., Ramirez, M., Hernandez-Rivera, S.P., (2007). Characterization of layers of
Tetryl, TNB and HMX on metal surfaces using fiber optics coupled grazing angle-
FTIR. Proc. SPIE Int. Soc. Opt. Eng. 6542, 65423-65410.
Pacheco-Londoño, L.C., Primera-Pedrozo, O.M., Hernández-Rivera, S.P., (2010). Evaluation
of Samples and Standards of Energetic Materials on Surfaces by Grazing Angle-
FTIR Spectroscopy in “Fourier Transform Infrared Spectroscopy: Developments,
www.intechopen.com
Techniques and Applications”, Rees, O.J., ed., Chemical Engineering Methods and
Technology Series, Nova Science Publishers, Inc. Hauppauge, NY, ISBN: 978-1-
61668-835-6.
Primera-Pedrozo, O.M., Pacheco-Londono, L.C., De la Torre-Quintana, L.F., Hernandez-
Rivera, S.P., Chamberlain, R.T., Lareau, R.T. (2004). Use of fiber optic coupled FT-IR
in detection of explosives on surfaces. Proc. SPIE Int. Soc. Opt. Eng. 5403, 237-245.
Primera-Pedrozo, O.M., Pacheco-Londoño, L., Ruiz, O., Ramirez, M., Soto-Feliciano, Y.M.,
De La Torre-Quintana, L.F., Hernandez-Rivera, S.P., (2005-a). Characterization of
thermal Inkjet technology TNT deposits by fiber optic-grazing angle probe FTIR
spectroscopy. Proc. SPIE Int. Soc. Opt. Eng. 543-552.
Primera-Pedrozo, O.M. (2005-b). MS Thesis, University of Puerto Rico, Mayaguez Campus,
Mayaguez, PR.
Primera-Pedrozo, O.M., Rodriguez, N., Pacheco-Londono, L., Hernandez-Rivera, S.P.,
(2007). Detection of 2,4,6-trinitrotoluene on non-traditional surfaces using fiber
optic coupled grazing angle probe-FTIR. Proc. SPIE Int. Soc. Opt. Eng. 6542, 65423-
65410.
Primera-Pedrozo, O., Soto-Feliciano, Y., Pacheco-Londoño, L., Hernández-Rivera, S. (2008).
High Explosives Mixtures Detection Using Fiber Optics Coupled: Grazing Angle
Probe/Fourier Transform Reflection Absorption Infrared Spectroscopy. Sens
Imaging: Int. J. 9 (3), 27-40.
Primera-Pedrozo, O., Soto-Feliciano, Y., Pacheco-Londoño, L., Hernández-Rivera, S. (2009).
Detection of High Explosives Using Reflection Absorption Infrared Spectroscopy
with Fiber Coupled Grazing Angle Probe/FTIR. Sens Imaging: Int. J. 10 (1), 1-13.
Primera-Pedrozo, O.M., Soto-Feliciano, Y.M., Pacheco-Londoño, L.C., Hernández-Rivera,
S.P., (2010). Fiber Optic-Coupled Grazing Angle Probe-Fourier Transform
Reflection Absorption Infrared Spectroscopy for Analysis of Energetic Materials on
Surfaces, in “Fourier Transform Infrared Spectroscopy: Developments, Techniques
and Applications”, Rees, O.J., ed., Chemical Engineering Methods and Technology
Series, Nova Science Publishers, Inc. Hauppauge, NY, ISBN: 978-1-61668-835-6.
Salleras, L., Donguez, A., Mata, E., Taberrer, J.L., Moro, I. and Salva, P. (1995). Public Health
Rep., Vol 110 pp338-342.
Schrader, B. (1995). Infrared and Raman spectroscopy: Methods and applications, Schrader, B., Ed.
VCH, New York, NY, pp. 215.
Schulte-Ladbeck, R.; Kolla, P.; Karst U. (2003). Trace Analysis of Peroxide-Based Explosives.
Anal. Chem. 75,731-735.
Schulte-Ladbeck, U, Karst. A field test for the detection of peroxide-based explosives. (2002).
Analyst. 127, 1152-1154.
Statgraphics Plus for Windows (1999). Statistical graphics System, User’s Guide version 3.0.
Statistical Graphics Corporation.
Smith, B. Fundamentals of Fourier Transform Infrared Spectroscopy, CRC Press. Boca
Raton, Florida, 2000.
Soto-Feliciano, Y., Primera-Pedrozo, O.M., Pacheco-Londono, L., Hernandez-Rivera, S.P.,
(2006). Temperature dependence of detection limits of TNT on metallic surfaces
using fiber optic coupled FTIR. Proc. SPIE Int. Soc. Opt. Eng. 6201, 62012-62019.
www.intechopen.com
Urbanski, T. (1964). Chemistry and Technology of Explosives. Vol. 1, Macmillan Co.: New York,
NY.
Zitrin, S.; Kraus, S.; Glasttstein, B. (1984). in: Proceedings of the International Symposium,
US Government Printing Office, Washington, DC. pp 137-141.
White, G.M. (1992). An Explosive Drug Case. J. Forensic Sci. 37, 652-656.
www.intechopen.com
13
Fourier Transform Photocurrent Spectroscopy

on Non-Crystalline Semiconductors
Jakub Holovský
Institute of Physics AS CR, v. v. i.
Czech Republic
1. Introduction
Spectroscopic methods are massively used in material physics, because as seen already from
Einstein’s explanation of photoeffect phenomenon material response to light is wavelength
sensitive. Such sensitivity is one of key characteristics for semiconductors and therefore
spectral dependence of conductivity on light excitation is of strong interest. Band structure
of electron configuration in semiconductors projects into the spectra of light absorption,
emission or photocurrent. In Fig. 1. such projection into spectrum of absorption coefficient
(details will be explained in paragraph 4.1) can be seen. Although absorption can be
measured optically, very low light absorptance cannot easily be measured so the method of
measurement of photocurrent upon light illumination is used instead. Basically the aim of
the photocurrent spectroscopy is to obtain the curve like the one on the right of the Fig. 1.
Fig. 1. On the left: band structure of density of electronic states of amorphous silicon
On the right: optical absorption coefficient curve with indicated regions attributed to
different electron transitions - desired result of photocurrent spectroscopy
Photocurrent method based on the Fourier transform (F-T) and called FTPS (Fourier
Transform Photocurrent Spectroscopy) is a spectroscopic method where monochromator is
replaced by FTIR (Fourier Transform Infrared) spectrophotometer. This method was firstly
www.intechopen.com
reported in 1997 (Tomm et al., 1997) and later (Poruba et al., 2001), (Vaněček et al., 2002)
well promoted mainly for the purposes of R&D of thin film silicon photovoltaics where the
quality of semiconducting layers are monitored. In earlier method CPM (Constant
Photocurrent Method) keeping the photocurrent constant during the measurement by
complicated modulation of light intensity was necessary (Vaněček et al., 1981). In FTPS
method this condition is fulfilled automatically and also the measurement is much faster.
The method can be more sensitive and more reproducible. Due to its first publications
mainly on microcrystalline silicon people usually understand the FTPS simply as a method
used for quality analysis of this material although its potential is not limited only to
microcrystalline silicon and its use on amorphous silicon and many other non-crystalline
semiconductors was successfully proven. Many practical issues of the use of FTPS are
discussed in this chapter too.
In general Fourier-transform (F-T) represents in optical spectroscopy strong alternative to
the classical approach based on light dispersion (monochromator). One advantage (Felgett)
is that in F-T spectroscopy light of single wavelength is not isolated but only “labeled” or
“encoded” so that the measurement can be performed for all wavelengths simultaneously.
The spectral distribution of measured effect is obtained subsequently after mathematic
decoding. Second advantage (Jacquinot) is much higher limit for resolution with the same
light throughput than in monochromators. These are main reasons why F-T spectroscopy is
used for example in FTIR vibrational analysis. The FTIR measurements have standardized
procedures and usually not deep understanding of principles is necessary. Since FTPS is not
a standard application of FTIR instrument, many other issues except the main advantages
have to be considered for correct measurement and interpretation. Understanding of them is
based on some fundamental principles that are in condensed form discussed in next section.
2. Principles of Fourier-transform spectroscopy

In this paragraph we want to discuss in condensed form the principal issues important for
correct measurement. For F-T spectrometers many alternatives exist, but most common and
instructive is the F-T spectrometer based on Michelson’s interferometer, or ‘modulator’ as
depicted in Fig. 2. Incoming light from source is partly transmitted and partly reflected by
beamsplitter. Each part continues into separate delay line with mirrors at the ends. After
phase shift given by product of wavenumber1 ν and retardation Δ, i.e. mutual path
reflection on the mirrors two beams superpose at the beamsplitter again but with mutual
retardation is time dependent: Δ=u(t-t0). It follows from theory of light coherence that the
difference of the two beams. One of the mirror moves linearly with velocity u, so that
intensity of light superposition will depend on retardation and thus will change in time t as
cosine function:
I (ν , t ) = B(ν ) ( 1 + cos(2π ⋅ 2ν u(t − t0 ))) (1)
The factor B(ν) is baseline and represents compound spectrum of additional effects i.e.: lamp
radiance, transmittance and reflectance of beamsplitter and effects of other optical elements
wavenumber ν harmonic modulation in time with frequency f that depends linearly on the
in the instrument. Formula 1 can be regarded as ‘coding key’ that attributes to each
1 Used in infrared spectroscopy, unit is inverse centimeter cm-1, ν(cm-1)=107/λ(nm)
www.intechopen.com
Fourier Transform Photocurrent Spectroscopy on Non-Crystalline Semiconductors 259
wavenumber according to formula 2. The modulation is for typical conditions (λ=500-

1600nm, u=0.16cm/s) in the range from 2 kHz to 6 kHz. If some of the factor in the
experiment exhibits frequency dependence correction to that has to be done, see paragraph 3.6.
f = 2ν u (2)
When the vavenumber is ‘encoded’ into modulation frequency the light beam is ready for
measurement. Then other elements in the measurement should be introduced: optical filter
with spectral transmittance F(ν) and detector that transforms light intensity into electrical
current. Its spectral dependence can be labeled as D(ν). We make one more step and we will
of integration over ν. Then we get formula 3 where the electrical current J is a function only
also go from one discrete wavenumber to continuous spectrum so that we will add the signs
of time t. The example of such time evolution called interferogram is in Fig. 2.
∫ D(ν ) ⋅ F(ν ) ⋅ B(ν ) ⋅ cos ⎣⎡2π ⋅ 2 vu(t − t0 )⎦⎤ dν

∞
J (t ) = J 0 + (3)
−∞
It is very important that the detector is linear so that the current is a linear function of
intensity and can be therefore represented in formula 3 linearly (y=D·x). Non-linear
parasitic contributions to the modulation at higher multiples of ν (higher harmonics) !

detectors (y=D·x+D´·x2+…) would introduce higher powers of cosine and would lead to
as t’=(t-t0)·4πu ), so that we can directly write bidirectional formulae between time domain
In formula 3 we can easily recognize Fourier-transform (for precision we renormalize time
and wavenumber domain (with FT as a symbol for Fourier transform):
J (t′) = π /2 ⋅ FT ( B(ν ) ⋅ F(ν ) ⋅ D(ν )) (4)
B(ν ) ⋅ F(ν ) ⋅ D(ν ) = 2/π ⋅ FT ( J (t′)) (5)
Formula no. 5 finally gives simple instruction to calculate spectra from recorded
interferogram. However only theoretically, because the interferogram is obviously not
infinite in time and is sampled only in finite steps. The finite length has then effect on
resolution through Rayleigh criterion: Two beams can be distinguished only if they differ in
that the theoretical resolution (in cm-1) is inverse of max. retardation ΔMAX (in cm). In reality
full retardation by full wavelength. Which transformed to domain of wavenumbers means
resolution will for given max. retardation ΔMAX (ΔMAX is twice the mirror path) be ~ 2/ΔMAX,
due to the Gibbs phenomenon and often used triangular apodization2 the theoretical limit of
that can go well below 0.1cm-1. These issues are important in FTIR but not in FTPS (typically
resolution 32cm-1 and triangular apodization is used). According to Jacquinot advantage the
F-T spectrometer can have up to 100 times higher light throughput than dispersive
instrument for given resolution (Griffiths 1977). In reality the built-in sources are optimized
with low required resolution (for thin film Si resolution Δν/ν or Δλ/λ around 0.01 is
only for high resolution where the light throughput is comparable with monochromator
2In F-T spectroscopy the spectra are not corrupted by convolution with instrument function but with F-
T of envelope function that can be artificially reshaped by so called apodization so that the convolution
has better representation of sharp features, but lower resolution.
www.intechopen.com
enough) and thus lower resolution with high throughput is possible only with external light
source, see paragraph 3.3.
Fig. 2. On the left: Michelson’s interferometer. Light enters from left side, is splitted by
beamsplitter (B), reflected at fixed (mF) and movable (mM) mirrors and then exits
downwards. On the right: Example of interferogram – amplitude of signal in domain of
time/mirror position
The relation between resolution and spectral range is bidirectional and so it concerns the
finite sampling density (sampling ‘resolution’). Due to Nyquist-Shannon-Kotelnikov
theorem that says roughly that the maximal distinguishable frequency is half of the
sampling frequency, F-T spectrometer is unable to distinguish between multiples of 15798/g
cm-1, where 15798 is wavenumber of red laser and g is parameter called sample spacing3.
This constraint unfortunately requires division of the spectrum by appropriate cut-off filters
into parts from m·15798/g cm-1 to (m+1)·15798/g cm-1 that have to be measured separately.
Fortunately modern instruments have g=0.5 and allow measurement up to 31600cm-1 which
with using halogen light as a source needs no filtering and spectrum can be measured at
once. But for example measurement of FTPS of thin films silicon with g=1 and halogen
source requires adding long-pass filter with edge above 633nm (‘red glass’).
Felgett (or multiplexing) advantage is basically the fact that we can measure all wavelengths
simultaneously and thus measure much faster. It is also the reason why the FTPS can
replace CPM method (see paragraph 4.5). These advantages have one important
disadvantage. Like most measuring instruments F-T spectrometer has limited dynamic
range, i.e. the ratio between highest and lowest values measured together and with small
relative error. This is especially important in the case of steep absorption edge4 in
photocurrent measurement of semiconductors that is investigated always in logarithmic
scale. The effective dynamic range is approximately 100 but can still vary a little according
to the shape. In the case of slow slopes the dynamic range can be higher whereas in abrupt
steep edges the dynamic range can be lower. Therefore for measurement of higher dynamic
range spectra have to be measured with optical filters that reduce or eliminate strong parts
of the spectra against weaker parts. Regarding the noise we can bring one argument that
3 In FTIR the sampling is made with the help of red laser (15798cm-1) interference. Sample spacing
expresses distance of sampling points as a fraction of distance between 2 maxima of red laser
interference.
4 Due to the Gibbs phenomenon sharp steps are not perfectly represented by truncated F-T.
www.intechopen.com
also supports the Felgett advantage: Noise is proportional to the square root of spectral
range, but signal as we can see from formula 3 is proportional linearly to the spectral range,
so the signal to noise ratio in turn increases with spectral range (Griffiths 1977). Therefore it
is better for enhancing the dynamic range to use filters that are not cut-off so they don’t
totally eliminate high signal region but are ‘smooth’ and only reduce the high signal region
and maintain nonzero signal in broad region.
each wavelength is modulated by different frequency f=2νu

Summary:
-
- resolution is limited by mirror path length and aperture dimensions and can easily go
below 0.1cm-1
- signal linearity is essential condition
- optical cut-offs are necessary for combination of halogen source with sample spacing
greater than 0.5
- dynamic range is ~100 and has to be enhanced by additional optical filtering
- filtering by smooth filters is better than by cut-off filters
3. FTPS experiment
FTPS method is atractive due to its simple implementation to the research grade FTIR
(Fourier-transform infrared) spectrometer that has option to external (e.g. photoacoustic)
detector. FTIR spectrophotometers are since 1990’s widely used for optical vibrational
spectroscopy (range from 400 to 25000 cm-1). They are user-friendly compact instruments
including source, modulator and detector. Advantage is if the instrument has an availability
of sample spacing 0.5. Then, only a high-quality low-noise current preamplifier (with
voltage output), suitable optical filters and sample holder and cables are necessary for
measurement of solar cells. For layers on glass additional voltage source has to be used.
Fig. 3. On the left: ‘simplest’ FTPS setup where sample S is inside the sample area, reference
R is measured without sample and with filter F (generally different for sample and
reference). On the right: ‘richest’ FTPS setup with beam reflected out by sliding mirror M1 to
external focusing mirror M2 , two different positions for filter F0 , F1 , external source H,
external mirror M4 , rotating mirror M3 , external A/D convertor and voltage source
www.intechopen.com
Other components such as external focusing mirror, external A/D converter or external
light source can be used optionally, see Fig. 3. Special features (not shown here) are internal
motorized filter wheel and grey filter wheel. Measurement of a sample (layer on glass) in
sample area has advantage of possibility of measurement of transmittance directly by
internal reference detector. Measurement is performed as a sequence of inteferogram scans.
The time for one scan is based on mirror velocity and max. retardation. Recorded
interferograms are corrected for the phase and summed together. Signal to noise of the
calculated spectrum increases with number of scans. 1 scan at velocity 0.16cm/s and
resolution 32 cm-1 takes approx. 1 second and typically few hundreds of scans are necessary
for the most ‘difficult’ samples. The evaluation is performed according to simple logic that is
obvious from the formula 5: We compare two measurements of compound spectrum
B(ν)·F(ν)·D(ν) one from sample (indexed by 1) and one from reference (indexed by 0), see
Fig. 3 and formula 6. The sample and reference can play a role of any factor in the formula.
In FTIR the sample (and reference) plays a role of optical filter F(ν), in FTPS the sample
plays a role of detector D(ν). Baseline is for both measurement the same and its effect cancels
out, as seen in formula (6). Filters can be for both measurements generally different. We
correct mathematically afterwards their effects by dividing the signal by the transmittance
spectrum of the filter that was used for the measurement. If we don’t know the
transmittances of the filters, we have to use same filters for sample as for reference which in
general means a limitation.
FT ( J 1 (t′)) F0 (ν )
D1 (ν ) = ⋅ ⋅ D0 (ν )
FT ( J 0 (t′)) F1 (ν )
(6)
Obviously, the response of the reference detector D0(ν) has to be known too. Since in optics
we calculate with photon fluxes rather than energy fluxes D means quantum efficiency of
the photon - electron generation in the sample or detector. Absolute quantum efficiency of
sample and reference detector is difficult to find out, so we always measure it relatively and
additional procedures have to be used for absolute scaling, e.g. according to additional
optical transmittance and reflectance measurement in medium absorption region. For the
purposes of FTIR measurement, the knowledge of quantum efficiency of reference detector
is not necessary and therefore it is usually unknown and for purposes of FTPS has to be
found out. Typically the detectors are pyrodetectors, so they have flat response in energy
flux (amperes per watt). By multiplying their response by photon energy (in electronvolts)
we obtain quantum efficiency (no. of electrons per one photon). By this we obtain quantum
efficiency for modulation frequencies close to zero. For obtaining quantum efficiency at real
conditions of wavelength dependent modulation at 2-6 kHz frequency we could either do
an approximate frequency dependence correction (paragraph 3.6) or rather do a calibration
by frequency independent detector5.
3.1 Sample preparation

Generally three types of samples can be measured: 1) Layer of semiconductor on low
alkaline glass (low-cost glass will cause problems with charging of impurities) is measured
5 Commercial calibrated c-Si detectors are unfortunately strongly frequency dependent, but can be used
for low frequency calibration of e.g. thin film solar cells with low frequency dependence and these can
then be used for calibration of reference detector in FTPS.
www.intechopen.com
in coplanar configurations, i.e. two contacts on layer are evaporated by Al or NiCr or

alternatively drawn by graphite paste. For smooth layers interdigitated contacts with thin
gaps can be used, but for scattering layers spacing at least 1.5 mm is necessary (Poruba et. al
2000), details in paragraph 5.1. Voltage typically in range from 50V to 500V is applied. Light
is perpendicularly focused to sample so that it has to entirely fill the gap between contacts.
Sample should be illuminated from layer side, only for ideal smooth layers or for
comparative measurement illumination from substrate side is possible (paragraph 4.2).
Advantage of these samples is possibility of transmittance measurement (Fig. 3) at the same
time and easy and accurate interpretation of the results, details in paragraph 4.2.
Disadvantage is that these samples do not correspond to material grown on real substrates.
2) More close to real conditions e.g. in solar cell technology might be layers grown on
conductive substrates as ZnO or SnO2 coated glass. These have to be measured in sandwich
configuration when measured semiconductor layer is sandwiched between the conductive
substrate and another planar electrode deposited on top. At least one of the electrodes has to
be transparent for illumination. In this case the current flows perpendicularly to surface and
thus only very small distance is between the electrodes. Usually also band bending effects
are presented and so small voltages (up to 2V) have to be carefully chosen to compensate
such effects (Poruba et al. 2003), (Holovský et al. 2010). 3) Solar cells on the other hand
don’t require any preparation and typically are measured without any voltage applied
(Poruba et al. 2001). Application of voltage is used to simulate real working conditions of
solar cells (Bailat 2004). Monolithic multijunctions e.g. tandems or modules have to be
selectively light-biased, more information in paragraph 5.3.
3.2 Choice of the FTIR instrument

The main requirements for the FTIR instrument used for FTPS are related to sample spacing,
mirror velocity, beamsplitter, light source, optical windows and detector. As described
above, the lowest possible sample spacing of 0.5 is strongly recommended and spacing of 1
be 0.16 cm/s. FTPS measurement is in the visible and near-infrared range (0.4-2.5 μm) and
is necessity. Lowest mirror velocity at the lowest sample spacing in true linear mode6 should
so the choice of material for beamsplitter and windows is mainly quartz or sapphire,
detector should be pyrodetector (e.g. thermoelectrically cooled deuterated triglycin
sulphate), mirrors should be aluminum and light source should be halogen lamp. Choice of
the halogen lamp is especially advantageous due to its intensity increase towards regions of
low absorptance of semiconductors. Usually beamsplitter, source and detector are
exchangeable so that the spectrophotometer can still be used for FTIR infrared range.
3.3 External focusing mirror, external A/D converter and external light source
The external A/D converters supplied by the same manufacturers can have better
performance that the built-in A/D converters in terms of number of bits, possibility of gain
ranging and signal to noise ratio that can be higher significantly. The use of external A/D
converter can be for some instruments impossible for measurement in the sample area and
has to be combined only with external focusing mirror. External focusing mirror is more
flexible in terms of size of samples, additional light biasing etc. see Fig. 4. Also the beam can
6Sometimes low speed modes are not true linear scans, but in principle fast step-scan modes (for step-
scan mode see paragraph 6)
www.intechopen.com
be focused to smaller spot to get higher intensity. Typical focal length of internal mirror is
~15 cm with the focused beam spot size 2x8 mm2.
Fig. 4. On the left: FTPS setup with the use of external focusing mirror for measurement of
maps of density of defects of PV modules. For selecting one spot shadowing stripes and bias
light was used. On the right: sample holder for layers on glass and solar cells that have both
contacts on the same side of glass substrate
The external light source can significantly increase intensity of light and also enhance its
intensity in blue region. Internal halogen source has typically power of 25W, dimensions
of filament approximately 2x8 mm2, is intentionally kept at lower voltage and is coupled
to the bench by mirror with high focal length for better resolution (~15cm). The photon
flux at focus is around 2x1017cm-2s-1. If we use halogen lamp with 75W power with
filament of comparably equal size7 and with combination of a mirror with lower focal
length (~10cm) we get approximately 7 times higher intensity. Then the resolution is no
more guaranteed by automatically controlled internal aperture and has to be checked
focal distance and νMAX is maximal wavenumber in the spectrum. For details see (Griffiths
according to formula 6 where d is largest dimension of light source (or entrance slit), f is
& de Haseth 1986).
Δν = ( d / 2 f )2ν MAX (6)
For described external source and for νMAX=20000cm-1 resolution is Δν~32 cm-1 that is what
is normally used. For maximizing intensity going to resolution Δν ~100cm-1 is possible and
intensity grows 4 times. Conservative estimate of potential intensity increase of external
source is 20 times. Advantage of external source is that the FTIR instrument is more
thermally stable. Disadvantage is that the bench alignment can be done only for either
source, see paragraph 3.7.
3.4 Current preamplifier, voltage source and sample holder

The choice of these components will strongly affect the signal to noise ratio and thus the
sensitivity and speed of the measurement. For the voltage source there is no necessity to buy
expensive sources. For low voltages simple battery source is better than expensive
programmable sources. For high voltages (up to 500V) some high voltage source has to be
used. The important issue is the scheme of serial connection of the voltage source, sample
7 12V halogen lamp nominally 100W designed for vehicles from manufacturer NARVA
www.intechopen.com
and current preamplifier in terms of electromagnetic noise. The use of BNC 50Ω coaxial
cables has proven to be much better compared to connection by twisted pair cables. The
connection in series can be easily realized by the design of a sample holder where both outer
wirings are connected to metal frame and each inner wire contacts one pole of the sample,
see Fig. 4. Grounding of the metal frame usually reduces noise level too. In the case of
measurement without voltage source (solar cells) shorting plug is used at one of the BNC
connector.
Most important is the choice of current preamplifier and state-of-the art instruments are
recommended. Important parameters are: noise level, frequency cut-offs, input impedances
and dynamic reserve. The noise level of preamplifier at high amplification is more critical
for highly resistive samples of layers on glass (1-100GΩ). For amplification 107 V/A
broadband noise should not be above 1 picoamper. On the other hand for measurement of
solar cells and modules with generally low resisitivity (10-100Ω) dynamic reserve8 is critical.
The frequency cut-offs are inequitable in all preamplifiers at high amplifications and cause
frequency dependence of signal (see paragraph 3.6) and prevent from using high
amplifications. Modulation frequency range is given by formula 2 and for our spectral range
and range of mirror velocities (0.16– 0.47cm/s) frequency range in low signal (=high
amplification) region is 1kHz – 10kHz. High amplifications cannot be used also due to
increasing input impedance. Two different preamplifiers one with lower noise level and one
with higher dynamic reserve were tested on solar cell and layers on glass and it was
impossible to make any unambiguous preference, see Fig. 5.
0
a-Si layer
(not measureable)
10
K 428
number of scans for S/N=1 (-)
3 SR 570 -1
10 10
μc-Si layer
signal (a.u.)
K 428 -2
2 SR 570 ln 10
10 SR 570 hbw
-3
10
1 lamp spectrum
10 -4
10 long-pass
a-Si solar cell short-pass
K 428 -5 signal w. long-pass
0 SR 570 ln 10 signal w. short-pass
10 SR 570 ln
final spectrum
-6
10
5
10 10
6
10
7
10
8
10
9
10000 20000 30000
-1
amplification (V/A) wavenumber (cm )
Fig. 5. Left: comparison of number of scans needed for signal to noise ratio S/N=1 for
preamplifiers Standford Research 570 and KEITHLEY 428, data are based on real
measurements of layers of amorphous (a-Si) and microcrystalline (μc-Si) silicon on glass and
for a-Si solar cell. (‘ln’ = low noise mode, ‘hbw’= high bandwidth mode).
Right: FTPS spectra of 250nm thick amorphous silicon measured only with 2 filters (long-
pass and short-pass), transmittances of the filters and source intensity spectrum included
8 Dynamic reserve express how larger can be broadband noise than the signal before overloading –
depends on setting and location of frequency band filters
www.intechopen.com
3.5 Optical filters

There are at least four reasons for using filters in FTPS: 1) Measurement with sample
spacing greater than 0.5 requires using cut-off filters for both sample and reference (for
details see paragraph 2). 2) Filters are necessary for improving dynamic range too. One
possibility is to use cut-off filters, such as 2mm polished Si wafer for measurement of
microcrystalline silicon (Poruba et al. 2002) or even set of filters for amorphous (Melskens et
al. 2007) to eliminate more and more from high absorption region and to get to lower and
lower absorptances. Alternative is to use less filters or even only one that only reduces light
intensity at high absorption region. If the minimum of signal S that we want to measure is
SMIN then transmittance of ideal filter would be 1 in region where S<100·SMIN and 100·SMIN/S
where S>100·SMIN then we would measure whole curve with dynamic range only 100. For
example see spectrum of long-pass filter in Fig. 5. 3) Third reason for filtering is that F-T
spectroscopy doesn’t reproduce spectrum where sharp step occur (absorption drop on the
bandgap in amorphous Si at 1.75eV) and better results are obtained when using the filter
that reduces the signal before the abrupt drop comes, see short-pass filter in Fig. 5. 4) Fourth
reason for filtering is when it is necessary to keep low signal conditions. For that color or
neutral density filters can be used (see section 4.5). The use of optical filters complicates
strongly the automation of the FTPS because generally for different materials different filters
have to be used.
3.6 Frequency dependence correction

Here we come to the main issue of FTPS. The frequency of modulation ranges from 1kHz to
10kHz and is wavenumber dependent as seen from formula 2. If performance of any part of
experiment (sample, reference, preamplifier) is frequency dependent a correction for this
dependence has to be done in order to obtain results same as measured at modulation
frequency close to zero. Because sample and reference has different frequency dependence,
wavenumber ν and certain frequency f can depend on a frequency either according to

formula 5 can be used only after such correction. We can suppose that the signal S at certain
formula 7 or formula 8, that means that the function Φ describing frequency dependence
can be wavenumber dependent and so has wavenumber as parameter.
S(ν , f (ν )) = S(ν ,0) ⋅ Φ( f (ν )) (7)
S(ν , f (ν )) = S(ν ,0) ⋅ Φ( f (ν ),ν ) (8)
Function Φ is unknown and can be investigated only experimentally for each sample. In
FTPS the way of investigating function Φ is to make additional measurements at different
0.16 cm/s only to higher ones, function Φ can be found only around some central frequency,
velocities and observe the signal change with velocity. But because we can go from velocity
measured signal spectrum S(ν,f(ν)) to new spectrum S(ν, 5kHz) that corresponds to
say 5kHz. Based on formulae 7 or 8 we can calculate signal at this frequency and so get from
spectrum as if measured at one wavelength independent frequency 5kHz. Obviously we
and from it follows that S(ν, 5kHz) = S(ν, 0)· constant and so we can get correct, but relative
can’t get further to zero frequency. For some samples we can suppose that formula 7 holds
spectrum. This explains why even for ‘good’ frequency dependence (formula 7) we can’t
measure absolutely. For high preamplification, major part of frequency dependence is
caused by preamplifier for which formula 7 holds. In practice it has only sense to assume
www.intechopen.com
wavelength independent frequency dependence obeying formula 7. As we will see in

paragraph 4.3 and 4.6 wavelength dependence as in formula 8 can occur only for
photothermal ionization effects or for samples where carrier diffusion dominates. For
correction we measure spectrum at three velocities and then in set of reference wavenumber
points we quantify and fit (typically by exponential) the relative decrease of signal with
frequency (calculated for each point by formula 2). In next step the average of the parameter
of relative decrease is made and finally one of the three curves is corrected (virtually to
5kHz). Only the curve that we want to correct (usually lowest velocity) has to be measured
finely with enough scans but for the two other much less scans are necessary. Frequency
dependence can be sometimes significant and can lead to shift of one end of the spectrum
compared to opposite end as high as by 50%, so it is clear that the accuracy is strongly
influenced by frequency dependence and so the error is usually at least few percents.
3.7 Bench alignment

For successful measurement correct bench alignment is necessary. Once properly done it
does not have to be done for months. Alignment is done in commercial instruments
automatically and instrument remembers alignment for each beamsplitter. It was found that
the shape of a baseline spectrum depends on alignment and if bench is properly aligned the
baseline (roughly spectrum of source) for quartz beamsplitter should monotonously
decrease towards high wavenumbers. To do so and to enhance intensity in visible range,
alignment in two steps first without filter and then with short-pass glass is necessary.
When using external source, alignement should be done as follows: 1) Make proper
alignment for internal source. 2) Mark on a screen that is placed into the focus the precise
position of the focal point. 3) Manually align the mirror (M4 in Fig. 3) into axis of Michelson
modulator. 4) Adjust the external source to obtain the same position and dimension of the
focal point on the screen. 5) Make new alignment with external source. Then the bench will
not be aligned for internal source.
Summary:
- research grade FTIR and low noise preamplifier are only necessary large investments
for setting up FTPS method
- high dynamic range or low noise of the preamplifier is more important for
measurement of solar cells or layers on glass, respectively
- proper choice of long-pass optical filters is main know-how in FTPS but prevents
simple automation of the method
- coaxial cables and grounded sample holder is necessary for low noise signal
- frequency dependence correction is necessary for FTPS on thin film silicon
- with external source the high troughput advantage allows 20 times higher intensity
- proper two-step bench alignement is necessary for accurate measurements
4. Interpretation of FTPS
In the section 1 and Fig 1 we outlined the aim of optical spectroscopy in disordered
semiconductors, e.g. amorphous silicon (a-Si) that is to get information about defect density
and disorder. In section 3 we covered most of the technical issues regarding the successful
measurement of a FTPS spectrum. Nice thing would be to show in this part a formula for
FTPS signal as a function of defects and disorder. Unfortunately situation is much more
www.intechopen.com
complex and to puzzle out its complexity a diagram (Fig. 6) is sketched where all the arrows
represent one specific relation that will be discussed hereafter. In spite of the complexity of
the diagram, the individual effects themselves are quite straightforward and well known in
the field of semiconductors.
Fig. 6. Diagram shows most of the possible effects playing between defects density and
disorder as input that we want to know and FTPS signal as output that we measure
4.1 From defects to absorption coefficient

Defects and disorder (their origin base on material microstructure is beyond the scope of
this article) are represented as electronic states in material and characterized by spatial
density and energy. In non-crystalline semiconductors these are only two parameters fully
characterizing the electronic structure of material, because due to high electron scattering
on irregularities no momentum quantum number and its conservation exist. So the
conditions for optical excitation between occupied and unoccupied state are only energy
conservation and spatial overlap of the initial and final state. Mathematically the absorption
coefficient can be expressed as convolution of spatial density of occupied and unoccupied
states by formula 9, where W(ħw) is matrix element the form of which is not consensual9.
α ( ¥ω ) = W ( ¥ω )∫ NV (E) NC (E + ¥ω ) dE (9)
This frequently used formula holds for undoped material under low light conditions (Fermi
level EF is well defined and is close to the middle of the gap) and also low temperature limit
is considered (states above EF are unoccupied, states below EF are all occupied).
9 W(E)=const, e.g. in (Davis 1970), W(E)= const·E , e.g. in (Jackson et al. 1985), W(E)=const/E , e.g. in
(Vaněček et al. 1984)
www.intechopen.com
material defect typ. good scaling method reference

energy value constant (cm-2)
<1 cm-1 2.4-5x1016 CPM
a-Si 1.2 eV (Wyrsch et al. 1991)
<2 cm-1 1.2-2.5x1016 PDS
μc-Si
0.8 eV <0.1 cm-1 ~ 1.7x1017 CPM (Vaněček et al. 2000)
0.7 eV <0.1 cm-1 3.6x1016 PDS (Klein et al. 2007)
Table 1. Defect absorption energies, typical values for good quality materials and factors for
recalculation into defect density (cm-3) for values obtained from constant photocurrent
method (CPM) and photothermal deflection spectroscopy (PDS)
In the band structure of amorphous materials the Gaussian humps in the middle represent
defect states or unsaturated bonds and the tilt of the edges (band tails) represent mainly
angular disorder of chemical bonds. Two vertical lines are called mobility edges, separate
delocalized extended states and localized tail states and define electrical gap of the material.
Y-axis in the Fig. 1 is in logarithm scale and only strong transitions are identified, i.e. only
transition involving always at least one state from valence or conduction band. Transitions
C1, C2, B1 and B2 project into region of subbandgap absorption on absorption coefficient.
There exist more possibilities of defect density evaluation from this region (Wyrsch et al.
1991). Often used is just taking the value of absorption coefficient at some physically defined
energy and multiplying it by scaling constant, see Table 1. Transition B2 is much stronger
than B1 due to its slower slope. Exponential part of absorption coefficient is according to
(Shah 2010) function of both slopes of B1 and B2, but usually is attributed only to the slope
of valence band tail. For its slope the Urbach energy E0 is defined by formula 10 (Street
1991). For good quality non-crystalline silicon material E0 is never higher than 50 meV.
α ( ¥ω ) = c ⋅ exp( ¥ω / E0 ) (10)
4.2 From structure to optical absorptance

For evaluation of optical absorptance, as seen from Fig. 6 not only absorption coefficient
discussed in paragraph 4.1, but also refraction index, material and interface morphology
and obviously macroscopic sample structure have to be known. Exact knowledge of all
parameters and accurate calculation of absorptance for arbitrary structures is insoluble
problem and therefore simple well defined samples should be measured if absolute results
and not only relative comparison is desired. Simplest example is smooth weakly absorbing
layer on nonabsorbing semiinfinite substrate. For the purpose of absorption coefficient
measurement approximate formulae no 11, 12 and 13 presented in (Ritter & Weiser 1986)
can be used:
(1 − R1 ) ⎡⎣ eα d − 1 + R2 (1 − e −α d )⎤⎦
A≅
eα d + R1 R2 e −α d − 2 R1 R2 cos(2γ + δ 1 + δ 2 )
(11)
(1 − R1 )(1 − R2 )
T≅ αd
+ R1 R2 e − 2 R1 R2 cos(2γ + δ 1 + δ 2 )
−α d
(12)
e
www.intechopen.com
270
( { })
Fourier Transforms - New Analytical Approaches and FTIR Strategies
α= ln 0.5 (1 − R2 )(1 + A / T ) + (1 − R2 )2 (1 + A / T )2 + 4 R2
1
(13)
d
Meaning of the symbols are: r01= √R1 · exp(iδ1) , r01= √R1 · exp(iδ1) , where r01, r12 are Fresnell
coefficients of perpendicular reflectance on first and second interface, A, T, α, d are
absorptance, transmittance, absorption coefficient and thickness respectively. Formula 13 is
directly derived from the two previous and contains no cos function and so we can profit
from advantage of absence of interference patterns in ratio A/T. Therefore it is advantageous
for thin films to measure absorptance and transmittance spectra simultaneously. This
formula however requires absolute measurement of A and T, that is not always possible and
therefore scaling according to additional optical measurement is necessary. Also knowledge
of R2 is necessary. Mathematic fit of e.g. Cauchy formula parameters of refraction index or
assumption of some typical spectrum of refraction index (Vaněček 1995) is possible for
calculation of R2. The least sophisticated, but still used is to approximate the formula 11 into
simple form 14 or 15. Both formulae do not take interference pattern into account and
should therefore be applied to smoothened curve of A. Formula 14 is very often used and
indeed by analyzing its error we can find that neglecting the term with cos in formula 11 and
curve with averaged interference will result in error in α of -10% in low absorption region
further neglecting some other terms on a way to formula 14 and attributing the result to
for amorphous silicon on glass. This formula however still requires knowledge of R1 and
absolutely scaled A. Thus often R1 is assumed to be constant and then formula 15 is used.
Then only saturation value AREL,MAX at high absorption where 1-e-αd→1 and thickness d have to
be known. Then resulting α is correct except the region above and close to the maximum point.
A ≅ (1 − R1 )(1 − e −α d ) (14)
α ≅ − ln ( 1 − AREL / AREL , MAX ) d (15)
So far theory for smooth, non scattering layer on glass was used. For measurement of
samples with rough interfaces or even with bulk scattering as for example microcrystalline
silicon, more complicated theory has to be used (Poruba et al. 2000). Sometimes effect of
scattering can be well masked and for example accurate comparison between measurements
of total and specular transmittance has to be done (Vaněček et al. 1998). Even more
complicated is the evaluation of absorptance in solar cells where the transmittance of
transparent conductive oxide layers (TCO) comes into play, but even here correction exists
(Python 2009). Otherwise FTPS (and photocurrent in general) on real solar cells is useful
mainly for comparative measurements of cells with same optical structure (TCO, roughness,
back reflectors).
Presented formulae are for total absorptance, but characterizing light absorption by its total
value is not always precise because absorption is generally not homogeneous due to many
effects: 1) due to exponential decrease of intensity in absorbing material, 2) due to standing
waves in thin films 3) due to inhomogeneity of material and thus inhomogeneity of
absorption coefficient itself, 4) due to inhomogeneously distributed defects, for example
close to surfaces of grain boundaries. Last two cases are not accounted in above presented
calculations and so the measurement in A/T mode is not legitimate and has an effect of non-
vanishing interference patterns (Vaněček et al. 1995). Moreover this effect depends on
www.intechopen.com
position of highly absorbing sub-layer in the whole layer and if it is on the back side with
respect to illumination, the non-vanishing interferences are not observed, inhomogeneity is
masked but generally higher absorptance is observed, see Fig. 9.
4.3 From absorptance to excess carrier generation

Until now the analysis was based purely on optics. Now we want to look closely whether all
absorbed photons create free charge carriers that can eventually contribute to photocurrent.
In undoped amorphous silicon for example mobility of holes is much smaller and so only
the transitions terminating above the conduction mobility edge can in principle contribute to
photocurrent. But at room temperature the thermal excitation may release the carriers from
small depth below the mobility edge above the edge and so enable them to be electrically
collected. This two-step carrier generation is called photothermal ionization. The thermal
ionization however has to be fast enough to be registered by phase correlated detection,
because FTPS is (like other lock-in methods) based on measurement of modulated response
that is correlated with excitation pulses and carriers excited too late are not accounted. In
this point we get to the issue of frequency dependence dependent on wavelength (section
3.6, formula 8). Based on theory developed for Modulated Photocurrent Method (Abe et al.
mobility edge F down to energies ΔE below original mobility edge, analogical to Fermi-
1988), we can describe the effect of thermal excitation as a dispersion of the conduction
Dirac function as in formula 16 where symbols ω, f0, kB and T have meaning of modulation
frequency, attempt-to-escape frequency (in a range of 1012 Hz), Boltzmann constant and
temperature, respectively.
+
⎡ ⎛ 2 ΔE ⎞ ⎤⎥
−1
⎛ω⎞
F ( ΔE ) = ⎢ 1 ⎜⎜ f 0 ⎟⎟ exp ⎜⎝ kBT ⎟⎠ ⎥
2
⎢ ⎝ ⎠
(16)
⎣ ⎦
In the work of Abe the electronic states below mobility edge having overlap with F
contribute as frequency dependent photocurrent signal. But not only frequency dependent
signals, but also frequency independent signals from displacement current from the same
electronic state are considered in his theory. So, depending on parameters in formula 16 and
depending on the ratio of displacement versus real currents, different transitions will have
different effect in photocurrent. In his work mainly transitions C1 are studied at frequencies
from 10Hz to 10kHz with strongest frequency dependence around 500Hz. Finally for
studying the same effect directly for FTPS a comparison of photocurrent measurement by
FTPS, CPM at 13Hz and DBP10 method for wide range of frequencies was done (Holovský et
al. 2008), see Fig. 8. From the comparison we see the effects of frequency dependence and
displacement currents in defect region and beginning of Urbach edge in DBP at different
frequencies. We also see difference between FTPS and CPM at defect reference energy 1.2 eV
well below factor 2. According to previous papers (Wyrsch et al. 1991), CPM itself gives
values of defect absorptance approximately twice lower than optical absorption. Similarly,
our measurements systematically show difference in defect absorptance between FTPS and
Photothermal Deflection Spectroscopy by factor around 2 at energy 1.2eV for amorphous
silicon. It means that photocurrent generation is calculated according to same optical
10 DBP stands for Dual Beam Photocurrent, method is explained in paragraph 4.5.
www.intechopen.com
absorption coefficient as in paragraph 4.1 except the region of defect absorption where the
photocurrent generation in amorphous silicon can be approximately twice lower. This
principal discrepancy of different absorption coefficient for light propagation and for
differences are in region of low homogeneous absorption ( 1-exp(-αd) ≈αd ) we make no

photocurrent generation shall not complicate our interpretation, because since the
error by assuming light propagation according to the same absorption coefficient as

corresponds to photocurrent generation. Only we have to choose correct recalculation
constant in Table 1 (for FTPS same as for CPM method).
10
2 a-Si:H layer (2μm)
4 CPM 13Hz
DBP 3.5Hz 6.2nA 10
CPM 13Hz bias light
DBP 13Hz 6.2nA
10
1 FTPS
DBP 69Hz 6.2nA
alphaA/T (cm )
alphaA/T (cm )
-1
-1
DBP 380Hz 6.2nA FTPS bias light

2
DBP 2kHz 6.2nA 10
0
10 FTPS bias light
-1 0
10 10
higher DBP a-Si:H layer (2μm)
frequencies
-2
10 -2
10
0.8 1.0 1.2 1.4 1.6 1.0 1.2 1.4 1.6 1.8 2.0
photon energy (eV) photon energy (eV)
Fig. 8. Left: Dual Beam Photocurrent (DBP) spectra on 2μm amorphous Si on glass converge
at higher frequencies to FTPS under bias light, at 380Hz and 2kHz DBP show anomalous
deviation due to displacement currents. Right: Similar effect of light bias on FTPS and CPM
methods and their comparison on the same sample
4.4 From structure to mobility lifetime distribution

When the excess carriers are generated, their efficient conduction to the collection electrodes
depends on ability of the sample to conduct electrical current. As seen in Fig. 6 it depends
mainly on microscopic morphology, but also presence of surfaces and so on macroscopic
geometry. And also on excess carrier distribution itself, but it will be subject of next
paragraph. Photocurrent in semiconductor, i.e. difference between electrical current in dark
carrier mobility μ, see equation 17. Excess carrier density is result of competition between
and under illumination is in simplest case given by excess carrier density n multiplied by
photogeneration G described in previous paragraphs and carrier recombination that

depends in simplest approximation linearly on excess carrier density. Such simple equation
has steady-state solution saying that excess carrier density is a product of photogeneration G
jph = eE ⋅ μ ⋅ n = eE ⋅ G ⋅ μ ⋅ τ (17)
and value τ called lifetime. Lifetime is a measure of recombination that in non-crystalline
Carrier mobility μ on the other hand depends on material disorder. So there are basically
semiconductors is dominated by Shockley-Read-Hall recombination through active defects.
homogeneous samples usually also μ is homogeneous, but near surfaces τ is lower due to
two material parameters that can influence conduction of generated carriers. In
www.intechopen.com
surface defects. And of course in heterogeneous material both μ and τ are generally
inhomogeneous.
4.5 Carrier density and lifetime

In last paragraph we defined lifetime as a value which after multiplication with generation
gives excess carrier density. Now we will discuss the fact that lifetime is not a constant but a
function of the excess carrier density. We discuss this phenomenon separately, because it
makes the photocurrent measurement complicated. That is because activity of
recombination centers is approximately given by positions of quasi-Fermi levels that depend
on excess carrier density. As a consequence, conductivity on large scale is not linear with
illumination, but rather sub-linear. Due to this phenomenon it is not possible to measure
photocurrent for different monochromatic light of the same intensity because with different
wavelength we could get photogeneration changes over many decades and so the lifetime
would not be constant. On a small scale on the other hand we still assume linearity so that
applying small harmonic light modulation will result in harmonic modulation of
photocurrent. But for the non-linearity on the large scales there are at least three methods
that solve this problem: 1) Dual Beam Photocurrent (DBP) method applies large DC light
with much higher intensity than the modulated monochromatic light so that lifetime is fixed
by the level on DC intensity. Drawback of this method is that, as seen in Fig. 8 the spectra
still depend on level of illumination. That is due to violating the low light condition defined
on the beginning in paragraph 4.1. We assume that for monochromatic light this condition is
still satisfied. 2) Constant Photocurrent Method (CPM), (Vaněček 1981) adjusts the intensity
of modulated monochromatic light in order to keep photocurrent constant. It will for the
constant mobility mean constant excess carriers density and so constant lifetime11. 3) FTPS
method can be regarded as combination of both CPM and DBP, because as all the
wavelengths are measured simultaneously, the level of modulated photocurrent does not
change and even the illumination does not change. The only question is the low light
condition that would not be normally satisfied. Fortunately this condition is critical in
subbandgap region only where the long-pass optical filter eliminating majority of light has
to be used (paragraph 3.5).
4.6 From distributions to electrical current

We already discussed all physical properties that play role in photoconductivity: absorption
coefficient, amount of absorbed light, generated excess carriers, mobility lifetime product.
We already discussed their properties and mutual dependencies. In principle none of these
physical values are constant and can be distributed non-uniformly. In this paragraph we
should put together these distributions to get the distribution of photocurrent in our sample.
So far we have not accounted for time in our thinking because all the processes we have
discussed can be regarded as infinitely fast. Carrier conduction through sample is however
not always so fast with respect to modulation frequency in FTPS. Treatment of such
situation accurately is complicated. In simplified case of uniform doping and uniform
18 and continuity equation 19, where n , j , G , τ , E are excess carrier concentration, current
temperature without taking into account Poisson equation we solve drift-diffusion equation
11 This implication may not hold only if both electrons and holes play significant role and total
photocurrent is combination of both. This is not a case in non-crystalline silicon.
www.intechopen.com
density, generation rate, lifetime and electric field respectively, other symbols have their
usual meaning. Time enters the problem by time dependent generation G(t). Solving such
equations is not realistically possible and so conditions for any simplification are desirable:
1) time modulation is slow 2) lifetime is uniform in bulk, 3) absorption coefficient is uniform
in bulk 4) absorption is low so that (except standing waves) the light intensity is uniform,
5) samples are thick and so without standing waves, 6) there is no additional absorption due
to surface defects 7) there is no additional recombination due to surfaces.
j = μ ⋅ ( eE ⋅ n + kBT ⋅ ∇n ) (18)
∂n 1
= ∇ ⋅ j + G(t ) −
τ
n
∂t e
(19)
a. If all 7 conditions are satisfied, solution of equations 18 and 19 leads to simple equation
17 and total photocurrent density is equivalent to total generation and so equivalent to
total optical absorptance and according to paragraph 4.2 we can use A/T mode and
formula 13. The same will practically work even if samples are thin and condition 5 is
not satisfied.
b. If condition 7 is not satisfied and neither absorption is low (condition 4) we will get the
situation that was for electric field applied perpendicularly to illumination described by
theory of DeVore (DeVore 1956). It is the effect of signal loss at high absorption region
when carriers are generated close to surface. Same theory however describes well also
situation in high absorption for thin samples (when neither conditions 5 is satisfied) see
Fig 9.
c. If the conditions 7 and 5 are not satisfied, and absorption is low or moderate (and not
high as in previous case) the effect of standing waves with maxima close or far from
surface will affect modulation of interference maxima in photocurrent. In this case the
A/T mode may give some non-vanishing interference maxima in whole spectrum
depending on the direction of illumination with respect to defective surface, see Fig. 9.
Instead of A/T approach (formula 13) averaging of interference maxima and formula 14
should be rather used. If instead of condition 7 condition 6 is not satisfied then the
effects on non-vanishing inreferences in A/T mode are observed only in low absorption
regions where additional absorptance is comparable with bulk absorptance (Vaněček et
al. 1995). Arguing that these two effects compensate is not very safe, because they affect
enough (1 μm or more) so that the surface areas with low mobility or with additional
different regions, see Fig. 9. In these cases the sample thickness should be chosen big
absorption have negligible effects compared to bulk absorptance and bulk

photoconductivity.
d. If one of the conditions 2 or 3 are not satisfied, for example in microcrystalline silicon,
situation is overcomplicated and exotic effects can be studied e.g. by comparison of
coplanar and sandwich arrangement (Unold et al. 2000) or by light biased CPM (Siebke
et al. 1998).
τ << f -1 (lifetime much shorter than time period of modulation) and then G(t)≈G and
e. Concerning the 1st condition of slow modulation, we assume that it is satisfied when
thus ∂n/∂t≈0 in equation 18. For FTPS f--1=0.2ms and so for non-crystalline silicon where
τ =1-10μs condition is well satisfied whereas for crystalline silicon wafers where
www.intechopen.com
τ =0.1-10ms not and at least in combination with high or moderate absorption

(condition 4) this leads to strong frequency dependence, and so for example using FTPS
for photoelectric measurement on c-Si wafers is problematic, see Fig. 9.
absorption coefficient (cm )
-1
10
5 back contacted c-Si solar cell
100
FTPS signal (a.u.)

4
10 80
3
10 60
2
10 PDS glass side
40
0.16 cm/s (2kHz)
PDS film side
1 20 0.32 cm/s (4kHz)
10 FTPS film side
FTPS glass side 0.48 cm/s (6kHz)
0 0
10 400 600 800 1000 1200
1.0 1.5 2.0 2.5 3.0
photon energy (eV) wavelength (nm)
Fig. 9. Left: Result of A/T of FTPS and Photothermal deflection specroscopy (PDS) for a
350nm thick a-Si layer on glass with defective surface layer. We see non-vanishing
interferences in film side measurement at low absorption in PDS and in whole range in
FTPS. Deviation of PDS and FTPS in high absorption region is partly given by surface
recombination. Right: Effect of high frequency modulation for c-Si solar cell observed by use
of different mirror velocities
Summary:
- classical absorption coefficient in disordered materials results from convolution of
density of states in low temperature limit and under dark conditions
- for optically homogeneous smooth layer A/T ratio has no interference maxima
- generation of conductive excess carriers follows optical absorption coefficient except the
region far from mobility edge below the gap where its contribution is frequency
dependent
- FTPS gives in defect absorption approximately twice lower value
- photocurrent depends not only on optical absorptance, but also on distribution of
mobility lifetime product
- thus only for electrically (mobility, lifetime) and optically (absorption coefficient)
homogeneous samples photocurrent is equivalent to optical absorptance and also A/T
mode can be used
- to reduce inhomogeneity effects induced by surfaces, thicker samples (1 micrometer or
more) should be used
- even for homogeneous samples measured photocurrent still depends on modulation
frequency versus sample thickness and so FTPS measurement of thick samples with
lifetime well above 0.2ms (c-Si wafers) can be problematic
- moreover lifetime itself is generally dependent on excess carrier density and thus
‘constant photocurrent’ methods like CPM or FTPS with long-pass filter are used.
www.intechopen.com
5. Review of FTPS applications

Two papers give broad overview about variety of use of o FTPS: (Vaněček & Poruba 2007),
(Poruba et al. 2008). In the following we will discuss separately in detail some scientific and
also industrial applications.
5.1 Microcrystalline silicon

In the paragraph 4.5 we discussed the capability of use of FTPS for amorphous silicon due to
its specific properties. Microcrystalline silicon is heterogeneous material with significant
Probably the first impulse was relative simplicity and good sensitivity especially on μc-Si
amorphous fraction and therefore basically same arguments can be used in case of FTPS too.
that has advantageous shape of absorptance and with using halogen lamp as a source only
(Poruba et al. 2001) on μc-Si p-i-n solar cells, later (Vaněček et al. 2002) on μc-Si layers on
one additional measurement with silicon filter is needed. The first reports of FTPS was
glass and later (Poruba et al. 2003) also on solar modules and layers on ZnO coated glass. As
interpretation of FTPS spectra Urbach slope as a measure of disorder and absorption
coefficient at energy 0.8eV as a measure of defect concentration (see Table 1) can be used.
The effect of surface or bulk scattering in both p-i-n cell and layer on glass can be large and
has to be corrected, see Fig.10. For layers on glass the effect depends strongly on spacing of
known surface roughness (Poruba et al. 2000) is necessary. Method of evaluation of μc-Si
the electrodes and minimal spacing 1.5 mm is recommended, then correction based on
solar cells avoiding the effect of ZnO was well developed by Python (Python 2009).
Absolute scaling can be in the case of smooth layers on glass made by approach developed
for ‘absolute CPM’ (Vaněček et al. 1995). Without knowledge of thickness or for solar cells
the approximate scaling according to crystalline silicon can be used: value at 1.35 eV is
crystalline fraction (Python 2009) or is scaled to value 245cm-1*Φc , where Φc is crystallinity12.

either scaled directly to value of c-Si 245cm-1 and we can call it normalization back to
For strong effect of scattering scaling at 1.2eV to 25cm-1 can be approximately done because
the factor of enhancement due to light scattering changes only a little between 0.8eV and
1.2eV (Poruba et al. 2008). Broad study of microcrystalline silicon by FTPS was made at
Université de Neuchâtel: (Bailat 2004), (Sculati-Meillaud 2006), (Python 2009) and
correlation between FTPS and intragrain or grain boundary defects and solar cell
deterioration was well verified. Microcracks in solar cells as another type of defects in solar
cells are however not visible by FTPS method (Python et al. 2010).
5.2 Application of FTPS in industry at Oerlikon Solar

The power output of solar modules depends on several PECVD layers and also on several
manufacturing steps before and after the PECVD deposition. To improve the module
efficiency each layer and the interface between them has to be optimized. The optimization
is simplified if a parameter search for each layer can be done and evaluated independently.
A single layer is much faster to coat and in addition the evaluation is not affected by
variations in the process steps required to manufacture a cell. The FTPS measurement is a
12 In region around 1.35 eV  scales approximately linearly with crystallinity, however more accurate is
effective medium theory (Vaněček et al. 1998) and according to (Siebke et al. 1998) scaling in broad
range of crystallinity is rather exponential
www.intechopen.com
relatively fast method to evaluate the defect density, which is a necessary requirement for a
high quality solar cell layer.
The evaluation approach depends on the layer quality. On a coarse level poor depositions
are always identified when the absorption at 0.8eV is very high e.g. above 1.0cm-1. For fine
tuning of already good quality layers with absorption values in the range of 0.05cm-1,
a comparison of only the same deposition setup and sample material is possible. Changes in
the substrate e.g. with additional SnO2 layer or different glass type leads to a change in the
absorption value. This is true even if two similar glass types are compared: Schott AF32eco
and Schott AF45 are coated with mc-Si in the same run, however, the FTPS data in Figure 10
shows a higher absorption for AF45 glass compared to AF32 type.
1
apparent α
absorption coefficient (cm )
AF32eco
-1
4
10
c
3 d AF45
10
a b so rp tio n @ 0 .8 e V
e 0.1
true α
2
10 f
g
1
10 h a
0 b
10 0.01
-1
10
-2
10 L100723_7e
0.001
0.5 1.0 1.5 2.0
50% 60% 70% 80% 90%
photon energy (eV)
Raman Crystallinity
Fig. 10. Left: True absorption coefficient (a,b) and apparent absorption coefficient of
6 different materials (c-h) from different labs that all exhibit similar deviation due to the
effect of scattering. Right: Effect of substrate on defect absorption for layers made in a single
deposition as an example of using FTPS for module optimization in industry at Oerlikon
Solar
Deposition on different substrate materials can have slightly increased absorption while still
giving improved cell results. An explanation for this, is that the FTPS measurement is
currently measured on AF32 glass samples placed on top off the module glass substrate. The
module glass is of a different quality selected for the module requirements. The growth of
the layer is affected by the different substrate surfaces. So a deposition parameter leading to
a perfect layer on the Schott glass (for FTPS measurement) might lead to a non optimal layer
on the large size module glass.
FTPS on amorphous silicon is slightly more difficult than for μc-Si, because it requires
5.3 Amorphous silicon
different optical filters than just crystalline silicon, see Fig.5. Moreover Photothermal
μc-Si is usually enough sensitive for a-Si. The first published results of FTPS on a-Si
Deflection Spectroscopy (Jackson & Amer 1982) that is not enough sensitive for good quality
appeared later and originated mainly from University of Delft, e.g. (Melskens et al. 2008).
www.intechopen.com
Principally only FTPS can replace slower method CPM that was especially designed for
amorphous silicon (paragraph 4.5) and it was shown by comparison between CPM and
FTPS on a-Si (Holovský et al. 2008). FTPS was also performed on layers on a-Si codeposited
on different substrates including rough conductive and non transparent Al. Special
sandwich arrangement with glass covered by conductive oxide together with transparent
conductive liquid (glycerol) as front contact was used. Bias voltage -1.5 V on front contact
was used to create homogeneous electrical field and to compensate band bending effects.
Recalculation from optical absorption to absorption coefficient was done by Monte-Carlo
optical simulator, details in (Holovský et al. 2010). Measurement of multijunction solar cells
monolithically interconnected was performed with the help of light biasing of the cells that
are not measured, see Fig. 4. Results of such measurements can be found in (Poruba et al.
2001), (Vaněček et al. 2007) and especially in (Holovský et al. 2007) the conditions and limits
of measurements tandem cells are analyzed. In tandem cells it is not possible to spatially
separate the light bias or measurement beam for individual sub-cells and so the modulated
generation occurs in all sub-cells. Limits of discrimination of signal from not measured cells
depend on quality of diode and also on modulation frequency vs. capacity of measured cell.
This unfortunately might be a problem for FTPS where high frequency modulations are
used.
Together with development of FTPS on μc-Si the method was also used for measurement of
5.4 Nanocrystalline diamond, CIS and organic semiconductors
subbandgap absorptance of defects and dopants of nanocrystalline diamond layers

prepared by MW PECVD (Kravets et al. 2002), (Kravets 2005). Main issue in interpretation
of the FTPS on nanodiamond is the effect of photothermal ionization (paragraph 4.3). This
effect can be reduced by use of step-scan mode (paragraph 6) of FTIR spectrometer when
ultimate sensitivity is reached with much slower speed of measurement (Remeš et al. 2007).
There has been efforts to use FTPS also for subbandgap absorption of chalcopyrite
semiconductors (CIS, CIGS). FTPS spectra has been measured on solar cells only (Poruba et
al. 2008), whereas measurement on layers on glass under room temperature has seemed
impossible due to high dark conductivity. The Urbach slope as a measure of compositional
disorder for interpretation of subbandgap region can be used (Wasim et al., 2001).
FTPS has even been used for measurement of photogeneration down to very low values in
organic solar cells based on polymer-fullerene blends (Vandewal et al. 2009). Study of
structural changes induced by annealing has been done (Poruba et al. 2008). According to
(Vandewal et al. 2008), frequency dependence is not an issue in case of organic solar cells.
6. Discussion
In the presented text we did not pointed much to an alternative step-scan mode of FTIR. In
step-scan mode the linear motion of mirror is separated into steps and in between them the
mirror is stationary. Modulation is then realized by slow vibrations of the fixed mirror
(phase modulation) or external chopper with lock-in amplifier (amplitude modulation). This
option will certainly solve the problem of frequency dependence (paragraph 4.3), but will
slow down the measurement so that the advantage compared to CPM method will be
weakened. But in some cases this solution brings ultimate sensitivity that is usually not
necessary e.g. for thin film silicon. Usually disadvantage of step-scan mode is sample
www.intechopen.com
spacing (see paragraph 2) that is higher than 0.5. Very promising in the point of reducing
modulation frequencies are new approaches based on arrays of different luminescence
diodes that are used for measurement of spectra of quantum efficiency of solar cells (Young
et al. 2008) or LCD-based F-T spectrophotometer (US patent no 6031609).
7. Summary
The Fourier Transform Photocurrent Spectroscopy being firstly published in 1997 has
relatively short history and in this chapter we briefly reviewed its today’s level of use and
we more basically explained the general principles and conditions of use. Method is largely
determined by the use of commercial FTIR spectrophotometer that is basis of the method. It
can make the method attractive on one hand, but brings challenges on other hand. Future
modern technology might show larger employment of the method. Principal advantages of
F-T allow maintaining special condition of measurement such as constant photocurrent or
constant illumination or allow achievement of high intensities. Method has so far been used
for variety of materials: thin film silicon, nanocrystalline diamond, organic semiconductors
and CIS compounds. In a laboratory FTPS can be sometimes replaced by CPM method (in
most cases) or by Photothermal Deflection Spectroscopy (only non-absorbing substrates,
lower sensitivity requirements) and its advantage over these methods is speed and
sensitivity. Its disadvantage is wavelength dependent modulation frequency in range of
kHz that requires correction procedure, complicates the interpretation of results and may
even disqualify the method for specific types of materials. Despite that FTPS method has
been proven to be useful in many scientific and also one industrial application.
8. Acknowledgments
The origin of this method as well as many pioneering results are mainly merits of Milan
Vaněček and Aleš Poruba and many colleagues from Institute of Physics of AS CR v. v. i.
The valuable feedback from solar cell industry is a merit of Stephan Jost from Oerlikon
Solar, Trubbach.
This work was supported by Czech Science Foundation projects no. GA202/09/0417 and no.
202/09/H041 and 7th Framework program, project N2P, no. CP-IP-214134 and MSMT
support 7E09057.
9. References
Abe, K.; Okamoto, H.; Nitta , Y.; Tsutsumi , Y.; Hattori , K. & Hamakawa, Y. (1988). Gap
states in undoped amorphous silicon studied by below-gap modulated photo-
current spectroscopy, Philosophical Magazine B, Vol. 58, No. 2, pp. 171-184 , (August
1988) ISSN: 1364-2812
Bailat, J. (2004). Growth, microstructure and electrical performances of thin film
microcrystalline silicon solar cells, PhD thesis, Université de Neuchâtel, July 2004
Davis, E. A. (1970). Optical absorption, transport and photoconductivity in amorphous
selenium. Journal of Non-Crystalline Solids, Vol. 4, No. C, April 1970, pp. 107-116
ISSN: 0022-3093
www.intechopen.com
DeVore, H. B. (1956). Spectral distribution of photoconductivity, Physical Review, Vol. 102,

No. 1, (April 1956) pp. 86-91
Gordijn, A.; Hoďákova, L.; Rath, J. K. & Schropp, R.E.I. (2006). Influence on cell performance
of bulk defect density in microcrystalline silicon grown by VHF PECVD. Journal of
Non-Crystalline Solids, Vol. 352 No. 9-10 (June 2006) pp. 1868–1871 ISSN: 0022-3093
Griffiths, P. R.; Sloane, H. J. & Hannah, R. W. (1977). Interferometers vs Monochromators:
Separating the Optical and Digital Advantages, Applied Spectroscopy, Vol. 31, No. 6,
(1977) pp. 485-495 ISSN: 0003-7028
Griffiths, P. R.; De Haseth, J. A. (1986). Fourier Transform Infrared Spectrometry, Wiley, New
York
Holovský, J.; Poruba, A.; Bailat, J. & Vaněček, M. (2007). Separation of signals from
amorphous and microcrystalline part of a tandem thin film silicon solar cell in
Fourier transform photocurrent spectroscopy, Proceedings of 22nd EU-PVSEC, pp.
1851-1854, ISBN: 3-936338-22-1, Milano, September 2007, WIP Munich
Holovský, J. ; Poruba, A.; Purkrt, A.; Remeš, Z. & Vaněček, M. (2008). Comparison of
photocurrent spectra measured by FTPS and CPM for amorphous silicon layers and
solar cells, Journal of Non-Crystalline Solids, Vol. 354, No. 19-25 (May 2008) pp. 2167 –
2170, ISSN: 0022-3093
Holovský, J. ; Dagkaldiran, Ü. ; Remeš, Z. ; Purkrt, A. ; Ižák, T. ; Poruba, A. & Vaněček, M.
(2010). Fourier transform photocurrent measurement of thin silicon films on rough,
conductive and opaque substrates, Physica Status Solidi A, Vol. 207, No 3, (March
2010) pp. 578-581, ISSN: 1862-6319
Jackson, W. B. & Amer, N. M. (1982). Direct measurement of gap-state absorption in
hydrogenated amorphous silicon by photothermal deflection spectroscopy, Physical
Review B, Vol. 25, No. 8 (April 1982) ISSN 1098-0121
Jackson, W. B.; Kelso, S. M.; Tsai, C. C.; Allen, J. W. & Oh, S.-J. (1985). Energy dependence of
the optical matrix element in hydrogenated amorphous and crystalline silicon,
Physical Review B, Vol. 31 No. 8 April 1985 ISSN: 1098-0121
Melskens, J.; Elzakker, G. van; Li, Y. & Zeman, M. (2008). Analysis of hydrogenated
amorphous silicon thin films and solar cells by means of Fourier Transform
Photocurrent Spectroscopy. Thin Solid Films, Vol. 516, No. 20 , (August 2008) pp.
6877–6881, ISSN: 0040-6090
Klein, S.; Finger, F.; Carius, R.; Dylla, D. & Klomfass J. (2007). Relationship between the
optical absorption and the density of deep gap states in microcrystalline silicon,
Journal of Applied Physics, Vol. 102, No. 10, p. 103501 (November 2007) ISSN: 0021-
8979
Kravets, R.; Ogorodniks, V.; Poruba, A.; Moravec, P.; Nesládek, M.; Rosa, J. & Vaněček, M.
(2002). Fourier-Transform Photocurrent Spectroscopy of Defects in CVD Diamond
Layers. Physica Status Solidi (a), Vol. 193, No. 3, (October 2002) pp. 502–507 (2002)
ISSN: 1862-6319
Kravets, R. (2005). Fourier Transform Photocurrent Spectroscopy of CVD diamond, Ph.D.
Thesis, Czech Technical University in Prague, 2005
Poruba, A.; Fejfar, A.; Remeš, Z.; Špringer, J.; Vaněček, M.; Kočka, J; Meier, J.; Torres, P. &
Shah, A. (2000). Optical absorption and light scattering in microcrystalline silicon
thin films and solar cells, Journal of Applied Physiscs, Vol. 88, No. 1, pp. 148-160 (July
2000) ISSN: 0021-8979
www.intechopen.com
Poruba A.; Vaněček, M.; Rosa, J.; Feitknecht, L.; Wyrsch, N.; Meier, J.; Shah, A.; Repmann T.
& Rech B. (2001). Fourier Transform Photocurrent Spectroscopy in thin film silicon
solar cells, Proceeding of 17th European Photovoltaic Solar Energy Conference, pp. 2981-
2984, ISBN: 3-936338-08-6, Munich, October 2001, WIP Munich
Poruba, A.; Špringer,J.; Mullerová, L.; Vaněček, M.; Repmann, T.; Rech, B.; Kuendig, J.;
Wyrsch, N. & Shah, A. (2003). Fast and sensitive defect characterization and
spectral response measurement of thin film silicon solar structures. Proceeding of 3rd
World Conference on Photovoltaic Energy Conversion (WCPEC3), p. 5P-A9-07, ISBN:
4-9901816-3-8, Grand Cube, Osaka, May 2003, WCPEC, Osaka
Poruba, A.; Holovský, J.; Purkrt, A. & Vaněček, M. (2008). Advanced optical characterization
of disordered semiconductors by Fourier transform photocurrent spectroscopy,
Journal of Non-Crystalline Solids, Vol. 354, No. 19-25 (May 2008) pp. 2421-2425, ISSN:
0022-3093
Python, M. (2009). Microcrystalline silicon solar cells: growth and defects, PhD thesis, Institut
de Microtechnique, Université de Neuchâtel, March 2009
Python, M.; Dominé, D.; Söderström, T.; Meillaud, F. & Ballif, C. (2010) Microcrystalline
silicon solar cells: effect of substrate temperature on cracks and their role in post-
oxidation Progress in Photovoltaics: Research and Applications, Vol. 18, No. 7,
November 2010, pp. 491–499 ISSN: 1099-159X
Remeš, Z,; Nesládek, M.; Bergonzo, P.; Barjon J. & Jomard. F (2007). Amplitude modulated
step scan Fourier transform photocurrent spectroscopy of partly compensated B-
doped CVD diamond thin films Phisica status solidi (a) Vol. 204, No. 9, pp. 2950–
2956 (September 2007), ISSN: 1862-6319
Sculati-Meillaud, F. (2006). Microcrystalline silicon solar cells: theory, diagnosis and
stability, PhD thesis, Institut de Microtechnique, Université de Neuchâtel, July 2006
Shah, A. (2010) Thin-film silicon solar cells, EPFL Press, ISBN 978-2-940222-36-0, Switzerland
Siebke, F. Yata, S. Hishikawa, Y. Tanaka, M. (1998). Correlation between structure and
optoelectronic properties of undoped microcrystalline silicon, Journal of Non-
Crystalline Solids Vol. 227–230 No. 2 (May 1998) pp. 977–981, ISSN: 0022-3093
Street, R. A. (1991). Hydrogenated amorphous silicon, Cambridge University Press, ISBN: 0 521
37156 2, Cambridge
Tomm, J. W.; Jaeger, A.; Bärwolff, A.; Elsaesser, T.; Gerhardt, A. & Donecker, J. (1997). Aging
properties of high power laser diode arrays analyzed by Fourier-transform
photocurrent measurements. Applied Physics Letters, Vol. 71, No. 16 (August 1997)
pp. 2233-2235 ISSN 0003-6951
Unold, T.; Brüggemann, R.; Kleider, J.P. & Longeaud, C. (2000) Anisotropy in the transport
of microcrystalline silicon, Journal of Non-Crystalline Solids, Vol. 266-269, No. 1 (May
2000) pp. 325-330, ISSN: 0022-3093
Vaněček, M.; Kočka, J.; Stuchlík, J. & Tříska, A. (1981). Direct measurement of the gap states
and band tail absorption by constant photocurrent method in amorphous silicon,
Solid State Communications, Vol. 39, No. 11, September 1981, pp. 1199-1202, ISSN:
0038-1098
Vaněček M.; Abrahám, A.; Štika, O.; Stuchlík, J. & Kočka J. (1984). Gap States Density in a-
Si:H Deduced from Subgap Optical Absorption Measurement on Schottky Solar
Cells, Physica Status Solidi (a) Vol. 83, No. 2 p. 617-623 (June 1984) ISSN: 1862-6319
www.intechopen.com
Vaněček, M.; Kočka, J.; Poruba, A. & Fejfar, A. (1995). Direct measurement of the deep defect
density in thin amorphous silicon films with the “absolute” constant photocurrent
method. Journal of Applied Physics, Vol. 78, No. 10, (November 1995) pp. 6203-6210
ISSN: 0021-8979
Vaněček, M.; Poruba, A.; Remeš, Z.; Beck, N. & Nesládek, M. (1998). Optical properties of
microcrystalline materials, Journal of Non-Crystalline Solids Vol. 227-230, Part 2 (May
1998) pp. 967–972 , ISSN: 0022-3093
Vaněček, M.; Poruba, A.; Remeš, Z.; Rosa, J.; Kamba, S.; Vorlíček, V.; Meier, J. & Shah, A;
(2000). Electron spin resonance and optical characterization of defects in
microcrystalline silicon, Journal of Non-Crystalline Solids Vol. 266-269, No. pp. 519-
523 (May 2000), ISSN: 0022-3093
Vaněček, M. & Poruba, A. (2002). Fourier-transform photocurrent spectroscopy of micro-
crystalline silicon for solar cells. Applied Physics Letters, Vol. 80, No. 5, (February
2002) pp. 719-721 , ISSN 0003-6951
Vaněček, M.; Kravets, R.; Poruba, A.; Rosa, J.; Nesládek, M. & Koizumi, S. (2003). Fourier
transform photocurrent spectroscopy of dopants and defects in CVD diamond.
Diamond and Related Materials, Vol. 12, No. 3-7, (March-July 2003), pp. 521-525,
ISSN: 0925-9635
Vandewal, K.; Goris, L.; Haeldermans, I.; Nesládek, M.; Haenen, K.; Wagner, P. & Manca, J.
V. (2008) Fourier-Transform Photocurrent Spectroscopy for a fast and highly
sensitive spectral characterization of organic and hybrid solar cells, Thin Solid Films
Vol. 516, No. 20 (August 2008) pp. 7135–7138 ISSN: 0040-6090
Vandewal, K.; Tvingstedt, K.; Gadisa, A.; Inganäs, O. & Manca, J. V. (2009) On the origin of
the open-circuit voltage of polymer–fullerene solar cells Nature Materials, Vol. 8
No.11, pp. 904 - 909 (November 2009) ISSN: 1476-1122
Wasim, S. M.; Rincón, C.; Marín, G.; Bocaranda, P. & Hernández, E. (2001). Effect of
structural disorder on the Urbach energy in Cu ternaries. Physical Review B, Vol. 64,
No. 19, 195101 ( October 2001) ISSN 1098-0121
Wyrsch, N.; Finger, F.; McMahon, T. J. & Vaněček, M. (1991). How to reach more precise
interpretation of subgap absorption spectra in terms of deep defect density in
a-Si:H Journal of Non-Crystalline Solids Vol. 137-138 Part 1 pp. 347-350 (1991) ISSN:
0022-3093
Young, D.L.; Pinegar, S.; Stradins, P. & Egaas, B. (2008) New Real-Time Quantum Efficiency
Measurement System Proceedings of the 33rd IEEE Photovoltaic Specialists Conference,
San Diego, California May 11–16, 2008 Conference Paper NREL/CP-520-42509
www.intechopen.com
14
Fourier Transform Rheology: A New Tool to

Characterize Material Properties
Massimiliano Grosso1 and Pier Luca Maffettone2
1Dipartimento di Ingegneria Chimica e Materiali, Università degli Studi di Cagliari
Piazza D’Armi, I-09123, Cagliari,
2Dipartimento di Ingegneria Chimica, Università degli Studi di Napoli “Federico II”
Piazzale Tecchio, I-80125, Napoli

Italy
1. Introduction
Liquid multiphase systems such as polymer blends or emulsions are ubiquitous in many
applications, including plastic production, food processing, pharmaceutical and cosmetic
production. When the constituents of the multiphase system are incompatible the phases are
immiscible, and, depending on their relative amount, the microstructure can consist of
droplets in a matrix, elongated fibrils or a co-continuous structure (Utracki, 2003) as
schematically shown in Figure 1. The morphology of the liquid multiphase system is
important in the applications as it strongly affects processing properties, and the properties
of the final products. With the term “morphology” we here indicate not only the overall
form or shape of the physical structure of the system, but also the distribution and
orientation of the phases, the interfacial area, and the volume of the interphase.
Hence, a profound knowledge of the relation between processing parameters, material
properties and morphology is essential to optimize the performances of the liquid
multiphase systems.
Fig. 1. Different morphologies of immiscible polymer blends (a) dilute droplet blends; (b)
elongated fibrils; (c) co-continuous structure
Substantial efforts were done in the last decades to set up experimental protocols aimed at
evaluating the morphological properties of polymer blends and emulsions via rheological
measurements. So far, the most reliable strategy for morphological characterization through
rheological measurements is based on the dynamic small amplitude oscillatory shear
www.intechopen.com
(SAOS) experiment: the samples are subjected to small amplitude shearing oscillations, and
the measured shear stress response is used to gain information on the blend properties (e.g.
Palierne, 1990).
Here, we present an alternative technique we have recently proposed to characterize the
liquid two-phase system morphology. This methodology is based on Large Amplitude
Oscillatory Shear (LAOS) flows. This kind of analysis is often referred in the literature as
Fourier Transform Rheology (FTR) (Wilhelm et al., 1998), since the stress response is usually
analyzed in the Fourier domain.
It will be shown that Fourier Transform Rheology possesses a high sensitivity in the
characterization of the morphology, thus allowing evaluation of properties that might
otherwise be hardly appreciated with traditional linear methodologies.
2. Rheological oscillatory experiences

2.1 Small Amplitude Oscillatory Shear
A typical tool used for the characterization of complex liquids is based on oscillatory
rheometry (Macosko, 1994). The basic working principle of an oscillatory rheological test is
to impose a sinusoidal shear deformation, and measuring the resultant shear stress
response. In a typical experiment, the sample is placed between two plates (or a cone and
plate geometry) (Figure 2): while the bottom plate remains stationary, a rotation is imposed
on the top plate, thereby allowing a time-dependent strain deformation on the sample:
γ ( t ) = γ 0 sin ωt (1)
where ω is the oscillation frequency and γ0 is the strain amplitude. The oscillation period is
thus T = 2π/ω. The resulting time-dependent shear stress, σ(t), is quantified by measuring
the torque on the top plate. At low strain amplitudes, the stress response can be assumed to
depend linearly on the strain deformation:
σ ( t ) = A sin ωt + B cos ωt = σ0 sin ( ωt + δ ) (2)
Torque
θ0
Fixed plate Pressure transducers
Fig. 2. Cone and plate geometry
www.intechopen.com
Fourier Transform Rheology: A New Tool to Characterize Material Properties 287
For this reason Small Amplitude Oscillatory Shear (SAOS) tests are usually referred as linear
rheological measurements. If the material behaves as an ideal elastic solid, then the stress is
in phase with the imposed deformation wave (i.e. proportional to sin(ωt)), and the
proportionality constant is the shear modulus of the material. On the other hand, if the
material is a purely viscous fluid, the stress is proportional to the rate of the strain
deformation (i.e. proportional to cos(ωt)), and the proportionality constant is the viscosity of
the fluid. The applied strain and the measured stress are in this case out of phase with phase
angle δ = π/2.
Complex materials usually show a response that contains both in-phase and out-of-phase
contributions. As a consequence, the total stress response at a given ω is characterized by
both the sine and cosine components:
σ ( t ) = G ' ( ω) γ 0 sin ωt + G " ( ω) γ 0 cos ωt (3)
In equation 3, G’(ω) is the storage modulus which characterizes the solid-like behavior,
whereas G”(ω) is the loss modulus that takes into account the fluid-like contributions. The
complex modulus to the frequency can be thus defined:
G * ( ω ) = G ' ( ω) + i G " ( ω ) (4)
where i is the imaginary unit. Using the relation between the complex modulus G*(ω) and
the complex viscosity η*(ω),
G * ( ω) = ωη* (5)
and
η* ( ω) = η '+ i η " (6)
The absolute value of the complex viscosity is of course given by:
η* ( ω) =
G '2 + G "2
=
G*
ω ω
(7)
The frequency dependence of G’ and G” provides some important information about the
microstructure of a material. For example, gels exhibit G’ that is larger than G” with both
moduli independent of frequency. Polymer melts show G’ and G” at low frequencies that
are dependent on ω2 and ω, respectively. For viscoelastic materials, the overlap frequency
(the frequency at which G’ and G’ curves intersect) gives information about the relaxation
time of the system. The plateau modulus, i.e. the value of G’ at high frequency, gives
information about the strength of the structures formed in the material.
For the case of dilute blends with Newtonian constituents, the dependence of G’ and G” can
be described in terms of the Palierne model which may quantitatively associate the linear
viscoelastic properties of polymer blends to its chemical-physical properties e.g. to the
interfacial tension (Palierne, 1990: Graebling et al., 1993a; Graebling et al., 1993b; Lacroix et
al., 1996).
www.intechopen.com
G * ( ω) = Gm* ( ω)
( )
1 + 3ϕ H RV , ω
1 − 2 ϕ H ( R , ω)
(8)
V
where
α
(
2Gm* ( ω) + 5Gi* ( ω) + )
( G ( ω) − G ( ω) ) ( 16G ( ω) + 19G ( ω))
4
( )
RV
H RV , ω =
( G ( ω ) + G ( ω) ) +
* * * *
α
i m m i
(9)
* *
( )( )
40
2 Gi* ( ω) + 3 Gm* ( ω) 16Gm* ( ω) + 19Gi* ( ω)

m i
R V
matrix and blend at frequency ω, α is the interfacial tension, φ is the volume fraction of the
and G*i(ω), G*m(ω) and G*(ω) are, respectively, the complex moduli of the dispersed phase,
dispersed phase and RV is the volume average drop radius of the included phase:
∫ R ψ ( R ) dr
∞
4
RV =
∫ R ψ ( R ) dr
∞
0
(10)
3
where ψ(R) represents the drop size distribution.

An example is reported in figure 3, where the elastic modulus for a blend composed by
Poly-DiMethylSiloxane (PDMS) in Poly-IsoButylene (PIB) is reported with respect to the
oscillation frequency, together with the elastic modulus of the neat constituents.
104
103
102
G' [Pa]
101
100
PIB
10-1
PDMS
10-2 PDMS/PIB blend (φ = 0.1)
10-3
10-1 100 Frequency [Hz] 101
Fig. 3. Elastic modulus as a function of frequency for the pure components (PIB: square
points; PDMS: triangles) and the polymer blend (full circles). The temperature is 30°C. The
strain amplitude is 50%
www.intechopen.com
The Palierne model is widely used to extract morphological or dynamical properties from
oscillatory data, for example the average drop radius can be estimated (e.g. Das et al.,
2005) or alternatively the surface tension (e.g. Huitric et al., 1998; Vincze-Minya and
Schausberger, 2007). As already remarked in the literature (Graebling et al. 1993a),
however, it is difficult to achieve a more detailed description of the morphology of the
included phase . Indeed, a complete description of the drop size distribution cannot be
reliably obtained with this technique. To our knowledge, the size distribution inference
for a polymer blend based on the Palierne method has been carried out only by Friedrich
et al. (1995). The methodology there proposed is based on a Tikhonv regularization, and
gave satisfactory results only for very dilute blends with unimodal drop radius
distributions.
2.2 Large Amplitude Oscillations

The above mentioned limits of the traditional linear oscillatory experiences motivated the
study of alternative experimental techniques that might be more sensitive to the material
morphology. In this regard, Large Amplitude Oscillatory Shear flows (LAOS) proved to be a
possible candidate for the morphological characterization. The capabilities of this technique
to pinpoint nonlinear material characteristics have been already analyzed in other contexts
(Neidhofer et al., 2004; Schlatter et al., 2005), proving to be quite effective.
In the following we will briefly review some basic issues on LAOS (see Wilhelm et al. 1999
the applied shear rate η ( γ$ ) can be expected to be important, and equation 3 is no longer
for details).When dealing with LAOS flows, the nonlinear dependences of the viscosity on
valid.
Due to the symmetry properties, the viscosity is independent of the shear direction and
therefore it can only depend on the absolute shear rate (Wilhelm et al., 1999):
( )
η ( γ$ ) = η ( −γ$ ) = η γ$ (11)
The Taylor expansion for the viscosity at small shear rates is given in equation (12):
η ( γ$ ) = η0 + η1 γ$ + η2 γ$ + ...
2
(12)
If the applied shear deformation is a harmonic oscillation with a given frequency ω1, strain
and strain rate are:
γ = γ 0 sin ω1t ⇒ γ$ = ω1 γ 0 cos ω1t (13)
Therefore, the absolute value of the shear rate signal γ$ can be represented in terms of a
proper Fourier series (Ramirez, 1985):
⎛ 2 4 ⎛ cos 2ω1t cos 4ω1t ⎞⎞

γ$ = ω1 γ 0 ⎜ + ⎜ − + ... ⎟ ⎟
⎝ π π ⎝ ⋅ ⋅ ⎠⎠
(14)
1 3 3 5
γ$ = a '+ b 'cos 2 ω1t + c 'cos 4ω1t + ... (15)
www.intechopen.com
By substituting Equation 15 into Newton’s equation for the viscosity leads to equation 16
( )
(Wilhelm et al., 1999)
σ = η ( γ$ ) γ$ = η0 + η1 γ$ + η2 γ$ + ... cos ω1t

2
(16)
⎛ η0 + η1 ( a '+ b 'cos 2 ω1t + ...) + ⎞

σ = η ( γ$ ) γ$ = ⎜ ⎟ cos ω1t
⎜ η2 ( a '+ b 'cos 2ω1t + ...)2 + ... ⎟
(17)
⎝ ⎠
The terms in brackets in Equation 17 can be thus simplified and written as a sum of even
harmonics:
σ = ( a "+ b "cos 2 ω1t + c "cos 4ω1t + ...) cos ω1t (18)
By multiplying the terms in the brackets by cosω1t one ends up with the shear stress
expression depending only on odd harmonics:
σ = A cos ω1t + B cos 3ω1t + C cos 5ω1t + ... (19)
where A, B, C are complex numbers. In a last step the non-linear torque signal is analyzed
towards frequency components by Fourier transformation. Eventually, the signal can be
described in terms of an odd function of the sinusoidal deformation
σ ( t ) = ∑ I Rk cos jω1t + I Ik sin jω1t =

∞
k =1
∑ I Ak cos ( kω1t + φk )
odd k
∞
(20)
k =1
odd k
In equation (20) IRk, IIk and IAk are real coefficients. Straightforwardly, one can easily express
the measured shear stress signal in the Fourier domain as:
σ ( t ) ⇔ σ# ( ω) = ∫ σ (t ) e ∑ I k δ ( ω − kω1 )
+∞ k =∞
− i ωt
=
FT
(21)
−∞ k =−∞
odd k
In equation 21, δ(ω-kω1) is the Dirac delta located at ω = k ω1 (k ∈ Z), i is the imaginary unit,
and Ik is the (complex) coefficient of the k-th harmonic. As the stress time series σ(t) is real
valued, the condition Ik= I*-k (with * denoting the complex conjugate) holds. As a
consequence of the assumption made in equation 19, only odd terms of the Fourier series
could be in principle accounted for in equation 21 (Wilhelm et al., 1998). It is easy to show
that the following relationship among the coefficients holds:
I Aj = I k = I Rk
2
+ I Ik2 (22)
significant. Incidentally, one can notice that, as γ0 tends to zero, the linear behaviour is
When dealing with SAOS flows only the first term of the summation in equation 21 is
recovered, thus II1 = G’γ0 and IR1 = G”γ0, and IRk ≈ IIk ≈ 0 for any k >1. The appearance of
www.intechopen.com
significant values for Ik (k > 1) marks the onset of nonlinearities in the stress response. As a
further remark, it has been shown that, at vanishing amplitudes the following scaling for
intensity of nth harmonic with strain has been observed for the constitutive equations so far
investigated (Nam et al. 2008; Yu et al. 2008):
I An ∝ γ n0 (23)
This allows the definition of new scalars, based on the ratios of intensities of higher
harmonics and first harmonic of the stress response. For example, coefficient Q is defined as
(Hyun & Wilhelm, 2009)
Q=
I A3 1
I A1 γ 02
(24)
This coefficient has been claimed to be helpful in distinguishing molecular architecture of

polymers based on LAOS (Hyun & Wilhelm, 2009).
the rheometer, is usually performed discretely at finite sample intervals (Δt). Based on the
From a practical point of view, the measurement of the stress, through the torque sensor of
sampling frequency, (r =1/Δt, number of data points collected per second), we obtain a time
series σ(n) of discrete measurements collected at NP instants.
Discrete Fourier transform of this time domain series will be a series of NST complex
numbers evaluated through well consolidated FT techniques (Bracewell, 1986):
i 2 π ( k − 1 )( n − 1)
Σ ( k ) = ∑ σ (n) e
∞
, 1≤ k ≤N
−
N
(25)
n=1
2 π/Δt. With the property of the Fourier transform leading to meaningful N/2 (symmetric)
The maximum frequency in the Fourier domain will correspond to the Nyquist frequency =
terms, the resolution in the frequency domain is 2π/T. Therefore, sampling interval
determines maximum frequency to which information can be obtained, while the duration
of measurements determines the resolution of frequency. Larger T values also lead to higher
signal to noise ratio. It should be remarked that some techniques are introduced in the
literature in order to improve the sensitivity to the signal of the measurement (Wilhelm et
al., 1999).
3. FTR on polymer blends

3.1 Theory
In this section, we will focus on the theoretical aspects concerning the characterization
through FTR of immiscible blends with low fraction of the dispersed phase. In this case the
morphology of the included phase is globular: the basic element of such a dilute blend is
thus a single drop dispersed in a matrix. Therefore, the study of single droplet behaviour is
regarded as a reasonable starting point to model the complex behaviour of immiscible
polymer blends. The overall rheological response, in fact, could be determined just on such a
basis.
The dynamic behaviour of dilute polymer blends subjected to LAOS flows can be modeled
as recently proposed in the literature by considering the single droplet dynamics together
www.intechopen.com
with a proper stress expression (Rallison, 1984; Stone, 1994; Almusallam et al., 2000; Yu et
al., 2002; Jackson and Tucker, 2003; Yu and Bousmina, 2003). For what matters the dynamics
of the drop, a handy though effective phenomenological model has been proposed by
Maffettone and Minale (1998) and applies to generic flow fields. The model is formulated in
terms of at most six first-order, ordinary, differential equations, and is capable of describing
drop deformation up to the nonlinear range. This model is known to be quite accurate for
small-to-medium droplet deformation, but loses some quantitative accuracy as droplet
deformation becomes large. We use here this model for its simplicity, even though
significant distortion of drop shape is expected under LAOS. Still, the Maffettone and
Minale model provides a useful basis for analyzing and interpreting the experimental
results also when significant strain deformations occur (Guido et al. 2004). The drop is
described as an ellipsoid by a second rank symmetric, positive definite, and time dependent
tensor. The shape dynamics can be thus described by the evolution of tensor S which
follows the equation:
f ⎛ 3 ⎞
− ( Ω ⋅ S − S ⋅ Ω ) = − 1 ⎜ S − I ⎟ + f2 (S ⋅ D + D ⋅ S)
dS
τ⎝ I2 ⎠
(26)
dt
In Equation 26, τ is the emulsion time (τ=ηR/Γ) where η is the matrix viscosity, R the
undistorted drop radius and Γ is the interfacial tension; I is the second rank unit tensor, D
and Ω are the deformation rate and the vorticity tensors respectively, and I2 is the second
scalar invariant of tensor S. The shear flows here considered give the following forms for the
deformation and vorticity tensors:
⎛0 1 0⎞ ⎛ 0 1 0⎞
1 ⎜ ⎟ 1 ⎜ ⎟
D = Ca ⎜ 1 0 0 ⎟ , D = Ca ⎜ −1 0 0 ⎟
2 ⎜ ⎟ 2 ⎜ ⎟
(27)
⎝0 0 0⎠ ⎝ 0 0 0⎠
In equation 27 the Capillary number is introduced:
ηγ$
Ca = =
viscous stress
interfacial stress Γ R
(28)
This gives the ratio between the two competing forces affecting the drop shape in shear flow
experiences: the driving force of deformation (i.e. the shear stress), and the resistance force
supporting the shape of the drop, that is the interfacial tension. The dependence of the
capillary number on time is understood. The functions f1 and f2 appearing in Eq. 26 are given
by (Maffettone & Minale, 1998):
40 ( λ + 1 )
f1 ( λ ) =
( 2λ + 3 )( 19λ + 16 )
f2 (λ) =
(29)
+
5 3Ca 2
2 λ + 3 2 + 6Ca 2
In equation 29 the ratio λ = ηd/ηm is defined, where ηd is the viscosity of the dispersed
phase and ηm is the matrix viscosity. At rest the drop is spherical (S=I). Notice finally that
within this description drop break-up is absent under shear flow for λ ≥ 3.
www.intechopen.com
Once the state of the drop deformation is known, one can calculate the stress  of a dilute
polymer blend according to Batchelor (1970):
σ = − pI + ηm ∇v + ∇vT ( )
isotropic Newtonian
term
− ∫ ( nu + un ) dA + ∫ ⎜ nn − I ⎟ dA
Γ ⎛
contribution
η 1 ⎞
(30)
VS V S⎝ 3 ⎠
viscous term elastic term
In Eq. 30, p is the pressure, ∇v is the velocity gradient tensor and ∇vT its transpose, η is
the viscosity of the continuous phase, V is the total volume of the system, n is the unit
vector normal to the ellipsoid surface, representing the interface between the two phases,
u is the velocity at the interface, dA is the area of an interfacial element, and the integrals
are calculated over the whole interface of the system, S. Equation 26 can be used to predict
the stresses if n and u are known. Predictions are obtained by integrating equations (26)
and (29) for the drop morphology. The elastic interfacial term in equation (30) is
calculated as suggested by Almusallam et al. (2004). The viscous term in the interface
stress is neglected.
Equation 30 is the sum of two conceptually different terms: the first one is due to the
Newtonian matrix contribution and depends linearly on the velocity gradient, whereas the
second term (the viscous and the elastic term) corresponds to the sum of interfacial
contributions related to the entire drop population. The first part depends linearly on the
applied shear rate, whereas the interfacial contribution is the only nonlinear term appearing
in Equation 29. Under LAOS, the first term will not contribute to higher harmonics in the
higher harmonics in the power spectrum of σ(t) (Grosso and Maffettone, 2007).
shear stress for its linear nature. Conversely, the interface contribution will give rise to
Consequently, the contribution to the higher harmonics in the Fourier spectrum of each
drop with radius R is directly related to the interfacial contribution. In the frequency domain
this can be written as:
Ik ( R) = ∫ σ ( t ) e 1 dt
2π − ikω t
T T
2π ⎛ η 1 ⎞ ⎞ − ikω t
∫ ( nu + un )dA +
η ⎛
∫⎜ ∫ ⎜ nn − 3 I ⎟⎠dA ⎟ e 1 dt
(31)
=−
T T⎝V A V A⎝ ⎠
with k odd and > 1.
Simulations are performed by mimicking realistic conditions for PDMS in PIB samples with
the relevant parameters reported in the Table 1. These parameter values are consistent with
to be φ = 0.1.
experiments carried out at a temperature T = 35 °C. The volume fraction is always assumed
Figure 4 shows the tangential stress σ of two simulated polymer blends both in the time and
set respectively equal to γ0 = 800% and ω = 0.1 s-1. The two blends differ for drop radius.
in the Fourier domains. The imposed deformation amplitude and oscillation frequency are
Figure (4.a) and (4.b) show the time evolution and the Fourier transform (namely the
I(ω1)=1) of a polymer blend consisting of equal drops with radius R1 = 1 μm, respectively,
absolute values of power spectrum rescaled with respect the fundamental harmonic, thus
www.intechopen.com
whereas figure (4.c) and (4.d) refer to a blend with drop radius R2=5 μm. It is apparent that
no significant difference can be appreciated in the time domain, the signals looking very
close to a sinusoidal waveform in both cases. This result is not unexpected since, as already
mentioned, the linear Newtonian matrix contribution dominates the response when
observed in the time domain.
principal harmonic (corresponding to the forcing frequency ω = 0.1 s-1) is not reported
On the contrary, the nonlinear features appear more evidently in the Fourier domain. The
entirely in order to magnify the harmonics appearing at higher frequencies. It is evident that
Fourier analysis allows a clear detection of the nonlinearities that are otherwise not
appreciable in the time domain. By comparing Fig. (4.b) with Fig. (4.d), it can also be noted
the significant dependence of higher harmonics of the shear stress on drop size.
Molecular
Density Viscosity Interfacial tension
Polymer Formula Weight
[Kg/m3] [Pa⋅s] [mN/m]
[Da]
PDMS [ –Si(CH3)2O– ]n 200000 971 175

3
PIB [ –CH2C(CH3)2– ]n 1300 894 57
Table 1. Main physical properties of PDMS/PIB system
400 (a) 400 (b)
200 200
stress [Pa]
stress [Pa]
0 0
-2 0 0 -200
-4 0 0 -400
0 2 4 time 6 8 10 0 2 4 time 6 8 10
1e-1 1e-1
(c) (d)
1e-2 1e-2
|I(w)|
|I(w)|
1e-3 1e-3
Y D ata
1e-4 1e-4
ω1 3 ω1 ω 5 ω1 7 ω1 ω1 3 ω1
ω
5 ω1 7 ω1
1e-5 0 50 100 150 200 250 300 350 400
1e-5
X D ata
Fig. 4. Tangential stress of a simulated polymer blend in a LAOS experience in the time
different drop radii: R1 = 1 μm (Figures 4.a and 4.c) and R2 = 5 μm (Figures 4.b and 4.d). The
domain (Figures 4.a and 4.b) and in the Fourier domain (Figures 4.c and 4.d) for two
deformation amplitude is γ0=800% and ω1 = 0.1·2π s-1. The physical parameters are in Tab. 1
www.intechopen.com
μm versus the strain amplitude γ0. As the strain deformation tends to zero, both quantities
Figure 5 reports the absolute values I3A/γ03 and I5A/γ05 for a fixed value of the radius R=10
approach a constant value thus confirming the asymptotic behaviour previously observed
for other constitutive models (Nam et al. 2008; Ewoldt et al. 2008). It should be remarked
that the limiting values depend on the blend properties (i.e. the phase viscosities, the surface
tension and the drop radii).
1e+1
1e+0
1e-1
5
I3/γ0 , I5/γ0
1e-2
3
1e-3
1e-4
1e-5
γ0
1e-6
1 10
Fig. 5. Scalars I3/γ03 and I5/γ05 vs the strain deformation γ0 for a simulated monodisperse
polymer blend with radius R = 10 μm
3.2 Experiments
In this section we will show some experimental results that demonstrate the sensitivity of
the FTR methodology when analyzing blend or emulsion morphology. The details of the
experimental part can be found elsewhere (Carotenuto et al. 2008).
The polymer blend is prepared with PDMS and PIB that are immiscible at room
temperature. PIB/PDMS emulsion is a widely used model system largely studied in the
literature by means of both rheological and optical techniques (Jansseune et al., 2000; Guido
et al. 2004; Wannaborworn et al. 2002). All the experiments were performed at constant
temperature T = 30°C. The main physical properties of the polymers are reported in Table 1.
The value of the interfacial tension for the very same polymers is found in the literature
experiments were carried out with a volumetric fraction, φ, of the dispersed phase fixed to
(Sigillo et al., 1997). PIB is the continuous phase and PDMS is the dispersed phase. All the
0.1 thus leading to a globular morphology. This value is small enough to consider
coalescence negligible. The viscosity ratio is equal to 3, and it is large enough to avoid
www.intechopen.com
significant break-up phenomena under pure shear flow. Thus, the blend can be assumed to
be stable, and its microstructure should not significantly vary in time during the
the drop size distribution ψ(R), is then assumed to remain unchanged during LAOS
experiments (negligible breakup and negligible coalescence). The blend morphology, i. e.,
experiments.
The experiments were conducted on three different blend samples, which hereafter will be
indicated with a capital letter A, B and C. The morphologies of the samples are supposed to
have a similar (but not equal) morphology since their preparation followed the same
protocol.
Oscillatory shear measurements (both SAOS and LAOS) were performed in a conventional
strain controlled rheometer (ARES, TA Instruments). Linear viscoelastic measurements were
analyzed using the software provided by the rheometer manufacturer. LAOS experiments
required a modification and improvement of the traditional rheometer data acquisition
system. The raw data coming from both motor and transducer were acquired and digitized
with a 16-bit analog-to-digital converter (National Instrument, PCI_6251). The motor signal
was correlated to the imposed strain deformation, γ, while the transducer signal was
associated with the measured torque. In order to maximize S/N, the rheometer was
equipped with a very sensitive torque transducer (2KFRTN1) that could detect a torque
ranging from 0.002 to 200 mN·m.
Before starting the acquisition, two main parameters were set: the scan rate, r [=] pts/s, and
the number of data points, Np. They were the same for both the channels (motor signal and
transducer signal). The ratio between Np and r gives the time required for the entire
acquisition, tacq = Np/r. The oscillation cycles collected during tacq depend on the imposed
deformation frequency (ω1). Typical values of r and Np are 1000 pts/s and 80000 pts,
respectively, thus tacq = 80 s. Thus, for an imposed deformation frequency ω1=0.1 Hz, 8
complete cycles were acquired. It should be noted that the higher values of r and Np, the
higher the S/N ratio (Wilhelm et al., 1999). It was however checked that acquisitions with
larger amount of data (r = 5,000 pts/s and Np = 400,000 pts) did not show any significant
increase in the quality of our data.
Raw data coming from transducer were collected and subsequently transformed into the
corresponding Fourier spectra. Odd multiples of the fundamental harmonic appear in the
nonlinear regime (LAOS). For the polymer blend under investigation, the 3rd and the 5th
overtones could be clearly detected in the shear stress Fourier spectrum for deformation
amplitudes γ0 > 100%. The electric signal measured by the torque transducer is supplied in
terms of potential difference units.
LAOS data were analyzed according to the FTR protocols. The imposed sinusoidal
deformation is γ(t)= γ0 sin(ω1 t), where ω1 = 2πΩ1=2π/T is the characteristic angular frequency
with T the oscillation period.
Linear viscoelastic measurements were carried out for a preliminary characterization of the
microstructure of the samples. Oscillatory measurements were performed with frequency
ranging from 0.1 to 10 Hz. Strain amplitudes up to 50% gave shear stress responses well
within the linear region. It was found that traditional SAOS measurements did not give a
clear discrimination between different blends, and the G’ curves for the three emulsions are
almost overlapping, thus indicating that SAOS suggest that the three blends have similar
morphologies.
www.intechopen.com
Blend RV [μm]
A 5.6
B 6.7
C 8.0
Table 2. Average drop radii for the three blends estimated with the Palierne method
From SAOS measurements one can obtain an estimation of the average dimension of the
dispersed phase, namely the volume-average drop radius, RV . According to Palierne (1990),
one can estimate the volume average drop radius for the emulsions. Table 2 contains the
values of the estimated average drop radii for the blends A, B, C. As expected, the volume
averaged drop radius, RV , for the three samples is very similar. A more detailed description
LAOS measurements were performed with γ0 ≥ 200%, where nonlinearities in the response
of blend morphology cannot be attained with the linear rheological measurements.
become clearly appreciable. A typical experimental result is shown in Fig. 6, where the
domain (Fig. 6.b) for γ0 = 800% and Ω1 = 0.1 Hz (or, equivalently, ω1=2π 0.1 rad/sec).
tangential stress response is reported both in the time (Fig. 6.a) and in the frequency
Fourier spectra report the absolute value of the overtones, normalized with the first
harmonic (Ik/I1 or equivalently Ik1) as commonly done in the FTR literature (e.g. Wilhelm
et al., 1998). The nonlinear shear stress response cannot be easily detected in the time
domain, but the corresponding power spectrum clearly shows the occurrence of a third
and a fifth peak.
3 0.010
(a) (b)
abs(I/I1)
2
Torque [Volt]
0 0.005
-1
-2
-3 0.000
0 10 20 30 40 50 60 0.1 0.3 0.5 0.7 0.9 1.1
time [s] Frequency [Hz]

Fig. 6. Transducer signals in the time domain (a) and the corresponding Fourier spectra (b)
for a polymer blend, at 30°C
In figure 6.b a peak I(2ω1) at an even multiple of the fundamental harmonic is also
observable. It should be reminded that this occurrence is unexpected since the stress signal
is demonstrated to be an odd function of the time. Several explanations for the presence of
even overtones in the spectra have been proposed in literature. Quite often, the occurrence
www.intechopen.com
of even harmonics is attributed to some artefact in the experiments as e.g. wall-slip

phenomena (Hatzikiriakos & Dealy, 1991). In the case here reported, the second harmonic
seems to be material-independent, and it can be attributed to an imperfect alignment of the
upper and lower plates of the rheometer (Carotenuto et al., 2008): it reasonably comes from
the instrument itself and results unrelated to the measured sample, for this reason it is
simply neglected.
Figure 7 shows the Q coefficient defined in equation 24 as a function of the strain amplitude
for the blend C. The third harmonic is clearly detected for the polymer blend under
investigation. The value of I31 is small but reproducible with an experimental error lower
than 3%. For the sake of comparison, data of the neat PIB and PDMS are also reported in Fig.
7. The pure component I31 is weighted by the corresponding amount in the blend (i.e., 0.1 for
the PDMS and 0.9 for the PIB). It is apparent that the I31 values of the pure components are
extremely low, according to their quasi-Newtonian behavior, and negligible when
compared with the I31 values of the blend. This experimental evidence unequivocally
suggests that the observed nonlinear response of the blend does not derive from simple
superposition of the nonlinear contribution of the neat polymers, but it seems essentially
due to the interface stress contribution. Such behaviour confirms the validity of the
assumptions made in Equation 29.
1e-2
PIB
PDMS
Blend
Q=I31/γ02
1e-3
1e-4
Strain [%] 1000

Fig. 7. The Q coefficient as a function of the imposed strain deformation for the pure
components (PIB: dashed line with square; PDMS: dashed-dotted line with triangles) and
for the blend C (solid line with circles). The temperature is 30°C. The oscillation frequency is
0.1 Hz
Figure 8.a shows the coefficients Q = I31/γ02 and P = I51/γ04 for the three blends A, B and C as
a function of the strain amplitude. It is shown that, as the strain deformation decreases, the
curves seem to tend to an asymptotic plateau value Q0 and P0, accordingly with the
theoretical predictions.
www.intechopen.com
The curves do not superimpose, thus suggesting that LAOS experiences could discriminate
between different morphologies. Indeed, as reported from Palierne results, the upper curve
refers to the blend A ( RV = 5.6 μm), the medium to the blend B ( RV = 6.7 μm), and the lower
to the blend C ( RV =8 μm). Hence, the Q0 coefficient of the blend seems to decrease with the
mean size of the inclusions. In Figure 8.b, the ratio P =I51/γ04 for the blends A, B and C are
also reported. Since the fifth overtones are significantly smaller than the third ones, they are
more affected by experimental noise. Analogously to Q behaviour, the fifth peaks are larger
for blend with smaller volume averaged drop radius.
0.001
Q=I31/γ02
1e-4
P=I51/γ04
1e-5
1000
Strain [%]
Fig. 8. The Q and P coefficients as a function of strain amplitude for the blend A (solid line
with circles), B (dashed line with triangles) and C (dashed-dotted line with squares). The
oscillation frequency is 0.1 Hz
4. Conclusion
Fourier Transform Rheology is a valuable tool to characterize the microstructure of dilute
immiscible polymer blend as it was shown both theoretically and experimentally. We
analyze the case of a blend with Newtonian constituents, and in such a case the nonlinearity
of the response under LAOS comes exclusively from the presence of a polymer-polymer
www.intechopen.com
interface. Indeed, distinct odd multiples of the fundamental harmonic are clearly evident in
the power spectrum of the emulsion, while are only barely distinguishable in the spectra of
the pure components (PIB and PDMS). FTR greatly enhances the sensitivity of the
experiments to the blend morphology, thus allowing the evaluation of details that are
otherwise difficult to be appreciated with time domain analysis.
5. Acknowledgments
Authors acknowledge Dr. Claudia Carotenuto for the helpful discussions on the
experimental technique.
6. References
Almusallam A.S., Larson R. & Solomon M.J. (2004). Comprehensive constitutive model for
immiscible blends of Newtonian polymers, Journal of Rheology, 48(2), pp 319-348
ISSN: 0148-6055
Batchelor G.K. (1970) The stress system in a suspension of force-free particles, J. Fluid Mech.
41, pp 545-570 ISSN: 0022-1120
Bracewell R.N. (1986) The Fourier-transform and its application. McGraw-Hill, New York,
ISBN: 978-0073039381
Carotenuto C., Grosso M., Maffettone P.L. (2008) Fourier Transform Rheology of Dilute
Immiscible Polymer Blends: A Novel Procedure To Probe Blend Morphology,
Macromolecules, 41, pp 4492-4500 ISSN: 0024-9297
Das N.C., Wang H, Mewis J. & Moldenaers P. (2005), Rheology and Microstructures
Formation of Immiscible Model Polymer Blends Under Steady State and Transient
Flows, J. Pol. Sci.: Part B: Polymer Physics, 43, pp. 3519–3533 ISSN: 1099-0488
Ewoldt RH, Hosoi AE & McKinley GH (2008) New measures for characterizing nonlinear
viscoelasticity in large amplitude oscillatory shear. Journal of Rheology 52(6), pp.
1427–1458 ISSN: 0148-6055
Friedrich C., Gleinser W., Korat E. Maier D. & Weese J. (1995). Comparison of sphere-size
distributions obtained from rheology and transmission electron microscopy in
PMMA/PS blend, Journal of Rheology, 39(6) pp. 1411-1425 ISSN: 0148-6055
Graebling D,, Muller R. & Palierne J.P. (1993). Linear viscoelastic behavior of some
incompatible polymer blends in the melt - interpretation of data with a model of
emulsion of viscoelastic liquids Macromolecules 26, pp. 320–329 ISSN: 0024-9297
Graebling D,, Muller R. & Palierne J.P. (1993). Linear viscoelasticity of incompatible polymer
blends in the melt in relation with interfacial properties, Journal de Physique IV, 3
pp. 1525–1534, ISSN: 1155-4339
Grosso M. & Maffettone P.L. (2007) A new methodology for the estimation of drop size
distributions of dilute polymer blends based on LAOS flows, J. Non Newt. Fluid
Mech. 143, pp. 48-58 ISSN: 0377-0257
Guido S.; Grosso M.; Maffettone P.L. (2004). Newtonian drop in a Newtonian matrix
subjected to large amplitude oscillatory shear flows, Rheologica Acta, 43, pp. 575-583
ISSN: 0035-4511
Hatzikiriakos, S. G., Dealy J. M. (1991) Wall slip of molten high density polyethylene. I.
Sliding plate rheometer studies. Journal of Rheology 35(4), 497-523 ISSN: 0148-6055
www.intechopen.com
Huitric J., Mederic P., Moan M., & Jarrin J. (1998) Infuence of Composition and morphology
on rheological properties of polyethylene/polyamide blends, Polymer 39(20) pp.
4849-4856 ISSN: 0032-3861
Hyun K, Wilhelm M (2009) Establishing a New Mechanical Nonlinear Coefficient Q from
FT-Rheology: First Investigation of Entangled Linear and Comb Polymer Model
Systems. Macromolecules 42 pp. 411–422 ISSN: 0024-9297
Jansseune T.; Mewis J.; Moldenaers P.; Minale M. & Maffettone P.L. (2000) Rheology and
rheological morphology determination in immiscible two-phase polymer blends, J.
Non-Newtonian Fluid Mech. 93, pp. 153-165 ISSN: 0377-0257
Lacroix C., Bousmina M., Carreau P.J., Favis B.D. & Michel A.(1996) Properties of
PETG/EVA blends .1. Viscoelastic, morphological and interfacial properties
Polymer 37 (1996), pp. 2939 ISSN: 0032-3861
Macosko W. (1994) Rheology: Principles, Measurements and Applications, Wiley-VCH,
New-York, ISBN: 978-0471185758
Maffettone P.L. and Minale M. (1998) Equation of change for ellipsoidal drops in viscous
flows, J. Non-Newtonian Fluid. Mech., 78, pp. 227-241 ISSN: 0377-0257
Minale M. (2010) Models for the deformation of a single ellipsoidal drop: a review, Rheol.
Acta, 49(8), pp. 789-806, ISSN: 0035-4511
Nam JG, Hyun K, Ahn KH, Lee SJ (2008) Prediction of normal stresses under large
amplitude oscillatory shear flow. J Non-Newtonian Fluid Mech 150:1–10 ISSN: 0377-
0257
Neidhöfer T., Siuola S., Hadjichristidis N., Wilhelm M. (2004) Distinguish Linear from Star
Branched Polystyrene Solutions with Fourier Transform Rheology, Macromol. Rapid
Comm., 25, pp. 1921-1926 ISSN: 1022-1336
Palierne J.F. (1990) Linear rheology of viscoelastic emulsions with interfacial tension, Rheol.
Acta 29 pp. 204-214 ISSN: 0035-4511
Rallison J.M. (1984) The Deformation of Small Viscous Drops and Bubbles in Shear Flows,
Ann. Rev. Fluid Mech. 16, pp. 45-66 ISSN: 0066-4189
Ramirez RW (1985) The FFT Fundamentals and Concepts. Prentice-Hall, Englewood Cliffs,
NJ, ISBN: 978-0133143867
Schlatter G., Fleury G., Muller R. (2005) Fourier Transform Rheology of Branched
Polyethylene: Experiments and Models for Assessing the Macromolecular
Architecture, Macromolecules, 38, pp. 6492-6503 ISSN: 0024-9297
Sigillo I.; Di Santo L.; Guido S.; Grizzuti N. (1997) Comparative Measurements of Interfacial
Tension in a Model Polymer Blend, Polymer Eng. and Sci. 37(9),pp. 1540-1549 ISSN:
0032-3888
H.A. Stone, Dynamics of Drop Deformation and Breakup in Viscous Fluids (1994) Ann. Rev.
Fluid Mech. 26, pp. 65-102 ISSN: 0066-4189
Tucker C.L. and Moldenaers P. (2002) Microstructural Evolution in Polymer Blends, Ann.
Rev. Fluid. Mech. 34, pp. 177-210 ISSN: 0066-4189
Utracki L.A. (2003) Polymer Blends Handbook, Kluwer Academic Publishers, Dordrecht-
Boston-London ISBN: 978-1402011146
Vincze-Minya K.A., Schausberger A. (2007) Characterization of Phase Morphology of
Polymer Melts(PP/PE Blends) Via Rheology, Journal of Applied Polymer Science, 105,
pp. 2294–2298, ISSN: 0021-8995
www.intechopen.com
Wannaborworn S; Mackley M.R; Renardy Y. (2002) Experimental observation and matching

numerical simulation for the deformation and breakup of immiscible drops in
oscillatory shear. Journal of Rheology 46(5), pp. 1279-1293 ISSN: 0148-6055
Wilhelm M, Maring D, Spiess HW (1998) Fourier-transform Rheology, Rheol. Acta, 37, pp.
399-405 ISSN: 0035-4511
Wilhelm M., Reinheimer P., Ortseifer M (1999) High sensitivity Fourier Transform
Rheology. Rheol. Acta 38 pp. 349-356, ISSN: 0035-4511
Yu W, Wang P, Zhou C (2008) General stress decomposition in nonlinear oscillatory shear
flow. Journal of Rheology 53(1):215–238 ISSN: 0148-6055
www.intechopen.com
15
Application of Fast Fourier Transforms in Some

Advanced Electroanalytical Methods
Parviz Norouzi1, Morteza Pirali-Hamedani2,3, Tayebeh Mirzaei Garakani1
and Mohammad Reza Ganjali1
1Center of Excellence in Electrochemistry, Faculty of Chemistry,
University of Tehran, Tehran,
2Department of Medical Chemistry, Faculty of Pharmacy,
Tehran University of Medical Sciences, Tehran,

3Pharmaceutical Sciences Research Center, Tehran,
Iran
1. Introduction
In the most of electrochemical (EC) experiments, measurements mostly are performed in the
time domain. However, in some cases, we require more information for the obtained data
such as knowledge about the frequency content and behavior of the electroanalytical signals
and of complete systems. Fortunately, there exists a defined method for transforming data
from the time domain into the frequency domain, where information exist about the spectral
content of EC data. The method for this propos is Fourier Transform (FT), which has the
ability to convert a time domain data to the complex frequency domain, meaning the
spectral data contains information about both the amplitude and phase of the sinusoidal
components that make up the signal. In addition, the inverse FT, converts the generated
complex frequency-domain signal data back into the time-domain without losing wanted
information. Accordingly, it can say that both the time- and frequency-domain data
complement and the two domains can provide a different view of the same EC data.
Application of fast Fourier transformation (FFT) algorithm for numerical EC data provides
the complex spectrum according to magnitude and phase, which can be used for real time
analysis. In this direction, in modern electrochemistry, FFT has been used for digital signal
processing and filtering. Also, the FFT process returns a vector of real and imaginary
elements, which represent the various resolved harmonics in impedance spectroscopy, AC
and square wave voltammetry (Popkirov, 1996).
On the other hand, it must be noted that in all EC data collection, to hold on the sampling
theorem for FFT, the bandwidth of the input signal is limited by an analog low pass filter
(cutoff frequency fc = fin,max) ahead of the Analog to Digital (A/D) converter. In fact, after
collecting data in the computer memory, they are used for calculating the signal in the
frequency domain.
This chapter serves as summary application of the FFT analysis techniques implemented in
EC measurement platform. By reading through this document, you will receive a
comprehension of the fundamental concepts in FFT-based measurements used throughout
EC application, providing you insights to better understanding of the measurement
www.intechopen.com
parameters, procedures, and resulting EC data. In addition, this chapter describes the
general operation of the FFT analysis accompanied with modern EC methods.
2. Basic FFT theory

The FT, a pervasive and adaptable tool, is used in many fields of science as a mathematical
technique to alter a problem into one that can more easily be solved. Scientists consider FT
theory as a physical phenomenon, not simply as a mathematical tool. Based on the Fourier
theory, any signal in periodic manner in the time domain can be derived from the sum of
sine and cosine signals of different frequencies and amplitudes, which is called as a Fourier
series (Weaver, 1983). Thereby, it is notable to calculate the frequency spectrum of a periodic
signal according to Equation 1,
x (t ) = + ∑ An ⋅ sin(n ⋅ ω0 ⋅ t ) + ∑ Bn ⋅ cos(n ⋅ ω0 ⋅ t )
A0 ∞ ∞
(1)
2 n= 1 n= 1
where, x(t) is data in time domain, An and Bn are the amplitude, wo is the frequency of the
waveform and n is the harmonic number. Each of these elements leads to a discrete
component in the frequency domain, and periodic signals exhibit discrete line spectra.
However, signals with a non-periodic characteristic in the time domain cannot be described
by FT, and those signals exhibit a continuous frequency spectrum with a frequency-
dependency. Therefore, the frequency spectrum of such signals is not composed of discrete
spectral components. The signal in the frequency domain is calculated by means of a FT
(Equation 2).
( ) ∫ x ( t )e
X f ( f ) = F x (t) =
∞
− j 2 π ft
dt (2)
−∞
Also, for measuring the harmonic section of the EC data, it is more useful to examine the
signal in the frequency domain (Rauscher et al., 2001). It has been shown that the signal in
the frequency domain of the fundamental (1st order harmonic) is superimposed by several
higher-order harmonics with the aid of a spectrum analyzer. As a matter of fact, this
information cannot be simply obtained by examining the signal in the time domain.
Practically, the higher order harmonics are not possible, and limited number of the data
samples can be used for FFT calculation.
3. Fundamentals of electroanalytical signals

It is well known that many fundamental microscopic processes take place on the electrode
surface, which can lead to the overall electrical signals. They may include the transport of
electrons through the electronic conductors, the transfer of electrons at the
electrode/electrolyte interfaces to form species which are originated from the cell materials,
and also, the stream of charged atoms. Indeed, the current depends on the ohmic resistance
of the electrodes and the electrolyte and also on the process rates at the electrode/electrolyte
interfaces.
In practical point of view, there are three different types of electrical signals in EC
measurements. Each basic electrical measurement of current (i), resistance (R), and potential
(V) has been used alone or in combination for analytical measurements (Brett & Brett, 1993).
www.intechopen.com
Application of Fast Fourier Transforms in Some Advanced Electroanalytical Methods 305
First, in transient measurements a waveform function of potential may be applied at

electrode surface and then the resulting time varying current measured. The ratio voltage
to current often called the time varying resistance, measures the impedance resulting from
the voltage perturbation at the electrode/solution interfaces. If a FFT is used, a distortion arises
because of the non-periodicity of excitation. Such transformation is only valid when the
applied potential waveform is sufficiently small so that system response becomes linear .
The second type contains signals containing random noise, and measure the resulting
current and voltage, and application of FFT to the results to obtain the frequency domain
data. This process can be used in electrochemical noise analysis (ENA) method for
determination of corrosion. This approach offers the advantage of fast data collection
because only one signal is applied to the interface for a short time .
The last type, the most common and standard one, is to measure EC data by applying a
single-frequency voltage or current to the electrode interface and measuring the phase shift
and amplitude, which leads to measure the real and imaginary parts of the resulting current
at a certain frequency. The most important advantage of such FFT analysis is in AC and
Square Wave Voltammetry (SWV), in which combines the first and third techniques.
4. Application of FFT in electroanalytical methods

As mentioned above, most electrical signals in EC measurements may be examined in the
time domain with the aid of a potentiostat and in the frequency domain with the aid of the
computer digital spectrum analyzer. The two display modes are related to each other, where
each signal variable in the time domain has a frequency spectrum characteristic (Gavaghan
& Bond, 2000). This calculation would be obtained in a continuous data collection, so the
frequency resolution would be unlimited. Noticeably such exact calculations are not
possible in practice, where by given certain prerequisites, the spectrum can be determined
with sufficient accuracy. In practice, the FT data is constructed with the aid of digital signal
processing, as a result, the signal to be analyzed has to be sampled by an A/D converter and
quantized in amplitude.
Most of the generated electrical signals in electroanalytical methods are continuous, in
which for every time value there is a defined data. However, in order these continuous
signals to be analyzed, it is needed a computer-based measurement system employed, such
as Labveiw interfacing program or other interfacing system. Those systems can convert the
EC electrical signal into a stream of digital data, which each data represents a numeric value
that is proportional to the measured data at a specific time. This process is known as data
sampling: converting the analog signals into a discrete-time signal (a process handled by
A/D converter) to allow electroanalytical data in a wide level range to be simultaneously
processed by FFT program and to be displayed on the computer screen (Norouzi et al.,
2003).
In the computing process, the FFT method operates by decomposing N data point in time
domain signal into N frequency domain signals each composed of a single point of the
electroanalytical measurement. The next step is to calculate the N frequency spectra
corresponding to these N time domain signals. Finally, the N spectra are synthesized into a
single frequency spectrum. If N is an integer power of 2, thus N= 2 p (p =1, 2, 3, ….), use the
fastest FT function that uses the decimation-in-time algorithm. Given N samples of a
periodic function f(t) by a normalized sampling rate (T=1), { f(0), f(1), f(2),....f(N-1) }, the
discrete Fourier transform (DFT) is defined by:
www.intechopen.com
F( k) = Fr ( k) + i ⋅ Fi ( k) = ∑ f ( n) ⋅ ⎡⎣ cos ( 2π nk / N ) − i ⋅ sin ( 2π nk / N )⎤⎦

N −1
(3)
n= 0
Normally, at first, an electrode response was recorded and then, DFT was applied on the
collected data and the existing frequencies, phase angle, real and imaginary parts are
calculated. Based on this calculation, the modern methods are established, such as ENA,
FFT Cyclic voltammetry, FFT SW voltammetry and FFT impedance spectroscopy.
4.1 Application of FFT electrochemical measurements based on noise analysis

The frequencies of unwanted signals in EC data, with noise or random characteristics,
cannot be found easily (Aballe et al., 1999; Sang et al., 2009; Dai, 2000). The analyst requires
a plot of the intensity at each individual frequency in order to make identification and the
EC signal to be interpreted. A means of “decoding” for calculating the individual
frequencies is needed. This procedure can be accomplished via a well-known mathematical
technique such as DFT. This transformation is performed by the computer based programs,
which then presents the user the desired spectral information. When the frequency region
of the noise is found, it can be used for two aims in the EC analysis; data filtering for
enhancing signal-to-noise (S/N), and corrosion monitoring (Darowicki & Zieliski, 2001;
Safizadeh& Ghali, 2010).
4.2 Application of FFT in cyclic potential sweep voltammetry based on filtering

In modern EC methods, the FFT analysis voltammetry were developed in order to overcome
some existing limitations encountered with electronic instrumentations. In fact, the
selectivity and percision characteristics of the classic electroanalytical methods depend on
the number of filter circuits. Normally, the most potentiostat typically has many filter
circuits that are arranged before and after the amplifier. The main problem here is to apply a
fast excitation signal for fast EC measurement. Actually, application a fast excitation signal
can produce a large charging current due to existing capacitor in the analog filters.
On the other hand, by using FFT filtering method, instead of the analog filters, the signal can
be measured very quickly. Consequently, the time element per sample is reduced to a
matter of less than second ( 10 −9 s) rather than several minutes which happened in the
classical analog measurements. As cyclic voltammetry finds its greatest use in the study of
fast electrochemical process, and transient intermediates, we are supposed to choose this
method for its description and the FFT application. Through the use of analog filters, the
maximum permissible sweep speed is limited by the transient time of the applied filter. The
maximum span that can be analyzed at a specific resolution by means of FFT is limited by
the sampling rate of the A/D converter and by the memory available for saving the sampled
data. In fact, to allow shorter sweep times, FFT digital filters are advantageous for narrow
resolution bandwidths. Basically, the EC digital analyzer is designed for bandwidths from
100 Hz to 10 MHz. This digital filtration provides condition that can be used for very high
potential scan rates on working electrode for electrochemical measurement in flowing
solution.
Figure 1 shows an example for FFT filtration for fast cyclic voltammetric measurement
(Norouzi et al. 2007). In this method at the beginning, a CV of the electrode was recorded
(see Fig. 1a) and then by applying FFT on the collected data, the existing high frequency
noises were indicated (see Fig. 1b). Finally, by using this information, the cutoff frequency of
www.intechopen.com
the analog filter was set at a certain value (where the noises were removed from the CV). The
resulted CV in Fig. 1c shows successfulness of the filtering procedure. This kind of filtration
and also current integration significantly reduces the noise level in the obtained data.
FFT
b
Cutoff Frequency
Invert FFT
Fig. 1. Application of FFT filtration to smooth a noisy CV, a) original CV, b) FFT spectrum of
the CV (the inset shows the cutoff frequency that is selected for filtration), c) the resulted CV
after removing the noise frequencies
The EC measurement based on the FFTdigital filters is widely used by Norouzi group for
determination of several organic compounds and drugs such as; Diphenhydramine
(Norouzi et al. 2010 b), Lidocaine (Norouzi at al. 2007), Methyldopa (Norouzi et al. 2009),
and Salbutamol(Ganjali et al. 2005). The EC method under reported condition was named
FFT Continuous Cyclic Voltammetry (FFTCCV). In this method the potential waveform was
www.intechopen.com
continuously applied during an experiment run, where the collected data was digitally
filtered by the technique, before using them in calculation for the analyte response (Norouzi
et al., 2010).
It must be noted that, this digital analyzing, also, offers the possibility of the frequency-
dependent gain. Recorded spectra can automatically be displayed with the correct levels. In
these cases a reduction of the displayed noise by decreasing the resolution bandwidth is not
permitted. Due to this fact that, the sensitivity is also important for the fast measurement
speeds the program of the FFT digital analyzer featuring a low noise figure leads to the use
of greater resolution bandwidths, and also with manual setting of the resolution and
bandwidths, the sweep time can be adapted automatically.
An example of application of that waveform is shown in Figure 2. This figure shows a
sequence of CVs recorded during the flow injection of 50 µL of 1.0 × 10–6 M Cl- (in 0.05 M
H3PO4) into the eluent solution containing 0.05 M H3PO4. The potential axis on this graph
represents potential applied to the working electrode during each sweep. The time axis
represents the time passing between the beginning of the flow injection experiment and the
beginning of a particular sweep (i.e. it represents a quantity proportional to the sweep
number). The characteristic element of CVs at gold electrode is a set of peaks associated with
the formation and dissolution of a surface oxide layer at about 1600 and 400 mV (when
potential sweep rate is 20 Vs-1), respectively. The process is also initiated by the
electrosorption of the hydroxyl ion, which at more positive potentials undergoes
deprotonation and structural rearrangement. The surface oxidation can be initiated by
adsorption of water molecule and then at more positive potential AuOH forms leading to
the formation of a two-dimensional phase of gold oxide;
Au ( H 2 O ) → AuOH + e + H+
At more positive potentials we have AuO according to the following reaction;
AuOH → AuO + e + H+
Figure 2b shows the absolute current changes in the CVs curves after subtracting the
average background of 10 CVs (in the absence of the analyte). As can be seen, this way of
presentation of the electrode response gives more details about the effect of adsorbed ion on
currents of the CV. The curves show that current changes mainly take place at the potential
regions of the oxidation and reduction of gold. When the electrode-solution interface is
exposed to Cl − , which can adsorb on the electrode surface, the oxide formation process is
inhibited.
All CV curves observed during the entire experiment (typically 100 to 1000 curves) were
always stored in computer memory and could be saved onto a hard drive for future analysis
(See Fig. 2a). The important point in FFTCCV method is that parts of the adsorbed analyte
surface. In this method, ΔQ is calculated based on the all-current changes at the CV (See Fig.
still remain on the electrode surface that can inhibit the oxidation process of the electrode
2c). A total absolute difference function can be calculated by using the following equation:
⎡ E= Ev ⎤
Δ QTA ( sτ ) = Δ t ⎢ ∑ i ( s , E) − i ( sr , E) + ∑ i ( s , E) − i ( sr , E) ⎥
E= Ei
⎣ E= Ei ⎦
(4)
E= Ev
www.intechopen.com
Current / μA
0,01
0,01
-0,01
-0,01
-0,03
-0,03
80
-400 70
-100
60
200
Po 500 50
te /s
nt 800 40 e
ia m
l/
m
1100
30 Ti
V
20
0,003
nt / μA
0,003
-0,001
Curre
-0,001
-0,005
-0,005
-400 70
-100
60
200
Po 500 50
/s
te 800 40
nt
e
m
ia 1100
l
Ti
/m 30
V 1400
20
0.4
0.35
c
0.3
0.25
0.2
ΔQ / nC
0.15
0.1
0.05
0
-0.05
0 25 50 75 100 125 150 175
Time /s
Fig. 2. a) Cyclic voltammograms at Au ultra-microelectrode recorded during the flow
injection. The eluent was 0.05 M H3PO4 with the flow rate of 0.5mL/min., b) Curves resulted
from subtracting the CVs in fig. a, from the average of 10 CVs (in the absence of analyte),
c) Response of Au ultramicroelectrode to 5 consecutive injections of analyte
www.intechopen.com
Where, s is the sweep number, is the time period between subsequent sweeps, Δt is the
time difference between two subsequent points on the CV curves, i (s, E) represents the
current of the CV curve recorded during the s-th sweep and i (sr, E) is the reference
current of the CV curve. Ei and E are the initial and the vertex potential, respectively. The
reference CV curve was obtained by averaging a few CV curves recorded at the beginning
of the experiment (i.e. before injection of the analyte). These equations show that for the
same flow injection experiment the analyte response can be obtained using different
integration limits.
It should be noted that in this method, all studied processes involve adsorption of analyte;
hence both charging and faradic currents may potentially carry useful analytical
information. To get such information, it was important to sample current at a frequency at
least two times higher than the current transducer bandwidth. In order to fulfill this
requirement the sampling frequency was always adjusted at 100 kHz. In addition FFT
digital low pass filters with 0.5-30 kHz cutoff frequencies lead to remove noise from the
data. If the main contribution to the baseline noise is from the “white” noise generated by
the potentiostat, the integration procedure usually provides a 3 to 60 fold improvement in
S/N compared to the simple monitoring of the current at a fixed potential. However, in the
case of severe environmental noise (e.g. power line noise) the improvement may be much
larger. Signals with weak level are thus shown more distinctly in the voltammogram and
the measured level values are thereby stabilized and reproducible. In the case of a
sinusoidal signal, the displayed level is not influenced by a reduction of the bandwidth. To
obtain stable and reproducible results of noise measurements, a narrow bandwidth should
be selected. The noise bandwidth is thus reduced and high noise peaks are better to be
averaged.
5. Application of FFT in electrochemical noise analysis

As mentioned above the identification of noise frequencies in the electrochemical data by
FFT method, can be used for study of some processes that occur on the electrode surface.
This type of noises, which are created by EC process, called electrochemical noise (EN). For
many years, EN has been observed during corrosion and other electrochemical reactions,
and the phenomenon is well established.
The theoretical behavior of EN is not, completely, discussed, but there have been many
useful applications, both in scientific surface and in corrosion monitoring studies. In an EC
system, noise of potential and of current can be made independently or together during
corrosion reaction that can be take placed on the electrode surface. The important point is
that EN measurement technique does not involve any external perturbation of the corroding
system. In fact, the instruments required to perform EN measurements are reasonably
simple, particularly with FFT digital analyzer and modern computer-based data acquisition
techniques. More commonly, the EN curve for potential and current has an appearance
similar to that shown in Figure 3.
Generally, localized corrosion processes tend to give particularly strong EN signals, which
consist of low frequency (< 1 Hz) and small amplitude signals. Those signals are
spontaneously generated by electrochemical reactions occurring at corroding or other
current are of the 1/f type. The range of frequencies depends on the sampling interval Δt
surfaces (Zaveri et al., 2007). However, In this process, both the noise of potential and
(typically 0.5 s) and on the number of readings M of a time record. The maximum and
www.intechopen.com
minimum frequencies that can be analyzed are: fmax = 1 /2Δt and fmin = 1/ Mt). The limiting
value at a frequency approaching zero is defined as the spectral noise resistance.
In practice, spectral noise plots can be obtained only in a frequency range that is more
limited than in EIS. On the high frequency side, the limit is imposed by the instrumental
noise, whereas in the low frequency region, the time of acquisition becomes very long
(Lafront et al., 2010).
In a different approach, the time record of potential or current is converted into a power
spectral density (PSD), which is the distribution of the power in the frequency domain. This
transformation is usually made by means of the FFT method. Figure 4 shows the calculated
power spectral density (PSD =1/fn, where n is a constant). The figure shows that PSD falls
with increasing frequency, giving a straight line on a log-log plot, implying a relationship. In
addition to the sloping 1/fn region of the spectrum, there are often indications of plateaus at
the high and/or low frequency ends of the spectrum. Potential noise frequency relation
changes have been noticed in the form of a component of character 1/f 2, this being
attributed to pit initiation.
Fig. 3. The sample graph for EN measurements of an Al palate in NaCl 0.01 M, top) Current
noise, bottom) Potential noise
The character of PSD changes as a function of frequency depend on the type of corrosion
and has also been the subject of investigations. The area under the curve is the total power
in the signal, and is identical to the standard deviation calculated from the time record.
Thus, as the frequency spectrum moves to higher PSDs, so the rate of reaction may be
expected to increase .The details of the calculation are beyond the scope of this book.
www.intechopen.com
Fig. 4. The sample of PSD graph for EN measurements of an Al plate in NaCl 0.01 M
6. Application of FFT in AC voltammetry

The determination of the characteristics of the EC data system by AC techniques requires
measuring the impedance at various frequencies, to result a frequency spectrum (Bond et al.,
1997). The most dominant application of FFT methods in AC voltammetry is in calculation
of the impedance of the electrode response, which is known as electrochemical impedance
spectroscopy (EIS).
By definition, AC voltammetry (ACV) utilizes a small-amplitude sine wave which is added
to a potential ramp to modulate the current output. In fact, ACV is an extension of classical
linear sweep techniques such as cyclic voltammetry. A DC ramp with a comparatively slow
sweep rate and an AC signal are superimposed and applied to a working electrode, and the
response of the AC current and its phase angle are registered. In general, modulation
potential amplitude up to 20 mV is used; higher amplitudes are not used to specifically
avoid contributions from higher order harmonics. This may be considered as a limiting case
for ACV. In fact, all of the information is intentionally contained in only the lowest order
harmonics. Its main strength lies in the quantitative characterization of electrode processes,
and it can also be used for analytical purposes.
A second general approach has employed AC voltammetry and the mathematics of the FFT.
This technique was introduced and used extensively by (Smith, 1976). In it, the perturbation
signal is composed of a sum of selected sinusoids. This approach, although powerful, is
quite expensive and complex as well, and has not been widely employed. Both experimental
implementation and data analysis employ a small-amplitude excitation waveform.
www.intechopen.com
The applied excitation waveform consists of a fundamental harmonic frequency f0 and a

number of odd harmonics (2n + 1) f0. This arrangement is superior to other excitation
waveforms. All these frequencies are applied at the same time and the data to each frequency
is found by the FFT. Depending on the method used for the data acquisition, different
methods are used for data validation. In case of a sinusoidal signal were applied to a nonlinear
system, the data function would contain multiples harmonics of the excitation waveforms.
The techniques commonly employed for AC-impedance measurements in modern
equipment can be subdivided into two main groups: single-sine and multiple-sine methods.
The lock-in technique and frequency-response analysis will be described as representatives
of the single sine techniques and FFTs will be introduced as an example for multiple-sine
techniques. In single-sine methods, a small-amplitude sinusoidal signal with a fixed
frequency is applied to the test cell. The response signal is then analyzed to extract the two
components of the impedance (real and imaginary parts or magnitude and phase).
Here, invalid impedance data cannot only be caused by nonlinearity but very frequently by
a lack of stability of the system under investigation. Impedance test equipment usually
comprises an AC measurement unit and a potentiostat or galvanostat. For many
applications, such as biomedical investigations or the characterization of thin films in which
it is not essential to maintain a DC-voltage level during the impedance measurement, a
potentiostat is not required (Házì et al., 1997).
EIS is frequently used to characterize systems that are changed during the time. Another
way to reduce the effects of a system changing during the measurement is to reduce the
total measurement time by using a multi-sine technique, which is FFT method. Figure 6
shows typical curve obtained by this technique for electrodeposition of Cd on a gold
electrode in 0.1 M Cd SO4.
In the case of multi-sine techniques, a measurement is carried out at several frequencies
simultaneously. The phases of the superimposed signals are randomized to minimize the
amplitude of the composite signal (see Fig. 5b). In contrast to single-sine techniques, multi-
sine techniques do not require waiting for a full cycle to be completed for each of the
frequencies used. The time-domain signals are digitized and transferred into the frequency
domain by carrying out the FFT. The resulting data for each discrete frequency can be
treated the same way as the impedance data obtained with a single sine technique. Repeated
application of the waveform and averaging of the signal before FFT is applied can improve
the S/N ratio of the multi-sine technique, although it increases the measurement time
required. The impedance data obtained in this manner can be presented in several different
formats.
Impedance spectra Zreal versus Zimg measured at a number of fixed DC potentials are
suitable for quantitative studies of redox reaction kinetics whereas potential-dependent
admittance values (1/Z) obtained at fixed frequencies can provide AC voltammograms that
are more readily interpreted in electroanalysis. The ratio of the two is the impedance of the
test electrode. The measurement is then repeated at another frequency. The impedance
spectra measured in this work were usually shaped as shown in the following Nyquist- and
Bode-plot. Once real and imaginary parts of the input signals have been determined by
correlation, the complex impedance of the test object can be calculated. It can be
mathematically proven that all the spurious components are rejected by this technique of
correlation provided that a sufficiently large number of cycles have been used for the
integration (Garland et al., 2002).
www.intechopen.com
20 20
18 18
a 16
16
14 14
m
12 12
Zimg /koh
10 10
8 8
6 6
4
4
2
2 0
0
-500
1
0
0.5
500
0
z
Po
H
ten 1000
)/
-0.5
q.
tia
re
l /m 1500 -1
(F
V
g
Lo
-1.5
-100 -100
-120
-120 b -140
ift / deg
-140
-160 -160
-180 -180
-200 -200
-220
Phase sh
-220
-240 -240
-260 -260
-280
-280 -300
-300
-500
1
0
0.5
500
z
H
0
Po
)/
ten 1000
q.
-0.5
tia
re
l /m
(F
1500 -1
V
g
Lo
-1.5
Fig. 5. Typical FFT EIS graph for electrodeposition of Cd on gold electrode, a) Z imaginary,
b) phase shift, changes in deterrent frequency and potential
www.intechopen.com
The advantage of modern FFT technique is that the information is obtained quickly;
therefore it may be used to study impedances evolving with time. The limitation of the FFT
technique is that the response to individual frequencies is usually weaker than that when
only one frequency is used. It should be added that other types of analysis of system
responses were also used, for example, Laplace transform of the applied perturbation and
the response to determine the impedance spectra (Carstensen et al., 2008). The time-domain
signals are digitized and transferred into the frequency domain by carrying out a FFT. The
resulting data for each discrete frequency can be treated the same way as the impedance
data obtained with a single sine technique.
In modern EIS analysis, lower frequency data are usually measured in the time domain.
The current response is then measured using an A/D computer. In this case the FFT is used
to convert the current signal into the frequency domain as carried out for other techniques.
The FFT capabilities have subsequently been incorporated into several commercial
instruments, primarily to speed up the acquisition of impedance data at low frequencies by
exploiting the multiplex character of the technique. Such determinations are normally
carried out at a single, fixed DC potential. In order to obtain potential-dependent
impedance data, repeated experiments at different applied DC potentials are therefore
required (Arundell et al., 2004).
The use of the FFT in combination with a controlled sequence of potential steps or pulses
has been shown to offer an approach by which these time-consuming repetitions can be
avoided and impedance measurements can be collected over a wide range of frequencies
and DC potentials in a single experiment. However, here, the voltammetric waveforms
composed of a sequence of potential steps are ideally suited in mathematical modeling
based on the techniques of numerical integration. This approach is elegant in its generality,
can be made arbitrarily precise, and is extremely efficient (Baranski et al., 1996). Baranski
developed a technique in which a small amplitude square-wave potential perturbation is
superimposed upon a potential staircase and in which the FFT is employed to convert
current measurements taken as a function of time during several cycles of the square wave
at each step potential to the frequency domain. Higher harmonics can be detected by FRAs
or lock-in amplifiers, which can be tuned to detect a multiple of the excitation frequency. An
alternative is to extract the harmonic signals from the response using FFT.
7. Application of FFT in SWV

The idea of obtaining electrode admittance from transient current time curves was
investigated previously by Pilla (Pilla, 1972). The speed and sensitivity of SWV is its main
advantage (Osteryoung & O’Dea, 1986). In the last years, data acquisition boards in the
electroanalytical instrumentation have improved significantly for carrying out SW
voltammetric analysis. The SWV has been used for study of the kinetics of electrode process
(Winston et al., 1988). Nevertheless such applications of this method are relatively rare,
which may be as a result of the rather complicated equations relating the current response of
the electrode to the kinetic parameters of the electrode processes. Also, the theory of SWV
does not take into account the effect of uncompensated solution resistance and a distortion
of the EC signal by a slow response of a current transducer. These problems are not easy to
be solved in any time domain voltammetric techniques. Because the electrode response
under AC voltammetric conditions is represented as an admittance (i.e. in the frequency
www.intechopen.com
domain), all data manipulations needed for obtaining kinetic information are relatively
simple.
The principles of this technique are simple. Normally, FFTSWV experiments are done
under conditions identical to the traditional SWV, but the electrode response at each DC
potential is converted into the frequency domain via FFT. Therefore, FFTSWV measures
the admittance of the electrode as a function of potential (Baranski & Szulborska, 1994).
The resultant data are almost similar to those obtained under AC voltammetric
conditions.
Indeed, in comparison with traditional AC voltammetry the equipment is much simpler
and less expensive, measurements are carried out much faster and it is possible to obtain
information about the admittance of the electrode at different frequencies from a single
run.
The potential waveform used in the FFTSW voltammetric measurement consists of many
SW pulses were superimposed on a staircase potential function, which was changed by a
small potential step of ΔE (Norouzi et al., 2008). The values of potential pulse of SW (ESW)
and ΔE were in a range of few mV (10 to 50 mV). In the computer program, the number of
SW cycles, Nc, in each staircase potential step was calculated based on the SW frequency as
follows, Nc= f0 ⁄1400Hz, for f0 >1400Hz, and Nc=1 for f0 ≤1400Hz. The values of Nc, fo, Esw,
Einitial and Evertex were the variable parameters of the technique, which were optimized for
achieving to best detector performance. It should be noted that in this method all processes
studied involve adsorption of analytes hence both charging and faradic currents may
potentially carry useful analytical information.
To get such information, it was important to sample data current at a frequency at least two
times higher than the current transducer bandwidth. In order to fulfill this requirement the
data sampling frequency was always adjusted between 50 and 100 kHz (depending on scan
rate). In addition a second order low pass filter with a 20 kHz cutoff frequency was placed
between the current output of the potentiostat and the data acquisition board. In the
computer program, the discrete FFT analysis was used for data processing. If one SW cycle
per potential step is applied, the time domain response resulted with this method is similar
to that which obtained using Osteryoung SW voltammetry. Here either four data points per
SW cycle were collected. If there is more than one cycle at one potential step, the current
recorded in different cycles at the same DC potential is averaged (i.e. the first data points in
every cycle are added together and divided by the number of cycles then the second and
subsequent data points are treated in the same way) (Baranski & Szulborska, 1994).
The first component in Eq. 5 gives the imaginary part (Zimg) of the impedance and the
second part gives its real component (Zrel). A full discussion for the determination of Zrel
and Zim based on the sampled currents (Is) will be given in the next section. Theoretically,
the detector impedance,
Z = Zimg
2
+ Zrel
2
(5)
where|Z| in a specific frequency is equal to E/I.

Application of discrete FFT analysis on the sampled current requires a specific method in
current sampling. The admittance of the electrode is calculated at each potential step by the
DFT method. High frequency components are removed by placing an analog low pass filter
www.intechopen.com
between the current transducer and the A/D converter (Van Valkenburg, 1982). It required
the number of sampled currents at each pulse cycle which must be represented by 2 n
(where n is an integer and greater than 1). Therefore the currents, Is, were sampled at even
time intervals, ts,
ts = 1 +
s
(6)
4 f0
where s is an integer number and changes from 0 to 7. Therefore if currents are sampled at
even time intervals, ts, ts+1/4f0, ts+2/4f0 and ts+3/4f0, then the values of the sampled currents
will be,
i0 = ∑ An sin(2nπ f0tc − φn ) (7)

n= 1
i1 = ∑ An sin(nπ / 2 + 2 nπ f0ts + φn ) (8)

n= 1
i2 = ∑ An sin(nπ + 2 nπ f0ts + φn ) (9)

n= 1
i3 = ∑ An sin(n3π / 2 + 2 nπ f0ts + φn ) (10)

n= 1
The equations show that in the first harmonic (n=1) the current components i0 and i2 (as
well as, i1 and i3) have a phase shift equal to π. However, their absolute values are the
same with an opposite sign. As mentioned above, the currents were sampled four times
per SW cycle, i0, i1, i2 and i3. In each step, ΔE, of staircase potential ramp, the total sampled
currents were 4Nc (Nci0, Nci1, Nci2 and Nci3), which were reduced to four by averaging
each Nci. Because of dependence of Nc on frequency, at SW frequencies lower than 1400
Hz lager number of currents were averaged, which could be helpful for reducing the
noise level. At the end of each potential ramp, the data were stored in an array matrix as
follows,
⎧i01 E3 ⎫
⎪ ⎪
i11 i21 i31 E0 E1 E2
⎪⎪ . . . . . . . . ⎪
⎪
Data array = ⎨ . . ⎬
⎪. . ⎪
. . . . . . (11)
⎪ ⎪
. . . . . .
⎪⎩i0n i1n i2n i3n E0 E1 E2 E3 ⎪⎭
where n is the number of the potential step and E0 to E4 are the electrode potentials at which
the current is sampled.
To calculate the admittance of the detector response, first the real and imaginary
components of the alternating current need to be calculated. The real component of I’ and E’
are given by,
www.intechopen.com
I’= i2 –i0 (12)
E’ =E2 –E0 =-2Es (13)

and the equation for the imaginary components are,
I” = i1 –i3 (14)
E” =E1 –E3 =2Es (15)

Now, the real, Y’ , and imaginary, Y” , components of the detecting admittance can be
calculated as follows,
I '− jI " I "− I '− j ( I '+ I " )

Y ' − jY " = =
E '− jE "
(16)
4ES
and
i0 + i1 − i2 − i3
Y' =
e0 + e1 − e2 − e3
(17)
i0 − i1 − i2 + i3
Y"=
e0 + e1 − e2 − e3
(18)
Also, the average current, Is at each potential step is given by
i0 + i1 − i2 − i3
Is =
(19)
4
The results from application of FFT analysis showed that measurement based on the first
harmonic component offered better detection limits. Therefore, the components at higher
frequencies than fundamental frequency were removed from the current response. The
filtration was initially done by utilizing analog filters. A series of low pass filters were
located before the A/D converter board. However, such filter may cause a small distortion
in the magnitude and phase of the fundamental harmonic (which can be determined by
calibrating). Also, a digital filtration was occurred during data acquisition. If the excitation
potential and the electrode response can be represented by periodic functions, the electrode
admittance can be calculated. Briefly in order to avoid problems with the interpretation of
the electrode admittance, it is necessary to remove all components of the electrode
response at frequencies higher than half the data acquisition frequency. This condition
is known as the Nyquist sampling theorem (Weaver, 1983).
The examples of the SW voltammetric responses on the Au UME in the FIA measurement
are shown in Figure 6. The analyte signal appears as a current decline in certain potential at
the SW voltammogram. It results the inhibition of the electrode surface processes by the
analyte adsorption process (Norouzi et al., 2009). To visualize the dependence of the analyte
signal to the electrode potential in Figure 6a, the differential form of the SW voltammograms
are shown (Figure 6b). In the differential graphs, also, it can be noted that the analyte signal
extends over a potential range of the SW voltammogram.
www.intechopen.com
0.06 0.06
μ AV-1
0.05 a 0.05
0.04 0.04
Admittance/
0.03 0.03
0.02 0.02
0.01 0.01
0 0
1100 100
800 90
po 500 80
ten 200
tia
l /m -100 70 e/s
m
V
-400 60 Ti
-0.025 -0.025
-0.02 b -0.02
-1
Admittance/ AV
-0.015 -0.015
μ
-0.01 -0.01
-0.005 -0.005
0 0
0.005 0.005
1100 100
800
po 90
ten 500
200 80
tia
l/m
V -100 70 e/s
m
-400 60 Ti
Fig. 6. a) FFT SW voltammograms at Au ultra-microelectrode recorded during the flow

injection. The eluent was 0.05 M H3PO4 with the flow rate of 0.5mL/min., b) Curves
resulted from subtracting the SWs in fig. a, from the average of 10 SWs (in the absence of
analyte)
www.intechopen.com
8. Conclusion
Electroanalytical techniques can offer rapid and low cost analysis of electroactive
compounds and heavy metal ions in aqueous systems with a parts-per-billion sensitivity
range. Electrical signals may be examined in both the time and in the frequency domain. The
two display modes are related to each other by FTT, so each signal variable in the time
domain has a characteristic frequency spectrum and vice versa. In the FFT based EC method
(such as ENA, FFT Cyclic voltammetry, FFT SW voltammetry and FFT impedance
spectroscopy), initially, an electrode response was recorded. Then, FFT was applied on the
collected data and the existing high frequency noises were indicated. Based on this
information, the cutoff frequency of the analog filter was set at a certain value (where the
noises were removed from the electrode response). The smoothing function is, effectively, a
moving average filter which is applied to the transfer function data before it is displayed in
order to minimize the presence of jagged edges and discontinuities in the displayed data.
Finally with the aid of this function a displayed trace can be smoothed by averaging over
several electroanalytical measurements. Therefore the signal at the spectrum analyzer input
may give rise to unwanted components which do not show any relationship to the input
signal.
Some of the major advantages of FFT-voltammetry over other electrochemical techniques
•
include:
Speed: Because all of the frequencies are measured simultaneously, most measurements
•
by FFT-voltammetry are made in a matter of nano seconds rather than several minutes.
Sensitivity: Sensitivity is dramatically enhanced with FFT-voltammetry for many
reasons. The detectors employed are much more sensitive, the electrical throughput is
much higher which results in much lower noise levels, and the fast scans enable the co-
addition of several scans in order to reduce the random measurement noise to any
desired level (referred to as signal averaging).
Finally, the sensitivity and accuracy of electroanalytical methods based on FFT, along with a
wide variety of software algorithms, have dramatically increased the practical use of
voltammograms for quantitative analysis. Quantitative methods can be easily developed
and calibrated and can also be incorporated into simple procedures for routine analysis.
Thus, the FFT-electroanalytical techniques have brought significant practical advantages to
other electroanalytical methods. It has made possible the development of many new
sampling techniques which were designed to tackle challenging difficulties which were
impossible by older technologies. It has made the application of electroanalytical analysis
virtually limitless.
9. References
Aballe, A., Bethencourt, M., Botana, F.J. & Marcos, M. (1999). Using wavelets transform in
the analysis of electrochemical noise data. Electrochim. Acta, 44, 26, 4805-4816.
Arundell, M., Patel, B.A., Yeoman, M.S., Parker, K.H. & O'Hare, D. (2004) Hilbert transform
of voltammetric data . Elecrochem. Commun., 6, 4, 366-372.
Baranski, A., & Szulborska, A. (1994) Fourier transform square-wave voltammetry. J.
Electroanal. Chem., 373, 157-165.
Baranski, A.S., Norouzi, P. & Nelsson, L.J. (1996). Proc. Electrochem. Soc., 9, 41.
www.intechopen.com
Bond, A.M., Schwall, R.J. & Smith, D.E. (1977). On-line FFT faradaic admittance
measurements application to A.C. cyclic voltammetry. J. Electroanal. Chem., 85, 2,
231-247.
Brett, C.M.A. & Brett, A.M.O. (1993). Electrochemistry: Principles, Methods, and
Applications, Oxford University Press, Oxford.
Carstensen, J., Foca, E., Keipert, S., Foell, H., Leisner M. & Cojocaru, A. (2008) New modes of
FFT impedance spectroscopy applied to semiconductor pore etching and materials
characterization . Phys. Status Solidi A, 205, 11, 2485-2503.
Dai, Y. (2000). The time–frequency analysis approach of electric noise based on the wavelet
transform. Solid-State Electron., 44, 12, 2147-2153.
Daneshgar, P., Norouzi, P., Ganjali, M.R. & Zamani, H.A. (2009). Ultrasensitive flow-
injection electrochemical method for detection of anticancer drug tamoxifen.
Talanta, 77, 3, 1075-1080.
Darowicki, K. & Zieliski, A. (2001). Joint time–frequency analysis of electrochemical noise. J.
Electroanal. Chem., 504, 2, 201-207.
Ganjali, M.R., Norouzi, P., Ghorbani, M. & Sepehri, A. (2005). Fourier transform cyclic
voltammetric technique for monitoring ultratrace amounts of salbutamol at gold
ultra microelectrode in flowing solutions. Talanta, 66, 5, 1225-1233.
Garland, J.E., Assiongbon, K.A., Pettit, C.M., Emery, S.B. & Roy, D. (2002) Kinetic analysis of
electrosorption using fast Fourier transform electrochemical impedance
spectroscopy: underpotential deposition of Bi3+ in the presence of coadsorbing
ClO4− on gold. Electrochim. Acta, 47, 25, 4113-4124.
Gavaghan, D.J. & Bond, A.M. (2000). A complete numerical simulation of the techniques of
alternating current linear sweep and cyclic voltammetry: analysis of a reversible
process by conventional and fast Fourier transform methods. J. Electroanal. Chem.,
480, 1-2, 133-149.
Házì, J., Elton, D.M., Czerwinski, W.A., Schiewe, J., Vicente-Beckett, V.A. & Bond, A.M.
(1997) Microcomputer-based instrumentation for multi-frequency Fourier
transform alternating current (admittance and impedance) voltammetry. J.
Electroanal. Chem., 437, 1-2, 1-15.
Lafront, A.M., Safizadeh, F., Ghali, E. & Houlachi, G. (2010) Study of the copper anode
passivation by electrochemical noise analysis using spectral and wavelet
transforms. Electrochim. Acta, 55, 7, 2505-2512.
Norouzi, P., Ganjali, M.R. & Meibodi, A.S.E. (2008) A novel adsorptive square wave
voltammetric method for pico molar monitoring of lorazepam at gold ultra
microelectrode in a flow injection system by application of fast Fourier transform
analysis. Anal. Lett., 41, 7, 1208-1224.
Norouzi, P., Ganjali, M.R., Daneshgar, P., Dinarvand, R., Moosavi-Movahedi, A.A. &
Saboury, A.A. (2007). Development of fast Fourier transform continuous cyclic
voltammetry at Au microelectrode in flowing solutions as a novel method for sub-
nanomolar monitoring of lidocaine in injection and biological fluids. Anal. Chim.
Acta, 590, 1, 74-80.
Norouzi, P., Ganjali, M.R., Nouryousefi, E., Shahtaheri, S.J. & Dinarvand, R. (2010). Trace
Detection of Diphenhydramine by Adsorption on a Microelectrode at Flow
Injection System by Fast Fourier Transform Continuous Cyclic Voltammetry .
Croatica Chem. Acta, 83, 2, 135-142, b.
www.intechopen.com
Norouzi, P., Ganjali, M.R., Shahtaheri, S.J., Dinarvand, R. & Hamzehpoor, A. (2009).
Monitoring of Methyldopa by Fast Fourier Transform Continuous Cyclic
Voltammetry at Gold Microelectrode .Chinese J. Chem., 27, 4, 732-738.
Norouzi, P., Garakani, T.M., Rashedi, H., Zamani, H.A. & Ganjali M.R. (2010) Ultrasensitive
Flow-Injection Electrochemical Method Using Fast Fourier Transform Square-Wave
Voltammetry for Detection of Vitamin B-1. Int. J. Electrochem. Sci., 5, 5, 639-652, c.
Norouzi, P., Rashedi, H., Mirzaei Garakani, T., Mirshafian, R., Ganjali, M. R. (2010). Fast
Fourier transformation with continuous cyclic voltammetry at a Pt-Au dual
microelectrode for the determination of Chloramphenicol in a flow injection
system, Int. J. Electrochem. Sci., 5, 377 – 39, a.
Osteryoung, J. & O’Dea, J.J. (1986). Electroanalytical Chemistry, in Bard, A.J., (ed.) Vol. 14,
Dekker, New York, p. 209.
Pilla, A.A. (1972). Computers in Chemistry and Instrumentation, in Mattson, J.S., Mark Jr.,
H.B. & MacDonald Jr., H.C. (eds.) Vol. 2, Dekker, New York.
Popkirov, G.S. (1996). Fast time-resolved electrochemical impedance spectroscopy for
investigations under nonstationary conditions. Electrochem. Acta, 41, 7-8, 1023-1027.
Rauscher, C., Janssen V. & Minihold R. (2001). Fundamentals of Spectrum Analysis, First
edition, Rohde & Schwarz GmbH & Co. KG, Mühldorfstrasse 15, 81671 München,
Germany.
Safizadeh, F. & Ghali E. (2010). Monitoring Passivation of Cu-Sb and Cu-Pb Anodes During
Electrorefining Employing Electrochemical Noise Analyses. Electrochim. Acta, In
Press, Accepted Manuscript, Available online 21 September.
Sang, Y.F., Wang, D., Wu, J.C., Zhu, Q.P. & Wang, L. (2009). The relation between periods’
identification and noises in hydrologic series data. J. Hydrol., 368, 1-4, 30 165-177.
Smith, D.E. (1976) Anal. Chem., 48, 221A-517A.
Van Valkenburg, M.E. (1982). Analog Filter Design, Holt, Rinehart and Winston, New York.
Weaver, H.J. (1983). Applications of Discrete and Continuous Fourier Analysis, Wiley, New
York.
Weaver, H.J. (1983). Applications of Discrete and Continuous Fourier Analysis, Wiley, New
York.
Winston, S., O’Dea, J.J., & Osteryoung, J. (1988). Square wave voltammetry for the
determination of kinetic parameters: The reduction of zinc(II) at mercury
electrodes. J. Electroanal. Chem., 255, 21-44.
Zaveri, N., Sun, R., Zufelt, N., Zhou, A. & Chen, Y.Q. (2007). Evaluation of microbially
influenced corrosion with electrochemical noise analysis and signal processing.
Electrochim. Acta, 52, 19, 5795-5807.
www.intechopen.com
16
Fourier Transforms Infrared Spectroscopy

Applied in Selective Catalytic Reduction
of NO by Acetylene
Xinping Wang
Dalian University of Technology
China
1. Introduction
Selective catalytic reduction of NO by hydrocarbons (HC-SCR) is believed to be one of the
most promising ways to remove nitric oxide from the exhaust gas of diesel and lean-burn
engines. Since HC-SCR was reported individually by Iwamoto and Held groups (Iwamoto
et al., 1999; Held et al., 1990), many studies were carried out in this field both on new
catalyst research and on reaction mechanism. In these studies, Fourier Transforms Infrared
Spectroscopy (FTIR) was extensively used for interpreting the relationship between surface
structure and the catalytic performance of the catalysts, especially, for disclosing the
reaction mechanism over the catalysts. In this chapter, we summarize the FTIR studies used
in the investigation of selective catalytic reduction of NO by acetylene (C2H2-SCR).
2. To characterize surface acidity of the catalyst

For HC-SCR, many zeolites have been investigated as catalysts. However, it seems that only
limited types of zeolites including ZSM-5 (Li et al., 2004; Li et al., 2008; Wang et al., 2007a;
Wang et al., 2007b; Wang et al., 2005; Niu et al., 2006), ferrierite (Seijger et al., 2003; Lee et al.,
2003; Kubacka et al., 2005; Kubacka et al., 2006) and mordenite (Pieterse & Booneveld, 2007;
Mosqueda-Jiménez et al., 2003a; Córdoba et al., 2005; Lónyi et al., 2007; Dorado et al., 2005),
which are called as pentasil zeolites, displayed high selectivity to NOx reduction in HC-SCR.
For instance, Pieterse et al. (Pieterse et al., 2003) investigated the selective catalytic reduction
of NO by methane (CH4-SCR) over several zeolite-based catalysts and reported an activity
order of Co-Pd-ZSM-5 > Co-Pd-MOR > Co-Pd-FER > Co-Pd-BEA. Abreu and his co-workers
(Torre-Abreu et al., 1999) found that zeolites modified by Cu used for selective catalytic
reduction of NO by propene (C3H8-SCR) are ordered in NO conversion at 400 oC with
CuMFI (58%) > CuMOR (43%) > CuY (20%). Sultana et al. (Sultana et al., 2008) reported that
zeolites are ordered with Pt-MOR (90%) > Pt-FER (77%) > Pt-ZSM-5 (74%) > Pt-BEA (70%) >
Pt-HY (63%) in terms of NO conversion for selective catalytic reduction of NO at ~300 oC
using diesel as reductant. Shibata et al. (Shibata et al., 2004) reported an order of Ag-MFI
(56%) > Ag-BEA (38%) > Ag-MOR (11%) >> Ag-Y (2%) in NO conversion at ~300 oC for the
C3H8-SCR assisted by H2. It was also reported that Co-ZSM-5 (Ivanova et al., 2001) and Ni-
ZSM-5 (Mihaylov et al., 2004) are more active than the corresponding Y zeolite promoted by
www.intechopen.com
Co or Ni for the CH4-SCR. All these results indicated that Y is inferior to the pentasil zeolites
for the HC-SCR.
100
80
NO conversion to N2 (%)
60
40
20
0
200 250 300 350 400 450
Temperature (oC)
Fig. 1. Conversion of NOx as a function of temperature over HZSM-5 (Æ), HFER (■), HMOR
(t) and HY (æ). Reaction conditions: 1600 ppm NO + 800 ppm C2H2 + 9.95% O2 in He with
a total flow rate of 50 ml/min over 0.2 g of catalyst
1540
1450
a
Absorbance
1700 1600 1500 1400

-1
W av en u m b er (cm )
Fig. 2. FTIR spectra of pyridine adsorbed on HZSM-5 (a), HFER (b), HMOR (c) and HY (d) at
500 oC in evacuation
www.intechopen.com
Fourier Transforms Infrared Spectroscopy Applied in
Selective Catalytic Reduction of NO by Acetylene 325
On the other hand, Shichi and co-workers (Shichi et al., 1998; Shichi et al., 2000; Shichi et al.,
2001a; Shichi et al., 2001b; Shichi et al., 2004) have emphasized intracrystalline diffusion of
reductants limited by the zeolites’ channels, based on their investigation results. They found
that NO conversion over MFI and MOR appeared to be significantly influenced by both
hydrocarbon molecular size and zeolite particle size in some instances. In our investigation
of HC-SCR using acetylene as reductant (C2H2-SCR), we found that the C2H2-SCR over
HZSM-5 is greatly affected by the intracrystalline diffusion of NO2 as well (Wang et al.,
2007c). Thus, we suppose that the limited diffusion of the reactants in narrow channels of
the pentasil zeolites is the main reason leading to the low NO conversion at high GHSV in
HC-SCR. Due to wider channels compared to the pentasil zeolites (Elzey et al., 2008), Y
zeolite may be favorable for intracrystalline diffusion of reactants. Besides of this, Y zeolite
is stable in severe hydrothermal conditions and economical (Furusawa et al., 2002).
Therefore, the zeolite could be expected to be a candidate for preparing a practical catalyst
for HC-SCR working at high GHSV, if Y can be effectively modified, due to its larger pore
size (0.74 nm × 0.74 nm) than HZSM-5 (0.53 nm × 0.56 nm), HFER(0.42 nm × 0.54 nm) and
HMOR (0.65 nm × 0.70). It lead us to study the zeolite in C2H2-SCR investigation. However,
a result in contrast with the expectation was obtained in the C2H2-SCR. HY displayed much
low activity compared to HZSM-5, HFER, and HMOR, as shown in Fig 1. To answer the
question why HY with the ideal larger pore size exhibited much inferior catalytic
performance to the pentasil zeolites in the HC-SCR, we studied the surface acidity of HY in
comparison with the pentasil zeolites (Ma et al., 2009). Fig.2 shows FTIR of pyridine
adsorbed on different zeolites, obtained by pyridine adsorption over the zeolites and a
degassing at 500 oC. Compared to HZSM-5, HFER and HMOR, HY gave considerable
weaker IR band at 1540 cm-1 being associated with pyridine adsorption on protons,
indicating that strong Brönsted acids over HY are much less than those of the pentasil
zeolites in amount. In literature, zeolites with Na or H form were usually used in HC-SCR
investigation as catalyst or support, and opposite results were obtained by the authors on
the acidity or basicity of the zeolites favorable for HC-SCR activity, for different HC-SCR
catalytic systems. For instance, it was reported that Ag-NaZSM-5 catalyst was more active
than Ag-HZSM-5 for the selective catalytic reduction of NO by methane (CH4-SCR) at 450
°C (Shi et al., 2002), and that Fe-ZSM-5 catalyst with precursor of NaZSM-5 was far more
active than that with precursor of NH4-ZSM-5 for selective catalytic reduction of NO by urea
(Sullivan & Keane, 2005). Similarly, high activity for the selective catalytic reduction of NO
by propene (C3H6-SCR) on Ce-NaZSM-5 (Niu et al., 2006; Seijger et al., 2003) and for
CH4-SCR on Pt-CoNaFER washcoated cordierite monolith (Lee et al., 2003) was obtained.
Whereas, Brönsted acids have been suggested by the other authors to be essential for HC-
SCR over many catalytic systems (e.g. ZSM-5 modified by Pd, Ga, In, Ce and Ag) (Shibata et
al., 2004; Loughran & Resasco, 1995; Kikuchi & Yogo, 1994; Nishizaka & Misono, 1994; Li &
Armor, 1994; Narbeshuber et al., 1997; Berndt et al., 2003; Gutierrez et al., 2005).
The authors found that Brönsted acids contributed to the aimed reaction in different steps.
For instance, Stakheev and coworkers reported that exchange of partial protons by
sodium with a level of 32% resulted in a nearly complete disappearance of the activity of
the zeolite for oxidation of NO to NO2, and a significant decrease of the activity for C3H6-
SCR (Gutierrez et al., 2005). In the C2H2-SCR, we found that the proton form of ZSM-5
based catalyst were much active, whereas the sodium form of ZSM-5 based catalyst
almost inactive for the reaction (Wang et al., 2007a; Wang et al., 2005), as shown in table 1.
We also found that, with SiO2/Al2O3 ratio of HZSM-5 increasing, C2H2-SCR activity of the
www.intechopen.com
HZSM-5 based catalysts (e.g. Ce-HZSM-5, 2%Mo/HZSM-5 and 2%Zr/HZSM-5(Wang et

al. 2008) as shown in table 2) decreased in same reaction conditions. These results
indicated that protons presenting in the zeolites is indispensable for C2H2-SCR and
activity of the catalyst strongly depends on the population of protons in the zeolites. Thus,
one could believe that it is the weaker surface acidity of HY compared to those of pentasil
zeolites which leads to the inferior catalytic performance of HY in comparison with
HZSM-5, HFER and HMOR.
Reaction NO conv. to N2 Total C2H2

Catalysts
temperature (oC) (100%) conv. (100%)
HZSM-5 300 58.9 86.6
NaZSM-5 325 ~0 8.0
Ce-HZSM-5 300 83.0 98.9
Ce-NaZSM-5 300 ~0 15.3
Mo-HZSM-5 350 82.7 100
Mo-NaZSM-5 350 ~0 14.7
Table 1. Catalytic performance of Na-, and HZSM-5 based catalysts in C2H2-SCR*

*Note: All of the catalyst were prepared from ZSM-5 with SiO2/Al2O3 of 25
Reaction conditions: 1600 ppm NO + 800 ppm C2H2 + 9.95% O2 in He with a total flow rate
of 50 ml/min over 0.2 g of catalyst.
Catalyst with Reaction NO conv. to Total C2H2

SiO2/Al2O3 ratio temperature (oC) N2 (100%) conv. (100%)
Ce-25 300 83.0 98.9

Ce-38 300 64.2 100
Ce-50 300 ~0 79
Mo-25 350 82.7 100
Mo-38 350 57.7 100
Mo-50 350 48.0 100
Zr-25 350 89 100
Zr-38 350 73 100
Zr-50 350 66 100
Table 2. Catalytic performance of Ce-HZSM-5, 2%Mo/HZSM-5 and 2%Zr/HZSM-5

prepared from the zeolite with different SiO2/Al2O3 ratio*
*Note: Ce-HZSM-5 was prepared by ion exchange, and 2%Mo/HZSM-5 and 2%Zr/HZSM-5 were
prepared by impregnation
www.intechopen.com
3. To confirm rate-determining step

Over many kinds of catalysts, HC-SCR was believed to be initiated from NO oxidation to
NO2, and the step was generally accepted to be crucial step (Kubacka et al., 2006; Pieterse &
Booneveld, 2007). To elucidate why HY displayed inferior catalytic performance in C2H2-
SCR, activity of HY for NO oxidation to NO2 was compared to those of HZSM-5, HFER and
HMOR. As shown in Fig. 3, all of the pentasil zeolites exhibited considerably higher activity
reaction of NO + O 2 ⎯ ⎯→ NO 2 is catalyzed by acid sites of the zeolite. It was strongly

compared to HY, in particular, in the case of HFER at 250 oC. The results implies that the
supported by experimental results that when sodium form of ZSM-5 was used as catalyst
for the reaction of NO oxidation, almost no catalytic activity could be observed. Combined
the catalytic activity of the zeolites both for NO oxidation and for C2H2-SCR of NO, it can be
found that the curves of NO oxidation to NO2
vs. reaction temperature resemble much to those of NO conversion to N2 in C2H2-SCR in
NO + O 2 ⎯ ⎯→ NO 2 over HY.
patterns, indicating that C2H2-SCR is significantly confined by the step of
70
60
NO conversion to NO2 (%)
50
40
30
20
10
0
200 250 300 350 400 450 500
Temperature (oC)
Fig. 3. Catalytic performance of HZSM-5 (Æ), HFER (▲), HMOR (t) and HY (æ) in
oxidation of NO with O2 at different temperatures. Reaction conditions: 200 ppm NO + 10%
O2 in N2 was fed to 0.200 g at a total flow rate of 100 ml/min
For C2H2-SCR over HZSM-5 and HFER, it can be speculated that the step of NO oxidation to
NO2 may not be the rate-determining step, as the two zeolites are rather active to NO
oxidation, due to their lager amount of protons characterized by FTIR of pyridine
adsorption (Fig. 2). The speculation was confirmed by the following experimental results.
As shown in Fig. 4, although HZSM-5 itself is more active for NO oxidation with O2 at the
reaction temperature ranged from 200~400 oC than all of xY/HZSM-5 (x = 0 ~ 9), it gave a
rather lower NO conversion to N2 compared to 3%Y/HZSM-5, as shown in Fig. 5 (Wang et
www.intechopen.com
al., 2008). To investigate the rate-determining step of C2H2-SCR over these catalysts, and
effectively research more active catalyst for the reaction of C2H2-SCR, several possible step
for the reaction over the catalysts, such as NOx adsorption over the catalysts and the activity
of the nitrous species toward reduction over the catalysts were investigated. At first, FTIR
spectrum arising from nitrous species on HZSM-5 and 3%Y/HZSM-5 after saturated co-
adsorption of NO+O2 was measured at 250 °C, as shown in Fig. 6. Compared to HZSM-5,
3%Y/HZSM-5 gave significantly stronger band at 1585 cm–1 due to bidentate nitrates
(Ivanova et al., 2001; Yu et al., 2004; Shimizu et al., 2001; He et al., 2004; Poignant et al., 2001)
and a new band at 1609 cm–1 caused by NO+O2 saturated co-adsorption at 250 °C. The
results indicated that the nitrous species adsorption capacity of 3%Y/HZSM-5 is much
stronger in comparison with that of HZSM-5. NO- and NO2-TPD on the catalyst samples
gave also the same conclusion. As shown in Fig. 7, substantially larger amounts of NO and
NO2 were measured in the temperature range of 250-550 °C on 3%Y/HZSM-5 in the NOx-
TPD compared to that on HZSM-5. It should be noted that the corresponding nitrous species
which desorbed from the catalyst surface in the temperature range quite seems to be those
reacting with C2H2 and contributing C2H2-SCR of NO, as large parts of them could be
speculated to stay on the catalyst surface at the C2H2-SCR reaction conditions. By combining
the C2H2-SCR activity with the nitrous species adsorption capacity of the catalysts, a
consistent relationship was obtained for Mo-(Wang et al., 2007a), W-(Wang et al., 2007b), Zr-
promoted HZSM-5 as well as Zr-promoted HFER, i.e. by these metals incorporating into
HZSM-5 and HFER, nitrous species adsorption capacity of the catalyst was significantly
enhanced, and at the same time, C2H2-SCR activity of the catalyst was largely improved.
80
70
NO conversion to NO2 (%)
60
50
40
30
20
10
0
0 50 100 150 200 250 300 350 400 450 500 550
Temperature (oC)
Fig. 4. Catalytic activity of xY/HZSM-5 for NO oxidation with O2: x=0 (○), x=0.5 (■), x=3 (♦)
and x=9 (▲). Reaction conditions: 200 ppm NO, 10 % O2 in N2 with a total flow rate of 100
ml/min over 0.200 g catalysts
www.intechopen.com
90
80
70
NO conversion to N2 (%)
60
50
40
30
20
10
200 250 300 350 400 450 500
Temperature (oC)
Fig. 5. NO conversion to N2 as a function of reaction temperature over HZSM-5 (●), and
3%Y/HZSM-5 (▲). Reaction condition: 1600 ppm NO, 800 ppm C2H2, 9.95% O2 in He with a
total flow rate of 50 ml·min-1 over 0.200 g of catalyst
0 .0 5 1585
1609
Absorbance
1637
b
1800 1700 1600 1500 1400 1300

-1
W a v e n u m b e r/c m
Fig. 6. FTIR spectra arising from nitrous species on HZSM-5 (a) and 3% Y/HZSM-5 (b) after
saturated co-adsorption of NO+O2 at 250 oC
www.intechopen.com
120
A
100 A
b
NO concentration (ppm)
80
60
40
a
20 b
0
0 100 200 300 400 500 600
Temperature (oC)
250
B
200
B
NO2 concentration (ppm)
b
150
100 b
a
50
0
0 100 200 300 400 500 600
Temperature (oC)
Fig. 7. TPD profiles of NO (A) and NO2 (B) over HZSM-5 (a) and 3% Y/HZSM-5 (b) after
saturated co-adsorption of NO+O2
www.intechopen.com
90
80
NO conversion to N2/ % 70
60
50
40
30
20
200 300 400 500
Temperature/oC
Fig. 8. The NO conversion to N2 as a function of temperature over HZSM-5 (●),

2%Zr/HZSM-5 (○), HFER (■) and 2%Zr/HFER (□). The reaction conditions are the same as
that in Fig. 5
1629
0.2 1598
a
b
Absorbance
2000 1800 1600 -1

1400
Wavenumber/cm
Fig. 9. Transient FTIR spectra at 250 oC upon HFER (solid curve) and HZSM5 (dashed curve)
exposing to 1000 ppm NO + 10 % O2 in N2 (A) for 1 min (a), 2 min (b), 5 min (c), 10 min (d),
15 min (e) and 30 min (f). For comparison, the transient FTIR spectrum recorded by
exposing 2%Zr/HZSM-5 to 1000 ppm NO + 10 % O2 in N2 at 250 oC for 30 min (dotted
curve) was given as m
www.intechopen.com
These results imply that the rate-determining step for the C2H2-SCR over HZSM-5 and
HFER is the nitrous species formation from NOx adsorption. Due to the restriction in length
for the chapter, here, we will interpret only the rate-determining step of C2H2-SCR over
HZSM-5 and HFER deduced from FTIR study on Zr and Y promoted zeolites.
0.05
1629
1598
Absorbance
HZSM-5
HFER
2%Zr/HZSM-5
2%Zr/HFER
2000 1900 1800 1700 1600 1500 1400 1300

-1
Wavenumber/cm
Fig. 10. FTIR spectra arising from surface nitrate species on HZSM-5, 2%Zr/HZSM-5, HFER,
and 2%Zr/HFER upon the catalyst samples exposing to gas mixture of 1000 ppm NO + 10 %
O2 in N2 for 30 min at 250 oC
The catalytic performance of HFER, HZSM-5, 2%Zr/HFER and 2%Zr/HZSM-5 in C2H2-SCR
was shown in Fig. 8. The conversion of NO into N2 was strongly influenced by the type of
zeolites. The parent zeolite of HFER displayed much higher activity than HZSM-5 for C2H2-
SCR in the temperature range of 250-375 oC. For instance, NO conversion to N2 over HFER
at 250 oC was 69.6%, which is much higher than that over HZSM-5 under the same reaction
conditions. On the other hand, significant doping effect of zirconium incorporation into the
zeolites on C2H2-SCR was observed, especially in the case of HZSM-5. For instance, NO
conversion to N2 was increased to 78.0% from 69.6% at 250 oC by 2% of zirconium
incorporation into HFER, whereas it sharply increased to 56.0% from 25.4% in the case of
HZSM-5. Why HFER was so much active at the lower reaction temperature compared to
HZSM-5? Why the doping effect of zirconium on C2H2-SCR was so much larger for HZSM-5
than HFER? To answer these questions, we tentatively combined their catalytic performance
in C2H2-SCR with their NOx adsorption capacity. As shown in Fig. 9, Two bands at 1598 and
1629 cm-1 due to bidenate and bridging nitrates (Li et al., 2007; Tsyntsarski et al., 2003; Li et
al., 2005a) respectively appeared on HZSM-5 and HFER when the zeolites were exposed to
NO+O2 at 250 oC. By comparing the band increasing in intensity at 1629 and 1598 cm-1 over
the zeolites, it can be known that the nitrate species formation was faster over HFER than
that over HZSM-5. Obviously, the relative nitrate species formation rate coincides well with
www.intechopen.com
the relative C2H2-SCR activity for the two parent zeolites. Similar relationship of C2H2-SCR
activity versus nitrate species formation rate can be also obtained on HZSM-5 and
2%Zr/HZSM-5. The bands due to nitrate species on 2%Zr/HZSM-5 (spectrum m in Fig. 9)
were obviously strong in intensity compared to that obtained at the same exposing time
over HZSM-5. Figure 10 gave the stable FTIR spectrum at 250 oC obtained by exposing
HZSM-5, HFER, 2%Zr/HZSM-5 and 2%Zr/HFER respectively to gas mixture of 1000 ppm
NO + 10 % O2 in N2 until the corresponding spectra no further change. By combining the
results given in this figure with those in Fig. 8, the relative activity of the zeolites and those
promoted by zirconium can be well understood. The lager nitrate species formation capacity
of HFER compared to HZSM-5 may be the reason leading to the higher C2H2-SCR activity of
HFER in comparison with HZSM-5. Accordingly, as 2%Zr/HFER have the further large
nitrate species formation capacity, due to 2% of zirconium incorporation, it displayed more
active than the zeolite itself. Also, the drastically larger C2H2-SCR activity of 2%Zr/HZSM-5
than that of the zeolite itself can be attributed to the substantially increased nitrate species
formation capacity of the resulting material prepared from zirconium impregnation on the
zeolite. To confirm the supposition that nitrate species formation on the catalyst surface is
the rate-determining step for the C2H2-SCR over HZSM-5 and HFER, it must be validated
that the nitrate species are important intermediate of the total reaction, i.e. they are active
toward reduction, which will be discussed in the section 4. It should be also validated that,
the step of nitrate species formation on the catalyst surface is most slow among all of the
steps in the route of C2H2-SCR over the zeolites.
A 1691
1691 B
0.05 1651 1651
0.05 1386
1580 1386
1580
1 min
Evacuation
Absorbance
Absorbance
3 min 1 min
3 min
5 min 5 min
10 min
30 min
30 min
2000 1800 1600 1400 2000 1800 1600 1400

-1 -1
Wavenumber/cm Wavenumber/cm
Fig. 11. FTIR spectra of surface species on HFER at 250 oC upon the fresh zeolite being
exposed to 500 ppm C2H2 + 1000 ppm NO + 10 % O2 in N2 (A) and then to 1000 ppm NO +
10 % O2 in N2 (B)
Note: The cell was evacuated briefly before the zeolite was exposed to NO+O2 in N2
www.intechopen.com
Figure 11 gave the transient in situ FTIR spectrum of surface species on HFER at 250 oC
recorded when the zeolite was exposed to 500 ppm C2H2 + 1000 ppm NO + 10 % O2 in N2 and
then reductant gas of acetylene was cut off thereafter. Although bands at 1598 and 1629 cm-1
due to nitrate species with strong intensity appeared in the spectra recorded when the zeolites
were exposed to 1000 ppm NO + 10 % O2 in N2 (Fig.10), no one of them appeared in the
transient in situ FTIR spectra as well as in the steady spectrum when 500 ppm C2H2 was
introduced into the gas mixture, i.e. in the C2H2-SCR reaction condition. Furthermore, the
nitrate species even could not be detected in the in situ FTIR in the followed period of 30 min
(Fig 11B), even though C2H2 was completely removed out from the gas mixture. These
experimental results indicate that no nitrate species (produced by NO+O2 co-adsorption over
the zeolite) can cumulate on the zeolite surface under the C2H2-SCR reaction condition.
Instead, the species produced by adsorption of acetylene or its derivates obviously cumulated
on the zeolite surface, being characterized by the bands at 1691, 1651 and 1380 cm-1. It means
that formation of the reductant species is faster by far than that required by nitrate species
reduction. In other words, the rate of nitrate species reduction is limited by that of nitrate
species formation. Thus, it can be concluded that the rate-determining step of C2H2-SCR over
the HFER is the nitrate species formation step. Then, it can be further deduced that the C2H2-
SCR reaction over HZSM-5 must be controlled also by this step, as discussed above, nitrate
species formation capacity of HZSM-5 is much lower than that of HFER, and C2H2-SCR
activity enhanced by zirconium impregnation was more significant on HZSM-5 than on HFER.
Finally, it should be noted that, the quantitive evidence obtained by FTIR must be based on a
strict experiment, e.g., the surface species must be measured over the self-supporting wafer
with the same weight. Otherwise, accurate FTIR result may be hardly obtained. To compare
the amount of surface species formed on the catalysts in FTIR, catalyst wafers with 14 ± 0.7 mg
were selected and certain parameters of the IR spectrophotometer were set in our experiments.
4. To identify reaction intermediates

The identification of reaction intermediates is much important not only for investigating the
reaction route over a catalyst, but also for understanding the effect of doping material being
introduced to the catalyst. As motioned above, in the C2H2-SCR investigation, we found that
Mo, W, Zr, Y impregnated on HZSM-5 and HFER significantly enhanced the formation of
nitrate species over the catalyst and correspondingly increased the C2H2-SCR activity. These
results could be well understood by the rate-determining step discussed in the section 3. On
the other hand, we found that, NaZSM-5 is almost inactive for C2H2-SCR, though it can be
considered that NaZSM-5 has much stronger nitrate species formation capacity compared to
HZSM-5. Certainly, one can considered that due to lack of protons and inert for catalyzing
NO oxidation, it can not initiate the aimed reaction. We also found that when NO in the gas
mixture was completely changed to NO2, the NO2 conversion to N2 over NaZSM-5 was still
less much than that over HZSM-5. These results imply that NaZSM-5 hardly catalyzes C2H2-
SCR due to not only inactive for NO oxidation, but also the other inferior action displayed in
the subsequent steps. Thus, the role of protons in the NO reduction by acetylene over
HZSM-5 was studied.
4.1 Characterization of nitrous species on HZSM-5 and NaZSM-5

The nitrous species formation and variation with temperature, as well as their reactivity
towards to reduction by acetylene were investigated by FTIR. The spectra obtained upon
www.intechopen.com
exposing HZSM-5 and NaZSM-5 to NO and O2 at 40 oC were depicted in Fig. 12. On HZSM-5
(Fig.12 A), the bands at 1622 cm-1 due to weakly adsorbed NO2 (Shimizu et al., 2001;
A 1622 1311 B 1389

c 1407
1389
0.2
Absorbance
Absorbance
1622
1715 c
b
1663 b 1311
a
0.1 a
1663
1800 1700 1600 1500 1400 1300 1800 1700 1600 1500 1400 1300
-1 -1
c
3700 C
b
Absorbance
2884
0.05 2477
3800 3600 3400 3200 3000 2800 2600 2400

-1
W avenum ber/cm
Fig. 12. FTIR spectra of surface nitrous species formed upon exposing HZSM-5 (A,C) and
NaZSM-5 (B) to 1000 ppm NO and 10 % O2 in N2 at 40 oC for 1min (a), reaching saturated
adsorption (b), and then a brief evacuation (c)
Pirngruber & Pieterse, 2006) or bridging nitrates (Ivanova et al., 2001; Yu et al., 2004; Sedlmair
et al., 2003b; Li et al., 2005b) and that at 1311 cm-1 due to unidentate nitrates(Mihaylov et al.,
2004; Shimizu et al., 2001; He et al., 2004) with three negative bands at 3700, 2884 and 2477
cm-1 were observed upon exposing the zeolite to NO+O2 for 1 min. The band at 3700 cm-1 is
due to asymmetric stretching vibration of molecular adsorbed water (Nakamoto, 1996;
www.intechopen.com
Hadajiivanov et al., 1998), and the bands at 2884 and 2477 cm-1 are due to the well-known A-
B-C structure produced by hydrogen-bonded hydroxyls (Mihaylov et al., 2004; Hadajiivanov
molecular adsorbed water through the reaction 2NO 2 + H 2 O → HNO 3 + HONO (Li et al.,
et al., 1998). The results indicate that the formation of nitrates occurred at the expense of
2005b; Szanyi et al., 2004). Prolonged exposure of the sample to NO+O2 led to an increase in
intensity only at 1311 cm−1, which can be explained by the reaction
M n+ − O 2 − + 2NO 2 → M n+ − NO 3− + NO 2− on cation defect sites such as extra-framework
alumina, similar to the reaction pathway suggested by Larsen et al. (Li et al., 2005b). Part of
the nitrous species (adsorbed NO2 and some unidentate nitrates) corresponding to the bands
at 1622 and 1311 cm−1 were so weakly adsorbed on the zeolite that they disappeared in the
subsequent brief evacuation. At the same time, two bands at 1663 and 1389 cm−1 due to
nitrites and/or nitrates associated with a very low concentration of Na+ ions in HZSM-5 (Li.
et al., 2005b; Satsuma et al., 1997; Szanyi & Paffett, 1996) were observed. Identical to HZSM-
5, the band at 1622 cm−1 appeared rapidly with NO+O2 co-adsorption on NaZSM-5 (Fig.
12B). However, different from HZSM-5, prolonged exposure of NaZSM-5 to NO+O2
primarily resulted in the appearance of bands with strong intensity at 1407 and 1389 cm-1
due to nitrates banding to Na+ (Li et al., 2005b; Satsuma et al., 1997; Szanyi & Paffett, 1999).
It is clear by comparing Fig. 12 A and B that much more stable nitrous species could form on
NaZSM-5 than on HZSM-5, which are in good accordance with the results obtained by NOx-
TPD as shown in Fig. 13. Thermal stability of the nitrous species on the zeolites was
investigated by FTIR, as shown in Fig. 14. On HZSM-5, the band at 1311 cm−1 due to
unidentate nitrates disappeared at 150 ◦C, and a band appeared at 1585 cm−1 due to
bidentate nitrates (Ivanova et al., 2001; Yu et al., 2004; Shimizu et al., 2001; He et al., 2004;
Poignant et al., 2001). It indicates that the bidentate nitrates were produced from unidentate
nitrates at the elevated temperature. Compared with the bridging nitrates (1622 cm−1), the
bidentate nitrates (1585 cm−1) were more thermally stable; the intensity at 1585 cm−1 slightly
decreased with the temperature increasing. Similar to those on HZSM-5, unidentate nitrates
Fig. 13. TPD profiles of NO (A) and NO2 (B) in N2 flow after saturated co-adsorption of
NO+O2 on HZSM-5 (a) and on NaZSM-5 (b)
www.intechopen.com
A 0.02
1585
d 1484
Absorbance
1622
1663 1389
2884
a
2477
3610
3600 3300 3000 2700 2400 2100 1800 1500

-1
Wavenumber/cm
B 1389
0.1
1407
Absorbance
1622
3679 3577
1715
a
1638
3800 3600 3400 1800 1700 1600 1500 1400 1300
-1
Wavenumber/cm
Fig. 14. The nitrous species on HZSM-5 (A) and on NaZSM-5 (B) at 150 oC (a), 250 oC (b), 350
oC (c) and 400 oC (d)
(1311 cm−1) on NaZSM-5 disappeared at 150 ◦C, indicating that the unidentate nitrate species
are less thermal stable than the other kinds of nitrates, irrespective of the adsorbents. It
implies that the unidentate nitrates are not involved in the C2H2-SCR, as the reaction
significantly occurred at the temperature above 250 ◦C. Quite different from the findings on
HZSM-5, although the bridging nitrates (giving band at 1622 cm−1) on NaZSM-5 changed
little at 150 ◦C, most of them desorbed when the temperature increased to 250 ◦C. Band at
1715 cm−1, observed exclusively on NaZSM-5, could be assigned to nitrates associated with
Na+ ions, because it appeared concomitantly with the band at 1407 cm−1 during the co-
adsorption of NO+O2 (Fig. 12 B) and disappeared at 350 ◦C together with the same band
(Fig. 14). On NaZSM-5, the most stable nitrous species are the nitrate species associated with
Na+ (1389 cm−1), which could remained on the zeolite at 400 ◦C. The FTIR results reveal that
the adsorption of NOx on ZSM-5 was strongly affected by the cations in the zeolite, and that
protons were essential for the bidentate nitrates formation on the zeolites.
www.intechopen.com
4.2 Characterization on the activity of nitrous species towards to reduction on the

zeolites
The bands due to nitrate species at 1626 and 1585 cm−1 disappeared within 1 min upon
exposure of HZSM-5 to C2H2+O2 at 250 ◦C (Fig. 15A). Concomitantly, the bands at 1693 cm−1
due to carbonyl-containing compound and at 1605 cm−1 due to the C=O vibration of
carboxylic groups (Li et al., 2005b) appeared. It indicates that the bridging and bidentate
nitrates are very active with reductant. However, no significant change in the spectra was
observed when the NaZSM-5 was exposed to the gas mixture of C2H2+O2 at 250 ◦C, as
shown Fig. 15 B), indicating that the nitrate species bonding to Na+ on NaZSM-5 are inert to
the reductant. The quite different nitrous species formed on HZSM-5 and on NaZSM-5 as
well as their disparate reactivity with reductant could explain the distinct catalytic
performance of the two zeolites in C2H2-SCR. Thus, it is rational to draw the conclusion that
protons are responsible not only for catalyzing NO oxidation to NO2, but also for the
formation of active nitrate intermediates in the C2H2-SCR.
0.02 1605 A B 1389

0.1
1484
1693
g
1715
1638 g
Absorbance
Absorbance
a a
1626 1585
1800 1700 1600 1500 1400 1800 1700 1600 1500 1400 1300
-1 -1
Fig. 15. FTIR spectra of surface species on HZSM-5 (A) and NaZSM-5 (B) at 250 oC when the
zeolite was subjected to saturated coadsorption of NO+O2 and a subsequent brief
evacuation (a), and then an exposure to C2H2+O2 for 1 min (b), 2 min (c), 4 min(d), 6 min (e),
8 min (f), and 10 min (g)
4.3 Characterization of carbonous species and the reactivity with nitrous species on
the zeolites
The evolution of carbonous species and their reactivity toward NOx were investigated by
FTIR. The IR spectra of carbonous species due to saturated adsorption of acetylene on the
HZSM-5 and NaZSM-5 at 80 ◦C are depicted in Fig. 16. The adsorption of acetylene on
HZSM-5 resulted in the appearance of positive bands at 1673 and 1628 cm−1 and negative
bands at 3700, 2884, and 2477 cm−1 (Fig. 16, spectrum c). As discussed in Section 4.1, the
three negative bands are arisen from the consumption of adsorbed water on the Brønsted
www.intechopen.com
acid site. No negative band at 3610 cm−1 due to consumption of acidic zeolite Al(OH)Si
hydroxyls (Poignant et al., 2001; Brosius et al., 2005) was observed caused by acetylene
adsorption. In an energy standpoint, acetylene would adsorb on the free Brønsted acid sites,
which were characterized by band at 3610 cm−1 as shown in Fig. 17, rather than adsorb on
the ones taken up by water. Therefore, it is reasonable to propose that acetylene reacted with
the adsorbed water on the Brønsted acid site, but did not simply take up the Brønsted acid
site by breaking the well-known A–B–C structure.
Based on above conclusion that acetylene reaction with water on Brønsted acid sites when
adsorbed on HZSM-5, it can be further reasonably speculated that vinyl alcohol (CH2=CH–
OH) species were formed by acetylene adsorption on the zeolite as depicted by model 1 as
shown in scheme 1: In principle, acetylene can be adsorbed on ZSM-5 in two ways. One way
does by reacting with water to form vinyl alcohol and the later can be strongly bonded to
Brønsted acid sites with hydrogen bond (model I), being characterized by a blue shift of
ν(C=C) with respect to the general C=C double band. Due to hydrogen bond of the –OH
group with the Brønsted acid sites, the blue shift of ν(C=C) may be further aggravated, and
it is the primary case of acetylene adsorption on HZSM-5, which is strongly supported by
the band of 1628 and 1723 cm-1. Another way does by binding to the cations (e.g. H+ and
Na+) in zeolite channels through weak static attraction to form a π-complex, which may
decreases the electron density in the highest occupied molecular orbital, leading to a red
shift of ν(C=C) with respect to the general C=C double band (model II). Obviously, it is the
primary case of acetylene adsorption on NaZSM-5, which is strongly supported by the band
at 1621 cm-1.
1621
3683 1630
a
b
Absorbance
3582 1628
1673
0.1
2884
2477 c
3700
3600 3300 3000 2700 2400 2100 1800 1500

-1
Wavenumber/cm
Fig. 16. FTIR spectra of carbonous species formed by saturated adsorption of C2H2 (500
ppm) in N2 at 80 oC on NaZSM-5 (a), HZSM-5 (c), and those when the sample of NaZSM-5
(b) and HZSM-5 (d) was subsequently subjected to a brief evacuation at this temperature
www.intechopen.com
3610
0.1
Absorption
H Z S M -5
N aZ S M -5
3750 3500 3250 3000 2750 2500 2250 2000 1750 1500
-1
W avenum ber/cm
Fig. 17. FTIR spectra of fresh HZSM-5 and NaZSM-5
model I:
H
H H H
O HO
+ + HC CH + H
H H
O O 1673 cm-1
Al Si Al Si
model II: H
H M = H, Na
H
O HO CH2
+ HC CH M+
M+
O O
Al Si 1621 cm-1
Al Si
Scheme 1. C2H2 adsorption model on the HZSM-5 and NaZSM-5

Evolution of the carbonous species on HZSM-5 with temperature is shown in Fig. 18. It is
obvious that the intensity of the band at 3659 cm−1 due to ν(–OH) of alcohol hydroxyl
groups increased with the temperature increase from 80 to 150 ◦C. At the same time, the
bands at 2884 and 2477 cm−1 due to the A–B–C structure, and the band at 1352 cm−1 due to
δ(OH) of the H-bonded zeolite hydroxyls were recovered on the HZSM-5 (Fig. 18),
accompanied by the appearance of the negative band at 3610 cm−1. With the temperature
further increase, the bands relating to adsorbed water on the Brønsted acid sites increased in
www.intechopen.com
intensity. It indicates that water was formed and adsorbed on the Brønsted acid sites at
elevated temperatures. Above 300 ◦C, the strong band centered at 1628 cm−1 disappeared,
which correlates well with the end of desorption peak (around 215 ◦C) of acetylene in C2H2-
TPD (not shown). As a result, a new band at 1646 cm−1 was clearly observed. Based on these
results, a scheme 2 for the evolution of carbonous species with temperature on HZSM-5 can
be proposed.
1628
0.05 2884 1479
a 2477
1352
Absorbance
3700
3659
1646
3610 g 1585
3600 3300 3000 2700 2400 2100 1800 1500

-1
Wavenumber/cm
Fig. 18. Spectra of carbonous species on HZSM-5 when evacuated at 80 oC (a), at 150 oC (b),
at 178 oC (c), (d) at 200 oC (d), at 300 oC (e), at 360 oC (f), and at 410 oC (g) after saturated
adsorption of acetylene
The vinyl alcohol adsorbed on the Brønsted acid sites in the models both I and II was
converted to a carbenium ion (HO–HC+–CH3), which gave bands at 3659 and 1352 cm−1 due
to the OH stretching vibration and the CH3 deformation vibration, respectively. The
carbenium ion decomposed to water and another carbonium ion (CH2=HC+) that gave the
band at 1646 cm−1. The formed water adsorbed on Brønsted acid sites led to a recovery of
intensity and an increase of the bands at 2884 and 2477 cm−1, as well as the negative band at
3610 cm−1. In addition, some of the carbenium ion (HO–HC+–CH3) was oxidized by oxygen
(ca. 5000 ppm) in the pure N2 (99.995%) to acetate species, giving the bands at 1585 and 1479
cm-1 due to the νas(COO) and νs(COO) of acetate species (Yu et al., 2004; Shimizu et al., 2001;
He et al., 2004; Poignant et al., 2001; Wu et al., 2005).
Evolution of the carbonous species on HZSM-5 with temperature is shown in Fig. 19.
Temperature increase from 80 to 200 ◦C resulted in recovery of the band at 1635 cm−1 as well
as a shift of the band (from 1623 to 1628 cm−1). Scheme 3 explains that acetylene was
released by the decomposition of vinyl alcohol, leaving water on the cation sites in zeolite,
leading to the disappearance of the negative band. Above 300 ◦C, almost no carbonous
surface species were detected on Na-ZSM-5, which is in line with the results obtained in
C2H2-TPD (not shown).
www.intechopen.com
3659 cm-1
1352 cm-1 1585 cm-1
H
H 1479 cm-1
HO H CH3
+ H HO + H3C COOH
H C
O O 1646 cm-1
Al Si Al Si [O] CH2
+ + H2O (ads)
H H C
O
Al Si
HO + CH2
H
O
Al Si
Scheme 2. The evolution of carbonous species on HZSM-5
3583 a 1623 0.02

1635
3682
Absorbance
1628
f
3800 3600 34001700 1650 1600 1550
-1
Wavenumber/cm
Fig. 19. Spectra of carbonous species on NaZSM-5 when evacuated at 150 oC (a), at 178 oC
(b), at 200 oC (c), (d) at 300 oC (d), at 360 oC (e), and at 420 oC (f), after saturated adsorption
of acetylene
H
H H
HO + CH2 O
Na +
+ HC CH
Na
O O
Al Si Al Si
Scheme 3. The decomposition of carbonous species on NaZSM-5
www.intechopen.com
The reactivity of carbonous species on HZSM-5 and on NaZSM-5 with NO and NO+O2 was
studied at 200 ◦C. Acetate species (1585, 1479 cm−1) were detected at steady state on HZSM-5
in NO/N2 and N2 (Fig. 20 A), indicating that the acetate species could exist in NO/N2. This
means that acetate species did not react with NO at the temperature. However, in
NO+O2/N2, no acetate species could be detected, and instead, a new band at 2131 cm−1
appeared. This band can be assigned to ν(C≡N) of cyanide (Shimizu et al., 2001; Mosqueda-
Jiménez et al., 2003b) produced by the reaction of acetate species with NOx. The result
indicates that the acetate species formed by carbenium ion (HO–HC+–CH3) oxidation on
HZSM-5 are active with nitrous species arising from the co-adsorption of NO+O2. Thus it
can be reasonably considered that acetate species is one of the important intermediates for
the C2H2-SCR. In contrast to the case of HZSM-5, although vinyl alcohol (1628 cm−1) was
also formed on NaZSM-5 (Fig. 20 B), no significant change of the band in intensity was
observed when the gas mixture was switched form NO/N2 to NO+O2/N2, indicating that
the vinyl alcohol bonding to Na+ in NaZSM-5 is inert under the reaction conditions.
Based on above discussion, it is clear that, the carbonous species formed by acetylene
adsorption on HZSM-5 and on NaZSM-5 as well as their reactivity with nitrous species are
also quite different.
A B 1389
1628
0.02
1407
0.05
Absorbance
1628
Absorbance
1715
2131 c c
b
b 1352
a
a 1585
1479 1638
2200 2000 1800 1600 1400 2200 2000 1800 1600 1400
-1 -1
Fig. 20. FTIR spectra of surface species on HZSM-5 (A) and NaZSM-5 (B) at 200 oC taken in
(a) N2, (b) NO, (c) NO + O2. Before the measurements, the catalysts were pretreated in
C2H2/N2 at 80 oC
5. To propose possible reaction mechanism

To design a catalyst more active for the HC-SCR in real lean-burn conditions, extensive
studies were also carried out on mechanism of the reaction. There have been different
opinions concerning the mechanism of HC-SCR in literature, which can be roughly
classified as “dissociative” (Goula et al., 2007; Burch & Watling, 1996) and “reduction” ones
www.intechopen.com
(Mihaylov et al., 2004; Mosqueda-Jiménez et al., 2003b; Hadjiivanov et al., 2003). The
“dissociative” mechanism proposed by Burch and Watling (Burch & Watling, 1996) has been
widely accepted by the authors being concerned with noble metal catalysts in the HC-SCR
field, and it can be expressed as follows (Goula et al., 2007):
NO ads → N ads + O ads (1)
N ads + N ads → N 2 (g) (2)
N ads + NO ads → N 2 O(g) (3)
For the “reduction” mechanism of HC-SCR, some authors have claimed that activation of
hydrocarbon occurs first, and some partially oxidized hydrocarbons (CxHyOz) produced by
the step then react with NO and/or NO2 to form the secondary intermediates (Sasaki et al.,
1992). For example, formate and acetate were proposed to be active species of HC-SCR over
CoOx/Al2O3 (He & KÖhler, 2006), Ga2O3/Al2O3 (He et al., 2005), Ag/Al2O3 (Shibata et al.,
2003), SnO2/Al2O3 (Liu et al., 2006), Cu-Al2O3 (Shimizu et al., 2000; Satsuma & Shimizu,
2003), In2O3/Al2O3 (Luo et al., 2007), Ag/Al2O3 (Zhan et al., 2007) and Pd/Al2O3 catalysts
(Huuhtanen et al., 2002). Acetaldehyde deriving from propene was also proposed to be
main active species of the HC-SCR over sulfated titania-supported rhodium catalyst (Flores-
Moreno et al., 2005). However, some authors have claimed that activation of NOx occurs
first, forming nitrous surface species, such as nitro (Meunier et al., 2000), nitroso (Poignant
et al., 2001, Gerlach et al., 1999), nitrosonium ions (Gerlach et al., 1999, Ingelsten et al., 2005),
nitrate or nitrite (Luo et al.,2007, Anunziata et al., 2007) over the catalyst. For instance,
bridging and bidentate nitrates were reported to be produced first by co-adsorption of
NO+O2 on Co/SO42−-ZrO2 (Tsyntsarski et al., 2003), BaY (Sedlmair et al., 2003a) and
Ag/Al2O3 (Bentrup et al., 2005). Tsyntsarski et al. (Tsyntsarski et al., 2003) have suggested
that both bridging and bidentate nitrates are active species of the HC-SCR. Mihaylov et al.
(Mihaylov et al., 2004) have reported that monodentate nitrates on Ni-HZSM-5 are highly
reactive towards methane. Lónyi et al. (Lónyi et al., 2007) have studied selective catalytic
reduction of NO by CH4 over Co-, Co,Pt-, and H-mordenite catalysts and suggested that
nitrosonium ions are surface intermediates of the reaction. There are also some other
suggestions about the reaction intermediate of HC-SCR in literature, including nitrile
(Poignant et al., 2001; Delahay et al., 2007), isocyanate (Mihaylov et al., 2004; Mosqueda-
Jiménez et al., 2004b; He & KÖhler, 2006), R-NOx (Mosqueda-Jiménez et al.,2003; Cowan et
al., 1998), amine (Poignant et al., 2001), acetonoxime (Shimizu et al.,2000; Resini et al., 2003)
and ammonia (Lónyi et al., 2007).
Although most of the “reduction” mechanisms were supported by Fourier transform
infrared (FTIR) identification of reaction intermediates (Joubert et al., 2006), none of them
has been widely accepted because of the complexity of the process (Mosqueda-Jiménez et
al., 2003b) being concerned with different catalysts, reductants and reaction conditions.
More investigation on the reaction mechanism is required for understanding the real
reaction route of HC-SCR over different catalysts and using different reductants.
5.1 Reaction mechanism of C2H2-SCR over Mo/HMOR

In the this section, a possible reaction mechanism of C2H2-SCR over Mo/ HMOR （Li et al.,
2008） and Zr/HFER (Xing et al., 2008) investigated by in situ FTIR was summarized.
www.intechopen.com
Fig 21 shows conversion of NO and C2H2 in C2H2-SCR over the mordenite-based catalysts
(HMOR, 0.5%Mo/HMOR, NaMOR) as a function of temperature. The activity of
0.5%Mo/HMOR for C2H2-SCR was considerably high compared to HMOR and NaMOR.
70% of NO conversion to N2 at 350 oC over 0.5% Mo/HMOR catalyst was obtained. It
indicates that molybdenum has a significant promotional effect on C2H2-SCR. The peak of
the “volcano” curve in NO conversion to N2 versus reaction temperature seems to be in line
with the temperature where C2H2 was nearly completely consumed. Hence, the drop of NO
conversion above 350 oC can be considered to be caused by the lack of reductant.
80
Conversion of NO to N2 (%)
70
A
60
50
40
30
20
10
0
250 300 350 400 450 500 550
Temperature (oC)
100
B
Conversion of C2H2 (%)
80
60
40
20
0
250 300 350 400 450 500 550
Temperature (oC)
Fig. 21. Conversion of NO (A) and C2H2 (B) as a function of reaction temperature in C2H2-SCR.
Reaction condition: 1600 ppm NO, 800 ppm C2H2, 9.95 % O2 in He with a total flow rate of 50
ml/min over HMOR 0.100 g (□), 0.5 % Mo/HMOR 0.100 g (▲) or NaMOR 0.200 g (♦)
www.intechopen.com
A 1629
0.025 2229 1592
3600
3654
250oC
Absorbance
300oC
350oC
400oC
4000 3500 3000 2500 2000 1500

Wavenumber/cm-1
B 1629
CC 1410
0.05
0.05
1592
2229 1629
250oC
Absorbance
Absorbance
o
250 C
300oC
o
300 C
o
350 C
o
350 C
o
400 C
o
400 C
1388
2200 2000 1800 1600 1400 2200 2000 1800 1600 1400
Wavenumber/cm-1 Wavenumber/cm-1
Fig. 22. Steady state in situ FTIR spectra of surface species on HMOR (A), 0.5%Mo/HMOR
(B) and NaMOR (C) in 1000 ppm NO + 10% O2 + N2 at different temperature
Figure 22 shows steady state in situ FTIR spectra of nitric species formed by NO+O2 co-
adsorption on the mordenite-based catalysts at different temperatures. Three main bands at
2229, 1629 and 1592 cm-1 associated with nitric species were observed on HMOR (Fig. 22 A).
The band at 2229 cm-1 is due to N-O stretching mode in NO+ (Li et al., 2005a; Pirngruber &
Pieterse, 2006; Gerlach et al., 1999), and the bands at 1629 and 1592 cm-1 can be assigned to
bridging and bidentate nitrates (Poignant et al., 2001; Li et al., 2005a; Sedlmair et al., 2003;
Yu et al., 2007), respectively. The NO+ (2229 cm-1) species was also detected by Gerlach et al.
(Gerlach et al., 1999) at 120 oC when NOx was adsorbed on the zeolite. As shown in Fig. 22
www.intechopen.com
A 1629
3600 2229
0.02 1592
a
3654
Absorbance
f 1698
4000 3600 2800 2400 2000 1600

Wavenumber/cm-1
0.05 B 1629
a
3654 2229 1592
Absorbance
1698
f
4000 3600 3200 2800 2400 2000 1600
Wavenumber/cm-1
0.05
C
3654 1629
a
Absorbance
f 1388
4000 3600 3200 2800 2400 2000 1600

Wavenumber/cm -1
Fig. 23. In situ FTIR spectra on HMOR (A), 0.5%Mo/HMOR (B) and NaMOR (C) at 250 oC: a
brief evacuation after saturated adsorption of NO+O2 (a), and subsequently exposing to
C2H2+O2 for: 1 min (b), 3 min (c), 5 min (d), 8 min (e), 30 min (f)
www.intechopen.com
A, a positive band at 3654 cm-1 due to adsorbed water (Mihaylov et al., 2004; Chafik et al.,
1998) and a negative band at 3600 cm-1 due to Brønsted acid sites (Mihaylov et al., 2004;
Gutierrez et al.,2007) were observed after co-adsorption of NO and O2 on the zeolite at 250
oC. It can be well interpreted by the NO+ formation pathway NO + NO2 + 2H+ → 2NO+ +
H2O, proposed by Hadjiivanov et al. (Hadajiivanov et al., 1998) and Gerlach et al. (Gerlach
et al., 1999). Band at 1629 cm-1 due to bridging nitrate and band at 2229 cm-1 due to NO+ on
0.5%Mo/HMOR are obviously greater in intensity respectively compared with those on
HMOR, particularly above 300 oC (Fig. 22 B). It indicates that molybdenum loading on the
HMOR zeolite have a promotional effect on the nitric species formation at higher
temperature. On NaMOR (Fig. 22 C), band (1629 cm-1) due to this type of bridging nitrate
was rather weak, and bands due to bidentate nitrates (1592 cm-1) and NO+ species (2229 cm-
1) even could not be observed. Instead, a broad band at 1410-1388 cm-1 due to nitrate ions
attached to Na+ sites (Mihaylov et al., 2004; Li et al., 2005a; Yu et al., 2007) appeared after
NO+O2 co-adsorption on the sample under the same condition.
A 1635 B 1635
0.025 1595
0.025
1595
250 oC
1479 1479
o
Absorbance
Absorbance
250 C
300 oC
300 o C
350 oC o
350 C
400 o C
400 o C
1663
2200 2000 1800 1600 1400 2200 2000 1800 1600 1400
-1
Wavenumber/cm W avenum ber/cm -1
Fig. 24. Steady state in situ FTIR spectra of adsorbed species in 500 ppm C2H2 + 10 % O2 +
N2 on HMOR (A), 0.5 % Mo/HMOR (B) at different temperature
Reactivity of the nitric species towards C2H2+O2 over the mordenite-based catalysts was
examined by in situ FTIR at 250 oC (Fig. 23). When C2H2+O2 was introduced into the FTIR
cell, bands due to bidentate nitrates (1592 cm-1) and NO+ species (2229 cm-1) on HMOR
arisen from NO+O2 pre-adsorption (Fig. 23A) rapidly decreased. Concomitantly, a new
band at 1698 cm-1 appeared and reached its maximum intensity within 3 min with
disappearance of bands at 2229 and 1592 cm-1. Similar result was obtained when C2H2 was
used instead of C2H2+O2 in the above experiment. The results indicate that NO+ and
bidentate nitrate species are fairly reactive towards acetylene at this temperature. It was
evidenced by the following changes of bands associated with water formation during the
process: A positive band at 3654 cm-1 due to water appeared, and at the same time, a
negative band at 3600 cm-1 arisen from water adsorption on Brønsted acid sites
www.intechopen.com
correspondingly increased in intensity. Unfortunately, the corresponding reactivity of

bridging nitrate species (1629 cm-1) could not be directly evaluated on the zeolite because
of water formation. The band due to bending mode of water appears at the identical wave
number with that of bridging nitrate species at 1629 cm-1. Similar experimental results as
that on HMOR was obtained on 0.5%Mo/HMOR (Fig. 23 B). However, quite different
results were obtained on NaMOR. The nitrate species attached to Na+ seem to be
completely inert towards the reactant. No change in intensity of band at 1388 cm-1 due to
the species could be observed on NaMOR (Fig. 23 C). Meanwhile, as expected, band at 1698
cm-1 did not appear on NaMOR. Instead, strong bands at 3654 and 1629 cm-1 due to water
adsorbed on the Na-form zeolite were observed, which may be simply resulted from
combustion of acetylene.
Figure 24 shows steady state in situ FTIR spectra of the surface species on HMOR and on
0.5%Mo/HMOR in gas mixture of 500 ppm C2H2 + 10% O2/N2 at different temperatures. No
band at 1698 cm-1 could be observed on the catalysts. Instead, bands at 1635, 1595 and 1479
cm-1 due to ν(C=O), νas(COO) and νs (COO) of carboxylic groups (Hadajiivanov et al., 1998;
Shimizu et al., 2007) appeared. Combined the result with that observed in Fig. 23, the band
at 1698 cm-1 can be attributed to nitrogen containing organic species, because this band
could appear only in the following conditions: both the nitric species (nitrogen oxides
and/or nitric surface species) and the reductant were present together in the reaction
system.
1629
2242 0.05
2208 a 1592
Absorbance
h 1698
2200 2000 1800 1600
-1
W avenumber/cm
Fig. 25. Steady state in situ FTIR spectra of adsorbed species on 0.5%Mo/HMOR in gas
mixture of 1000 ppm NO + 500 ppm C2H2 + 10 % O2 + N2 at 250 oC (a), 300 oC (c), 350 oC (e),
400 oC (g) and that on HMOR at 250 oC (b), 300 oC (d), 350 oC (f), 400 oC (h)
www.intechopen.com
Figure 25 shows the steady state in situ FTIR reaction spectra of the surface species on
HMOR and 0.5%Mo/HMOR in gas mixture of 1000 ppm NO + 500 ppm C2H2 + 10% O2/N2
at different temperatures. Two overlapped bands respectively centered at 2242 and 2208
cm-1 were appeared in the spectra, in addition to the bands at 1698, 1629 and 1592 cm-1. The
band at 2242 cm-1 can be assigned to -NCO vibration of isocyanate (2242 cm-1) (Satsuma &
Shimizu, 2003; Kameoka et al., 2000) and that at 2208 cm-1 can be assigned to N-O
stretching of NO+ (Pirngruber & Pieterse, 2006; Gerlach et al., 1999). It was reported that
the stretching frequency of NO+ on HMOR is influenced by an interaction between NO+
and some other surface species formed on the zeolite (Gerlach et al., 1999). Herein, it
should be noticed that although the band at 1698 cm-1 for 0.5%Mo/HMOR (spectrum a)
was slightly weaker in intensity in comparison with that of HMOR (spectrum b) at 250 oC,
it became much stronger than that of HMOR above 300 oC. The relative intensity of the
band observed on HMOR and 0.5%Mo/HMOR at different temperatures could be
correlated well with the relative activity of the catalysts for C2H2-SCR (Fig. 21). Good
accordance of NO reduction with the band at 1698 cm-1 could also be obtained on NaMOR
(Fig 26). The band at 1698 cm-1 just appeared at the temperature (spectrum e), where NO
conversion to N2 became significant over the zeolite in C2H2-SCR. Based on the above
findings, we believe that the species with the band at 1698 cm-1 is a crucial intermediate for
C2H2-SCR over the mordenite-based catalysts.
1388
0.05
1720
3654 1629
a
Absorbance
g
1698
4000 3600 3200 2800 2400 2000 1600
Wavenumber/cm-1
Fig. 26. Steady state in situ FTIR spectra of adsorbed species in gas mixture of 1000 ppm NO
+ 500 ppm C2H2 + 10% O2 + N2 on NaMOR at different temperature: 150 oC (a), 200 oC (b),
250 oC (c), 300 oC (d), 350 oC (e), 400 oC (f), 450 oC (g)
To further study the reaction route of C2H2-SCR over the catalysts, reactivity of the
intermediate (1698 cm-1) was investigated. After a pre-exposure of HMOR to 1000 ppm NO
www.intechopen.com
+ 500 ppm C2H2 + 10% O2/N2 at 250 oC and a followed brief evacuation, the catalyst was
exposed to gas mixture of 1000 ppm NO + 10% O2/N2. As a result, the intensity of the band
at 1698 cm-1 rapidly decreased (Fig. 27, spectrum a). It indicates that the intermediate is
rather reactive towards NO+O2 at the temperature. On the other hand, no decrease in
intensity of the band at 2242 cm-1 due to -NCO species could be observed during the period
(Fig. 27), indicating that -NCO species is inert towards NO+O2. When the sample was then
exposed to gas mixture of 500 ppm C2H2 + 10% O2 in N2, as shown in Fig. 28 (c~f), the bands
both at 2242 cm-1 due to -NCO and at 2229 cm-1 due to NO+ disappeared within 1 min.
Concomitantly, a band at 1698 cm-1 and bands at 3654 and 1629 cm-1 due to adsorbed water
appeared. It should be noticed that the intensity of the three bands as well as that of the
negative band at 3600 cm-1 continued to increase within three minutes. The result agrees
well with the proposition in literature that isocyanate species can be easily hydrolyzed to
amines (Poignant et al., 2001; Bion et al., 2003). Thus, the band at 1698 cm-1 can be assigned
to acid amide species on the zeolite. It is in accordance with that reported by Poignant et al.
(Poignant et al., 2001), who found the species with band at 1694 cm-1 in reaction of NO +
C3H8 + O2 over HZSM-5 at 350 oC.
1629
0.05 3600 a 2242 1592
Absorbance
g
2229 1698
3600 2800 2400 2000 1600

Wavenumber/cm-1
Fig. 27. In situ FTIR spectra of surface species on HMOR at 250 oC: a brief evacuation after a
pre-exposure of the catalyst to NO+C2H2+O2 for 30 min (a), and then when the catalyst was
exposed to NO+O2 for: 1 min (b), 3 min (c), 5 min (d), 8 min (e), 30 min (f), and finally
evacuated briefly (g)
In Fig. 22, we showed that molybdenum loading on HMOR zeolite considerably promoted
the formation of bridging nitrate species. However, neither higher NO conversion to N2 in
C2H2-SCR nor stronger band due to the acid amide species (1698 cm-1) could be observed on
0.5%Mo/HMOR compared to HMOR at 250 oC. It leads us to speculate that bridging nitrate
species (1629 cm-1) may make no contribution to C2H2-SCR at the lower temperature (< 250
oC) over the mordenite-based catalysts. The speculation was validated by the following
experimental results: Although the bridging nitrate species (1629 cm-1) was detected by FTIR
www.intechopen.com
after co-adsorption of NO+O2 on NaMOR at 250 oC (Fig. 22C, spectrum a), no NO

conversion to N2 could be obtained at the temperature (Fig. 21).
Above 300 oC, both C2H2-SCR activity (Fig. 21) and the intensity of the band due to acid
amide species detected by FTIR (1698 cm-1, in Fig. 25) were larger on 0.5%Mo/HMOR
compared to those on HMOR, which corresponds well with the larger population of
bridging nitrate species (1629 cm-1, in Fig. 22) given by 0.5%Mo/HMOR in comparison
with those given by HMOR at the temperature. The results indicate that bridging nitrate
species (1629 cm-1) are also involved in C2H2-SCR at higher temperature. Same conclusion
can be drawn on NaMOR. The band at 1629 cm-1 due to bridging nitrate species arisen from
1629
0.05 3600 a
2242 1592
3654 2229
Absorbance
1698
f
3700 3650 3600 3550 3500 3450 2800 2400 2000 1600
Wavenumber/cm-1
Fig. 28. FTIR spectra of the surface species on HMOR at 250 oC: a brief evacuation after
exposing the catalyst to NO+C2H2+O2 for 30 min (a), a brief evacuation after exposing the
catalyst to NO+O2 for 30 min (b), then exposed the catalyst to C2H2+O2 for 1 min (c), 3 min
(d), 5 min (e), 30 min (f)
NO+O2 pre-adsorption on the zeolite at 350 oC (Fig. 29, spectrum a) rapidly disappeared
when gas mixture of 500 ppm C2H2 + 10% O2/N2 was introduced to the FTIR cell at this
temperature. Correspondingly, band at 1698 cm-1 due to acid amide species appeared (Fig.
29, spectrum b). No change of the band at 1388 cm-1 in intensity could be observed during
this procedure. The results again indicate that bridging nitrate species are involved in the
desired reaction at the higher temperature, whereas nitrate species attached to Na+ have no
contribution to C2H2-SCR under the reaction condition.
It explains why the C2H2-SCR activity of NaMOR became significant when the reaction
temperature increased to 350 oC. Thus, the considerably larger activity of 0.5%Mo/HMOR
in comparison with that of HMOR for C2H2-SCR above 300 oC (Fig. 21) can be rationally
attributed to the larger bridging nitrate formation capacity of the catalyst.
www.intechopen.com
0.025
1698
Absorbance
1388
a 1629
2000 1800 1600 1400

W avenum ber/cm -1
Fig. 29. In situ FTIR spectra of surface species on NaMOR at 350 oC: The catalyst was
subjected a brief evacuation after saturation adsorption of 1000 ppm NO and 10 % O2 in N2
(a), and then exposed to 500 ppm C2H2 + 10 % O2 / N2 for 1min (b), 10min (c)
On the basis of above discussion, a possible reaction mechanism of C2H2-SCR over the
mordenite-based catalysts can be outlined in Scheme 4. Nitrosonium ions (NO+), bidentate
and bridging nitrate species formed by NO+O2 co-adsorption react first with C2H2, leading
to isocyanate species (2242 cm-1) formation. The isocyanate species is then rapidly
hydrolyzed to the acid amide species (1698 cm-1) that is a crucial intermediate for C2H2-SCR
over the mordenite-based catalysts.
NO+ and bidentate nitrate C2H2+O2

-NCO
NO+O2 species (active at 250 oC)
(isocyanate)
bridging nitrate species
(active above 300 oC)
hydrolyze
NO+O2
N2, CO2, H2O acid amide species
Scheme 4. A possible reaction pathway of the C2H2-SCR over the mordenite-based catalysts
www.intechopen.com
As a summarization for C2H2-SCR over the mordenite-based catalysts, following reaction

mechanism can be drawn: The nitric species, including nitrosonium ions (NO+) and
bidentate nitrate, are fairly active towards the desired reduction in the temperature range of
250-450 oC. However, bridging nitrate species begin to make its significant contribution to
the reaction above 300 oC. Molybdenum incorporated into HMOR zeolite considerably
improved the bridging nitrate formation capacity of the catalyst. It explains the promotional
effect of molybdenum on C2H2-SCR at the higher temperatures. Isocyanate (-NCO), as an
active species produced from the reaction of the nitric species with C2H2, can be rapidly
hydrolyzed to the acid amide species (1698 cm-1) that is a crucial intermediate for C2H2-SCR
over the mordenite-based catalysts.
5.2 Reaction mechanism of C2H2-SCR over Zr/HFER

Surface species formed on 2%Zr/FER by exposing the catalyst to NO+O2/N2 at 250 oC for 30
min and subsequently to C2H2/N2 characterized by FTIR spectra are shown in Fig. 30. NO+,
bridging and bidentate nitrate species, which give the bands at 2188, 1629 and 1598 cm-1
(Poignant et al., 2001; Li et al., 2007; Li et al., 2005a; Brosius et al., 2005; Yu et al., 2007)
respectively, were presented by the co-adsorption of NO+O2 in N2 (spectrum a). The three
Wavenumber/cm-1
Fig. 30. FTIR spectra of surface species on 2%Zr/FER: the catalyst was exposed to 1000 ppm
NO + 10 % O2 + N2 at 250 oC for 30 min (a), and subsequently to 500 ppm C2H2 + N2 for 1
min (b), 2 min (c), 3 min (d), 5 min (e) and 10 min (f); a adsorption steady state got by
exposing the fresh catalyst to formamide vapor in N2 at 150 oC (g)
www.intechopen.com
bands rapidly decreased in intensity with an appearance of bands at 2240, 1691, 1651, 1580,
1566 and 1386 cm-1 when the gas mixture was switched to C2H2+N2. Apparently, the bands
at 1651 and 1566 cm-1 due to -ONO and -NO2 vibrations [Yeom et al., 2006] increased with
time, along with the band at 1691 cm-1. The weak band at 2240 cm-1 due to -N=C=O
vibration of isocyanate species (Liu et al., 2006) increased in intensity during the first
several minutes, and decreased then after, and finally disappeared when the band at 1691
cm-1 reached its maximum in intensity. The intensity change of bands at 2240 and 1691 cm-1
quite seems that the species giving the band at 1691 cm-1 is produced by the isocyanate
species.
In literature, Poignant et al. have found the band at 1694 cm-1 during the reaction of
NO+C3H8+O2 over HZSM-5 at 350 oC (Poignant et al., 2001) and assigned the band to
acetamide species. Larrubia et al. also found a band at 1690 cm-1 in FTIR on Fe2O3-TiO2
when they exposed the catalyst to acetamide vapor at 350 oC (Larrubia et al., 2001).
However, no band at around 1690 cm-1, but at 1717 cm-1 was found when we exposed the
2%Zr/HFER catalyst sample to acetamide vapor in N2 in the temperature range of 30-300
oC. Instead, a spectrum that much close to spectrum f (in fig. 30) in shape, with the bands
at 1691 and 1386 cm-1 along with 1651 cm-1 (-ONO vibration) was obtained when the
catalyst was exposed to formamide vapor (spectrum g). In addition, the number of carbon
in the amine species is also supported by a simultaneously formed formate species with
the band at 1580 cm-1 (Haneda et al., 2002) (spectra c-f). Hence, we propose that
formamide species, which gives the band at 1691 cm-1, was formed during the reaction of
C2H2-SCR over 2%Zr/FER catalyst by nitric species reacting with acetylene.
Correspondingly, a possible formation route of the formamide species can be proposed as
follows:
O
H C CH2 ONO
(1651 cm-1)
OH O O
O
CH CH + HO NO2 H C CH NO2 H C CH2 NO2 H C CH N
OH
(1566 cm-1)
O O
(H 2 O)
H C NH2 + CO2 H C N C O
(1691 cm-1) (2240 cm-1)
In the reaction of C2H2-SCR over 2%Zr/FER, formate species may be produced by the
formamide species further reaction with nitric species:
O O
H C NH2 + NO OH H C OH+ N2+ H O
2
(1691 cm-1) (1580 cm-1)
In literature, it is widely accepted that amine species are curtail intermediates of HC-SCR
(Poignant et al., 2001; Joubert et al., 2006; Larrubia et al., 2001; Arve et al., 2007). Fig. 31
shows FTIR spectrum of surface species on 2%Zr/FER at 300 oC recorded when a
www.intechopen.com
saturated state was reached on the catalyst in the gas mixture of C2H2+NO+O2 in N2, and
the transient spectra recorded when C2H2 was cut off from the gas mixture. Obviously,
the bands at 1691 cm-1 due to formamide species instantly decreased in intensity by
switching the gas mixture of C2H2+NO+O2 to NO+O2, indicating that formamide species
is much reactive for reacting with NO+O2 (and/or nitrate species) over the catalyst at the
temperature. The result is in line with the assumption concerning the formate species
formation.
0.02 1691
1386
1580
Absorbance
f
2000 1900 1800 1700 1600 1500 1400 1300
-1
Wavenumber /cm
Fig. 31. FTIR spectra of surface species on 2%Zr/FER at 300 oC when the catalyst was
exposed to 500 ppm C2H2 + 1000 ppm NO + 10 % O2 + N2 for 25 min (a) and subsequently
exposed to 1000 ppm NO + 10 % O2 + N2 for 1 min (b), 5 min (c), 10 min (d), 15 min (e) and
30 min (f)
Steady state FTIR spectra of surface species formed on 2%Zr/HFER and HFER in gas
mixture of C2H2+NO+O2 in N2 at some desired temperatures were compared in Fig. 32. The
band at 1691 cm-1 due to formamide species on 2%Zr/HFER was higher in intensity than
that on HFER at each temperature, which may be the result of the higher concentrated
nitrate species on 2%Zr/HFER compared to HFER, as discussed in the section 3. It is in good
accordance with their order in activity for C2H2-SCR (Fig. 8).
www.intechopen.com
0.05 1691
1580
1386
a
b
Absorbance
1800 1700 1600 1500 1400 1300

-1
Wavenumber /cm
Fig. 32. Steady state in situ FTIR spectra of surface species on 2%Zr/FER (solid curve) and
on HFER (dash curve) in gas mixture of 500 ppm C2H2 + 1000 ppm NO + 10 % O2 + N2 at the
temperature: 250 oC (a), 300 oC (b), 350 oC (c), 400 oC (d) and 450 oC
Scheme 5. Possible reaction mechanism of C2H2-SCR over Zr/HFER
www.intechopen.com
0.05 (A)
a 1598
1629
Absorbance
2000 1900 1800 1700 1600 1500 1400 1300

-1
Wavenumber/ cm
0.025 1598 (B)

1629 1386
1691
a
Absorbance
1580
g
2000 1900 1800 1700 1600 1500 1400 1300

-1
Wavenumber/ cm
Fig. 33. FTIR spectra of the species recorded on two 2%Zr/FER catalyst wafers at 300 oC: (A)
In gas mixture of 500 ppm C2H2 + 10 % O2 + N2 (a), followed by a brief evacuation (b), then
exposed to gas mixture of 1000 ppm NO + 10 % O2 + N2 for 1 min (c), 2 min (d), 5 min (e)
and 30 min (f). (B) In gas mixture of 1000 ppm NO +10 % O2 + N2 (a), followed by a brief
evacuation (b), then exposed to gas mixture of 500 ppm C2H2 +10 % O2 + N2 for 1 min (c), 2
min (d), 5 min (e), 10 min (f) and 30 min (g)
www.intechopen.com
To have an insight into the formation route of formamide species, the reaction of acetylene
in gas phase with the species formed on catalyst surface and that between the species were
studied in different reaction conditions. No bands due to formamide species could be
observed in the following conditions: (1) Gas mixture of C2H2+NO+O2/N2 was flowed
through empty FTIR cell (without catalyst) at 300 oC (not shown). (2) Gas mixture of
NO+O2/N2 was flowed through 2%Zr/HFER at 300 (Fig. 33 A) or 250 oC (not shown),
before which the catalyst was exposed to C2H2+O2/N2 for 30 min and evacuated briefly.
These results indicate that formamide species can be produced by NO+O2 reacting neither
with C2H2 in gas phase nor with the surface species arisen from C2H2+O2 co-adsorption.
However, the bands at 1691, 1580 and 1386 cm-1 due to formamide and formate species
clearly appeared when the catalyst was exposed to gas mixture of C2H2+O2/N2 at 300 oC
after a pretreatment in NO+O2+N2 for 30 min and briefly evacuated (Fig.33 B). It reveals
that formamide species can only be produced by C2H2 in gas phase reacting with nitrate
species.
Based on above discussion, the reaction mechanism of C2H2-SCR over Zr/HFER can be
outlined in Scheme 5.
6. The end words of the chapter

For the convenience of reading, we would like to give a note describing the experimental
results obtained by FTIR in the chapter to the readers.
The in situ FTIR studies were carried out in a quartz IR cell equipped with CaF2 windows on
a Nicolet 360 FTIR spectrophotometer. All of the spectra were obtained by accumulating 32
scans at a resolution of 2 cm-1. The IR adsorption arising from gas phase on the pathway of
infrared laser in the cell were recorded at desired temperature, which was coded as Sg(T).
For each experiment, the self-supporting wafer of the catalyst sample was first activated in
the cell at 500 oC in N2, and then its IR absorption was measured at each temperature, which
was coded as Sb(T). Thus the transient and steady states in situ IR spectra of surface species
on the catalyst samples in gas mixtures given in this chapter are those in which the
corresponding Sg(T) and Sb(T) were strictly subtracted.
7. Acknowledgments
This corresponding investigation was financially supported by the National Natural Science
Foundation of China (Grant No. 20833011 and 20877015) and the State Hi-tech Research and
Development Project of the Ministry of Science and Technology of China (Grant No.
2008AA06Z319).
8. References
Anunziata O. A.; Beltramone A. R.; Requejo F. G.. J. Mol. Catal. A, 267 (2007) 194.
Arve K.;Backman H.; Klingstedt F.; Eränen K.; Murzin Yu D.. Appl. Catal. B, 70 (2007) 65.
Bentrup U.; Brückner A.; Richter M.; Fricke R.. Appl. Catal. B, 32 (2001) 229.
Bentrup U.; Richter M.; Fricke R.. Appl. Catal. B, 55 (2005) 213.
Berndt H.; Schütze F.-W.; Richter M.; Sowade T.; Grünert W.. Appl. Catal. B, 40 (2003) 51
Bion N.; Saussey J.; Haneda M.; Daturi M. J. Catal., 217 (2003) 47.
Brosius R.; Bazin P.; Thibault-Starzyk F.; Martens J. A. J. Catal., 234 (2005) 191.
www.intechopen.com
Burch R.; Watling T. C.. Catal. Lett., 37 (1996) 51.

Chafik T.; Kameoka S.; Ukisu Y.; Miyadera T. J. Mol. Catal. A, 136 (1998) 203.
Córdoba L.F.; Fuentes G.A.; Correa C.M.. Micropor. Mesopor. Mater., 77 (2005) 193.
Cowan A.D.; Cant N.W.; Haynes B.S.; Nelson P.F. J. Catal., 176 (1998) 329.
Delahay G.; Guzmán-Vargas A.; Coq B. Appl. Catal. B, 70 (2007) 45.
Dorado F.; de Lucas A.; García P.B.; Romero A.; Valverde J. L. Ind. Eng. Chem. Res., 44 (2005)
8988.
Elzey S.; Mubayi A.; Larsen S.C.; Grassian V.H.. J. Mol. Catal. A, 285 (2008) 48.
Flores-Moreno J. L.; Delahay G.; Figueras F.; Coq,B.. J. Catal., 236 (2005) 292.
Furusawa T.; Seshan K.; Lefferts L.; Aika K.. Appl. Catal. B, 39 (2002) 233.
Gerlach T.; Schütze F.W.; Baerns M.. J. Catal., 185 (1999) 131.
Goula G.; Katzourakis P.; Vakakis N.; Papadam T.; Konsolakis M. Catal. Today, 127 (2007)
199.
Gutierrez L.; Ulla M.A.; Lombardo E.A.; Kovács A.; Lónyi F., Valyon J. Appl. Catal. A, 292
(2005) 154.
Gutierrez L.B.; MiróE.E.; Ulla M.A.. Appl. Catal. A, 321 (2007) 7.
Hadajiivanov K.; Saussey J.; Freysz J.L.; Lavalley J.C.. Catal. Lett., 52 (1998) 103.
Hadjiivanov K.; Penkova A.; Daturi M.; Saussey J.; Lavalley J.C.. Chem. Phy. Lett., 377 (2003)
642.
Haneda M.; Bion N.; Daturi M.; Saussey J.; Lavalley J.; Duprez D.; Hamada H. J. Catal., 206
(2002) 114.
He C.H.; KÖhler K.. Phys. Chem. Chem. Phys., 8 (2006) 898.
He C.H.; Paulus M.; Find J.; Nickl J.A.; Eberle H.J.; Spengler J.. J. Phys. Chem. B, 109 (2005)
15906.
He H.; Zhang C.; Yu Y.. Catal. Today, 90 (2004) 191.
Held W.; Konig A.; Ritcher T.; Puppe L.. SAE Paper, 900496 (1990).
Huuhtanen M.; Kolli T.; Maunula T.; Keiski R.L.. Catal. Today, 75 (2002) 379.
Ingelsten H.H.; Zhao D.; Palmqvist A.; Skoglundh M.. J. Catal., 232 (2005) 68.
Ivanova E.; Hadjiivanov K.; Klissurski D.; Bevilacqua M.; Armaroli T.; Busca G.. Micropor.
Mesopor. Mater,46 (2001) 299.
Iwamoto M.; Yahiro H.; Yu-u Y.; Shundo S.; Miauno N.; Shokubai, 32(1990) 430.
Joubert E.; Courtois X.; Marecot P.; Canaff C.; Duprez D.. J. Catal., 243 (2006) 252.
Kameoka S.; Ukisu Y.; Miyadera T.. Phys. Chem. Chem. Phys., 2 (2000) 367.
Kikuchi E.; Yogo K.. Catal. Today, 22 (1994) 73.
Kubacka A.; Janas J.;Włoch E.; Sulikowski B.. Catal. Today, 101 (2005) 139.
Kubacka A.; Janas J.; Sulikowski B.. Appl. Catal. B, 69 (2006) 43.
Larrubia M. A.; Ramis G.; Busca G.. Appl. Catal. B, 30 (2001) 101
Lee T.J.; Nam I.S.; Ham S.W.; Baek Y.S.; Shin K.H.. Appl. Catal. B, 41 (2003) 115.
Li G.; Larsen S.C.; Grassian V.H.. Catal. Lett., 103 (2005) 23.
Li G.; Larsen S.C.; Grassian V.H.. J. Mol. Catal. A, 227 (2005) 25.
Li G..; Wang X.; Jia C.; Liu Z.. J. Catal., 257 (2008) 291.
Li L.; Chen J.; Zhang S.; Guan N.; Richter M.; Eckelt R.; Fricke R.. J. Catal., 228 (2004) 12
Li L.; Zhang F.; Guan N.; Richter M.; Fricke R.. Catal. Commun., 8 (2007) 583.
Li L.; Zhang F.; Guan N.. Catal. Commun., 9 (2008) 409.
Li Y.; Armor J.N.. J. Catal., 145 (1994) 1.
Liu Z.; Woo S.I.; Lee W.S.. J. Phys. Chem. B, 110 (2006) 26019.
www.intechopen.com
Lónyi F.; Valyon J.; GutierreL. Z; Ulla M.A.; Lombardo E.A.. Appl. Catal. B, 73 (2007) 1.
Loughran C.E.; Resasco D.E.. Appl. Catal. B, 7 (1995) 113.
Luo C.; Li J., Zhu Y.; Hao J.. Catal. Today, 119 (2007) 48.
Ma Xiaofei; Wang Xinping; Bi Ran; Zhao Zhen; He Hong. J. Mol. Catal. A, 303(2009) 90.
Meunier F.C.; Zuzaniuk V.; Breen J.P.; Olsson M.; Ross J.R.H.. Catal. Today, 59 (2000) 287.
Mihaylov M.; Hadjiivanov K.; Panayotov D.; Appl. Catal. B, 51 (2004) 33.
Mosqueda-Jiménez B.I.; Jentys A., Seshan K.; Lercher J.A.. Appl. Catal. B, 43 (2003) 105.
Mosqueda-Jiménez B.I.; Jentys A.; Seshan K.; Lercher J.A.. Appl. Catal. B, 46 (2003) 189.
Nakamoto K.. Infrared Spectra of Inorganic and Coordination Compounds (Mir, Moscow.
1996)
Narbeshuber T.F.; Brait A.; Seshan K.; Lercher J.A.. J. Catal., 172 (1997) 127.
Nishizaka Y.; Misono M.. Chem. Lett., (1994) 2237.
Niu J.; Yang X.; Zhu A.; Shi L.; Sun Q.; Xu Y.; Shi C.. Catal. Commun., 7 (2006) 297.
Pieterse J.A.Z.; van den Brink R.W.; Booneveld S.; de Bruijn F.A.. Appl. Catal. B, 46 (2003)
239.
Pieterse J.A.Z.; Booneveld S.; Appl. Catal. B, 73 (2007) 327.
Pirngruber G.D.; Pieterse J.A.Z.. J. Catal., 237 (2006) 237.
Poignant F.; Freysz J.L.; Daturi M.; Sauaaey J.. Catal. Today, 70 (2001) 197.
Resini C.; Montanari T.; Nappi L.; Bagnasco G.; Turco M.; Busca G.; Bregani F.. J. Catal., 214
(2003) 179.
Sasaki M.; Hamada H.; Kintachi Y.; Ito T.. Catal. Lett., 15 (1992) 297.
Satsuma A.; Yamada K.; Sato K.; Shimizu K.; Hattori T.; Murakami Y.. Catal. Lett., 45 (1997)
267.
Satsuma A.; Shimizu K.. Prog. Ener. Com. Sci., 29 (2003) 71.
Sedlmair C.; Gil B.; Seshan K.; Jentysa A.; Lercher J.A.. Phys. Chem. Chem. Phys., 5 (2003)
1897.
Sedlmair C.; Seshan K.; Jentys A.; Lercher J.A.. J. Catal., 214 (2003) 308.
Seijger G.B.F.; van Kooten Niekerk P.; Krishna K.; Calis H.P.A.; van Bekkum H.; van den
Bleek C.M.. Appl. Catal. B; 40 (2003) 31.
Shi C.; Cheng M.; Qua Z.; Yang X.; Bao X.. Appl. Catal. B, 36 (2002) 173.
Shibata J.; Shimizu K.; Satokawa S.; Satsuma A.; Hattori T.. Phys. Chem. Chem. Phys., 5 (2003)
2154.
Shibata J.; Takada Y.; Shichi A.; Satokawa S.; Satsuma A.; Hattori T.. Appl. Catal. B, 54 (2004)
137.
Shichi A.; Satsuma A.; Iwase M.; Shimizu K.; Komai S.; Hattori T.. Appl. Catal. B, 17 (1998)
107.
Shichi A.; Katagi K.; Satsuma A.; Hattori T.. Appl. Catal. B, 24 (2000) 97.
Shichi A.; Satsuma A.; Hattori T.. Appl. Catal. A, 207 (2001) 315.
Shichi A.; Satsuma A.; Hattori T.. Appl. Catal. B, 30 (2001) 25.
Shichi A.; Satsuma A.; Hattori T.. Catal. Today, 93–95 (2004) 777.
Shimizu K.; Kawabata H.; Maeshima H.; Satsuma A.; Hattori T.. J. Phys. Chem. B, 104 (2000)
2885.
Shimizu K.; Okada F.; Nakamura Y.; Satsuma A.; Hattori T.. J. Catal., 195 (2000) 151.
Shimizu K.; Shibata J.; Yoshida H.; Satsuma A.; Hattori T.. Appl. Catal. B, 30 (2001) 151.
Shimizu K.; Sugino K.; Kato K.; Yokota S.; Okumura K.; Satsuma A.. J. Phys. Chem. C, 111
(2007) 6481.
www.intechopen.com
Stakheev A.Y.; Lee C.W.; Park S.J.; Chong P.J.. Catal. Lett., 38 (1996) 271.
Sullivan J. A.; Keane O.. Appl. Catal. B, 61 (2005) 244.
Sultana A.; Haneda M.; Fujitani T.; Hamada H.. Micropor. Mesopor. Mater., 111 (2008) 488.
Szanyi J.; Paffett M.T.. J. Catal., 164 (1996) 232.
Szanyi J.; Kwak J.H.; Moline R.A.; Peden C.H.F.. Phys. Chem. Chem. Phys., 5 (2003) 4045.
Szanyi J.; Kwak J.H.; Peden C.H.F.. J. Phys. Chem. B, 108 (2004) 3746.
Torre-Abreu C.; Henriques C.; Ribeiro F.R.; Delahay G.; M Ribeiro.F.. Catal. Today, 54 (1999)
407.
Tsyntsarski B.; Avreyska V.; Kolev H.; Marinova Ts.; Klissurski D.. J. Mol. Catal. A, 193
(2003) 139.
Wang C.; Wang X.; Xing N.; Yu Q.; Wang Y. Applied Catalysis A. General 334 (2008) 137.
Wang X.; Xu Y.; Yu S.; Wang C.. Catal. Lett., 103 (2005) 101.
Wang X.,; Yu S.; Yang H.; Zhang S..Appl. Catal. B, 71 (2007) 246.
Wang X.; Zhang S.; Yu Q.; Yang H.. Micropor. Mesopor. Mater., 109 (2007) 298.
Wang X.; Yang H.; Yu Q.; Zhang S.. Catal. Lett., 113 (2007) 109.
Wang X.; Yu Q.; Li G.; Liu Z.; Catal. Lett., 123 (2008) 289.
Wu Q.; He H.; Yu Y.. Appl. Catal. B., 61 (2005) 107.
Xing N.; Wang X., Zhang A.; Liu Z., Guo X.. Catal. Comm., 9 (2008) 2117.
X Zhan.; He H.; Ma Z.. Catal. Comm., 8 (2007) 187.
Yeom Y. H.; Li M.; Sachtler W. M. H.; Weitz E.. J.Catal., 238 (2006) 100.
Yu Q.; Wang X.; Xing N.; Yang H.; Zhang S.. J. Catal., 245 (2007) 124.
Yu Y.; He Hong; Feng Q.; Gao H.; Yang X.. Appl. Catal. B, 49 (2004) 159.
www.intechopen.com
17
Description of Topography of Surfaces and

Thin Films with the use Fourier Transformation,
Obtained from Non-Standard
Optical Measurements
Janusz Jaglarz
Institute of Physics, Cracow University of Technology
Poland
1. Introduction
This chapter explains how to interpret the results obtained from optical scattering study to
characterize topography surfaces and films. In particular, it focuses on measurement of
bidirectional reflectance distribution function (BRDF) and on optical profilometry (PO).
These techniques allow to determine two main functions describing surface topography,
namely: power spectral density (PSD) and autocorrelation function (ACF). They characterize
any real surface quantitatively and qualitatively. The PSD and ACF functions are commonly
calculated from Fourier transform (FT) of the surface profiles determined in AFM, SEM or
PO) measurements.
The optical scattering is directly related to material and surface properties. We can
distinguish three types of scatterings base on their features: 1) Topographic scattering
resulting fom roughness causing phase fluctuations impressed on the reflected wavefront
by the surface height variation [Beckmann and Spizinochino 1963] 2) material scattering
created by fluctuations in the composition or density of the surface material [Elson 1984].
Defect scattering is resulting from a presence of sparse distribution of some surface features
responsible for scattering, is distributed broadly and continuously over surfaces such as pits
or bumps in case of topography and patches of different reflectivity in the case of material
scattering [Stover 1995a].
One should emphasize that the roughness is the main source of scattering on surfaces at
visible wavelengths range.
Topography of real surface may be described using some fundamental parameters
described in national and international norms [ANSI/ASME B46.1, ISO 2517]
2003]): 1) Root mean square roughness σ (rms). It is calculated by the vertical deviations of a
For macroscopic characterization of surface the following terms is applied [Whitehause
by the ratio: T = 21/2 σ /s. 3) The power coefficient the type of qualifying statistical
real surface from its ideal form. 2) The rms slope s and correlation length which is expressed
distribution of surface heights. The distance between respective surface features is defined
as spatial wavelengths. In topographic analysis the term spatial frequency which is a reverse
of spatial wavelength is used.
www.intechopen.com
The given statistical terms are calculated from the surface-profile data in a specified spatial
bandwidths. The range of measured spatial wavelengths is determined by lateral resolution
of measuring tool and sampling length in short and long wavelength limit respectively.
2. The power spectral density and autocorrelation function

The real surface and films are described by two statistical functions, namely autocorrelation
(ACF) and power spectral density (PSD) functions [Bennett and Mattsson 1999]. PSD
function is defined in spatial frequency domain and it expresses the roughness power per
unit frequency over the sampling length. From mathematical point of view, PSD is
evaluated from Fourier transform (FT) of surface profile h(r). For one dimensional profile
h(r) PSD is described as:
L − L∫/2
PSD( f r ) = lim
2
h(r )e i 2π f dr
2 L/2
(1)
L →∞
Where f – is spatial frequency and L is sampling length. The units one dimensional PSD are
length to third power.
The roughness and average slopes are first and second statistical momentums of PSD
function so they may well be directly calculated from PSD frequency spectra by the
following integrals:
σ2 = ∫ PSD ( f ) df
f max
(2)
f min
s2 = ∫ (2π f )2 PSD ( f ) df
f max
(3)
f min
If values of PSD are known, one can determine the statistical parameters, such as root-mean
square (rms) roughness, slopes and correlation length by using ABC model which describe
PSD in simple analytical formula [Elson and Benett 1995]:
PSD( f ) = A[1 + ( Bf )2 ]−C / 2 (4)
The A parameter is a PSD(f) value for low frequency formula, B/2π is correlation length and
where A, B, C are model parameters related with basic quantities characterizing a surface.
C determine type of power law in high spatial frequency. The ABC model applies for single
surface.
The other way of surface characterization is describing it by means of autocorrelation
function ACF(r) defined in spatial wavelength domain. Generally the autocorrelation
function of surface is used to compare data sets of heigths h(r) to a translated version of itself
and the averaging.
ACF ( r ) = lim ∫ h(r )h(r + τ )dr

1 L/2
(4a)
L →∞ L
− L /2
www.intechopen.com
Description of Topography of Surfaces and Thin Films with the use Fourier Transformation,
Obtained from Non-Standard Optical Measurements 365
Where translation τ is so called lag distance. The ACF function is usually calculated from
surface profiles obtained in AFM and SEM studies. For many random surfaces the ACF(r)
may be expressed in the following form [Bendat and Piersol 1971, Whitehead 2003]:
2α
ACF(r ) = ACF(0)exp( −
r
) (5)
T
specifies a type of statistical distribution of surface irregularities. So for α = ½ and α = 1 the

Where ACF(0) is the value of autocorrelation function in point r = 0. The value of coefficient
ACF(r) describes Lorentzian and Gaussian distribution respectively. For distance r= T the
ACF(r) decreases e – fold. Thus the value of r = T defines autocorrelation length. Such
defined ACF is equal statistically averaged surface spatial wavelength.
In some surface investigations obtained by the use of optical methods, as in optical
profilometry, the following useful relation between average slope s and ACF function could
be applied [Stover et. al 1984]:
d 2 ACF(τ )
m2 = lim
dτ 2
(6)
τ →∞
According to Parseval’s theorem PSD and ACF provide the same information about surface
statistic expressed in space frequency and wavelength domain.
Therefore the PSD and ACF may express by using Wiener – Kninchin relations [Bendat and
Piersol 1986] as Fourier transform and inverse FT:
∫ PSD ( f x )e
∞
ACF(r ) = j 2π f τ
df (7)
−∞
∫ ACF (τ ) e
∞
PSD( f ) = − j 2π f τ
dτ (8)
−∞
The calculating procedure of macroscopic surface parameters is shown on Fig. 1.
Fig. 1. Scheme of calculating surface parameters procedure from optical measurements
www.intechopen.com
3. The BRDF technique

Optical measurements allow to find macroscopic parameters of surfaces and films from
angular measurements of scattered light. The very good tool for single surface and
multilayered films characterization is bidirectional reflection function method (BRDF)
In BRDF the differential power of scattered beam dP per solid angle of receiver aperture dΩ
[Nicodemus 1965].
in the θs direction and per incident power Pi coming from the θi direction is measured. In
Fig. 2a the geometry of incident and measured reflected radian beams have been presented.
a)
b)
Fig. 2. Geometry of BRDF measurement (a) and scheme of BRDF measurement setup (b)
Practically dP/dΩ is equal to the measured scatter power Ps per acceptance angle Ω of a
detector namely:
dP / dΩ
BRDF = = ⎡sr −1 ⎤
Pi cosθ s Pi Ω cosθS ⎣ ⎦
PS
(9)
If studied surface is relatively flat, the excellent approach give vector The Raleigh-Rice vector
perturbation theory [Elson and Benett 1979] is best applied. It shows simple dependence
between scattered radiation expressed by BRDF and PSD of investigated surface.
16π 2
BRDF = cosθ i cos θ sQ PSD( f )
λ2
(10)
www.intechopen.com
and λ where λ is the light wavelength. PSD is represented in spatial frequency f that is
where Q is a factor dependent on polarization state of the light source, the optical constants
related with angles θi and θs by following formula:
sin(θ s ) − sin(θ i )
f =
λ
(11)
If the surface roughness is spatially isotropic, as is usually assumed for randomly polished
surfaces, the PSD depends only on the magnitude of the surface spatial frequency and is
independent of its direction on the surface plane. All information about the surface
spectrum can be obtained from scattering measurements made in the plane of incidence.
phenomena, which means that BRDF and PSD (except for the factor Qcosθs) are directly
This allows to extract the topographic structure of the single surface from scattering
proportional. The so called “golden rule” is a principle of surface investigations by the use
angular scatterometry [Church et al. 1977, Stover 1995b].
BRDF measurements presented in this chapter were performed with an automatic
scatterometer (Fig. 2b). It consists of a 650 nm laser diode as the light source (with the beam
diameter of 2 mm) mounted on a goniometric table with 0.01 deg resolution. The light
scattered at the sample surface is measured with a Si photodiode detector. The rotations are
obtained by a computer controlled stepper motors. For a fixed angle of incidence, the
scattered intensity in the plane of incidence are measured by varying the detector
orientation. All measurements are carried out with the s-polarized incident beam. In any
case, the sample surface size was much larger than the beam diameter. Moreover, the
minimal illuminated area (4 mm2) is large enough to yield meaningful statistical description
of the surface [Stover 1984].
4
2
0 a
-2
-4
50 55 60 65 70
log(BRDF)
4
2
0
-2 b
-4
-6
-2 50 55 60 65 70
-3 c3
-4
angle of incidence θi
50 55 60 65 70
Fig. 3. BRDF vs. angle of incidence for 3 different samples: flat silicon–a, polished steel – b,
SWM7 and Spectralon – c surfaces
www.intechopen.com
samples. The image in Fig. 3 consists of polished silicon with roughness σ = 1nm (Fig.
Fig. 3 shows the results of BRDF measurements as a function of scattering angle for 3
3a), mechanically polished steel σ = 69 nm SWM7 with (Fig. 3b) and lambertian white
diffuser Spectralon (Fig. 3c). Because scattered intensity strongly depends on scattering
angle and intensity ratio for near specular to far specular scattering may reach several
same θi = 600 for all samples. As it is easy to notice the angular BRDF’s are different. The
orders so generally the BRDF is shown in logarithmic scale. The incidence angle was the
BRDF of polished Si as other good mirrors has high narrow specular peak. Outside specular
reflection mirror reflectors scatter diffusely light at very low level. Polished steel with
moderately high roughness has BRDF with lower and wider specular peak and intensity of
scattered light monotonically decrease in larger scattering angle. In Spectralon [Workmann
1998] BRDF specular reflection disappears and light is scattered diffusively and
homogeneously in all direction.
In the case of polished surfaces the power spectrum of the surface errors is generally a
smooth and broad function of spatial frequency, and the smoothed value of the BRDF is
independent of the system pupil function.
In a situation when optical constants are unknown, the polarization factor Qs for “s”
given by the geometric mean of the sample specular reflectances R(θi) and R(θs) at angles θi
polarization could be found from angle dependence of specular reflection coefficient. It is
and θs. For scattered radiation measured in plane of incident Qs is equal [Church et al. 1989,
Stover 1995b]:
Qs = { Rs (θ i )Rs (θ s )} 2
1
(12)
This equation allows to determine Qs without knowing the optical constants of the sample.
Because Qs is a smooth function, a good curve fit may be found by measuring sample
reflectance for several angles of incidence. Of course, if optical indices are known, the Qs
factor can be calculated directly from the definition.
specular reflectance measurements performed for 7 incidence angles θi from range 600 to 810.
Fig. 4 Shows PSD(f) calculated from BRDF. The polarisation factor Qs was determined from
The ABC model was fitted to experimental data.
0,6
PSD [ μm ]
-3
3
0,3
0,0
1
0,2 0,4 0,6 0,8 1,0 1,2 1,4
spatial frequency [ 1/μm]
Fig. 4. PSD function extracted from BRDF measurements for TiN films on different
substrates on polished: 1 —Si, 2 —SWM7 steel and 3 — on WC plate respectively
www.intechopen.com
Roughness σ of TiN films from equation (4) and autocorrelation length T, and power factor
C from ABC model (5) have been found. The values of determined parameters are shown in
Table 1. The result strongly suggest the influence of substrate type on topography of TiN
films obtained under the same technological conditions.
Log-Log analysis of PSD(f) is very convenient and easy to interpretation of PSD(f)
dependence. Many optical surfaces are described by the means ABC model. On Fig. 5 the
log-log PSD(f) for Duralcan alloy before and after CO2 laser annealing [Kaczmar et al. 2000]
has been presented.
100 frequency cutoff

T2=1/(2π*f2)
σ2 = π*f*PSD2(0)
1/2
Log(PSD) [μm ]
-3
σ1 = π*f*PSD1(0)
10
1/2
1
frequency cutoff
T1=1/(2π*f1)
1E-3 0,01 0,1
Log(f) [1/μm]
Fig. 5. The values of topographic parameters have been obtained as following:

For low spatial frequencies Bf << 1 the PSD(f) has constant value. For high frequencies Bf >>
1 and the formula (3) may be approached by the use ABC model as follows:
Log ( PSD ) = Log ( A ) − CLog( Bf ) (13)
Substituting for y = Log(PSD) and for x=Log(Bf), we obtain linear dependence: y = -C x + b,

where C qualify type of random distribution of irregularietes. For the special case C=2 or
C=4, the distribution of surface highs is Lorentzian or Gaussian respectively.
If we try to extrapolate PSD(f) curves in low and high spatial frequency bandwidth by
straight lines one then we will obtain break point on cut of them. It allows to determine
autocorrelation length according formula: T=1/2π*fc. According to (2) the surface roughness
from this point so called cut-off frequency. By knowing its value one may to calculated
frequency fc one may estimate the σ follows to approached formula: σ1/2 = π*fc*PSD(0),
is equal field under the PSD curve. For low frequency bandwidth i.e from zero to cut-off
where PSD(0) denote power spectral density at low frequency.

For some surfaces surfaces in their PSD or BRDF versus frequency the break point do not
occur (that is Bf>>1). In that case the PSD and BRD may be approximated linear with slope
C. For random one dimensional surface from ABC model we obtain from (4):
PSD( f ) = ABf −C (14)
The formula (14) expressed in logaritmic terms is the same as (12). Surfaces without break
point are called fractal surface [Church 1988]. The autocorrelation length T of fractal profile
www.intechopen.com
is equal to the length of investigated surface. On Fig. 6 the BRDF versus spatial frequency
for 3 fractal surfaces: 1-glass BK7, 2-polished stainless steel, 3-BaSO4 white standard is
shown.
1000
10
3
Log(PSD)[a.u.]
0,1
1E-3
2
1
1E-5
1E-7
0,01 0,1
Log(f) [a.u.]
Fig. 6. The power spectral density for fractal profile, 1-BK7 glass, 2-stainless steel, 3-BaSO4
The curves 1 and 2 represents fractal glass and steel surfaces. The C coefficients of power
spectrum slopes are nearly 4, so their highs of fractal surface have Gaussian distribution.
Curve 3 presents power spectrum for barium sulphate which is applied as Lambertian
diffuser [Workmann 1998]. As may easy conclude white standard diffuser are fractal
surfaces with zero slope of PSD spectra.
The many practical surfaces in optics or semiconductor application exhibit the fractal-like
power spectral density over bandwidths of interest.
As it is known the roughness value depends on spatial bandwidth which is specified by
measurements setup and other conditions such as sample size etc. Because PSD spectra of
measured surfaces may differ considerably results of roughness may be influenced by
position and width of chosen bandwidth. Fig. 7 presents the calculated roughness versus
spatial frequency area for SWM7 steel, stainless steel and silicon surfaces respectively.
0,02
0,01 2
σ [μm]
0,00
f [1/μm]
-0,02 0,00 0,02 0,04 0,06 0,08 0,10 0,12 0,14
Fig. 7. Surface roughness as a function of spatial frequency bandwidth obtained from BRDF
for, SWM7 steel, stainless steel and silicon surface - curve 1, 2, 3 respectively
www.intechopen.com
As it may be concluded from the above figure, for some surfaces roughness values reach
constant value at low frequencies and topographic features observed at high frequencies
essentially do not contribute to total roughness.
If the surface roughness contains periodic components, such as tool marks in precision-
machine surfaces, the BRDF will contain sharp diffraction lines. The positions and intensities
of those lines are related to the feed rates and amplitudes of the tool marks, while their
widths are determined by the system pupil function. The case of periodic surface is shown
on Fig. 8. The BRDF for ST3SY steel concurrent (curve 2) and reverse concurrent (curve 1) in
milling process is shown.
-2
1 reverse concurrent milling

-3
2 concurrent milling
-4
f+f0 f+4f0
-5
ln(BRDF)
-6 2
-7
1
-8
-9
f [1/μm]
0,00 0,05 0,10 0,15
Fig. 8. BRDF for periodic steel surface obtained in milling process

Spatial wavelength of fundamental mill calculated from BRDF is equal Λ0=154μm. All
interval frequency f0 = 1/Λ0.

remaining peaks from low frequency bandwidth occur on BRDF spectrum in the same
Some random optical surface may not be described by the use ABC model. It is for mono-
crystalline silicon wafer with (100) orientation etched in KOH solution. In this case
reduction of surface reflectivity is a key factor for improving the crystalline silicon solar cell
efficiency. In Fig. 9 the relative BRDF frequency spectra for that wafer have been shown.
100
90
80
70
θi = 60
BRDF [a.u]
60 0
50
θi = 45
40
0
30
1E-3 0,01 0,1
f [a.u]
Fig. 9. BRDF of texturized monocrystalline Si in KOH for two angles of incidence
www.intechopen.com
For high spatial frequency slopes have positive values unlike to most optical surfaces. This
peculiar behaviour in BRDF(f) dependce is due to surface texture of etched silicon wafer. It
consists of pyramid-like irregularietes differing in sizes and shapes [Lipiński and
Cichoszewski 2008]. So created texture scatters light into hight angles (frequencies). It allows
to substantially reduce reflection for near specular angles.
Additionally one influences of non-topographic features on power spectral density function
should be discussed. Such cases can be found in composite (alloys) structures. If
components are different the large variations in optical constants may occur. Apart from
large reflectance differences, the phase change of reflection for absorbing materials appear.
This phase change resulted from reflectance and may equal several tens of degrees for
conductors and can be interpret as pseudo-height differences [Bennet and Mattsson 1999]. In
other words a hypothetical, perfectly smooth surface made up of elements having different
optical constants may show shadowing effect ordinarily interpreted as height differences in
optical measurements. For flat surfaces dielectric function variation plays key role in light
scattering. This situation can be found in optical study of Al-Si-SiC composites [Kaczmar el
al. 2000]. Additionally in Al-Si-SiC structures such a metal-semiconductor-dielectric
composites particular monocrystals may occurr on the surface and may be considerably
larger than light wavelength. They have own surface direction and reflect light according to
geometrical law of reflection.
The phase changes of reflected light which are components of the investigated composites
(alloys) result from appreciable differences in their optical constants. In experimental data
analysis, the differences in n and k indices particular components could be treated as
for light with wavelength λ =550 nm, the pseudo-heights are for Al -24 nm, 2.5 nm for Si and
pseudo-height caused by phase differences. For example for the aboveAl-Si-SiC composites
0 nm for SiC for which is dielectric material and was treated as reference. Therefore the
pseudo-heights must be taken into consideration when predicting values of power spectral
density and autocorreation functions are obtained by the use of optical methods.
4. The Power spectral density of thin films from BRDF study

The BRDF results for film depend on topographies of its upper and lower side and in the
case of diffusive layers depend also on light scattering in the bulk.
1. If the topography of both upper and bottom interfaces are identical the PSD is given by
[Elson 1992]:
BRDF = F1 + F2 PSD
2
(15)
where F1 and F2 are optical function of interfaces film- air and film – substrate
respectively, and PSD represents power spectral density function of film system.
2. In the case completely uncorrelated surfaces the BRDF function should be expressed by
the formula:
BRDF = F1 PSD1 + F2 PSD2

2 2
(16)
where PSD1 and PSD2 denote power spectral density for top and bottom film surface
respectively. In this case, the scattered power from the different interfaces is added.
Anyway one can extract part of scattering coming only from the upper surface using
full internal reflection effect in the film.
www.intechopen.com
Fig 10. shows reflection from transparent film on rough substrate. Part of the light reflect
from the top surface, part after refraction on the boundary of the media traverse the film
twice due to reflection from substrate and goes out from the sample (see Fig. 10). This
surface is rough, the part radiation is scatter into angles θs that are larger than Brewster
situation occurs for specular reflection for any angle of incidence. If however bottom
angle θΒ. For this case the all radiation scattered at angles θs > θΒ is internally reflected.
Fig. 10. Propagation of scattered light in layer with rough bottom

It means that the reflected light at suitably great angles comes only from scattering from
upper interface of the film and allows to determine PSD from BRDF measurements for
upper surface of the film.
If a bulk scattering does not occur the optical losses originate from surface and interface
only. In that case determining power spectral density can be done from BRDF
measurements. Fig. 11 shows determined PSD function for bare polished silicon 1 – curve 1,
thermally obtained SiO2 on the same Si substrate – curve 2, porous silica and titania-silica
blend films on BK7 glass obtained by sol-gel technique [Karasiński 2005] –curve 2 and 3
respectively.
4
0
3
Log (PSD )
-2
2
1
-4
-6
-8 -6 -4 -2 0
Log (f )
Fig. 11. PSD function of thin films. Curve 1 – bare silicon, 2 –Thermal SiO2 on Si, 3 – porous
silica and 4 porous SiO2-TiO2 on BK7 glass
www.intechopen.com
As it may be noticed on Fig. 11 for films 2 and 3 the irregularities with low spatial frequency
(longer spatial wavelength) contribute larger fraction to total roughness than the higher
ones. Therefore autocorrelation lengths T for layers 1 and 2 are bigger than for thermally
obtained silica. The PSD obtained for silicon and SiO2 on the same Si before after annealing
identical. The values of σ determined from AFM and BRDF measurements are similar
are very similar. It results from the fact that upper and lower interfaces of SiO2 film are
however roughness calculated from BRDF study are larger. As a matter of fact in BRDF we
measure scattered radiation from much bigger surface area than in AFM. The total
roughness enlarge due to fraction of irregularities coming from longer spatial wavelengths
measured by the use BRDF technique.
The values of macroscopic parameters for surface and films presented in this work have
been shown in Table 1. The results given in Table 1 concern optical surfaces. Such surfaces
can meet I optics and optoelectronics applications. For those surfaces roughness is
commonly less than 5 nm and their autocorrelation length are order hundred manometers to
micrometers. So well average slopes are less than 10-2. In optics surfaces with slopes less
than 10-2 are regarded as flat [Ohlidal 1988].
sample Thickness Roughness σ T C

[nm] [nm] [nm]
Polished Si - 1.7 2730 3.91
Thermal SiO2 on Si 352 2.9 2890 3.88
Porous silica on Si 672 2.1 770 1.97
titania-silica on Si 211 1.2 455 2.11
TiN on Si 330 11.7 6780 3.61
TiN on steel 393 34 13890 3.81
Polished aluminium 7.3 3100 3.64
Table 1. The values of topographic parameters of random layers and surfaces
5. Determination of the PSD and ACF function from optical profilometry

Measurements of optical reflectance by means of classical reflectometry inform us about
optical properties on a large area, i.e. of the order of 1-5 cm2. The results obtained on a much
smaller will be similar if coatings and surfaces are homogenous over the investigated area
uniformities differ from tens μm to several mm, the measurements taken from the
and inside the layers. For inhomogeneous surfaces, when topographic or materials non-
integrating sphere measurement and standard reflectometry give rather an averaged

reflectance over a larger scale reflected samples.
The scattered radiation measured by optical profilometry (OP) is a function of heights of
irregularities and slopes of microfacets, but sensitivity of this method derives mainly from
detection of the slope change [Brown and Breitmeier 1988, Whitehead 2003]. The presence of
www.intechopen.com
long lateral irregularities is often caused by manual or mechanical treatments and may have
a periodical nature. The short spatial waves result rather from random process of the surface
formation and their contribution to the total profile is easy to determine from atomic force
microscopy (AFM) technique.
Resolution of OP depends on beam diameter where diameters changes from 1 μm to 1 mm.

Optical profilometry measures intensity of reflected radiation from the surface point.
In this chapter we presented result of profilometric studies. The long spatial wavelength
irregularities detected in OP investigations may contribute substantially to the total
roughness. Optical profilometry measurements complete the topography description in
long spatial wavelengths.
The optical profilometer described in this chapter is multifunctional experimental set up for
surface topography investigations. It work in two modes. The first mode – specular mode
light beam with 12 μm diameter. OP measurements have been normalized with the
employs with laser He-Ne as the light source with collimating system allowing to achieve
calibration sphere method [Mainsah et al. 1996]. A lead screw stepper motor actuated
optical profiles of surface with 20 μm lateral resolution. The scheme of optical profilometer
device may scan 30 mm × 30 mm surface with step of 0.02 mm. It allows one to obtain the
(OP) is shown in Fig. 12.

The second mode of OP – BRDF mode which is more preferred for rougher films surfaces is
based on reflection probe R-200-7. This mode is used to get reflectance over smaller areas at
different angles of incidence. The probe consists a bundle of 7 optical fibres, i.e. 6 fibres
μm. Of course the size of illumination defines lateral resolution of measurements. The laser
around one “illumination fibre” (which illuminates sample), each with a diameter of 200
diode (λ = 635 nm) is used as the light source. The reflected light is collected by six fibres
around the illumination fibre. The probe is coupled with a collecting fibres to a A/D
converter connected with computer. The axis of the probe may be angled at 15º, 30º and
normal to the sample plane.
Fig. 12. Scheme of optical profilometer: 1 – laser diode (λ = 635 nm), 2,3 – detectors, 4 - beam
splitter, 5 - colleting lens, 6 - objective, 7- sample, 8 -XY stage, 9 – controlling/collecting
unit
The analysis of two-dimensional Fourier transforms of surface profiles expressed as PSD
function of fx and fy coordinates is an effective tool for any texture characterization. Fourier
www.intechopen.com
transformations of surface profiles are determinate by the use of fast Fourier transform (FFT)
procedure. images allow to find distribution of surface heights in inverse space, distinct
their periodicity and evaluate anisotropy of surface films in.
In Fig. 13 a and b, the mechanically polished steel SWM7 block profile, and its two-
dimensional FFT is presented.
a)
b)
Fig. 13. Surface profile –a and its FT –b of mechanically polished steel SWM7
Grooves and scratches visible in Fig. 13a are the result of the polishing process. These
features cover the whole surface in a more or less uniform way and overlap. FFT allows
toassess the repeatability of the processes forming the periodic inequalities. As may be
noticed, in its two-dimensional FFT (Fig 13b) periodic dependence of the PSD along the
direction forming an angle 450 with the X and Y axes may be distinguished. It is result of
interference of spatial waves appearing in X and Y directions. The occurrence of spatial
waves interferences indicate the same processes along mutually perpendicular
directions.
Fig. 14a shows FFT of optical profile of polyvinylcarbazole (PVK) film deposited on
Corning glass 7059, obtained by spin -coating method [Patel et al. 2007], The scanning
area was 10x10 mm which enables detecting of surface features in longer space
wavelengths.
www.intechopen.com
a)
7
5
PSD [μm2*mm]
0
0 1 2 3 4 5
b) f [1/mm]
0,030
0,025
0,020
0,015
SDF
0,010
0,005
0,000
0 2 4 6 8 10 12 14
c) Slopes [arbitrary units]

Fig. 14. FFTprofile of PVK film on Si substrate obtained by means OP BRDF mode –a and
one dimensional PSD along X and Y axis –b, relative slopes distribution function for PVK
film –c respectively
www.intechopen.com
Because refractive index of PVK for 633 nm light wavelength is nearly the same as Corning
7059 glass [10], almost the whole scattered light originates from upper surface of the film.
FFT analysis of optical profiles shown in spatial frequencies, allows to distinguish periodic
features and specify interference of spatial waves and to evaluate ratio of surface anisotropy.
Because PVK was formed by centrifugal force in spin–coating process the PSD functions
obtained for X and Y axes are similar, what is presented in Fig 14b. The determined PSD
demonstrates as well periodic structure of surface. Also the relative slope density function
(SDF) was determined for this sample (Fig. 14c).
OP measures the local slope of microfacets (local surfaces), the autocorrelation function is
given as:
ACF(s ) = σ 2 p(s ) (17)
This relation may be applied to ACF determination directly from OP measurements if p(s) is
known.
Fig. 15a shows FT from optical profile of satinated glass texture used in glass houses. In Fig.
15b the calculated PSD function for x and y spatial frequency is presented.
4,5
B
PSD [jedn. umowne]
4,0
3,5
3,0
2,5
2,0
4,0 3,5 3,0 2,5 2,0 1,5 1,0
Czêstotliwoœæ[1/mm]
Fig. 15. FT of satinated glass surface and arbitrary one-dimensional PSD for x and y
direction
As results from Fig 15.b the PSD for x and y directions are nearly the same. Thus one may be
conclude that satinated glass surface exhibits very good isotropy. It is important factor for
glass use in green houses.
www.intechopen.com
On the other hand the OP measurements allow to detect anisotropy of surface creation process.
Optical profilometry could be applied to waviness detection occurring on surface or to
inspection of periodic variation of film thickness. As an example in Fig. 16a the profilometric
map of TiO2-SiO2 film on BK7 glass substrate obtained by dip coating sol-gel method is shown.
a)
b)
c)
Fig. 16. PO profile of TiO2-SiO2 film on –a and its FFT – b and ACF –c
www.intechopen.com
It presents periodic changes of reflection coefficient for TiO2-SiO2 film. In Fig. 16b the
Fourier Transform of OP profile from Fig. 16a is shown. This is due to two harmonic
vibrations caused by mechanical elements of technological equipment Elimination of
vibration allows to obtain very flat TiO2-SiO2 films [Karasiński 2005].
6. Conclusion
As it is shown in this chapter the optical methods may be used as effective tools for surfaces
and interfaces thin films and layers. Surface topographic parameters may be obtained from
biredirectional reflectance distribution function and optical profilometry measurements. The
BRDF and OP techniques allow to get information about surface topography. From these
optical investigations the surface power spectral density (PSD) and autocorrelation (ACF)
functions for surfaces and films may be found. Both functions are calculated from the Fourier
transformation of surface profiles. The root mean square (rms) surface roughness, slopes and
correlation length for studied samples from PSD function have may be determined.
The resolution of BRDF method allows to determine surface parameters in spatial
wavelength range from 0.1 to hundreds of micrometers. The OP may be applied as
complementary method for measurements of longer spatial wavelengths.
The most effective analysis of the surface statistic is calculating power spectral density or
autocorrelation functions from surface profiles in spatial frequency or wavelength domains
respectively.
The scattered radiation is a function of highs of irregularities and slopes of microfacets but
the sensitivity of PO studies is resulted mainly from sensitivity of detection of slopes
change. In PO BRDF mode the technique is sensitive for upper surface of a film when
substrate is optically flat. The presented techniques may be also take advantage in control
manual painting or varnishing. The presence of long lateral irregularities is often caused by
manual or mechanical treatments (i.e paintings, varnishing, manual polishing) and may
have periodical nature. The short spatial waves result rather from random process of surface
formation and their contribution in total profile is easy to determine from light scattering by
the using BRDF method.
If films are transparent and their interfaces are partly correlated, the description of film
topography is difficult.
The PO specular mode may be used for transparent film thickness determination if upper
and lower interfaces are flat. This technique allows measures thickness of layer If interfaces
correlated are in transparent films. If surface are partially correlated the BRDF -PO mode is
more preferred for their topography measurements
7. Acknowledgment
This work was supported by Grant N N503 137235 of Polish Science Committee.
8. References
Print Books
Stover J.C., Optical Scattering. (1995). Measurements and Analysis (2d ed.), SPIE, ISBN 0-8194-
1934-6, Belligham. (b)
Bennett J. M.,. Mattsson L. (1999). Introduction to surface roughness and Scattering (2 ed.) SA
ISBN 1-55752-609-5, Washington, D.C.
www.intechopen.com
Workmann J. ( st ed.) (1998). Applied spectroscopy, Academic Press, ISBN 0-112-764070-3.

Whitehouse D. J. (2003). Handbook of surface and nanometrology, (1st-ed.) IOP Publishing, ISBN
0-7503-0583-5, Bristol and Philadelphia.
Beckmann P, Spizinochino B. (1963) The scattering of electromagnetics waves from rough
surfaces, Pergamon Press Oxford, London, New York, Paris:
Papers in Journals
Elson J.M., Bennett J.M., (1979). Vector scattering theory, Opt. Eng., 18, 116.
Church E.L., Jenkinson H.A.,. Zavada J.M, (1977). Measurement of the finish of diamond-
turned metal surfaces by differential light scattering, Opt. Eng., 16 360.
Chandley P.J. (1976). Determination of the autocorrelation function of height on rough
surface from coherent light scattering, Opt. Quant. Elect. 8 329.
Church E. L. (1988). Fractal surface finish, Appl. Opt. 27 1518.
Ohlidal I., (1988). General formulas for the optical characterization of single layers with
spectroscopic reflectometry, J. Mod. Opt. 35, 1373.
Stover J.C., Serati S.A., Gillespie C.H., (1984). Calculation of surface statistic from light
scattering, Opt. Engng, 23 ,406.
Karasiński P. (2005) Influence of technological parameters on the properties of sol-gel silica
films, Opt. Appl. Vol. 35, No. 1 117.
ANSI/ASME (1985). Standard B46.1-1985, Surface Texture (Surface Roughness, Waviness, and
Lay), Am. Soc. Mech. Eng., New York.
Elson J. M. (1984). Theory of light scattering form rough surface with inhomogeneous
dielectric permittivity. Phys. Rev. B30 5460.
Elson J. M. Benett J. M. (1995) Calculation of the power spectral density from surface profile
data, Appl. Opt., 34, 20.
Stover J.C., Serati S.A., Gillespie C.H., (1984).Calculation of surface statistic from light
scattering, Opt. Engng, 23 406.
Bendat, J.S., Piersol, A.G. (1971) Random Data: Analysis and Measurement, Proceedures, Wiley-
Interscience, New York
Bendat, J.S., Piersol, A.G. (1986) Engineering Applications of Correlation and Spectral
Analysis, ( 2d ed.), John Willey &Sons, New York.
Nicodemus F. E. (1965). Directional Reflectance and Emissivity of an Opaque Surface Applied
Optics, 4 767.
Mainsah E., Dong W.P., Stout K.J., (1996) .Measurement 17 173.
Papers in Conference Proceedings
Kaczmar J.W., Pietrzak K., Wlosinski W. (2000) The production and application of metal
matrix composite materials. J Mater Process Technol., 106 58.
Karasiński P. (2005). Influence of technological parameters on the properties of sol-gel silica
films, Opt. Appl. 35, 117
M.J. Elson, (1992) Theory and Software for Light Scattering from Multilayer Optical
Components with Interfacial Roughness, Naval Air Warfare Center, Weapons
Division, China Lake, CA, Technical Publication 8084.
Stover J.C. (ed.), (1995). Optical Scattering in the Optics, Semiconductor and Computer Disk
Industries, Proc. SPIE 2541. (a)
Patel S., Ramrakhiani M., Bisen D., P. (2007) J. Appl. Polym. Sci, 104, 722.
www.intechopen.com
Church E. L., Leonard T.A., Takacs P. Z. (1989). The prediction of BRDF from surface profile
measurements. Proc. SPIE 1165 196.
Brown A.J.C., Breitmeier U., (1988) Industrial application of an optical profilometer.
Proceedings SPIE, 954.
Lipiński M., Cichoszewski J. (2008) in Proc. 23nd EU PVSEC, (eds. G. Willeke, H.
Ossenbrink, and P. Helm, WIPRenewable Energies, Munich, Germany, 1911.
www.intechopen.com
18
Application of FTIR Spectroscopy to

Agricultural Soils Analysis
Linker Raphael
Faculty of Civil and Environmental Engineering
Technion – Israel Institute of Technology
Israel
1. Introduction
Soil is a very complex medium that contains minerals, organic matter, micro-organisms, air
and water. Soil is one of the most important factors for agriculture and some soils are
deemed more fertile than others. Soil fertility is directly related to factors such as nutrients
concentrations or availability, organic matter content, acidity, moisture, etc., as well as to
agricultural practice such as till vs. no-till (Desbiez et al. 2004) (Sá et al. 2009). Ensuring soil
fertility is a basic requirement for any form of sustainable agriculture, yet in practice this
seemingly trivial goal is very difficult to achieve due not only to the complexity of the soil
medium itself, but also due to the complexity of the soil-crop-air interactions and to the fact
that some processes require years before having any visible impact. Various recent studies
have shown that soil fertility is declining in many farmlands due mainly either to
inadequate farming practices (Gobeille et al. 2006), insufficient fertilization, in which case
the soil reserves are depleted, or over-fertilization that results in pollution to the
groundwater or toxic accumulation of chemicals in the soil. Avoiding such under- or over-
fertilization is the chief goal of the so-called precision fertilization concept, which aims at
delivering exactly the amount of nutrients required to sustain optimal growth of the crop.
One of the main obstacles to the application of the precision fertilization concept, or the
more general precision farming concept, is soil heterogeneity. Hence, although it is technically
possible to perform a wide range of analyses and derive a soil fertility or health index such
as the one proposed by Idowu et al. (2008), most of the required analyses are time-
consuming, which in practice makes it impossible to map the soil properties of a field with
the required spatial and/or temporal resolution. The need for fast and cheap methods that
would enable the analysis of a large number of samples has been stressed in numerous
studies (Viscarra Rossel&McBratney 1998b) and infrared spectroscopy has long been
recognized as one of the most promising techniques (e.g. McCarty&Reeves 2006; Mouazen
et al. 2007; Cécillon et al. 2009). As in other applications, the initial works were conducted
mainly in the near-infrared (NIR) range for which instruments were readily available.
Technological achievements in the mid-infrared (mid-IR) range during the last decade have
made this spectral range much more attractive and an increasing number of soil studies are
conducted using this spectral range. Although there is no doubt that near-infrared
spectroscopy can be very useful for some analyses, several comparative studies have shown
www.intechopen.com
the superiority of the mid-infrared techniques in most cases (McCarty&Reeves 2006;

McCarty et al. 2002; Viscarra Rossel et al. 2006; Canasveras et al. 2010) and therefore the
present chapter will focus solely on mid-IR techniques.
2. Mid-Infrared techniques commonly used for soil analysis

Infrared (IR) spectroscopy is based on the interaction of molecules with electromagnetic
energy in the infrared spectral region, which is in the wavelength range of 0.8-1000 m. This
IR range is commonly divided into four regions, labeled near-, mid-, thermal- and far-
infrared, respectively (Figure 1). The particularity of the mid-infrared range is that it
includes the so-called fundamental vibrations of the molecules. When a molecule absorbs IR
radiation at frequencies matching that of its own molecular vibrations, it results in an
increase of the amplitude of the vibrations at these frequencies. Since each frequency
corresponds to a given amount of energy and a specific molecular motion (e.g. stretching,
bending or contracting of chemical bonds), the mid-IR spectrum can reveal the kind of
molecular motions and bonds (functional groups) that are present in the molecule and hence
can serve as a unique fingerprint of a specific compound. Furthermore, most functional
groups have characteristic IR absorption bands that do not change much from one
compound to another. By comparison, the near-infrared range is dominated by overtones
and combinations of these fundamental vibrations, which makes the interpretation of the
NIR spectra much more difficult.
Fig. 1. The electromagnetic spectrum with emphasis on the ultra-violet – infrared range
Reproduced from Viscarra Rossel et al. (2006).
Nowadays, Fourier Transform Infrared (FTIR) is the preferred method for mid-IR
spectroscopy as it provides quantitative information in a rapid and accurate fashion. A
typical FTIR spectrometer obtains an infrared spectrum by collecting the interferogram of a
sample signal, which contains all the infrared frequencies, applies the Fourier transform to
the digitized signal, and outputs the spectrum. Such a FTIR spectrometer relies on an
interferometer, which splits to radiation beam into two beams that are recombined after a
path difference has been introduced (Griffiths&de Haseth 1986). The most common
interferometer is the so-called Michelson interferometer, which consists of a beam splitter
located between two perpendicular mirrors, one of which can move along an axis
perpendicular to its plane (Figure 2). At the beam splitter, the radiation beam from the
infrared source is partially reflected to the fixed mirror and partially transmitted to the
moveable mirror. The moveable mirror is moved in a highly controlled fashion to create the
www.intechopen.com
Application of FTIR Spectroscopy to Agricultural Soils Analysis 387
path difference between the two beams. After the beams return to the beam splitter, they
interfere and are again partially transmitted and partially reflected to the detector. Due to
the effect of the interference, the intensity of each beam passing to the detector and
returning to the source depends on the path-length difference between the two beams. The
two beams can undergo constructive interference, destructive interference or a combination
of both, depending on the path-length difference. Constructive interference, which yields a
maximum detector signal, occurs when the optical path difference is an integer multiple of
the wavelength. The variation in the energy that reaches the detector as a function of the
path difference yields the interferogram, which is the integral of all interference patterns
produced by each wavelength. The detected interferogram can not be interpreted directly,
but has to be “decoded” using the well-known Fourier Transformation (Griffiths&de Haseth
1986). Fourier transform is typically thought of as decomposing a signal into its component
frequencies and their amplitudes. The Fourier transform is an integral transform that re-
expresses a function in terms of sinusoidal basis functions, i.e. as a sum or integral of
sinusoidal functions multiplied by some coefficients ("amplitudes"). The general idea is that
a multiplication of the input waveform of unknown amplitude and frequency
(interferogram signals) by a known reference frequency of unity amplitude (the analyzing
wave) can give us the unknown amplitude and original frequency. Thus, by using a
frequency-adjustable analyzing wave, each digitized point of the interferogram can be
transformed from the time (or optical retardation) domain to the frequency domain, which
results in the IR spectrum. When a single interferogram is thus Fourier-transformed, a so-
called single-beam spectrum is generated, which is the raw detector response versus the
wavelength. In order to produce the absorption spectrum of a sample, the sample single-
beam spectrum must be normalized against a background spectrum taken with no sample
in the beam path. The absorption spectrum can be presented equivalently as a transmittance
(T= I/I0) or absorbance (A= log10I0/I) spectrum, where I is the intensity measured with the
sample in the beam and I0 is the intensity measured from the background spectrum.
Direct transmittance is the oldest and most straightforward spectroscopic technique, which
is based on the absorption of the IR radiation as it passes through the sample. Clearly, this
technique is applicable only to samples that do not absorb all the incoming IR energy and
are sufficiently transparent in this spectral range. For highly absorbent samples, such as
soils, it is necessary to prepare a pellet that embeds the soil sample in a transparent matrix,
most usually KBr. The pellet preparation involves grounding 2-3 mg of soil with ~1 g of KBr
using a mortal and pestle and using a hydraulic press and die to create a thin, IR-
transparent disk. The main advantage of this technique is that it yields very clear and
information-rich signals. Its main disadvantages are the lengthy preparation required to
prepare the pellets and the fact that it is difficult to obtain quantitative results. Also, since
only a few mg of soils are used for the preparation of each pellet, the resulting spectrum
may not be representative the bulk soil. For these reasons, transmittance measurements are
rather rarely used in soil analysis and the reflectance and photoacoustic methods described
below are prefered.
2.1 Diffuse reflectance

Nguyen et al. (1991) pointed out to two main drawbacks of the transmittance technique in
addition to the time-consuming sample preparation: (1) possible reaction of the sample with
the halide matrix, and (2) scattering and/or total absorption for samples with high
concentrations and large particles relative to the infrared wavelengths. To circumvent these
limitations, Nguyen et al. (1991) introduced the use of diffuse reflectance (DRIFT) for soil
www.intechopen.com
Fig. 2. Schematic description of a Michelsen interferogram.

analysis. While the study of Nguyen et al. (1991) was purely qualitative and restricted to
band assignments, the subsequent articles of Janik et al. (1995) and Janik&Skjemstad (1995)
showed that mid-IR DRIFT could be used to quantify various soil components. Since then,
the use of mid-IR DRIFT has been investigated in numerous soil studies (see excellent
review paper of Reeves III 2010), and it is arguably the mid-IR spectroscopic method most
commonly used for soil analysis. DRIFT spectroscopy is mainly concerned with radiation
emerging from a non-mirror or matt surface after it has undergone absorption, refraction,
reflection and scattering in the bulk material. DRIFT spectra are subject to nonlinear scaling
of the intensity which reduces the intensity of strongly absorbing bands in comparison to
weaker bands. Thus, low intensity bands appear to be enhanced in DRIFT spectra. The
specular component becomes increasingly significant at high sample concentrations and for
large particles, and the occurrence of strong specular reflection explains for instance why the
presence of inorganic carbon interferes with the development of calibrations for soil organic
carbon (Reeves et al. 2005). In order to avoid such non-linearities and spectral distortions,
the use of samples diluted with KBr has long been advocated and such KBr dilution is still
used in many studies. However, using KBr-diluted samples increases the sample
preparation time and may cause interferences due to ion exchange between the sample and
the KBr matrix or adsorption of water onto the KBr (Bertrand et al. 2002). Janik&Skjemstad
(1995), Janik et al. (1998) and Reeves et al. (2001) have shown that quantitative analysis can
also be performed on mid-IR spectra of neat (undiluted) soil samples. These findings were
further supported by the comparative study of Bertrand et al. (2002) who compared results
obtained using neat and KBr-diluted samples and concluded that KBr dilution does not
improve the accuracy of the measurements.
2.2 Attenuated total reflectance

In the Attenuated Total Reflectance (ATR) method, the IR radiation propagates through a
crystal with a high refractive index that is in contact with the sample (Figure 3). Mirrors are
used to direct the IR beam toward the crystal at an angle that exceeds the critical angle for
This critical angle θC depends on the refractive indices of the sample and ATR crystal
internal reflection, so that the radiation undergoes multiple reflections within the crystal.
according to
θc = sin −1 ⎜⎜ 2 n ⎟⎟
⎛n ⎞
⎝ 1⎠
(1)
where n1 and n2 are the refractive indices of the crystal and the sample, respectively. Due to
the quantum mechanic properties of the light, the electromagnetic field extends beyond the
www.intechopen.com
Fig. 3. Attenuated total reflectance (ATR) configuration. I denotes the incoming light (from
the interferometer), D denotes the detector, L and M are lenses and mirrors. Reproduced
from (Etzion et al. 2004)
crystal surface for a short distance known as the evanescent field. If a sample is applied
directly onto the surface of the ATR crystal, some of the IR radiation (i.e., the evanescent
wave) is absorbed by this sample, so that the sample absorbance spectrum can be obtained.
The evanescent wave decays exponentially with the distance from the surface of the crystal.
The depth of penetration of the evanescent wave dp is defined as the distance from the
crystal-sample interface at which the intensity of the evanescent wave decays to 37% (1/e) of
its original value, and is given by
dp = λ
( )
⎡ 2⎤
2π n1 ⎢sin 2 θ
(2)
− n2 n1 ⎥
⎢ ⎥
⎣ ⎦
where is the radiation wavelength. The penetration depth and the total number of
reflections along the crystal can be controlled to some extend either by varying the angle of
incidence of the radiation (θ) or through the selection of the crystal material. The penetration
depth is typically less than 10 μm, so that very good contact between the sample and the
crystal is critical in order to obtain reliable and reproducible results.
The first use of ATR for soil analysis was reported by Ehsani et al. (2001), who attempted to
determine nitrate concentration using ATR spectra of dry soil samples. This study
demonstrated the difficulty of obtaining adequate contact between the soil and the ATR
crystal, but also showed how the results could be greatly improved by using a soil paste or
slurry. Such an technique was further developed by Shaviv et al. (2003), Linker et al. (2004),
Linker et al. (2005) and Borenstein et al. (2006) who used samples consisting of soil pastes
close to water saturation. Since nitrate is highly soluble in water and is not fixed in the soil
matrix, all the nitrate is present in the liquid phase of the soil paste, which has several
advantages. First, since the paste moisture content is less than 1 g [H2O]/g [soil], the nitrate
concentration in the liquid phase is higher than in the dry soil. Second, much better contact
is obtained between the ATR crystal and the liquid phase than could be obtained between
the ATR crystal and the dry soil. Finally, the original moisture of the soil sample has no
influence on the measurement. However, water exhibits very strong absorbance bands in
the mid-IR range, which may distort or hide bands of interest. This is illustrated in Figure 4
which shows the spectra of de-ionized water, de-ionized water with 1000 mg[N]/kg[water]
nitrate, and a paste of sandy soil with 870 mg[N]/kg[water] nitrate. Distortion and shift of
www.intechopen.com
the spectra can be observed, especially in the regions indicated by circles. The magnitudes of
these changes are comparable with the size of the nitrate signal and would cause significant
inaccuracies if straightforward estimation of nitrate was performed. Therefore an accurate
water subtraction procedure such as the one developed by Linker et al. (2004, 2005) should
be applied to the spectra prior to the quantitative analysis.
Fig. 4. Spectra of de-ionized water, de-ionized water with 1000 mg[N]/kg[water] nitrate,
and a paste of sandy soil with 870 mg[N]/kg[water] nitrate. The circles indicates the regions
in which shifts and distorsions of the spectra are most clearly visible. Reproduced from
Linker et al. (2005).
2.3 Photoacoustic spectroscopy

Fourier transform infrared photoacoustic spectroscopy (FTIR-PAS) differs from the previous
methods in the sense that it relies on complete absorption of the incoming radiation by the
sample. The sample is placed in a sealed enclosure that is purged with He to avoid
atmospheric interferences and to which is connected a highly sensitive microphone which
records the pressure waves that result from the local heating induced by the absorbed
radiation (McClelland et al. 2002) (Figure 5). The recorded spectrum depends on the
absorption properties of the sample, its thermal diffusivity, and the thermal penetration
depth. Therefore, photoacoustic spectra are more difficult to interpret than reflectance ones
and there is no one-to-one correspondence between the spectra obtained with both
techniques. Also, photoacoustic spectra are much more prone to measurement noise.
Nonetheless, since this method can be used with highly absorbing samples without any
pretreatment, it is well suited for soil samples analyses and its use has been reported in
several recent studies (Du et al. 2008a, 2008b, 2009b).
3. Data processing
Due to the large amount of data generated by FTIR spectrometers and the complexity of the
spectra, it is imperative to use chemometrics procedures to analyze the data. A detailed
www.intechopen.com
Modulated IR radiation IR Spectrum
KBr window
Helium Computer
Microphone
Sample
Fig. 5. Schematic description of FTIR photoacoustic spectroscopy. Reproduced from Du et

al. (2008a).
description of such procedures can be found in various textbooks such as Brereton (2003) or
Chau et al. (2004). Partial Least Squares (Wold et al. 2001) is the procedure most commonly
used for quantitative determination of one or several soil components (see for instance the
excellent compilation table in Viscarra Rossel et al. (2006)). PLS is an easy-to-use
straightforward procedure that is available in standard statistical and chemometrics
softwares, which probably explains its popularity. The main limitation of PLS is that it
assumes linearity of the data, but the good results reported in most soil studies show that
this does not appear to be too restrictive an assumption when analyzing soil spectra.
Exceptions to this generally linear behavior were reported for instance by Janik&Skjemstad
(1995) and Janik et al. (2009), especially at very high concentrations. However, Janik et al.
(2009) reported that in those cases where PLS was not capable of providing a good
prediction model, adding a non-linear element in the form of a neural network did not
improve the results significantly.
The a priori assumption of linearity can be avoided altogether by using procedures such as
wavelet decomposition and neural networks. An example of such a procedure is depicted in
Figure 6. In this procedure, wavelet decomposition is used to compress the data and obtain
a low-dimensional representation of the spectra. Although wavelet decomposition by itself
does not produce a compressed representation of the original data, data reduction can be
achieved by eliminating the wavelet coefficients that do not contain valuable information.
Various approaches have been reported in the literature for selecting the most relevant
coefficients, such as eliminating all “small” coefficients using for instance either
thresholding (Kai-man Leung et al. 1998; Ehrentreich 2002), entropy (Kai-man Leung et al.
1998), mutual information (Alsberg et al. 1998), maximum likelihood (Leger&Wentzell
2004), or genetic algorithms (Depczynski et al. 1999), or retaining only the coefficients with
the highest variance (Trygg&Wold 1998) as depicted in Figure 6. Once data compression has
been achieved, the remaining coefficients can be used as input variables for a neural
network that creates a non-linear mapping between these inputs and the property (or
properties) of interest. The neural network can also be replaced by a simpler linear
regression as in Viscarra Rossel&Lark (2009), in which case a hybrid model that contains
both linear and non-linear stages is obtained.
Since the wavelet transform decomposes the signal into components at different scales (or
loosely speaking, components of different widths), it is a very powerful tool for resolving
overlapping bands and separate the bands of interest from the background and
interferences. For instance, Jahn et al. (2006) used such an approach to distinguish between
the strongly-overlapping absorbance bands of nitrate and calcium carbonate in the ATR
spectra of calcareous soils (Figure 7).
www.intechopen.com
Fig. 6. Schematic representation of a prediction model combining wavelet decomposition of

the spectra and a neural network
Fig. 7. Overlapping bands of nitrate and carbonate in ATR spectra of calcareous soils (Top
frame) and the wavelet decomposed values at scale 2 (dashed, attributed to carbonate) and
at scale 3 (attributed to nitrate,14ppm and 120ppm N-NO3) (Bottom frame). Reproduced
from Jahn et al. (2006).
www.intechopen.com
4. Most common applications

4.1 Estimation of soil nutrients and organic matter
Soil nutrients, such as C, N, P, K, S, Ca and microelements, play a primordial role in the
development of agricultural crops and hence determination of their concentrations is
crucial to the application of the precision agriculture concept. Viscarra Rossel et al. (2006)
compiled an excellent review of the studies dedicated to the estimation of these
parameters using diffuse reflectance. Table 1 present a non-exhaustive list of studies that
focused on the determination of carbon, total nitrogen, nitrate, potassium, phosphorus
and organic matter using diffuse reflectance, ATR and photoacoustic spectroscopy. Table
1 clearly shows that carbon, and in particular organic C, is the property most commonly
determined, using DRIFT. Most studies reported very good results with correlation
coefficients (R2) between the actual and estimated values higher than 0.90. A noticeably
much poorer result was reported by Reeves&Smith (2009) (R2=0.60). This latter study was
not restricted to agricultural soils as the previous studies (some of which included a
number of samples similar to that of Reeves&Smith (2009)), but included soil samples
from national forests, rangelands, woodlots, etc. and Reeves&Smith (2009) hypothesized
that this wide range of soil uses was responsible for the poor performance of the
regression models. Acceptable results were also reported for estimation of total nitrogen
(R2>0.80), using either DRIFT or photoacoustic spectroscopy. For nitrate, Viscarra Rossel
et al. (2006) did not find any correlation between DRIFT spectra and nitrate concentration.
This finding contrasts sharply with the good results that can obtained using the ATR
technique, for which the typical determination errors range from 5 to 25 mg[N]/kg[dry
soil] for realistic agricultural field concentrations. The largest determination errors
correspond to calcareous soils for which nitrate determination is hindered by the
absorbance band of carbonate that overlaps the nitrate band (Linker et al. 2004, 2005, 2006;
Borenstein et al. 2006; Jahn et al. 2006). For potassium, phosphorous and organic matter
conflicting findings have been reported. Bertrand et al. (2002) reported that DRIFT could
be used to estimate potassium concentration (R2=0.85). Acceptable results were also
reported by Du et al. (2009b) who used photoacoustic spectroscopy (R2=0.79).
McCarty&Reeves (2006) and Reeves&Smith (2009) obtained relatively poor but still
potentially useful results using DRIFT (R2 of 0.60 and 0.76, respectively), but Viscarra
Rossel et al. (2006) concluded that DRIFT spectroscopy was not suitable for estimating
potassium concentration (R2=0.40). For phosphorous, Janik et al. (2009) and Du et al.
(2009b) reported R2 values of 0.87 and 0.81 using DRIFT and photoacostic spectroscopy,
respectively. However, Bertrand et al. (2002), McCarty&Reeves (2006) and Viscarra Rossel
et al. (2006) did not find a relationship between DRIFT spectra and phosphorous
(R2<0.40). For organic matter, very good results were obtained using photoacoustic
spectroscopy (Du et al. 2007, 2009b) and DRIFT (Masserschmidt et al. 1999), with R2 values
above 0.90 in all three studies. However, very poor correlation between DRIFT spectra
and organic matter content was reported by Canasveras et al. (2010).
The somewhat conflicting results reported in some of the DRIFT studies emphasize the need
for standardization of the analyses (both analytical reference analyses and DRIFT
spectroscopy) and underscore the problem of transferability of the results (See Section 5.1).
Photoacoustic spectroscopy appears to be a very promising technique which should
certainly be investigated in further studies.
www.intechopen.com
Attenuated total
Diffuse reflectance Photoacoustic
reflectance
Janik&Skjemstad (1995); Janik
et al. (1998); McCarty et al.
(2002); Reeves et al. (2002);
Reeves (2009, 2010);
Reeves&Smith (2009)
Organic C ;McCarty&Reeves (2006)
;Viscarra Rossel et al. (2006,
2008); Viscarra Rossel&Lark
(2009) ;Janik et al. (2009)
;Zimmermann et al. (2007)
;McBratney et al. (2006)
McCarty et al. (2002) ;Reeves
et al. (2001, 2002); Reeves
Total C (2009, 2010); Viscarra Rossel et
al. (2008); Minasny et al.
(2009); Reeves&Smith (2009)
Janik&Skjemstad (1995);
Reeves et al. (2001);
Total N Du et al. (2009b)
McCarty&Reeves ( 2006);
Viscarra Rossel et al. (2008)
Ehsani et al. (2001);
Shaviv et al. (2003);
Linker (2004);
Viscarra Rossel et al. (2006) Linker et al. (2004,
Nitrate
;Verma&Deb (2007) 2005, 2006);
Borenstein et al.
(2006); Jahn et al.
(2006);
Bertrand et al. (2002);
McCarty&Reeves (2006);
Potassium Viscarra Rossel et al. (2006); Du et al. (2009b)
Minasny et al. (2009);
Reeves&Smith (2009)
McCarty&Reeves (2006);
Viscarra Rossel et al. (2006) ;
Phosphorus Du et al. (2009b)
Janik et al. (2009); Minasny et
al. (2009)
Organic Masserschmidt (1999); Du et al. (2007,
matter Canasveras et al. (2010) 2009b)
Table 1. List of studies that used reflectance and photoacoustic spectroscopy for the
estimation of soil carbon, total nitrogen, nitrate, potassium phosphorus and organic matter
www.intechopen.com
4.2 Determination of soil moisture

Water in itself is essential for plant growth and its presence in the soil influences numerous
properties such as the availability of oxygen and nutrients, microbial activity, swelling and
shrinking of clays or soil stability. Since the methods traditionally used to estimate water-
related soil properties are time consuming, much research has been devoted to the
development of alternative methods. In particular, very good results have been reported for
some of these properties using near-infrared spectroscopy under laboratory conditions
(Mouazen et al. 2005, 2006; Viscarra Rossel&McBratney 1998a) and encouraging results have
been reported for in-situ field measurements (Mouazen et al. 2005). A summary of the main
results obtained with mid-IR DRIFT spectroscopy is presented in Table 2. With the exception
of Dataset #1 in Tranter et al. (2008), mid-IR spectroscopy was capable of estimating to some
extent water retention at -10kPa and -1500kPa, with R2 values ranging from 0.64 to 0.81 and
from 0.66 to 0.89, respectively. However, the study of Bertrand et al. (2002) clearly shows that
KBr dilution of the samples or high CaCO3 concentration deteriorate the results. Although this
has not been specifically verified, the very poor results obtained by Tranter et al. (2008) on
Dataset #1 were most probably due to the fact that the spectra were recorded from intact
soil cores rather than from grounded samples as in all the other studies. Regarding moisture
of air-dried and oven-dried soil, the few results available are very good, unless the soil has a
high CaCO3 concentration.
4.3 Soil characterization

The soil physical structure affects various physical, chemical and biological processes such
as for instance water infiltration and retention or root penetration and proliferation. With
this respect, clay is the component most often studied by mid-IR spectroscopy, although
some studies also attempted to estimate sand or silt content (McBratney et al. 2006; Viscarra
Rossel et al. 2008). For clay, R2 values of 0.85 or more were reported by Janik&Skjemstad
(1995), Viscarra Rossel et al. (2008) and Viscarra Rossel&Lark (2009). Reasonably good
results were reported by Janik et al. (2009), Canasveras et al. (2010) and Janik et al. (2009),
with R2 value of 0.84 and 0.82, respectively, but lower R2 values (0.74-0.79) were reported by
Bertrand et al. (2002), McCarty&Reeves (2006), Minasny et al. (2009) and McBratney et al.
(2006). No explanation has been offered to explain such a variability of the prediction
performances. In addition to the previous DRIFT studies, Du et al. (2007) reported very
good results (R2>0.90) using photoacoustic spectroscopy and Du et al. (2008b) further
showed that this method can be used to differentiate between different types of clays.
Another important soil component that can be estimated by mid-infrared spectroscopy is
carbonate. In this case the results are much more consistent, with R2 values of 0.95 or more
reported in the studies of Janik&Skjemstad (1995), Bertrand et al (2002), Canasveras et al.
(2010) and Du et al. (2008b). Linker et al. (2006) and Du et al. (2008a) showed that the
spectral information related to carbonate and other soil constituents can be used to
discriminate between soils, and Linker et al. (2006) further showed that such a automatic
discrimination can improve the estimation of nutrients such as nitrate.
5. Current challenges and emerging applications

5.1 Spectral libraries and model transferability
The results presented in the previous Section clearly demonstrated that mid-IR spectroscopy
can be used to determine accurately a series of soil characteristics and properties under
www.intechopen.com
Water retention Moisture of

Water retention Moisture of air-
Study at -10kPa or oven-dried
at -1500kPa dried soil
120cm soil
Bertrand et al.
0.92‡ 0.92
(2002)
0.86 (KBr
0.84‡ (KBr
diluted
diluted samples)
samples)
0.74‡ 0.68
(CaCO3>2%) (CaCO3>2%)
McBratney et
0.64† 0.66†
al. (2006)
Janik et al.
0.79 0.84
(2007)
Tranter et al.
(2008) (Dataset
0.02 0.50
#1 – intact
cores)
0.48† 0.71†
Tranter et al.
(2008) (Dataset
0.85 0.89
#2 – crushed
samples)
Janik et al.
0.81 0.89 0.89
(2009)
Table 2. R2 values between the actual soil moisture or water retention and the one estimated
from mid-IR spectra. The results were obtained using neat ground samples unless otherwise
indicated. † indicates that results obtained indirectly using properties estimated by mid-IR
spectroscopy and pedotransfer functions. ‡ indicates that the result correspond to water
retention at 120cm water tension
laboratory conditions. However, two questions still remain mostly unanswered: (1) could a
regression model developed by one group of researchers be readily applied by other
researchers and (2) would it be possible to establish spectral libraries that combine spectra
from different sources? Such issues are of course not restricted to mid-IR spectroscopy but
are also relevant to NIR spectroscopy (Cécillon et al. 2009). These two issues are related
since in the absence of such libraries each research team tends to work with spectra recorded
“in-house” using experimental procedures that are, at least to some extend, specific to that
specific laboratory. Furthermore, due to the extensive work required to collect soil samples
and to perform detailed chemical analyzes of these samples, most studies are still based on
soil samples belonging to a specific geographic area. The issue of mid-IR model
transferability was recently addressed by Minasny et al. (2009) who concluded that “Due to
the differences in the origins of the soil and in the laboratory measurement procedures,
calibration functions from one region do not perform well on another one”. They further
emphasized that in their opinion the cause for this lack of transferability lies mainly in the
difference in laboratory techniques, different analytical procedures, and inter-laboratory
www.intechopen.com
bias. The creation of a library of mid-IR DRIFT spectral of soils was considered by Viscarra
Rossel et al. (2008) who emphasized three main prerequisite for the development of such a
library: (1) it should describe adequately the soil variability in the region in which it is to be
used, (2) experimental procedures, and in particular sample handling, preparation, storing
and scanning must be standardized and (3) the analytical data used as reference should be
acquired using reliable and accredited analytical procedures. This issue was also addressed
by Reeves III in his recent review on the use of NIR and mid-IR DRIFT for soil analysis
(Reeves III 2010) in which he emphasized that “Spectral libraries for quantitative analysis
will be useless, or worse deceptive, if users base results on combining spectra from differing
sets of samples that have not been standardized.” However, as the study of Reeves&Smith
(2009) clearly demonstrated, standardization in itself does not guarantee accurate results,
even for properties that seem trivial to determine such as organic carbon. In this case,
Reeves&Smith (2009) explained their poor results by the fact that their database included
soils of various usages rather than to be restricted to agricultural soils. Finally, it must be
noted that since specular reflection occurs for both organic and non-organic fractions of soil,
libraries of spectra collected by other means than diffuse reflectance would be largely
useless for comparing mineral spectra to soil spectra (Reeves et al. 2005). All these factors
underline the complexity of the task at hand if one was to contemplate the creation of a truly
universal soil spectral library.
5.2 Indirect estimation of additional soil properties

In addition to those properties that can be measured directly by mid-infrared spectroscopy,
there is a large number of important soils properties that are not associated with distinct
absorbance bands. In such cases McBratney et al. (2006) propose to use the estimates
obtained by mid-IR spectroscopy in an inference system that uses pedotransfer functions.
This approach is illustrated in Figure 8: Once regression models based on mid-IR spectra
have been calibrated for properties such as clay, silt, sand and organic carbon, these models
are used to estimate the same properties of a new sample, together with their respective
uncertainties. These values are then used to predict the bulk density, the water content at
field capacity and wilting point, and finally the available water capacity.
An important feature of this approach is that the determination errors are propagated
through the various estimation stages so that a confidence interval can be associated with
each prediction (Figure 9). Tranter et al. (2008) showed how such an approach improved
considerably the estimation of soil water retention, which was very poorly estimated
directly from the mid-IR spectra (See Table 2). Minasny&McBratney (2008) also reported
more parsimonious models and greater accuracy when using an inference system then when
using straightforward PLS regression on the mid-IR spectra. Such an approach may help
solving the “model transferability” problem discussed in the previous section and further
work should be devoted to validating and improving this interesting approach.
5.3 On-the-go sensing and combination with GIS

In order to achieve “precision farming”, the information obtained from soil analyzes must
be combined with a Geographic Information System (GIS) platform in order to create
overlaying maps of soil characteristics, water content, nutrient availability, yield, etc. In
addition to measurement accuracy, a key issue is the number of sampling points and the
spatial resolution of the measurements. The development of “on-the-go” sensors is of
www.intechopen.com
Fig. 8. Top frame: Schematic description of the combination of mid-IR spectroscopy and
inference models as suggested by McBratney et al. (2006). Bottom frame: Quantiles of the
available water capacity estimated according to the inference scheme. Reproduced from
McBratney et al. (2006).
primordial importance (Viscarra Rossel&McBratney 1998b) and considerable research
efforts have been devoted to the development of fast, fully automated and geo-referenced
measurement systems (Sibley et al. 2009; Mouazen et al. 2007; Christy 2008; Adamchuk et al.
2004, 2005; Sinfield et al. 2010). Among the spectroscopy-based methods, the use of NIR has
been investigated more actively, mostly due to the lower price and higher robustness of the
equipment. Indeed, in their review paper Sinfield et al. (2010) concluded that the “the FT-IR
ATR technique, while very accurate and fast, makes use of expensive and delicate
equipment (e.g., moving mirror in interferometer) which is not readily amenable to an on-
the-go setting” but that development of sensors based on a limited set of wavebands, as
suggested by Linker (2004) and Jahn et al. (2006), could “enhance the potential to apply
reflectance based approaches in the field by limiting the sophistication of requisite
equipment“. At any rate, considerable improvements are still required before large-scale
implementation could be considered.
5.4 Monitoring of N isotopes

Nitrogen, either naturally present in the soil or added as fertilizer, may undergo a series of
complex transformations. These transformations are interdependent and depend on a large
number of variables, so that isolating specific processes or pathways is very challenging.
One of the methods used to study these processes consists of enriching the soil with stable
www.intechopen.com
15N isotopes and monitoring the concentration of 15N in the by-products of the various
reactions (Stevens et al. 1993; Stevens&Laughlin 1994; Baggs 2008). Such studies require the
quantification of 14N and 15N, which is done by isotope ratio mass spectrometry (IRMS).
However, expensive and laborious preparation of the samples is required to convert the N
species into forms suitable for the IRMS measurements and the method can not realistically
be used to analyze a large number of samples during an experiment. Du et al. (2009a)
showed how mid-infrared spectroscopy could provide a much faster and cheaper
alternative for N-isotope tracing, albeit with much lower accuracy than IRMS. The method is
based on the observation that since the mid-infrared range corresponds to the fundamental
vibrations of a molecule, which depend on the atoms’ mass, the absorbance bands of 14N-
based and 15N-based compounds are slightly shifted relative to each other, so that the
concentration of each species can be estimated. This spectral shift between 14N-NO3 and 15N-
NO3 is shown in Figure 9.
Fig. 9. Top frame: FTIR-ATR spectra of 14N-NO3 and 15N-NO3 in water. Bottom frame:
Nitrate formation during an incubation experiment as estimated from FTIR-ATR
measurements. Nine sub-samples were analyzed at each sampling point. ‘‘c’’ is the total
amount of nitrate formed. ‘‘d’’ and ‘‘e’’ are the formed 15N-NO3 and 14N-NO3, respectively.
Reproduced from Du et al. (2009a).
www.intechopen.com
Using PLS regression models calibrated using saturated soil pastes spiked with 14N-NO3
and 15N-NO3, Du et al. (2009a) were able to estimate the 14N-NO3 and 15N-NO3
concentrations in a non-calcareous soil with determination errors of less than 6 mg[N]/
kg[dry soil], which enabled them to monitor separately the formation of 14N-NO3 and 15N-
NO3 during an incubation experiment (Figure 9). Although the measurement uncertainties
reported in Figure 9 may seem rather large, the low cost and short time required for the
FTIR-ATR measurements would make it possible to perform and average more
measurements than as done in the study of Du et al. (2009a), and thus reduce the
measurement uncertainties. After further validation of the approach with other soils,
including with calcareous soils in which the carbonate absorbance band overlaps the nitrate
absorbance band and interferes with nitrate determination (Linker et al. 2004, 2005, 2006),
and extension of the method to monitoring soil NH4 and/or NO2 species, this method
would provide a very powerful and cheap tool for studying soil nitrogen transformations.
6. Conclusion
The abundant studies conducted in the mid-IR range have demonstrated the potential that
this technique holds for rapid and inexpensive soil analysis. A major advantage of
spectroscopy techniques in general and mid-IR in particular is that several properties can be
determined from a single spectrum, which greatly reduces the costs of analysis compared to
conventional laboratory techniques. In addition, the measurement is very rapid so that a
large number of samples can be easily screened or measurements could be conducted “on-
the-go”, at least in principle. However, several problems still need to be solved before this
technology could be upgraded from the research laboratory into routine analyses. The most
pressing issue appears to be the standardization of the FTIR and conventional procedures
used to analyze the samples, which would make it possible to consider the establishment of
spectral libraries similar to those existing in other FTIR fields. Such libraries would provide
the basis necessary for developing robust chemometrics models, based either solely on mid-
IR spectra or combining mid-IR spectra and pedotransfer functions, valid not only at the
local but also at the regional scale.
7. References
Adamchuk, V. I.; Hummel, J. W.; Morgan, M. T. & Upadhyaya, S. K. (2004). On-the-go soil
sensors for precision agriculture. Computers and Electronics in Agriculture 441: 71-91.
Adamchuk, V. I.; Lund, E. D.; Sethuramasamyraja, B.; Morgan, M. T.; Dobermann, A. &
Marx, D. B. (2005). Direct measurement of soil chemical properties on-the-go using
ion-selective electrodes. Computers and Electronics in Agriculture 483: 272-294.
Alsberg, B. K.; Woodward, A. M.; Winson, M. K.; Rowland, J. J. & Kell, D. B. (1998). Variable
selection in wavelet regression models. Analytica Chimica Acta 368: 29-44.
Baggs, E. M. (2008). A review of stable isotope techniques for N2O source partitioning in
soils: recent progress, remaining challenges and future considerations. Rapid
Communications in Mass Spectrometry 2211: 1664-1672.
Bertrand, I.; Janik, L. J.; Holloway, R. E.; Armstrong, R. D. & McLaughlin, M. J. (2002). The
rapid assessment of concentrations and solid phase associations of macro- and
micronutrients in alkaline soils by mid-infrared diffuse reflectance spectroscopy. Australian
Journal of Soil Research 408: 1339-1356.
www.intechopen.com
Borenstein, A.; Linker, R.; Shmulevich, I. & Shaviv, A. (2006). Determination of soil nitrate
and water content using attenuated total reflectance spectroscopy. Applied
Spectroscopy 60: 1267-1272.
Brereton, R. G. (2003). Chemometrics. Data analysis for the laboratory and chemical plant, Wiley &
Sons. Chichester.
Canasveras, J. C.; Barron, V.; del Campillo, M. C.; Torrent, J. & Gomez, J. A. (2010).
Estimation of aggregate stability indices in Mediterranean soils by diffuse
reflectance spectroscopy. Geoderma 158: 78-84.
Cécillon, L.; Barthès, B. G.; Gomez, C.; Ertlen, D.; Genot, V.; Hedde, M.; Stevens, A. & Brun,
J. J. (2009). Assessment and monitoring of soil quality using near-infrared
reflectance spectroscopy (NIRS). European Journal of Soil Science 605: 770-784.
Chau, F. T.; Liang, Y.; Gao, J. B. & Shao, X. (2004). Chemometrics. From basics to wavelet
transform, In: Chemical analysis: A series of monographs on analytical chemistry and its
applications, J. D. Winefordner (Ed.).
Christy, C. D. (2008). Real-time measurement of soil attributes using on-the-go near infrared
reflectance spectroscopy. Computers and Electronics in Agriculture 611: 10-19.
Depczynski, U.; Jetter, K.; Molt, K. & Niemoller, A. (1999). Quantitative analysis of near
infrared spectra by wavelet coefficient regression using a genetic algorithm.
Chemometrics and Intelligent Laboratory Systems 472: 179-187.
Desbiez, A.; Matthews, R.; Tripathi, B. & Ellis-Jones, J. (2004). Perceptions and assessment of
soil fertility by farmers in the mid-hills of Nepal. Agriculture Ecosystems &
Environment 1031: 191-206.
Du, C.; Linker, R. & Shaviv, A. (2008a). Identification of agricultural Mediterranean soils
using mid-infrared photoacoustic spectroscopy. Geoderma 143: 85-90.
Du, C. W.; Linker, R. & Shaviv, A. (2007). Characterization of soils using photoacoustic mid-
infrared spectroscopy. Applied Spectroscopy 61: 1063-1067.
Du, C. W.; Linker, R.; Shaviv, A. & Zhou, J. M. (2009a). In Situ Evaluation of Net
Nitrification Rate in Terra Rossa Soil Using a Fourier Transform Infrared
Attenuated Total Reflection N-15 Tracing Technique. Applied Spectroscopy 63: 1168-
1173.
Du, C. W. & Zhou, J. M. (2009). Evaluation of soil fertility using infrared spectroscopy: a
review. Environmental Chemistry Letters 72: 97-113.
Du, C. W.; Zhou, J. M.; Wang, H. Y.; Chen, X. Q.; Zhu, A. N. & Zhang, J. B. (2009b).
Determination of soil properties using Fourier transform mid-infrared
photoacoustic spectroscopy. Vibrational Spectroscopy 491: 32-37.
Du, C. W.; Zhou, J. M.; Wang, H. Y.; Mang, J. B. & Zhu, A. N. (2008b). Study on the soil mid-
infrared photoacoustic spectroscopy. Spectroscopy and Spectral Analysis 286: 1242-
1245.
Ehrentreich, F. (2002). Wavelet transform applications in analytical chemistry. Analytical and
Bioanalytical Chemistry 3721: 115-121.
Ehsani, M. R.; Upadhyaya, S. K.; Fawcett, W. R.; Protsailo, L. V. & Slaughter, D. (2001).
Feasibility of detecting soil nitrate content using a mid-infrared technique.
Transactions of the ASAE 446: 1931-1940.
Etzion, Y.; Linker, R.; Cogan, U. & Shmulevich, I. (2004). Determination of protein
concentration in raw milk by mid-infrared Fourier transform infrared/attenuated
total reflectance spectroscopy. Journal of Dairy Science 879: 2779-2788.
Gobeille, A.; Yavitt, J.; Stalcup, P. & Valenzuela, A. (2006). Effects of soil management
practices on soil fertility measurements on Agave tequilana plantations in Western
Central Mexico. Soil & Tillage Research 871: 80-88.
www.intechopen.com
Griffiths, P. R. & de Haseth, J. A. (1986). Fourier transform infrared spectrometry, Wiley-

Interscience. New York.
Idowu, O. J.; van Es, H. M.; Abawi, G. S.; Wolfe, D. W.; Ball, J. I.; Gugino, B. K.; Moebius, B.
N.; Schindelbeck, R. R. & Bilgili, A. V. (2008). Farmer-oriented assessment of soil
quality using field, laboratory, and VNIR spectroscopy methods. Plant and Soil
3071-2: 243-253.
Jahn, B. R.; Linker, R.; Upadhyaya, S. K.; Shaviv, A.; Slaughter, D. C. & Shmulevich, I. (2006).
Mid-infrared spectroscopic determination of soil nitrate content. Biosystems
Engineering 944: 505-515.
Janik, L. J.; Forrester, S. T. & Rawson, A. (2009). The prediction of soil chemical and physical
properties from mid-infrared spectroscopy and combined partial least-squares
regression and neural networks (PLS-NN) analysis. Chemometrics and Intelligent
Laboratory Systems 972: 179-188.
Janik, L. J.; Merry, R. H.; Forrester, S. T.; Lanyon, D. M. & Rawson, A. (2007). Rapid
prediction of soil water retention using mid infrared spectroscopy. Soil Science
Society of America Journal 712: 507-514.
Janik, L. J.; Merry, R. H. & Skjemstad, J. O. (1998). Can mid infrared diffuse reflectance
analysis replace soil extractions? Australian Journal of Experimental Agriculture 387:
681-696.
Janik, L. J. & Skjemstad, J. O. (1995). Characterization and analysis of soils using mid
infrared partial least-squares. 2. Correlations with some laboratory data. Australian
Journal of Soil Research 334: 637-650.
Janik, L. J.; Skjemstand, J. O. & Raven, M. D. (1995). Characterization and analysis of oils
using mid infrared partial least-squares. 1. Correlations with XRF-determined
major-element composition. Australian Journal of Soil Research 334: 621-636.
Kai-man Leung, A.; Chau, F. T.; Gao, J. B. & Shih, T. M. (1998). Application of wavelet
transform in infrared spectrometry: spectral compression and library search.
Chemometrics and Intelligent Laboratory Systems 431-2: 69-88.
Leger, M. N. & Wentzell, P. D. (2004). Maximum likelihood principal components regression
on wavelet-compressed data. Applied Spectroscopy 587: 855-862.
Linker, R. (2004). Waveband selection for determination of nitrate in soil using mid-infrared
attenuated total reflectance spectroscopy. Applied Spectroscopy 58: 1277-1281.
Linker, R.; Kenny, A.; Shaviv, A.; Singher, L. & Shmulevich, I. (2004). Fourier Transform
Infrared-attenuated total reflection nitrate determination of soil pastes using
principal component regression, partial least squares, and cross-correlation. Applied
Spectroscopy 58: 516-520.
Linker, R.; Shmulevich, I.; Kenny, A. & Shaviv, A. (2005). Soil identification and
chemometrics for direct determination of nitrate in soils using FTIR-ATR mid-
infrared spectroscopy. Chemosphere 61: 652-658.
Linker, R.; Weiner, M.; Shmulevich, I. & Shaviv, A. (2006). Nitrate determination in soil
pastes using attenuated total reflectance mid-infrared spectroscopy: Improved
accuracy via soil identification. Biosystems Engineering 94: 111-118.
Masserschmidt, I.; Cuelbas, C. J.; Poppi, R. J.; De Andrade, J. C.; De Abreu, C. A. & Davanzo,
C. U. (1999). Determination of organic matter in soils by FTIR/diffuse reflectance
and multivariate calibration. Journal of Chemometrics 13: 265-273.
McBratney, A. B.; Minasny, B. & Rossel, R. V. (2006). Spectral soil analysis and inference
systems: A powerful combination for solving the soil data crisis. Geoderma 136: 272-
278.
www.intechopen.com
McCarty, G. W. & Reeves, J. B. (2006). Comparison of NFAR infrared and mid infrared
diffuse reflectance spectroscopy for field-scale measurement of soil fertility
parameters. Soil Science 1712: 94-102.
McCarty, G. W.; Reeves, J. B.; Reeves, V. B.; Follett, R. F. & Kimble, J. M. (2002). Mid-infrared
and near-infrared diffuse reflectance spectroscopy for soil carbon measurement.
Soil Science Society of America Journal 662: 640-646.
McClelland, J. F.; Jones, R. W. & Bajic, S. J. (2002). Photoacoustic spectroscopy, In: Handbook of
Vibrational Spectroscopy, J. M. Chalmers and P. R. Griffiths (Ed.), Wiley & SonsNew
York
Minasny, B. & McBratney, A. B. (2008). Regression rules as a tool for predicting soil
properties from infrared reflectance spectroscopy. Chemometrics and Intelligent
Laboratory Systems 94: 72-79.
Minasny, B.; Tranter, G.; McBratney, A. B.; Brough, D. M. & Murphy, B. W. (2009). Regional
transferability of mid-infrared diffuse reflectance spectroscopic prediction for soil
chemical properties. Geoderma 153: 155-162.
Mouazen, A. M.; De Baerdemaeker, J. & Ramon, H. (2005). Towards development of on-line
soil moisture content sensor using a fibre-type NIR spectrophotometer. Soil &
Tillage Research 80: 171-183.
Mouazen, A. M.; Karoui, R.; De Baerdemaeker, J. & Ramon, H. (2006). Characterization of
soil water content using measured visible and near infrared spectra. Soil Science
Society of America Journal 70: 1295-1302.
Mouazen, A. M.; Maleki, M. R.; De Baerdemaeker, J. & Ramon, H. (2007). On-line
measurement of some selected soil properties using a VIS-NIR sensor. Soil & Tillage
Research 93: 13-27.
Nguyen, T. T.; Janik, L. J. & Raupach, M. (1991). Diffuse reflectance infrared Fourier-
transform (DRIFT) spectroscopy in soil studies. Australian Journal of Soil Research 29:
49-67.
Reeves III, J. B. (2010). Near- versus mid-infrared diffuse reflectance spectroscopy for soil
analysis emphasizing carbon and laboratory versus on-site analysis: Where are we
and what needs to be done? Geoderma 158: 3-14.
Reeves, J.; McCarty, G. & Mimmo, T. (2002). The potential of diffuse reflectance
spectroscopy for the determination of carbon inventories in soils. Environmental
Pollution 116: S277-S284.
Reeves, J. B. (2009). Does the Spectral Format Matter in Diffuse Reflection Spectroscopy?
Applied Spectroscopy 63: 669-677.
Reeves, J. B.; Francis, B. A. & Hamilton, S. K. (2005). Specular reflection and diffuse
reflectance spectroscopy of soils. Applied Spectroscopy 59: 39-46.
Reeves, J. B.; McCarty, G. W. & Reeves, V. B. (2001). Mid-infrared diffuse reflectance
spectroscopy for the quantitative analysis of agricultural soils. Journal of Agricultural
and Food Chemistry 49: 766-772.
Reeves, J. B. & Smith, D. B. (2009). The potential of mid- and near-infrared diffuse
reflectance spectroscopy for determining major- and trace-element concentrations
in soils from a geochemical survey of North America. Applied Geochemistry 24: 1472-
1481.
Sá, J. C. d. M.; Cerri, C. C.; Lal, R.; Dick, W. A.; de Cassia Piccolo, M. & Feigl, B. E. (2009).
Soil organic carbon and fertility interactions affected by a tillage chronosequence in
a Brazilian Oxisol. Soil and Tillage Research 1041: 56-64.
www.intechopen.com
Shaviv, A.; Kenny, A.; Shmulevitch, I.; Singher, L.; Raichlin, Y. & Katzir, A. (2003). Direct
monitoring of soil and water nitrate by FTIR based FEWS or membrane systems.
Environmental Science & Technology 3712: 2807-2812.
Sibley, K. J.; Astatkie, T.; Brewster, G.; Struik, P. C.; Adsett, J. F. & Pruski, K. (2009). Field-
scale validation of an automated soil nitrate extraction and measurement system.
Precision Agriculture 102: 162-174.
Sinfield, J. V.; Fagerman, D. & Colic, O. (2010). Evaluation of sensing technologies for on-
the-go detection of macro-nutrients in cultivated soils. Computers and Electronics in
Agriculture 701: 1-18.
Stevens, R. J. & Laughlin, R. J. (1994). Determining N-15 in nitrite or nitrate by producing
nitrous-oxide. Soil Science Society of America Journal 584: 1108-1116.
Stevens, R. J.; Laughlin, R. J.; Atkins, G. J. & Prosser, S. J. (1993). Automated-determination
of nitrogen-15-labeled dinitrogen and nitrous-oxide by mass-spectroscopy. Soil
Science Society of America Journal 574: 981-988.
Tranter, G.; Minasny, B.; McBratney, A. B.; Rossel, R. A. V. & Murphy, B. W. (2008).
Comparing spectral soil inference systems and mid-infrared spectroscopic
predictions of soil moisture retention. Soil Science Society of America Journal 72: 1394-
1400.
Trygg, J. & Wold, S. (1998). PLS regression on wavelet compressed NIR spectra.
Verma, S. K. & Deb, M. K. (2007). Nondestructive and rapid determination of nitrate in soil,
dry deposits and aerosol samples using KBr-matrix with diffuse reflectance Fourier
transform infrared spectroscopy (DRIFTS). Analytica Chimica Acta 582: 382-389.
Viscarra Rossel, R. A.; Jeon, Y. S.; Odeh, I. O. A. & McBratney, A. B. (2008). Using a legacy
soil sample to develop a mid-IR spectral library. Australian Journal of Soil Research
46: 1-16.
Viscarra Rossel, R. A. & Lark, R. M. (2009). Improved analysis and modelling of soil diffuse
reflectance spectra using wavelets. European Journal of Soil Science 60: 453-464.
Viscarra Rossel, R. A. & McBratney, A. B. (1998a). Laboratory evaluation of a proximal
sensing technique for simultaneous measurement of soil clay and water content.
Geoderma 851 19-39.
Viscarra Rossel, R. A. & McBratney, A. B. (1998b). Soil chemical analytical accuracy and
costs: implications from precision agriculture. Australian Journal of Experimental
Agriculture 38: 765-775.
Viscarra Rossel, R. A.; Walvoort, D. J. J.; McBratney, A. B.; Janik, L. J. & Skjemstad, J. O.
(2006). Visible, near infrared, mid infrared or combined diffuse reflectance
spectroscopy for simultaneous assessment of various soil properties. Geoderma 131:
59-75.
Wold, S.; Sjostrom, M. & Eriksson, L. (2001). PLS-regression: a basic tool of chemometrics.
Zimmermann, M.; Leifeld, J. & Fuhrer, J. (2007). Quantifying soil organic carbon fractions by
infrared-spectroscopy. Soil Biology & Biochemistry 39: 224-231.
www.intechopen.com
19
The Application of FT-IR Spectroscopy

in Waste Management
Ena Smidt, Katharina Böhm and Manfred Schwanninger
BOKU - University of Natural Resources and Life Sciences, Vienna
Austria
1. Introduction
This synopsis of FT-IR spectroscopic applications in waste management covers relevant
issues regarding monitoring, process and quality control. Quality in this context means low
reactivity, low gas forming potential and compliance with limit values of materials to be
landfilled, appropriate compost ingredients and improvement of the stable carbon pool by
humification of compost organic matter. The division into sections was carried out
according to specific materials and processes and the related questions to be answered.
Waste materials have not posed a problem in the past, as long as they were returned to the
natural cycle. The higher degree of utilisation, the long lasting life cycle of goods and their
re-use caused minor waste amounts and contributed to the balance of input and output
streams. As natural materials were the basis for good production, waste was integrated in
the natural cycle. Organic waste materials only escaped degradation if they were preserved
under particular conditions. Inorganic residues from ore mining have led to local
contamination by heavy metals. In contrast, fragments of ancient pottery or ruins have
gained in importance as historical witnesses. The economical and social increase in
prosperity and urban development has been paralleled by strongly rising amounts of
organic and inorganic waste and the acceleration of turnover rates. These have led primarily
to sanitation problems, especially in expanding urban areas. Disposal of these wastes by
spreading them over the surrounding countryside or filling dumps were only temporary
“solutions”. Natural waste such as foliage falling in autumn can be considered as an
intermediate product in a closed circle. Except for natural disasters, turnover rates and the
related contents of substances are subjected to regulations. Anthropogenic waste is in most
cases the end of a one-way street that causes system imbalances. Therefore new approaches
and waste management strategies aim at copying natural cycles.
The amount, changes in the chemical composition of our goods, the use of hazardous
substances and the careless landfilling of waste in dumps have caused serious
environmental problems and demonstrated the need for action. The awareness of this issue
emerged when soils and groundwater were contaminated by leachate emissions. Due to the
missing separation of hazardous waste, landfill remediation has often implied a complete
excavation of the landfilled material. Discussions on climate change, the role of waste
management in the global carbon cycle and its contribution to the carbon budget have
drawn attention to relevant gaseous emissions from landfills with a global warming
potential, especially of methane.
www.intechopen.com
Analytical approaches in the past differed in that the main analytical interest focused on the
effect of gaseous emissions on the environment and on soil and water pollution.
Environmental analytical methods were applied to find out the effects in these spheres
caused by waste materials. Waste materials themselves were not investigated. The first
attempts to encapsulate waste materials were aimed at prevention of contact with the
environment. Analytical investigations in the past concentrated on pollution control and on
the chemical and physical quality of landfill liners with regard to their density, tightness
and stability.
Later on the interest in the waste material itself came to the fore as it represents the source of
future emissions. The environmental impact and the economic implication caused the
government to react in terms of regulations regarding the handling of waste materials.
During the last decades waste management has become an important industrial sector in
countries with a high environmental awareness and adequate standards. The approach of
waste encapsulation was abandoned due to the missing knowledge about the durability of
landfill liners and technical facilities. Moreover, it was evident that the encapsulation would
have promoted the preservation of the material and postponed the problem of reactivity.
The concept of a “multi-barrier system” that includes pre-treatment of municipal solid
waste in terms of stabilisation besides the technical equipment and an adequate geological
basement was implemented. The reduction of waste reactivity within a manageable space of
time should avoid the long lasting after care period of several decades that does not comply
with the objective of the Austrian Waste Act (BMLFUW, 1990) not to burden the next
generations with environmental problems from the past.
Among the wide range of current research topics in waste management the development of
appropriate analytical tools is an overall concern at the national and the European level.
Waste materials represent an analytical challenge due to their widely varying composition
and structure contrary to all principles of natural order. The implementation of limit values
by national rules and the need for environmental compliance have stimulated the discussion
on adequate methods, their reliability and usefulness. Their availability is a prerequisite to
achieving the protection of the environment according to the principles of social welfare and
sustainability as required by the aims of the Austrian Waste Act (BMLFUW, 1990). For
waste management practice analytical methods should be fast, cheap, easy to handle, and
marginally error-prone. Many requirements in view of the complex matrix!
There are two main problem areas in waste management to be dealt with: toxic effects and
reactivity of the material. Heavy metals, organic pollutants such as hydrocarbons, polycyclic
hydrocarbons and chlorinated substances are quantified individually. The complex matrix
of waste materials makes the determination and quantification of single substances difficult.
Interferences of different compounds affect their extraction and separation for
quantification. The complex mixture of different materials regarding chemical components,
texture and behaviour implies the application of new analytical tools. With respect to
pollutants the determination of single compounds remains indispensable in many cases.
Analytical advances concentrate on improved extraction and disintegration methods and on
modern instruments for determination. For an overall waste characterisation including
chemical properties and behaviour more holistic approaches have gained in importance.
1.1 Fourier transform infrared (FT-IR) spectroscopy and data evaluation

FT-IR spectroscopy has proved to be a powerful tool to comply with the purpose of
comprehensive characterisation. The unique characteristic of the material presented by the
www.intechopen.com
The Application of FT-IR Spectroscopy in Waste Management 407
spectrum sheds light on material properties, its behaviour as well as on specific components
represented by their functional groups. The substantial progress regarding the infrared
spectroscopic measurement is achieved by the recording of the interferogram, the fast
detection systems and the Fourier transformation. This combination provides many
advantages and has caused a high interest in this method for process and quality control in
many areas (Mantsch & Chapman, 1996; Moron & Cozzolino, 2004; Pollanen et al., 2005;
Zhang et al., 2005). Infrared spectroscopy is based on interactions of infrared radiation with
matter. Infrared light causes functional groups to vibrate. The uptake of energy is indicated
by absorption bands in the spectrum. Measured band intensities depend on the content of
the substance to be determined and on the individual interaction of the functional group
with infrared radiation at a distinct energy level (Hesse et al., 1995; Smith, 1999; Socrates,
2001). The near-infrared (NIR) region from 14,000 cm-1 – 4000 cm-1 and the mid-infrared
(MIR) region from 4000 cm-1 – 400 cm-1 are commonly applied for process and product
control in many industrial fields. The spectral pattern of substances reveals inherent features
and allows the proof of their identity. These characteristics support the solution of frequent
problems in industry. Infrared spectroscopic investigations of waste materials presented in
this chapter have focused on the KBr (potassium bromide) pellet and the attenuated total
reflection (ATR) technique in the MIR area. Most experiences and insight regarding spectra
interpretation, assignment and behaviour of bands during waste degradation or
stabilisation have been gained from the KBr technique (Smidt et al., 2002; Smidt &
Schwanninger, 2005; Smidt & Meissl, 2007). Subsequently, investigations were extended to
the NIR area and the reflection mode using a fibre probe and the integrating sphere, as well
as to the ATR technique in the MIR area (Meissl et al., 2008a). During the last decade the
applicability for waste characterisation and assessment has been proved and gives impetus
to new ways in waste management practice (Chen, 2003; Michel et al., 2006; van Praagh et
al., 2009). In association with multivariate statistical methods these techniques provide an
additional benefit for application in practice in terms of time saving and handling.
Multivariate statistical methods extract a maximum of latent information from a huge data
pool (Brereton, 2002; Esbensen, 2002) and transform the complex spectral pattern into a
generally understandable result. The presented results for classification and prediction of
the applied multivariate statistical methods focus on Principal Component Analysis (PCA),
Soft Independent Modelling of Class Analogy (SIMCA), Partial Least Squares - Discriminant
Analysis (PLS-DA), and Partial Least Squares Regression (PLS-R). PCA is used to analyse
large data sets. This procedure extracts the information of the original data matrix which
leads to a smaller number of dimensions, called principal components (PCs). The required
number of PCs depends on the complexity or dissimilarity of the data-set. The variance
explained by each PC decreases with increasing number of PCs. PCA reveals the inherent
data structure and underlying features which supports the identification of similarities and
differences between materials. The loadings spectra, corresponding to the PCs, provide
information on the contribution of the spectral regions to the differentiation between
samples visible in the scores plot. Based on the PCA, classification and prediction models
can be developed. SIMCA is a well known pattern recognition method which describes each
class separately in a principal components space. New objects are considered to belong to
the class if their distance (e.g. Euclidean) to the constructed PC space is not significantly
larger than the distance of the class objects to their PC space. In a further classification
method PLS-DA, samples are distinguished by means of the partial least squares regression
according to their membership that is defined by the dummy variable (+1) and (-1).
www.intechopen.com
PLS-R relates the variation in one variable (questioned parameter = Y-variable) to the
variation of spectral data (X-variables). Based on this correlation models were calculated
and further used for prediction. It is a prerequisite that the questioned parameter is reflected
by the spectral pattern (Michel et al., 2006; Meissl et al., 2007). The quality of classification
and prediction models is expressed by several characteristics (model parameters). R2 is the
coefficient of determination. The root mean square error of cross validation (RMSECV)
reflects the error of the model. Cross validation is a validation method where samples out of
the sample set are alternately excluded from calibration and used for prediction. This
procedure is repeated until all samples have been kept out once. The bias indicates the
systematic difference between the measured and the predicted values. The ratio of standard
deviation to standard error of performance (RPD) provides information on the precision of
analyses for a specific purpose (Williams & Norris, 2004).
This study shows in many cases the development or changes of spectral characteristics,
therefore the spectra were shifted in parallel for a clear presentation. Most spectra were
recorded from KBr pellets except the ATR spectra shown in section 2.3. For multivariate
data analysis spectra were vector normalised.
1.2 Interpretation of infrared spectral data originating from waste materials

Waste is a complex material and heterogeneous regarding the chemical composition and
texture. Due to the low amount of sample needed for MIR spectroscopic measurement
adequate sample preparation is a prerequisite to obtain reliable and reproducible results. A
convenient procedure of sample preparation was described by Meissl et al. (2008b).
Pure substances are characterised by sharp distinct absorption bands that can be assigned to
functional groups. By contrast, waste materials display broad overlapping bands due to the
complex mixture and manifold interactions between degrading organic molecules. During
the last decade MIR and NIR spectroscopy has been applied with increasing success to
complex samples. New approaches have gained in importance due to the association with
multivariate statistical methods. The proof of identity and the classification of unknown
waste materials are based on similarities or differences of the spectral pattern. Identification
of functional groups and assignment to substances are an essential target to follow and to
understand the chemical changes during degradation and stabilisation processes. Due to the
complexity of the material and overlapping bands the identification is limited to a few but
significant indicator bands.
Spectra interpretation of waste materials is primarily based on theoretical locations of
functional groups cited in literature for pure substances as well as for waste materials or
components of waste materials (Hesse et al., 1995; Smith, 1999; Socrates, 2001; Smidt et al.,
2002). Due to interactions of degrading organic molecules band shifts in waste spectra occur
frequently. Several bands that are assigned to organic functional groups, such as aliphatic
methylene bands (Table 1) have a stable band position in waste spectra. The location of many
other bands may vary due to the stronger influence of the whole molecule and the waste
matrix. Bands of inorganic compounds are mostly sharp and support identification by their
characteristic shape and relatively stable position. For many functional groups wavenumber
regions are indicated. Band assignment is supported by recording the spectral signature of
waste ingredients or by addition of a specific component to waste materials of which the
presence should be verified. The most relevant indicator bands that provide information on
the stage of degradation are compiled in table 1. The list is limited to bands that are clearly
visible in the spectrum of the complex waste matrix even though many substances are
www.intechopen.com
represented by several bands. Intensities of bands that are assigned to organic functional
groups decrease with progressing degradation until a nearly constant level is reached that
indicates the slow-down of metabolic activities. Bands of some metabolites appear temporarily
and disappear. Degradation of organic matter leads to a relative increase of mineral
compounds that is reflected by the increase of the corresponding absorption bands (Hesse et
al., 1995; Naumann et al., 1996; Haberhauer et al., 1998; Grube et al., 1999; Smith, 1999;
Ouatmane et al., 2000; Socrates, 2001; Reig et al., 2002; Smidt et al., 2002; Zaccheo et al., 2002;
Chen, 2003; Madejova, 2003; Tan, 2003; Smidt & Schwanninger, 2005; Smidt & Meissl, 2007).
Location
Vibration Functional group or component
wavenumber (cm-1)
Organic compounds
2920 C-H stretching (as) aliphatic methylene group
2850 C-H stretching (s) aliphatic methylene group
2590-2520 S-H stretching thiols
1740-1700 C=O stretching aldehyde, ketone, carboxylic acids, esters
C=O, COO- stretching amide I, carboxylates
1685-1630
C=C stretching aromatic ring modes, alkenes
1600-1590 C=C aromatic skeleton
1570-1540 N-H in plane bending amide II and secondary amines
1515-1505 aromatic skeletal lignin from lignocellulosic materials
1430-1420 COO- stretching carboxylic acids
1350-1250 C-N stretching primary and secondary aromatic amines
C-O-C stretching esters
1265-1240
C-N stretching amide III
C-O-C, C-O polysaccharides
1250-900
C-O-P phosphodiesters
Inorganic compounds
3700-3200 SiO-H stretching silica
bonded and non-bonded hydroxyl groups
3700-3400 O-H stretching
and water
1635 O-H bending water
1450-1410 C-O stretching carbonate
1384 (1400-1340) N-O stretching nitrate (leachate)
1140-1080 S-O stretching sulphate
1080 Si-O stretching quartz
Si-O stretching clay minerals
1030
Si-O-Si stretching silica
875 C-O out of plane bending carbonate
713 C-O in plane bending carbonate
680-610 S-O bending sulphate
Table 1. Location of indicator bands, corresponding vibration and assignment to functional
groups and components (as = antisymmetric, s = symmetric)
The nitrate band at 1384 cm-1 in solid waste materials has been proved by addition of KNO3
(potassium nitrate) (Fig. 1a). The appearance of the nitrate band in the compost spectrum at
www.intechopen.com
an advanced stage of composting corresponds to the evolution of nitrate in the biological

process (Fig. 1b). The sharp band shape is an additional indicator of an inorganic compound
that does not interact with other molecules as degrading organic molecules do. Both facts
support identification of nitrate characterised by a stable band position in solid waste samples
(Smidt et al., 2002). It is evident that all background information available contributes to band
assignment and spectra interpretation. It should be mentioned that the position of the nitrate
band in liquid samples vary in a wider range (cp. freeze-dried leachate in Fig. 5).
(a) (b)
1384
1384
Absorbance
Absorbance
1.0
0.5
48 w
0.2
0.1 32 w
3w
1800 1300 800 1800 1300 800

W avenumber (cm-1 ) W avenumber (cm-1 )
Fig. 1. (a) Increasing nitrate band at 1384 cm-1 due to addition of KNO3, corresponding to
mass fractions of 0.1, 0.2, 0.5, and 1% nitrate; (b) emerging nitrate band during a composting
process (3, 32, 48 weeks)
(a) (b)
S-O stretch paper sludge
with calcite
after HCl
O-H stretch treatment
S-O
Absorbance
Absorbance
bends
O-H bends
construction waste
CaSO 4 *2H2 O cellulose
3400 2400 1400 400 3400 2400 1400 400

Fig. 2. (a) Infrared spectra of mineral lumps from a construction waste landfill and
CaSO4*2H2O; (b) spectral characteristics of the original industrial landfill sample (carbonate
bands indicated by arrows), the HCl treated sample and cellulose as reference
www.intechopen.com
Figure 2 shows approaches of band assignment by comparison of waste ingredients with a

pure substance and by sample treatment. Figure 2a displays spectra of mineral lumps from a
construction waste landfill and CaSO4*2H2O (calcium sulphate) to reveal the specific bands
assigned to sulphate that plays a relevant role in leachate. Apart from the sulphate (S-O)
stretching and bending vibrations at 1136 cm-1 and 620 cm-1, respectively, the mineral lumps
also show strong O-H stretching vibrations. Figure 2b illustrates an example of waste
material originating from an abandoned landfill with unknown composition. The sample
was treated with HCl (c = 0.1 mol/L) to convert carbonates into carbon dioxide and water,
and to reveal bands of other compounds that are overlapped and covered by strong bands
of carbonate. Cellulose was identified by comparison with a spectrum of pure cellulose. The
material was found to be industrial paper sludge mixed with carbonate for stabilisation
before landfilling.
2. Abandoned sites and old landfills

The environmental damages resulting from abandoned landfills and dumps have revealed
the problems of careless disposal. At the beginning of a well-regulated and organised waste
management system pollutants and toxic substances were in the foreground due to their
immediate impact on water and soil. They covered the more subtle contribution of
unsuspicious goods of daily use to relevant emissions and their effect on the imbalance in
the global carbon budget. The registration of abandoned landfills is part of exploration
programmes in terms of risk assessment and adequate remediation measures. The majority
of abandoned landfills under investigation contain municipal solid waste and construction
waste originating from the sixties and seventies of the last century. The measurement of
landfill gas emissions is still state of the art and part of monitoring programmes. Long-path
spectroscopic instruments developed for air pollution control are also applied for landfill
monitoring. Long-path FT-IR spectroscopic investigations provide several advantages such
as fast scanning of large surfaces at long distances without contact and online measurement
of emissions from areas that are difficult to access. Due to infrared absorption of many air
pollutants they can be detected and quantified by FT-IR long-path instruments (Bacsik et al.,
2005). Quantification of gaseous compounds is carried out according to Lambert-Beer’s law
(Weber et al., 1996; Galle et al., 2001; Hegde et al., 2003). Methane and nitrous oxide are the
main components to be measured in the context of landfill monitoring. The measurement of
gaseous emissions only provides information on the current microbial activity that depends
on environmental conditions such as water supply. The gas forming potential according to
the present chemical compounds that could be degraded under appropriate conditions
cannot be determined in this way. Preferential gas flows make the quantification of
emissions difficult. Due to the missing base seal, the leachate that provides additional
information is not available in most cases. Therefore investigations of the solid waste are
indispensable.
2.1 Classification of waste materials

The classification of landfilled materials should be a first step. The knowledge about the
landfilled material supports the interpretation of data as the composition influences the
behaviour, especially biological tests of reactivity. Municipal solid waste (MSW) comprises
all components of consumer goods and features a typical spectral pattern. Different pre-
treatment and landfilling conditions of MSW lead to divergent spectral characteristics of the
www.intechopen.com
waste material. Old landfills (old-LF) are characterised by the high portion of construction
waste. The 5-year-old recent reactor landfill (reactor-LF) differs from the landfill with
mechanically-biologically treated waste (MBT-LF) due to the missing pre-treatment. The
waste bank represents a special kind of deposit where scarcely biologically treated MSW
was piled up to form a constructional wall of about 6 m. Despite the differences in
composition and degradation conditions similarities are distinctive. Industrial waste (ind-
LF), by contrast, features a more one-sided composition. The following PCAs are based on
the mean-centred MIR spectra that cause zero-lines in the scores and loadings plots,
indicating the averages of all samples. The PCA supports the discovery of differences and
similarities that are visualised by long or short distances between the samples in the scores
plot (Figs. 3a, c). They illustrate the differences between landfills containing MSW and an
industrial landfill containing carbonate stabilised paper sludge. The loadings plots elucidate
how much the spectral regions contribute to the first (PC1, explained variance 69%) and the
second (PC2, 15%) principal component (Figs. 3b, d). The PCA calculated with the whole
spectrum (Figs. 3a, b) reveals the strong influence of inorganic components, especially of
carbonate and clay minerals that dominate the loadings spectrum of PC1. The position of the
ind-LF samples found at the positive side of PC1 in the scores plot corresponds to the
positive loadings (e.g. at 1420 cm-1). The position of most of the MSW landfill samples found
at the negative side of PC1 in the scores plot corresponds to the negative loadings (e.g. at
1030 cm-1). To sum it up: positive loadings plus positive scores = ind-LF with more
carbonate (1420 cm-1), negative loadings plus negative scores = MSW landfills with more
clay minerals (1030 cm-1). This means that ind-LF and MSW landfills differ basically in
composition especially in carbonate and clay minerals. Different types of MSW landfills are
distinguished along PC2 that reflects the development from reactor-LF to old-LFs. Apart
from inorganic compounds (clay minerals) spectral regions that can be assigned to organic
functional groups (2920 and 1640 cm-1) become more relevant in the loadings spectrum (Fig.
3b). The characteristics are in accordance with higher organic matter contents and reactivity
in the reactor-LF compared to old-LFs that display higher mineral contents due to
mineralisation and a portion of construction waste. However, the transition from one
landfill type to another one is not distinct as old-LFs can still feature more reactive sections.
Vice versa, the reactor-LF also contains sections of low organic matter content and moderate
reactivity. Most of the material from the 15-year-old waste bank is located in the scores plot
between the reactor-LF and the old-LFs. The variance within the bank samples is caused by
different environmental conditions and development during the 15-years of disposal. More
details were reported by Smidt et al. (2007).
Figs. 3c and 3d present the scores plots and the loadings spectra of a PCA that was based on
the organic indicator bands represented by C-H and C=O vibrations. Separation of MSW
landfill types due to organic compounds is performed along PC1. MSW landfill types and
the ind-LF are differentiated along PC2. In terms of these characteristics several old-LF
samples move closer to ind-LF samples in the scores plot. This indicates a weaker
discrimination power of the selected spectral regions.
The PCA is an appropriate tool to obtain a general idea on the similarities and differences of
waste materials and the contributing spectral regions. For practical application the
development of classification models is indispensable in order to immediately attribute
waste materials to defined classes depending on the problem to be solved. A classification
model according to SIMCA was reported by Smidt et al. (2008b) to distinguish different
waste materials.
www.intechopen.com
(a) (b)
0.4 0.20
old-LF 0.16
0.2
0.12
MBT-LF 1420
ind-LF
PC2 (15%)
Arbitrary unit
0.08
0
waste bank 0.04
PC1
0.00
-0.2
reactor-LF
-0.04
PC2
1030
-0.4 -0.08
-0.5 0 0.5 1 3400 2400 1400 400
-1
PC1 (69%) Wavenumber (cm )
(c)0.12 (d) 0.20
ind-LF PC1
0.16
2920
1575
0.12
0.06 2850
Arbitrary Unit
PC2 (19%)
old-LF 0.08
reactor-LF
0.04
0.00
0.00
waste
bank -0.04
MBT-LF
PC2 1640
-0.06 -0.08
-0.2 -0.1 0.0 0.1 0.2 3200 2700 2200 1700 1200
PC1 (56%) W avenumber (cm-1 )
Fig. 3. Scores plots and loading spectra of samples from landfilled wastes based on the
whole spectra (a, b) and on selected wavenumber regions from 3030 to 2800 cm-1 and from
1795 to 1530 cm-1 (c, d)
2.2 Risk assessment of abandoned landfills and dumps regarding the reactivity
The careless landfilling of waste in the past has caused extensive investigations to assess the
remaining risk and the necessity of remediation measures. In this context the stability of
waste organic matter is a fundamental criterion (Binner & Zach, 1999; Tesar et al., 2007; van
Praagh et al., 2009). The proof of stability can be carried out by means of time-consuming
biological tests that provide information on the current microbial activity under aerobic
conditions and the gas forming potential under anaerobic conditions. These tests were
primarily established for MBT waste to verify its stability prior to landfilling (Binner &
Zach, 1999; Adani et al., 2004). The dissolved organic carbon (DOC) and the total organic
carbon (TOC) are relevant parameters for the assessment of abandoned landfills according
to the Austrian Landfill Ordinance (BMLFUW, 2008). The DOC provides information on
www.intechopen.com
soluble organic compounds. Their portion decreases with progressing stabilisation. The
stipulated limit values of TOC restrict the organic matter content in all landfill types. Only
for MBT waste the TOC limitation is not valid. Figure 4 shows the correlation between
predicted and measured (references) parameters using selected regions of the MIR spectrum
and PLS-R. The model parameters are summarised in table 2.
(a) 2000 (b) 25
20
)
1500
-1
TOC predicted (% DM)

DOC predicted (mg C L
15
1000
10
500
5
0 0
0 500 1000 1500 2000 0 5 10 15 20 25 30
-1
DOC measured (mg C L ) TOC measured (% DM)
Fig. 4. Correlation between (a) the predicted and the measured DOC and (b) the predicted
and the measured TOC, both based on the spectral ranges from 3003 - 2816 and 1770 -
400 cm-1 and PLS-R
Model parameters DOC TOC

Wavenumber ranges (cm-1) 3003 - 2816 and 1770 - 400 cm-1
Calibrated range 15 – 1700 mg C L-1 2.6 – 23.5 % DM
R2 89% 90%
RMSECV 158 mg C L-1 1.9 % DM
RPD 3.0 3.1
No. of PLS components 6 7
Bias -5.94 -0.0719
Table 2. Parameters for the PLS-R models dissolved organic carbon (DOC) and total organic
carbon (TOC) content
2.3 Success control of remediation measures

In situ aeration of old reactive landfills is one measure to avoid methane emissions and to
accelerate mineralisation in that anaerobic conditions are exchanged for aerobic ones. The
spectral pattern of the solid waste matrix reflects the chemical changes mainly by decreasing
band heights of organic functional groups (Tesar et al., 2007). Aeration of the solid waste is
paralleled by oxidation of soluble components in the leachate. The most conspicuous
changes in the spectral pattern are caused by transformation of N-H and S-H groups
containing compounds to the mineralisation products nitrate and sulphate that are
www.intechopen.com
presented in the spectrum by strong bands of the N-O and the S-O vibrations (Smidt &
Schwanninger, 2005; Smidt & Meissl, 2007). ATR spectra (4000 – 600 cm-1) of freeze-dried
leachate from aerated landfill material and leachate from landfill material under anaerobic
conditions were classified by means of PLS-DA (Fig. 5). The corresponding loadings
spectrum of the 1st PLS component (Fig. 5b) shows the relevant regions, especially the S-O,
N-O, C-H, S-H and N-H vibrations (Table 1) that contribute to the discrimination of leachate
obtained from materials under different aeration conditions. Three PLS components explain
97% of the total variance, whereas the 1st one already explains 94%, indicating a distinct
differentiation and the efficiency of aeration.
(a) 1.5 (b) 0.15

1100
0.10
620
0.5 0.05
predicted ()
Arbitrary Unit
Aerated
0.00
-1.5 -0.5 0.5 1.5
-0.5 -0.05 2550-2530
-0.10 2920 2850 1550
Anaerobic 1360
-1.5 -0.15
class () 3600 2600 1600 600
W avenumber (cm-1 )
Fig. 5. PLS-DA results based on spectral data of freeze-dried leachate from landfill material
under anaerobic and aerated conditions: (a) correlation between predicted and class
(dummy variable) and (b) loadings spectrum of the 1st PLS component
2.4 Long-term behaviour of deposits

The long-term behaviour of landfills and dumps is a relevant issue with respect to
remediation activities and the re-use of these areas. Although organic matter degradation
leads to similar metabolic and mineralised products, the individual composition causes a
specific spectral pattern as shown in figure 3. MSW that had been piled up to a waste bank
(Fig. 3) showed different characteristics after 15 years corresponding to depths and air
supply. Aerobic conditions in the upper layer of the profiles accelerated degradation and
corresponding changes in the spectral pattern (Smidt et al., 2007). The classification of
samples according to specific stages of degradation by means of spectral characteristics is a
fast method for the risk assessment of abandoned landfills, besides the precise
determination of relevant parameters as shown in section 2.2. The long-term behaviour is
strongly influenced by both the chemical composition and environmental conditions.
The scores plot and loadings spectra in figure 6 illustrate the distance of samples and the
responsible spectral regions. The samples originated from an old landfill. They were taken as a
composite sample along the depth profile. Two samples (black dots) that were collected
between tight clayey layers, where degradation was inhibited, differ considerably from the
other ones. The loadings spectra of PC1 and PC2 reveal the strong influence of mineral
www.intechopen.com
compounds (1430 – 400 cm-1). The loadings spectrum of PC1 additionally shows the
contribution of organic components (1630 cm-1). Besides the poor mineralisation the enclosed
material was protected against degradation and subsequent mixing with construction waste.
(a) 0.3 (b) 0.20
0.2 0.16
0.12 1430
0.1
Arbitrary Unit
PC2 (35%)
0.08
0.0
0.04
-0.1 PC1
0.00
-0.2 -0.04
1630 875
PC2
-0.3 -0.08
-0.6 -0.4 -0.2 0.0 0.2 0.4 3400 2400 1400 400
Fig. 6. PCA based on spectra of samples from an old landfill; (a) scores plot with 2 samples
(black dots) between tight clayey layers, (b) loadings spectra of PC1 and PC2
3. Separation of recyclables and mechanical-biological or thermal pre-

treatment of municipal solid waste prior to landfilling
The maximum possible separation of recyclables is a crucial issue. In addition to the
individual separation by the waste producer, the recovery of recyclables is improved by
mechanical sorting systems in the plant. The pre-treatment of municipal solid waste prior to
landfilling comprises two main strategies: the mechanical-biological and the thermal
treatment. Both processes aim at reducing reactivity, either by mineralisation and
stabilisation or by incineration of organic matter.
3.1 Identification and separation of recyclables by NIR spectroscopy

The development of mechanical separation and sorting systems has become an important
technical approach in the improvement of the recycling proportion and the required quote
respectively. Identification of different materials is a prerequisite in order to achieve an
efficient and correct separation. NIR spectroscopy is an appropriate tool for a
comprehensive characterisation of wastes (Bonifazi et al., 2009) and differentiation of
polymers. In this field of application the measurement in the near infrared area ranks first
(Feldhoff et al., 1997; Van Den Broek et al., 1998; Kulcke et al., 2003; Leitner et al., 2003; Li et
al., 2005; Dou et al., 2006; Luiken & Bos, 2010).
3.2 Mechanically-biologically treated (MBT) waste - process control and compliance

with limit values according to the Austrian Landfill Ordinance
National and international regulations and limit values for MBT waste require verification
of compliance before final disposal. For MBT waste reactivity and stability are relevant
www.intechopen.com
properties. Process kinetics, the release of organic components and their long-term
behaviour are fields of intensive research in waste management (Barlaz et al., 1990; Barlaz,
1998; Tintner et al., 2010). Despite the scientific interest, the performance of industrial
processes strongly depends on this basic knowledge. Waste organic matter primarily
consists of non-hazardous organic molecules responsible for the reactivity of the material
and emerging gaseous and liquid emissions. Chemical and physical sum parameters are
preferably applied for waste characterisation in practice. Waste organic matter is usually
quantified by the determination of loss of ignition, TOC and nitrogen content. The
significance of these parameters is limited as they do not differentiate the particular
behaviour of the molecules subsumed therein. Mineralisation products of organic matter
such as ammonium and nitrate provide information on progressing biological degradation.
High ammonium contents are assigned to early stages of degradation, whereas nitrate
indicates the late phase when ammonium has been oxidised. Nevertheless, a wide range of
contents is observed during the biological treatment. The coexistence of ammonium and
nitrate and the repeated increase and decrease of ammonium contents make unambiguous
assessment difficult. Last but not least these parameters depend on the uniformity of
metabolic processes in the rotting unit. The equal progress is strongly related to process
operation in terms of mechanical mixing and evenly distributed aeration and moisture.
Biological tests provide more comprehensive information on reactivity in that chemical
properties of all compounds contribute to the behaviour of waste materials under aerobic
and anaerobic conditions. Biodegradability of organic compounds is revealed by the oxygen
uptake or CO2 release due to microbial activity, indicating mineralisation of organic
molecules. The gas sum over a period of 21 days, generated under anaerobic conditions, is
measured by means of the incubation test (Austrian Standards Institute, 2004). For MBT
materials to be landfilled biological tests replace the determination of the total organic
carbon content (TOC) because the limit value of 5% dry matter for this landfill type cannot
be achieved by biological degradation only. However, biological parameters are time-
consuming and require considerable expertise. In addition to biological tests limitation of
the calorific value by 6600 kJ kg-1 dry matter was stipulated for MBT waste. The calorific
value indicates the efficiency of plastic separation and progressing degradation of organic
matter by its decline (BMLFUW, 2008). Chemical changes during the biological treatment of
MSW are reflected by spectral characteristics that allow a fast process control, evaluation
and optimisation. The progress of organic matter degradation is influenced by the technical
system and process operation. The development of spectral characteristics is illustrated in
figure 7. Fresh input materials feature strong aliphatic methylene bands at 2920 and
2850 cm-1. Absorption bands of organic components (cp. Table 1, cellulose) decrease or
disappear as far as they indicate temporarily present metabolic products (1740, 1320 cm-1).
This process showed extensive mineralisation after 9 weeks. It can be assumed that the
chemical composition of MSW is related to the biological behaviour. PLS-R models based on
MIR spectra were developed for the prediction of the biological parameters “respiration
activity” and “gas generation sum” to get information on reactivity of MBT waste faster
(Böhm et al., 2010a). Bands that feature considerable changes during the biological treatment
such as the aliphatic methylene bands and the fingerprint region (1788 - 1532 cm-1, 1350 -
1028 cm-1), were selected for the prediction models.
The loadings spectrum of the first PLS component out of 8 (Fig. 8b) indicates that organic
(methylene bands) and inorganic (carbonate bands at 1430 and 875 cm-1) compounds
contribute to the thermal behaviour of the waste sample. Whereas waste organic matter
www.intechopen.com
causes the calorific value to increase, water and carbonates affect its decrease due to the
endothermic reactions of water evaporation and carbonate decay.
1030
1420
1630
875
Absorbance
2920 2850
9 weeks
6 weeks
0 weeks 1320
174 Cellulose
3900 3400 2900 2400 1900 1400 900 400

-1
Wavenumber (cm )
Fig. 7. Changing spectral characteristics of MSW during the biological treatment
Figure 8 illustrates the correlation between the predicted and the measured calorific value of
the PLS-R model (Model parameters: R2 = 76%, RMSECV = 711 kJ kg-1 dry matter (DM),
RPD = 2.1; bias = 2.1).
(a) 12000 (b) 0.08

DM)
2920
0.06 2850
-1
Calorific value predicted (kJ kg
9000 0.04
Arbitrary Unit
0.02
6000 0
-0.02
3000 -0.04
-0.06 875
1430
0 -0.08
0 3000 6000 9000 12000 3400 2400 1400 400
Calorific value measured (kJ kg-1 DM) W avenumber (cm-1 )
Fig. 8. (a) Correlation between the predicted and the measured calorific values based on the
whole MIR spectrum, (b) the corresponding loadings spectrum of the 1st PLS component
3.3 Thermal treatment of MSW - natural and accelerated ageing (carbonation) of MSW
incinerator bottom ash
Inorganic residues from incineration have gained in significance during the last decades due
to the increase of the thermal treatment of waste for energy recovery. The long-term
behaviour of landfilled bottom ash, the process of carbonation and the fate and release of
www.intechopen.com
heavy metals all dependent on the ageing process, have become relevant topics. In figure 9a
the main differences between spectra of MSW (I) and MSW incinerator bottom ash (II) are
elucidated. Spectra of incineration residues differ from MSW by the loss of bands that are
assigned to organic functional groups. Figure 9b illustrates two spectra of MSW incinerator
bottom ash before (III) and after CO2 exposure (IV), two spectra of a bottom ash deposit
from a depth of 50 cm (V), from the surface (VI), and a spectrum of the calcite reference. It is
evident that the progress of carbonation in abandoned deposits depends on the available
surface and the access of air. Carbonates that are part of MSW decay during incineration
above 650 °C. The reaction of CaO with water leads to Ca(OH)2 which causes the pH-value
of the inorganic residues to increase.
(a) (b)
1420 1030 1420
1550
2920 875
2850
Absorbance
875 Absorbance
calcite
I
VI
3610 V
II
IV
1635 III
1030
3400 2400 1400 400 3400 2400 1400 400

Fig. 9. Spectra of (a) MSW (I) and MSW incinerator bottom ash (II) and (b) MSW incinerator
bottom ash (III) before and (IV) after CO2 exposure, two samples from different depths (V = 50
cm and VI = surface layer) of a bottom ash deposit and the spectrum of the reference (calcite)
The uptake of CO2 from air effects carbonation. This process in the opposite direction leads
to stabilisation, immobilisation of heavy metals and the decrease of the pH-value to a
neutral level (Chimenos et al., 2000; Polettini & Pomi, 2004; Rendek et al., 2006). This process
can be accelerated by forced CO2 supply (Mostbauer et al., 2008).
The ageing and stabilisation of incinerator bottom ash are solely chemical processes
compared to the stabilisation of waste organic matter that is promoted by microbial activity.
Carbonation is reflected by the increase of typical “carbonate bands” (Table 1) in the
spectrum. Quantification of carbonates in MSW incinerator bottom ash by means of FT-IR
spectra leads to reliable results (Smidt et al., 2009). The standard method is based on the
measurement of the CO2 release from carbonates by acid treatment which can be affected by
the complex matrix.
4. Recovery of biogenic waste materials, secondary products

In the context of resource recovery biogenic waste materials have become important
ingredients for valuable composts that are applied as soil conditioners. Composting has
been practiced for many centuries all over the world at different technical levels (Ahmad et
www.intechopen.com
al., 2007). The variety of biogenic waste materials available from the public separate
collection and from industrial processes requires quality criteria and suitable input materials
for specific applications. This purpose has been stipulated by the Austrian Compost
Ordinance (BMLFUW, 1992). The limitation of heavy metals and pollutants is in the focus of
legislation. Nutrients (Courtney & Mullen, 2008), humic substances (Meissl et al., 2007;
Smidt et al., 2008d) and phytosanitary properties (Bruns et al., 1996; Erhart et al., 1999) are
additional quality criteria that emphasise the positive effects on plants and soils and
improve the reputation of composts. Biogenic waste and sewage sludge are usually
processed in composting plants whereas manure compost is produced locally by farmers.
4.1 Differentiation of various input materials and assignment of composts to specific

biogenic waste materials
Figure 10 displays five composting processes in different plants, grouped by a PCA.
(a) 0.3 (b) 0.12
Bio A 1425
3695
0.2
0.08
0.1 SSL compost

Arbitrary Unit
Bio B 0.04
PC2 (42%)
0.0 PC1
Bio C
0.00
-0.1
Bio D 2920
-0.04
-0.2
PC2 1600
1240
-0.3 -0.08
-0.2 0.0 0.2 0.4 0.6 3400 2400 1400 400
Fig. 10. (a) Scores plot and (b) loadings spectra (PC1 and PC2) of a PCA based on spectral
characteristics of five different composting processes (Bio = biowaste, SSL = sewage sludge)
Input materials are distinguished by their characteristic spectral pattern. Along PC1 the
particular composition of biowaste and sewage sludge compost contributes to the
separation. Biowaste composting processes (Bio A-D) are separated along PC2 according to
specific features such as anaerobic pre-treatment (Bio D) or addition of mineral compounds
(Bio A). Bio B and Bio C represent the typical mixture of yard and kitchen waste that is
processed in open windrow systems.
The loadings spectra of PC1 and PC2 indicate the relevant bands that contribute to the
differentiation of composts: mineral components, O-H vibrations of kaolin at 3695 and
910 cm-1 (additive to sewage sludge compost) and organic components at 1600 cm-1
(aromatic compounds such as humic substances) and at 1240 cm-1 (C-O vibrations).
4.2 Process control during the aerobic treatment (composting) of biowaste (yard
waste, market waste) and sewage sludge
The degradation of organic matter during composting goes through several phases. The
most intensive rotting phase is paralleled by high microbial activity indicated by the high
www.intechopen.com
oxygen demand and the release of metabolic products such as ammonium and volatile fatty
acids responsible for odour emissions. During the maturation phase the respiration activity
reaches a nearly constant low level. The increasing pH-value affects losses of ammonia.
Oxidation of ammonium causes nitrate contents to increase. Compost maturity is
characterised by a C/N ratio of 10 – 12 (Ouatmane et al., 2000), low microbial activity and
plant compatibility that has to be verified by plant tests according to the Austrian Compost
Ordinance (BMLFUW, 1992). Figure 11 and 12 show different stages of three composting
processes by spectral characteristics. The indicator bands (cp. Table 1) that are identifiable
by visual inspection are the same as indicated for MSW. Nevertheless, the complete pattern
of composts differ sufficiently to separate MSW from biogenic waste as shown by the
SIMCA classification model (Smidt et al., 2008b). Figure 11a illustrates a typical biowaste
composting process over a period of 121 days. Apart from organic matter mineral
compounds such as carbonates and clay minerals are usually occurring constituents in
biowaste from the separate collection. Their relative increase due to mineralisation of
organic substances is obvious. Figure 11b demonstrates the degradation of straw mixed with
liquid manure in a fast composter device during a period of nine days. The spectra are
dominated by straw characteristics. It is evident that straw is only partially degraded during
the short period of composting. Besides the disappearing band at 1720 cm-1 and the
decreasing band at 1260 – 1230 cm-1 that can be assigned to esters in e.g. hemicelluloses
(Stewart et al., 1995) and the decreasing bands at 1165 and 1060 cm-1 (cellulose marked by
arrows in the spectrum) intact cell structures are indicated by sharp and distinct bands in
the wavenumber region 1520 - 1200 cm-1. Figure 12 demonstrates the development of
primary sludge and anaerobically stabilised sludge from a wastewater treatment plant to
sewage sludge compost. The latter originates from a composting plant. The FT-IR spectrum
of the primary sludge is dominated by bands that can be assigned to cellulose,
hemicelluloses (1740 and 1240 cm-1) and to the microbial biomass that is identified by
aliphatic methylene and amide bands. The mature sewage sludge compost features low
methylene bands and a considerable nitrate band at 1384 cm-1. Due to the specific mixture in
the composting plant the portion of clay minerals is higher, the carbonate content lower.
(a) (b)
1430
1030 2920 2850

1640
875
Absorbance
Absorbance
1640
2920 2850
9d
121 d 3d
57 d 0d
0d 1320 straw 1720
3400 2400 1400 400 3400 2400 1400 400

Fig. 11. Development of spectral characteristics during composting processes of (a) biowaste
and (b) straw/liquid manure
www.intechopen.com
1030
1420
1640 1384
875
Absorbance 2920 2850
sewage sludge
compost
1540
anaer. stabilised 1240
sludge
primary sludge 1320

1740 Cellulose
3900 3400 2900 2400 1900 1400 900 400

Wavenumber (cm-1)
Fig. 12. Spectra of different process stages of sewage sludge treatment and composting
(a) 2920 2850 1320 (b)

60 12
260 d
71 d 50 10
57 d
RA 4 (mg O 2 g -1 DM)
Band height (%)

Absorbance
40 2920 cm-1 8
44 d
28 d 30 1320 cm-1 6
25 d
20 2850 cm-1 4
8d
2d 10 2
RA4
0 0
3100 2900 2700 1380 1280 0 70 140 210 280
W avenumber (cm-1 ) Composting time (days)
Fig. 13. (a) Decreasing aliphatic methylene bands at 2920 and 2850 cm-1, emerging and
disappearing band at 1320 cm-1; (b) development of corresponding relative band heights and
respiration activity (RA4) during a biowaste composting process (2-260 days)
Data evaluation can be performed in different ways. Changing heights or areas of bands
indicate the progressing mineralisation. Haberhauer et al. (1998) suggested the ratio of band
heights (2920 cm-1/1640 cm-1) as an indicator of stability. Figure 13a shows the development
of selected bands (2920, 2850, and 1320 cm-1), figure 13b the corresponding band heights
during a biowaste composting process over a period of 260 days. The individual band
height is expressed in per cent of ten measured band heights in the spectrum (Smidt et al.,
2002). Aliphatic methylene groups are part of many biomolecules with different
degradability. Degradation of easily degradable substances is indicated by the decreasing
band height at 2920 and 2850 cm-1 that reached a low constant level (Fig. 13). Recalcitrant
www.intechopen.com
molecules such as long chain fatty acids and waxes are responsible for the remaining weak
bands. They are still found in soil organic matter (Jandl et al., 2002). The emerging and
disappearing band at 1320 cm-1 that can be assigned to amines shows the typical behaviour
of metabolic products that are temporarily visible in the spectrum. The corresponding
microbial activity of the composting process is indicated by the respiration activity over a
period of 4 days (RA4). The interaction of microbial activity and changing chemical
composition is revealed by the conformity of the curves, each of them reflecting process
kinetics.
4.3 Quality assessment of the final product - prediction of parameters using PLS-R
For products such as compost, quality criteria have been defined and require analytical
control. Various parameters are involved depending on the waste material to be
investigated. Waste organic matter is affected by different mechanisms of degradation and
transformation. Composts that comply with quality standards according to the Austrian
Compost Ordinance (BMLFUW, 1992) leave the waste management regime and become
products for soil amelioration. Mineralisation and humification contribute substantially to
stabilisation. Whereas mineralisation causes enrichment of scarcely degradable substances,
humification leads to synthesis of new biomacromolecules. Therefore humification
counteracts carbon losses by CO2 release in that it is fixed in humic substances. Extractable
humic acids are a suitable parameter for assessing the increase of stable humic substances in
composts. Prediction of humic acid contents and respiration activity (RA4) by FT-IR
spectroscopy and PLS-R is an adequate approach for avoiding the time consuming
procedure of humic acid (HA) and RA4 determination. Prediction models being valid for
biowaste composts were developed by Meissl et al. (2007; 2008a). Additional PLS-R models
for TOC and total nitrogen (TN) were established by Böhm et al. (2010b). Table 3 compiles
the parameters for the TOC, TN, RA4, and HA prediction models. The quality of the models
depends on the precision of the reference analyses. Considering this fact all developed MIR
based PLS-R models are suitable for practical application, especially for time-consuming
parameters such as the determination of humic acid contents and respiration activity.
Model
TOC TN RA4 HA
parameters
Wavenumber
3000-2800, 1790-1492, 1373-1030 1745-1685, 1610-1567
ranges (cm-1)
Calibrated 9.4 - 27.3 % 1.0 - 60.6 mg O2 g-1
0.9 - 2.6 % DM 4.5 - 45.6 % ODM
range DM DM
R2 89% 83% 88% 88%
RMSECV 1.3 % DM 0.14 % DM 2.9 mg O2 g-1 DM 2.4 % ODM
RPD 3 2.4 2.9 2.9
No. of PLS
6 9 7 7
components
Bias 0.005 0.001 -0.01 0.004
Table 3. Parameters of the models for total organic carbon (TOC), total nitrogen (TN), humic
acids content (HA), and respiration activity (RA4); ODM = organic dry matter
Prediction of commonly used parameters in waste management has been reported by several
authors, especially for composts (Reeves & Van Kessel, 2000a; Reeves & Van Kessel, 2000b;
www.intechopen.com
Michel et al., 2006; Huang et al., 2007). Apart from organic matter, TOC, TN and nutrient
contents, both physical and biological parameters were predicted by FT-NIR spectroscopy.
4.4 Compost application on soils; development of soil organic matter by compost

application – characterisation of humic acids by FT-IR spectroscopy
The loss of organic matter in soils by agricultural activities has become a serious problem in
many countries (Montanarella, 2002). The application of composts is an appropriate
measure for improving the stable organic matter content long-term (Franko et al., 1997;
Jensen et al., 1997; Li et al., 1997) as confirmed by field experiments. Figure 14a visualises the
increase of humic acid contents (% ODM) in three different plots of agricultural soil where
well humified biowaste compost has been applied since 1994, 1999 and 2004 (C). These plots
were compared to the reference soils (R) without any amendment. The spectral region of
aromatic compounds where humic acids are expected (1610 - 1567 cm-1) were selected
(Meissl et al., 2007; Meissl et al., 2008a). The increase is visualised by the highlighted area in
figure 14b (1610 - 1567 cm-1) between the spectra of the reference soil (R) and of the soil with
compost application (C). It corresponds to the duration of compost application (Smidt et al.,
2008c). Humic acids belong to the stable organic matter fraction in soils with low turnover
rates. Regarding the age of soil humic acids and the diversity of these molecules the
question arises, if compost humic acids are similar regarding the composition and stability.
Figure 15a demonstrates the changing spectral pattern of compost humic acids after 4, 25
and 260 days of composting. In the scarcely composted material humic acids feature strong
aliphatic methylene bands and bands that can be assigned to polysaccharides (1150 –950
cm-1). In the mature compost, spectral characteristics become similar to soil humic acids. The
intensity of the band at 1030 cm-1 is caused by impurities of clay minerals that were not
completely separated by the extraction procedure of soil humic acids (Smidt et al., 2008d).
(a) 100 (b)

1994 1999 2004 1994 1999 2004
80
C
C
Absorbance
HA (% ODM)
60
C R
R
40
R
20
R C R C R C I II I II I II
0
Fig. 14. Increasing contents of stable organic matter in agricultural soil due to compost
application since 1994, 1999 and 2004, verified by (a) increasing humic acid (HA) contents
(shaded bar) and (b) corresponding spectral characteristics (C = compost amended soil, R =
reference soil, presented wavenumber region 1795 (I) – 1505 (II) cm-1)
Lignin is known to serve as a precursor in humic substance formation (Tan, 2003). A lab
scale composting process with biogenic waste was carried out to reveal the effect of lignin
on humic acid synthesis. Addition of 5% lignin from an industrial process could improve
www.intechopen.com
the humic acid content in the mature compost compared to the reference without lignin
application. Figure 15b shows the spectra of freeze-dried humic acids extracted from the
fresh biogenic waste with lignin addition (1 d), from the 42-day-old compost with lignin
(42 d) and from the reference without lignin (R-42 d). Whereas spectral characteristics of
lignin are clearly visible at the beginning (1 d), they disappear during the progressing
process (42 d). Humic acids become similar to the reference. Based on FT-IR spectra it can be
assumed that lignin building blocks were integrated in the humic acid molecule (Smidt et
al., 2008a).
(a) (b)
2920 1720 1620
2850 1230 1030
R-42 d
Absorbance
Absorbance
soil
42 d
260 d
1d
25 d
4d
lignin
3400 2400 1400 400 3400 2400 1400 400

Fig. 15. (a) Spectral pattern of humic acids originating from a biowaste composting process
(4, 25, 260 days) and from soil; (b) integration of lignin in compost humic acids
5. Conclusion
The application of FT-IR spectroscopy in waste management provides fast, cheap and
reliable process and product control. Moreover, environmental monitoring can be based on
spectral characterisation. Besides conventional methods of spectral data evaluation
multivariate statistical methods enable the extraction of the inherent information. The
development of classification and parameter prediction models is a prerequisite for using
this analytical tool in waste management practice. There are many perspectives for the
adoption of the method in this area. Depending on the questions to be answered tailor made
models can be developed in order to replace time-consuming analyses. The strength of FT-
IR spectroscopic investigations is the comprehensive characterisation of complex materials
by their unique signature. Due to the small sample amount used for infrared spectroscopic
measurements careful sample preparation is a crucial issue in view of the heterogeneous
mixture of waste materials.
6. References
Adani, F.; Tambone, F. & Gotti, A. (2004). Biostabilization of municipal solid waste. Waste
Management, 24, 775-783, 0956053X
www.intechopen.com
Ahmad, R.; Jilani, G.; Arshad, M.; Zahir, Z.A. & Khalid, A. (2007). Bio-conversion of organic
wastes for their recycling in agriculture: An overview of perspectives and
prospects. Annals of Microbiology, 57, 471-479, 15904261
Austrian Standards Institute (2004). Stability parameters describing the biological reactivity of
mechanically-biologically pretreated residual wastes - Part 2: Gas generation by incubation
test (GS21), OENORM S 2027-2
Bacsik, Z.; Mink, J. & Keresztury, G. (2005). FTIR spectroscopy of the atmosphere Part 2.
Applications. Applied Spectroscopy Reviews, 40, 327-390, 05704928
Barlaz, M.A. (1998). Carbon storage during biodegradation of municipal solid waste
components in laboratory-scale landfills. Global Biogeochemical Cycles, 12, 373-380,
08866236
Barlaz, M.A.; Ham, R.K. & Schaefer, D.M. (1990). Methane production from municipal
refuse: A review of enhancement techniques and microbial dynamics. Critical
Reviews in Environmental Control, 19, 557-584, 1040838X
Binner, E. & Zach, A. (1999). Biological reactivity of residual wastes and dependence on the
duration of pretreatment. Waste Management and Research, 17, 543-554, 0734242X
BMLFUW (1990). Abfallwirtschaftsgesetz, BGBl. Nr. 325/1990
BMLFUW (1992). Verordnung zur getrenten Erfassung von Bioabfällen - Austrian ordinance for
the seperate collection of biogenic waste from households, BGBl. 68/1992
BMLFUW (2008). Verordnung des Bundesministers für Land- und Forstwirtschaft, Umwelt und
Wasserwirtschaft über Deponien (Deponieverordnung 2008), BGBl. Nr. II 39/2008
Böhm, K.; Smidt, E.; Binner, E.; Schwanninger, M.; Tintner, J. & Lechner, P. (2010a).
Determination of MBT-waste reactivity - An infrared spectroscopic and
multivariate statistical approach to identify and avoid failures of biological tests.
Waste Management, 30, 583-590, 0956053X
Böhm, K.; Smidt, E.; Binner, E.; Tintner, J. & Lechner, P. (2010b). Prediction of various
quality parameters of biowaste composts by means of FourierTransform infrared
(FTIR) spectroscopy. Proceedings of 7th international conference ORBIT 2010,
9789606865251, Heraklion, Greece, Grafima publications, Thessaloniki, Greece
Bonifazi, G.; Serranti, S.; Bonoli, A. & Dall'Ara, A. (2009). Innovative recognition-sorting
procedures applied to solid waste: The hyperspectral approach. Proceedings of WIT
Transactions on Ecology and the Environment, pp. 885-894, 9781845641818,
Brereton, R.G. (2002). Chemometrics: Data analysis for the laboratory and chemical plant. John
Wiley & Sons Ltd., 0471489778, Chichester, England
Bruns, C.; Ahlers, S.; Gattinger, A.; Schuler, C.; Vogtmann, H. & Wolf, G. (1996). The
suppressive effects of composted seperately collected organic waste and yard waste
compost on two important soilborne plant pathogens, In: Science of composting, M.
de Bertoldi, P. Sequi, B. Lemmes and T. Papi (Ed.), 1094-1095, Chapman & Hall,
0751403830, England
Chen, Y. (2003). Nuclear Magnetic Resonance, Infra-Red and Pyrolysis: Application of
Spectroscopic Methodologies to Maturity Determination of Composts. Compost
Science & Utilization, 11, 152-168, 1065657X
Chimenos, J.M.; Fernández, A.I.; Nadal, R. & Espiell, F. (2000). Short-term natural
weathering of MSWI bottom ash. Journal of Hazardous Materials, 79, 287-299,
03043894
www.intechopen.com
Courtney, R.G. & Mullen, G.J. (2008). Soil quality and barley growth as influenced by the
land application of two compost types. Bioresource Technology, 99, 2913-2918,
09608524
Dou, Y.; Zhang, W.; Zhang, Y.; Zhou, H.; Wang, D. & Xu, J. (2006). Identification of plastics
by fourier transform raman and infrared spectroscopy. Fenxi Huaxue/ Chinese
Journal of Analytical Chemistry, 34, 1615-1618, 02533820
Erhart, E.; Burian, K.; Hartl, W. & Stich, K. (1999). Suppression of Pythium ultimum by
biowaste composts in relation to compost microbial biomass, activity and content of
phenolic compounds. Journal of Phytopathology, 147, 299-305, 09311785
Esbensen, K. (2002). Multivariate Data Analysis - in practice. Alborg University, 8299333032,
Esbjerg
Feldhoff, R.; Wienke, D.; Cammann, K. & Fuchs, H. (1997). On-line post consumer package
identification by NIR spectroscopy combined with a FuzzyARTMAP classifier in an
industrial environment. Applied Spectroscopy, 51, 362-368, 00037028
Franko, U.; Crocker, G.J.; Grace, P.R.; Klír, J.; Körschens, M.; Poulton, P.R. & Richter, D.D.
(1997). Simulating trends in soil organic carbon in long-term experiments using the
CANDY model. Geoderma, 81, 109-120, 00167061
Galle, B.O.; Samuelsson, J.; Svensson, B.H. & Borjesson, G. (2001). Measurements of methane
emissions from landfills using a time correlation tracer method based on FTIR
absorption spectroscopy. Environmental Science and Technology, 35, 21-25, 0013936X
Grube, M.; Zagreba, E.; Gromozova, E. & Fomina, M. (1999). Comparative investigation of
the macromolecular composition of mycelia forms Thielavia terrestris by infrared
spectroscopy. Vibrational Spectroscopy, 19, 301-306, 09242031
Haberhauer, G.; Rafferty, B.; Strebl, F. & Gerzabek, M.H. (1998). Comparison of the
composition of forest soil litter derived from three different sites at various
decompositional stages using FTIR spectroscopy. Geoderma, 83, 331-342, 00167061
Hegde, U.; Chang, T.C. & Yang, S.S. (2003). Methane and carbon dioxide emissions from
Shan-Chu-Ku landfill site in northern Taiwan. Chemosphere, 52, 1275-1285, 00456535
Hesse, M.; Meier, H. & Zeeh, B. (1995). Spektroskopische Methoden in der organischen Chemie.
Georg Thieme Verlag, 3135761053, Stuttgart
Huang, G.; Han, L. & Liu, X. (2007). Rapid estimation of the composition of animal manure
compost by near infrared reflectance spectroscopy. Journal of Near Infrared
Spectroscopy, 15, 387-394, 09670335
Jandl, G.; Schulten, H.R. & Leinweber, P. (2002). Quantification of long-chain fatty acids in
dissolved organic matter and soils. Journal of Plant Nutrition and Soil Science, 165,
133-139, 14368730
Jensen, L.S.; Mueller, T.; Nielsen, N.E.; Hansen, S.; Crocker, G.J.; Grace, P.R.; Klír, J.;
Körschens, M. & Poulton, P.R. (1997). Simulating trends in soil organic carbon in
long-term experiments using the soil-plant-atmosphere model DAISY. Geoderma,
81, 5-28, 00167061
Kulcke, A.; Gurschler, C.; Spöck, G.; Leitner, R. & Kraft, M. (2003). On-line classification of
synthetic polymers using near infrared spectral imaging. Journal of Near Infrared
Spectroscopy, 11, 71-81, 09670335
Leitner, R.; Mairer, H. & Kercek, A. (2003). Real-time classification of polymers with NIR
spectral imaging and blob analysis. Real-Time Imaging, 9, 245-251, 10772014
www.intechopen.com
Li, C.; Frolking, S.; Crocker, G.J.; Grace, P.R.; Klír, J.; Körchens, M. & Poulton, P.R. (1997).
Simulating trends in soil organic carbon in long-term experiments using the DNDC
model. Geoderma, 81, 45-60, 00167061
Li, H.; Kumagai, M.; Takahashi, Y.; Amano, T.; Fujiwara, K.; Yoshimura, N. & Ogawa, N.
(2005). Effect of color and shape on discrimination of plastic by a portable near-
infrared spectrometer. International Journal of the Society of Material Engineering for
Resources, 17, 33-38, 13479725
Luiken, A. & Bos, P. (2010). Automated identification and sorting of post-consumer textile
waste. Melliand International, 16, 50-51, 09479163
Madejova, J. (2003). FTIR techniques in clay mineral studies. Vibrational Spectroscopy, 31, 1-
10, 09242031
Mantsch, H.H. & Chapman, D. (1996). Infrared Spectroscopy of Biomolecules. Wiley,
0471021849, New York
Meissl, K.; Smidt, E. & Schwanninger, M. (2007). Prediction of humic acid content and
respiration activity of biogenic waste by means of Fourier transform infrared (FTIR)
spectra and partial least squares regression (PLS-R) models. Talanta, 72, 791-799,
00399140
Meissl, K.; Smidt, E.; Schwanninger, M. & Tintner, J. (2008a). Determination of humic acids
content in composts by means of near- and mid-infrared spectroscopy and partial
least squares regression models. Applied Spectroscopy, 62, 873-880, 00037028
Meissl, K.; Smidt, E. & Tintner, J. (2008b). Reproducibility of Fourier transform infrared
spectra of compost, municipal solid waste, and landfill material. Applied
Spectroscopy, 62, 190-196, 00037028
Michel, K.; Bruns, C.; Terhoeven-Urselmans, T.; Kleikamp, B. & Ludwig, B. (2006).
Determination of chemical and biological properties of composts using near
infrared spectroscopy. Journal of Near Infrared Spectroscopy, 14, 251-259, 09670335
Montanarella, L. (2002). Organic matter levels in European Agricultural soils. Proceedings of
„Biological Treatment of Biodegradable Waste – Technical Aspects“, Brussels, 8.-
10.4.2002, EUR 20517 EN, 2003,
Moron, A. & Cozzolino, D. (2004). Determination of potentially mineralizable nitrogen and
nitrogen in particulate organic matter fractions in soil by visible and near-infrared
reflectance spectroscopy. Journal of Agricultural Science, 142, 335-343, 00218596
Mostbauer, P.; Lenz, S. & Lechner, P. (2008). MSWI bottom ash for upgrading of biogas and
landfill gas. Environmental Technology, 29, 757-764, 09593330
Naumann, D.; Schultz, C.P. & Helm, D. (1996). What can infrared spectroscopy tell us about
the structure and composition of intact bacterial cells?, In: Infrared Spectroscopy of
Biomolecules, H. H. Mantsch and D. Chapman (Ed.), 279-310, Wiley, 0471021849,
New York
Ouatmane, A.; Provenzano, M.R.; Hafidi, M. & Senesi, N. (2000). Compost maturity
assessment using calorimetry, spectroscopy and chemical analysis. Compost Science
& Utilization, 8, 124-134, 1065657X
Polettini, A. & Pomi, R. (2004). The leaching behavior of incinerator bottom ash as affected
by accelerated ageing. Journal of Hazardous Materials, 113, 209-215, 03043894
Pollanen, K.; Hakkinen, A.; Reinikainen, S.P.; Rantanen, J.; Karjalainen, M.; Louhi-Kultanen,
M. & Nystrom, L. (2005). IR spectroscopy together with multivariate data analysis
as a process analytical tool for in-line monitoring of crystallization process and
www.intechopen.com
solid-state analysis of crystalline product. Journal of Pharmaceutical and Biomedical

Analysis, 38, 275-284, 07317085
Reeves, J.B. & Van Kessel, J.S. (2000a). Determination of ammonium-N, moisture, total C
and total N in dairy manures using a near infrared fibre-optic spectrometer. Journal
of Near Infrared Spectroscopy, 8, 151-160, 09670335
Reeves, J.B. & Van Kessel, J.S. (2000b). Near-Infrared spectroscopic determination of carbon,
total nitrogen, and ammonium-N in dairy manures. Journal of Dairy Science, 83,
1829-1836, 00220302
Reig, F.B.; Adelantado, J.V.G. & Moreno, M.C.M.M. (2002). FTIR quantitative analysis of
calcium carbonate (calcite) and silica (quartz) mixtures using the constant ratio
method. Application to geological samples. Talanta, 58, 811-821, 00399140
Rendek, E.; Ducom, G. & Germain, P. (2006). Carbon dioxide sequestration in municipal
solid waste incinerator (MSWI) bottom ash Journal of Hazardous Materials, 128, 73-
79, 03043894
Smidt, E.; Lechner, P.; Schwanninger, M.; Haberhauer, G. & Gerzabek, M.H. (2002).
Characterization of waste organic matter by FT-IR spectroscopy: Application in
waste science. Applied Spectroscopy, 56, 1170-1175, 00037028
Smidt, E. & Meissl, K. (2007). The applicability of Fourier Transform Infrared (FT-IR)
spectroscopy in waste management. Waste Management 27, 268-276, 0956053X
Smidt, E.; Meissl, K.; Schmutzer, M. & Hinterstoisser, B. (2008a). Co-composting of lignin to
build up humic substances-Strategies in waste management to improve compost
quality. Industrial Crops and Products, 27, 196-201, 09266690
Smidt, E.; Meissl, K.; Schwanninger, M. & Lechner, P. (2008b). Classification of waste
materials using Fourier transform infrared spectroscopy and soft independent
modeling of class analogy. Waste Management, 28, 1699-1710, 0956053X
Smidt, E.; Meissl, K. & Tintner, J. (2007). Investigation of 15-year-old municipal solid waste
deposit profiles by means of FTIR spectroscopy and thermal analysis. Journal of
Environmental Monitoring, 9, 1387-1393, 14640325
Smidt, E.; Meissl, K.; Tintner, J.; Binner, E. & Lechner, P. (2008c). Influence of composts and
digestates on soil properties - Retrospect and outlook. Proceedings of CODIS 2008 -
Compost and digestate: sustainability, benefits and impacts for the environment and plant
production, pp. 299, 9783037360163, CH-Solothurn, Research Institute of Organic
Agriculture FIBL, CH-Frick
Smidt, E.; Meissl, K.; Tintner, J. & Ottner, F. (2009). Interferences of carbonate quantification
in municipal solid waste incinerator bottom ash: evaluation of different methods.
Environmental Chemistry Letters, 1-6, 16103653
Smidt, E. & Schwanninger, M. (2005). Characterization of Waste Materials Using FT-IR
Spectroscopy – Process Monitoring and Quality Assessment. Spectroscopy Letters,
38, 247-270, 00387010
Smidt, E.; Tintner, J.; Meissl, K. & Binner, E. (2008d). Influence of input materials and
composting operation on humification of organic matter. Dynamic Soil, Dynamic
Plant, 2 (Special Issue 1), 50-59, 17496500
Smith, B. (1999). Infrared Spectral Interpretation. CRC Press, Boca Raton, 0849324637, London,
New York, Washington; DC
Socrates, G. (2001). Infrared and Raman Characteristic Group Frequencies. Tables and Charts. John
Wiley & Sons Ltd., 0471852988, Chichester
www.intechopen.com
Stewart, D.; Wilson, H.M.; Hendra, P.J. & Morrison, I.M. (1995). Fourier-transform infrared
and Raman spectroscopic study of biochemical and chemical treatments of oak
wood (Quercus rubra) and Barley (Hordeum vulgare) straw. Journal of Agricultural
and Food Chemistry, 43, 2219-2225, 00218561
Tan, K.H. (2003). Humic Matter in Soil and the Environment – Principles and Controversies.
Marcel Dekker, 0824742729, New York Basel
Tesar, M.; Prantl, R. & Lechner, P. (2007). Application of FT-IR for assessment of the
biological stability of landfilled municipal solid waste (MSW) during in situ
aeration. Journal of Environmental Monitoring, 9, 110-118, 14640325
Tintner, J.; Smidt, E.; Böhm, K. & Binner, E. (2010). Investigations of biological processes in
Austrian MBT plants. Waste Management, 30, 1903-1907, 0956053X
Van Den Broek, W.H.A.M.; Wienke, D.; Melssen, W.J. & Buydens, L.M.C. (1998). Plastic
material identification with spectroscopic near infrared imaging and artificial
neural networks. Analytica Chimica Acta, 361, 161-176, 00032670
van Praagh, M.; Heerenklage, J.; Smidt, E.; Modin, H.; Stegmann, R. & Persson, K.M. (2009).
Potential emissions from two mechanically-biologically pretreated (MBT) wastes.
Waste Management, 29, 859-868, 0956053X
Weber, K.; Lamp, T.; Weidemann, J.; Haren, G.V.; Eisenmann, T.; Mosebach, H.; Gärtner, A.
& Bröker, G. (1996). FTIR-langwegabsorptionsspektroskopie zur Messung von
Luftverunreinigungen: Grundlagen und ausgewählte Beispiele von Messungen bei
Industrieanlagen, im urbanen Umfeld und bei Deponien, In: VDI Berichte, 205-255,
00835560
Williams, P. & Norris, K. (2004). Near-Infrared technology in the agricultural and food industries.
American Association of Cereal Chemists, Inc., 1891127241, St. Paul
Zaccheo, P.; Ricca, G. & Crippa, L. (2002). Organic matter characterization of composts from
different feedstocks. Compost Science & Utilization, 10, 29-38, 1065657X
Zhang, A.D.; Zeng, W.X.; Niemczyk, T.M.; Keenan, M.R. & Haaland, D.M. (2005).
Multivariate analysis of infrared spectra for monitoring and understanding the
kinetics and mechanisms of adsorption processes. Applied Spectroscopy, 59, 47-55,
00037028
www.intechopen.com
20
Open-Path FTIR Detection of

Explosives on Metallic Surfaces
John R. Castro-Suarez1, Leonardo C. Pacheco-Londoño1,
Miguel Vélez-Reyes2, Max Diem3 and Thomas J. Tague, Jr.4
and Samuel P. Hernandez-Rivera1
ALERT DHS Center of Excellence for Explosives
Center for Chemical Sensors Development
1Department of Chemistry, University of Puerto Rico, Mayagüez, PR
2Department of Electrical Engineering, University of Puerto Rico, Mayagüez, PR
3Department of Chemistry, Northeastern University, Boston, MA
4Bruker Optics, Billerica, MA 01821
1,2Puerto Rico
3,4U.S.A
1. Introduction
Defense and security agencies are in constant demand of new ways of detecting chemical
and biological threats posed by terrorist organizations. Fundamental and applied research
in areas of interest to national defense and security focus in detection of highly energetic
materials (HEM), including homemade explosives (HME) that could be used as weapons of
mass destruction (Marshal and Oxley, 2009; Yinon and Zitrin, 1996; Schubert and Rimski-
Korsakov, 2006). The Department of Homeland Security of the United States of America has
even gone a step further and established a university based Center of Excellence in explosives
detection, mitigation and response to conduct transformational research, technology and
educational development for effective characterization, detection, mitigation and response to
the explosives-related threats facing the country and the world. The Awareness and
Localization of Explosives Threats (AWARE) is co-lead by Northeastern University (ALERT at
NU, Boston, MA) and University of Rhode Island (ALERT at URI, Kingston, RI).
Current detection methods of explosives are based on a wide variety of technologies that
focus on either bulk explosives or traces of explosives. Bulk explosives can be detected
indirectly by imaging characteristic shapes of the explosive charge, detonators, and wires or
directly by detecting the chemical composition or dielectric properties of the explosive
material. Trace detection methods rely on detection of vapors emitted from the explosives or
on explosive particles that are deposited on nearby surfaces (National Academy of Sciences
Committee, 2004). Although there are hundreds of publications about methods of detection
of HEM in water, soil, air, clothing, surfaces, etc. and these offer the advantage of providing
very low limits of detection at ppb levels (Caron, et al., 2010; Hilmi and Luong, 2000; Yinon,
1996; Szakal and Brewer, 2009; Miller and Yoder, 2010). They require, in the majority of the
cases, sampling at the scene followed by a sample preparation step, to be later analyzed by a
www.intechopen.com
particular technique. Sampling and sample preparation are among the main disadvantages
in HEM detection, in many cases threatening the health and life of analysts and first
responders. Vibrational spectroscopy, in its various modalities, has shown to be useful for
detection of dangerous chemicals, among them HEM. Near-infrared or mid-infrared
spectroscopies have shown to be powerful techniques for IR vibrational analysis, able to
detect organic and inorganic substances in any state: solid, liquid or gas (Gunzler, 2002;
Smith, 2000). IR vibrational spectra can to be used for identify and quantify samples in
complex matrices because each substance has its own fingerprint spectrum in the mid IR
(MIR). This means that IR spectroscopy can be used for discriminant analysis even when the
target analyte is in very small quantities (Bangalore, 1997).
Standoff detection is defined as where equipment and operator stay away from the sample
while measuring some property of the target (Parmenter, 2004). An area of IR spectroscopy
that has increased interest in defense and security applications is standoff IR (SOIR)
spectroscopy. In SOIR detection, vibrational signatures can to be recorded from several
meters to hundreds of meters in distances between the target and the instrument. Fourier
transform infrared (FTIR) standoff detection provides a means of doing real time analysis, in
which no sample preparation is needed, no human contact needs to be provided,
measurements are typically fast, and chemical information for each explosive can be
obtained in high detail which can allow identification and even quantification. This makes
the standoff IR detection a powerful technique for sensing of energetic materials at a
distance, thus preventing or minimizing the damage caused by terrorist action, in the case
that this comes to be detonated.
Open-Path Fourier Transform IR (OP/FTIR) spectroscopy has been used for atmospheric
gas analysis and environmental monitoring for over 40 years (Griffiths, et al., 2009). It is one
of the two methodologies devised for measuring concentration of gaseous trace components
in the atmosphere using infrared spectroscopy: extractive sampling analysis and in situ
open-path analysis. Although Russwurm and Childers credit Hanst for the initial
description of FTIR monitoring of atmospheric pollutants in the atmosphere by OP IR
(Russwurm and Childers, 2001; Hanst, 1971) Stephens and his group at the Environmental
Protection Agency had already made measurements of ambient concentrations of peroxy
acetyl nitrate (PAN) in the Los Angeles city basin before 1969 (Stephens, et al., 1969; Scott, et
al., 1957). Aside from the apparently inconsequent controversy (since Hanst was part of the
Stephen’s group) valid questions on why is OP/FTIR is rarely used nowadays and why its
development has been undermined with technological problems remain unanswered.
Among possible answers to these questions stands out a limited sensitivity of the technique
for atmospheric monitoring: 1-100 ppb by volume (contrasted to the requirements of parts
per trillion by volume on many pollutants) and the lack of development of algorithms that
can be incorporated into the instruments acquisition and analysis routines (software) that
can could make the use of the technique a more amenable and user friendly one. In a recent
article by Griffiths and collaborators, the authors point out the difficulties encountering
when using OP/FTIR that have led to a slow development of the remote sensing modality
(Griffiths, et al., 2009). The clear advantages of wide area sensing and long range capabilities
have been overshadowed by hardware and more so, by software limitations. Inadequacy in
compensating for variable atmospheric contributions, mainly by water vapor and carbon
dioxide has hampered the wide usage of OP/FTIR both in environmental studies as well as
in Defense and Security applications. In this study, two types of FTIR standoff detection
experiments were carried out: active mode OP/FTIR and passive mode OP/FTIR. The
www.intechopen.com
Open-Path FTIR Detection of Explosives on Metallic Surfaces 433
detection of particle dispersion is also relevant to the forensic community. When an airbag
deploys in an accident the lubricant is dispersed on the passenger. Many times the
occupants in the vehicle are ejected or end up in a different location in the vehicle. Such a
device could definitively place the occupants. Additionally, persons involved in the illicit
manufacture of drugs will frequently retain chemical residues on their clothing. In this case,
remote detection could be used to link a suspect to the crime scene.
There is a limited number of scientific contributions in the area of SOIR detection of HEM
deposited on surfaces. Work by Theriault and colleagues (Theriault, et al., 2004), Van Neste
and collaborators (Van Neste, et al., 2009), Blake and co-workers (Blake, et al., 2009) and
Pacheco-Londoño and colleagues (Pacheco-Londoño, et al. 2009) have helped to contribute
the development of this application of OP/FTIR. Theriault and collaborators made field
measurements of liquid contaminants deposited on a number of surfaces at a standoff
distance of 60 m using FTIR radiometry (Theriault, et al., 2004). Van Neste and collaborators
described standoff detection measurements of trace quantities of surface adsorbed high
explosives (Van Neste, et al., 2009). They used two quantum cascade lasers (QCL) operated
simultaneously in the MIR, with tunable wavelength windows that match with absorption
peaks of nitroexplosives tested. In this important contribution researchers demonstrated a
sensitivity of 100 ng/cm2 and a standoff detection distance of 20 m for surface adsorbed
analytes such as explosives and chemical agent simulant tributyl phosphate. The detection
of Explosives on metallic substrates is the first step in demonstrating the facility of passive
and active open path FTIR detection for general use. Other substrates such as textiles,
plastic, wood, and glass are less reflective and present a greater challenge. The emergence
of alternative bright sources, such as QCL’s, puts the active detection of explosive residues
on real life materials over significant pathlengths within reach. Blake and colleagues
recorded hyperspectral images of galvanized steel plates, containing
cyclotrimethylenetrinitramine (RDX), using a commercial long-wave infrared imaging
spectrometer at a standoff range between 14 and 50 m (Blake, et al., 2009). Pacheco-
Londoño and collaborators built an active IR standoff detection system by coupling a bench
FTIR interferometer to a gold mirror and external cryocooled detector assembly for
detection of explosives present as traces on reflective surfaces (Pacheco-Londoño, et al.
2009). Source-target distances in the range of 1 – 3.7 m were studied and limits of detection
(LOD) values obtained were 18 and 20 g/cm2 for TNT and RDX, respectively. The results
of the prototype built were attributed to the use of a modulated MIR beam source that was
able to cut down stray light from laboratory illumination.
2. Description of methodology
Open Path Infrared Spectroscopy is a well established technique for atmospheric sensing of
gases and condensable vapors. In the current application, after validating the spectroscopic
system in detection of gases and condensable vapors, are more challenging application was
addressed: detection solid samples deposited as trace contaminants on metallic surfaces
were detected by OP/FTIR. Sample preparation is a critical task in the development of any
analytical technique. Three steps were performed for standoff detection of explosives and
•
other highly energetic materials deposited on Al plates:
•
TNT samples were weighted and dissolved in dichloromethane.
Mixtures were deposited on the Al plates using a Teflon stub and were then allowed to
air-dry.
www.intechopen.com
• OP/FTIR standoff detection experiments were carried out, both in active mode using a
mid infrared (MIR) source and passive mode using a thermal excitation (utilizing a
tungsten lamp).
These steps are illustrated in Fig. 1.
Background spectra of Al plates with no TNT deposited on them were run for every
standoff distance tested in active mode. In the case of passive mode standoff detection,
background spectra were acquired at every range value and every plate temperature tested.
Then, statistical routines were applied using chemometrics. In particular, partial least
squares (PLS) regression analysis was used to perform quantification studies of HEM
surface loadings at all standoff distances. Standoff detection of solid samples present as
traces on metallic substrates required a sample preparation methodology that would be able
to deposit solid samples on a solid substrate, with high coverage uniformity and
reproducibility. Due to the size of the substrates, sample smearing technique was used to
shown in Fig. 2a, aluminum plates of areas 30.5 cm × 30.5 cm were used as material support
deposit the HEM TNT at trace amounts on metallic substrates (Primera-Pedrozo, 2008). As
for HEM samples. Dichloromethane was used to clean the aluminum surfaces. Plates were
dichloromethane was used to dissolve TNT sample to be deposited. A 3 cm × 15 cm Teflon

allowed to air-dry before of depositing the desired HEM surface loading. A small amount of
stub was used to smear the HEM sample on the aluminum substrates (Fig. 2b). The amount
of HEM that remained on the Teflon stub after sample smearing was negligible. The
nominal surface concentrations obtained by the smearing technique were 50, 100, 200, 300,
and 400 µg/cm2 of HEM. Figs. 2b and 2c show how TNT was deposited on aluminum plates
for 50 (Fig. 1b) and 100 g/cm2 (Fig. 1c), respectively.
Fig. 1. Steps for remote detection of nitroexplosives and other HEM deposited on surfaces
using OP/FTIR
www.intechopen.com
a) b) c)
Fig. 2. Sample preparation: (a) Clean Al plate ready for sample smearing technique; (b) 50
µg/cm2 of TNT loading concentration; (c) 100 µg/cm2 of TNT surface concentration
A model EM27 (Bruker Optics, Billerica, MA) OP/FTIR spectrometer, was used to obtain the
MIR spectral information of TNT samples. Table 1 contains the specifications and technical
data of the EM27. Fig. 3 illustrates the difference in operation between a benchtop FTIR
spectrometer used in absorption mode (Fig. 3a) and an IR interferometer configured for
open-path measurements (Fig. 3b). The optical bench consisted of a compact, enclosed, and
desiccated Michelson type interferometer equipped with ZnSe windows, internal black
body calibration source, KBr beamsplitter, f/0.9 and a field of view (FOV) of 30 mrad (1.7°).
Its main features are: permanently aligned, vibration insensitive, and friction-free
mechanical bearing. The system was capable of acquiring 32 spectra per second at 1 cm-1
resolution.
PARAMETER SPECIFICATION
700 – 1300 cm-1 (useful for passive measurements)

Spectral Range:
700 – 4000 cm-1 (useful for active measurements)
Resolution: 1.0 cm-1 (Option: 0.5 cm-1)
NEDT:
up to 0.08°C for one scan with a resolution of 1 cm-1 and
(Noise Equivalent Delta
a mirror speed of 40 kHz, depending on the detector
Temperature):
Optical Bench: compact, enclosed, desiccated, purge interferometer
Beamsplitter: KBr-substrate with multi-layer coating (Option: ZnSe)

vibration insensitive, friction-free mechanical bearing;
Interferometer: permanently aligned; symmetrical interferogram
acquisition at 4 scanning velocities up to 40 kHz
Standard Detector: MCT-detector (narrow band), liquid nitrogen cooled,
Field of View: 30 mrad (10 mrad with receiver telescope)
Table 1. Specifications of EM27 Open Path FTIR spectrometer
www.intechopen.com
a
Interferometer Sample
Source Detector
b
Sample Interferometer
Source Detector
Fig. 3. FTIR interferometer configured as: (a) absorption spectrometer: detects source
radiation only; (b) open-path spectrometer: detects source and sample radiation
High sensitivity measurements were achieved by using a high sensitivity closed cycle cryo-
cooled photoconductive mercury cadmium telluride (MCT) MIR detector. Both, the MIR
source and interferometer were focused on the target by MIR telescopes for active mode
measurements (Fig. 3). Telescopic sights with mounts were used to align both source and
FOV ≥ 7.5 mrad; the receiver IR telescope was also a 6 in. diam. F/3 gold coated reflector
collector. The transmitter source telescope was a 6 in. diam. F/4, gold coated mirrors with
with a FOV of 10 mrad. For passive mode measurements, the experimental setup was the
same that in active mode, but the telescope coupled MIR source for excitation of the target
was not used. Instead, a 500 W tungsten lamp was used to heat the aluminum plates with
and without target analytes deposited on them. For both active mode and passive mode,
multivariate calibrations were obtained and relevant statistical parameters were calculated
and used as criteria to judge the quality of the method.
3. Application of multivariate calibrations to OP/FTIR data analysis

The mathematical modeling and detection of explosives and other threats in a complex
environment can be complex. The large spectral bandwidth of a FTIR spectrometer
facilitates the analysis. The purpose of calibration techniques is to correlate measured
quantities such as the absorption of infrared radiation with properties of the system, for
example, the concentration of one component in a multicomponent system. The accepted
method of data analysis of gas phase contaminants present in a complex multicomponent
mixtures as is the case of atmospheric pollutants present in air is classical least squares
(CLS) regression analysis, also termed linear regression analysis or least squares (Russwurm
and Childers, 1999). For quantification studies, CLS calibration curves can be generated
using two methods: measurement of the absorbance peak heights and integration of areas in
the spectral region of interest. Calibration plots using peak areas represents a better choice
www.intechopen.com
for quantitative analysis when compared to peak height analysis (Lavine, 2002; Kramer,
1998). Usually, two steps are required: the calibration of the method and the analysis to
determine a value of an unknown sample. It is important to emphasize that measuring
surface concentrations using the peak area method is conceptually simple and easy to use,
but it has some limitations. The method is univariate (the concentration is determined
with a single spectral peak) and depends on a linear correlation between the concentration
and the spectral response. The results can, therefore, be undermined by perturbations
such as fluctuations caused by detector noise, temperature variations, or molecular
interactions.
In a series of recent important papers by Griffiths and collaborators, the authors have
demonstrated three important and novel aspects of data analysis for open path FTIR
detection at a distance (Hart and Griffiths, 1998; Hart, et al., 200; Griffiths, et al. 2009; Shao,
•
et al. 2010). Specific contributions can be summarized as follows:
Establishment that multivariate data analysis techniques are required to exploit all the
benefits that having a wealth of spectral information immersed in a congested
multicomponent spectrum as that contained in SOIR experiments, not only for gas
phase measurements but also for solid phase OP/FTIR detection at a distance, as Castro
•
and collaborators have recently demonstrated (Castro, et al., 2010).
Demonstration that a single background spectrum measured at a fixed source-target
distance, temperature, pressure and ambient gases partial pressures can be used for all
•
ranges and combinations of other relevant variables.
Demonstration that the representation of OP/FTIR spectra in absorbance or percent
transmission is equivalent. The output of any FTIR spectrometer is a single-beam
spectrum that must be ratioed or compared (subtracted) against a appropriate
background spectrum resulting in the transmittance spectrum of the sample, T( ). In
quantitative measurements, the transmittance should be converted to absorbance,
A(ν), i.e. −log10{(T(ν)} since the absorbance is a linear function of the concentration of
the species absorbing, thus rendering it more amenable to use of chemometrics
routines of analysis. For the current application: OP/FTIR detection of solid threat
chemicals deposited on surfaces as traces, representation as the spectral difference
between the sample spectrum and the background spectrum is even more critical due
to the possibility of specular reflectance and scattering from the sample (Castro, et al.,
2010).
Multivariate calibrations make use of not only a single spectral point but take into account
spectral features over a wide range. Therefore, the analysis of overlapping spectral bands or
broad peaks becomes feasible. The information contained in the spectra of the calibration
samples will be compared to the information of the concentration values using a PLS
regression. The method assumes that systematic variations observed in the spectra are a
consequence of the concentration change of the components. However, the correlation
between the components concentration and the change in the infrared signal does not have
to be a linear one.
Calibrations are typically constructed using chemometrics methods of data analysis such
as the partial least squares (PLS) regression algorithm. The PLS algorithm is commonly
incorporated in spectroscopic software such as OPUS™, Pirouette™ and Matlab
Toolboxes™, among others. The advantages of using chemometrics for the quantification
of organic compounds on glass, aluminum and stainless steel and other surfaces have
www.intechopen.com
been discussed in the literature (Mehta, et al., 2003; Hamilton, et al., 2005; Perston, et al.,
2007; Primera-Pedrozo, et al., 2004; Primera-Pedrozo, et al., 2005; Primera-Pedrozo, et.al.
2007; Soto-Feliciano, et al., 2006; Primera-Pedrozo, et al., 2008; Primera-Pedrozo, O. M.;
2009).
PLS is a multivariate method that uses spectral features over a wide spectroscopic range. It
is a spectral decomposition method that is intimately related to principal component
analysis (PCA). In PCA, the spectral matrix is first decomposed into a set of eigenvectors
and scores, and then a regression is performed against the concentrations as a separate step.
However, PLS uses the concentration information during the decomposition process. In the
case of OPUS™ (Bruker Optics), Quant2 software is used to find the best correlation
function between the spectral information and the loading concentrations. Quant2 uses a
partial least squares-1 (PLS-1) regression method. Calibrations are performed using PLS-1 in
which only one component can be analyzed separately, instead of simultaneously analyzing
multiple components, as in the PLS-2 routine of chemometrics. Then, cross validations are
performed and the root mean square errors of the cross validations (RMSECV) and the root
mean square errors of estimations (RMSEE) are used as criteria to evaluate the quality of the
correlations obtained. In the standard ‘‘leave-out-one’’ cross validations, each spectrum is
omitted from the training set and then tested against the model built with the remaining
spectra. As illustrated in Tables 1 and 2, some explosives require spectroscopic
preprocessing (except mean centering) and more PLS evaluations in order to obtain a good
model. In the case of the first derivative, the Savitzky-Golay algorithm is actually used to
obtain the derivative. The number of smoothing points used can be adjusted to suppress the
effect of noise (Beebe, 1998). Other details of the advantages of using a chemometrics model
such as PLS to correlate the loading concentration with IR spectra have been discussed in
the literature (Hamilton, et al., 2005).
Multivariate calibrations require a large number of calibration samples and yield a large
amount of data. In order to conveniently handle the data, the spectral information and the
concentration information are written in the form of matrices, where each row in the
spectral data matrix represents a sample spectrum. The concentration data matrix contains
the corresponding concentration values of the samples. The matrices are then broken down
into their eigenvectors which are called factors or principal components. Only the relevant
principal components are used instead of the original spectral data, thus leading to a
considerable reduction of the amount of data. A PLS regression algorithm will be developed
to find the best correlation function between spectral and concentration data matrix
(OPUS™, Bruker Optik, 2006). The OPUS/QUANT software package (OPUS™, Bruker
Optik, 2006) is designed for quantitative analysis of spectra consisting of bands showing
considerable overlap. The software allows determining the concentration of more than one
component in each sample simultaneously. For this purpose, QUANT uses a partial least
squares (PLS) regression method.
The residual (Res) is the difference between the true and the fitted value. Thus the sum of
squared errors (SSE) is the quadratic summation of these values:
SSE = ∑ ⎡⎣Resi ⎤⎦
2
(1)
The coefficient of determination (R²) gives the percentage of variance present in the true
component values, which is reproduced in the regression. R² approaches 100% as the fitted
concentration values approach the true values:
www.intechopen.com
⎛ ⎞
R2 = ⎜ 1 − ⎟ × 100
∑ ( i − y m )
SSE
⎜ ⎟
(2)
⎝ ⎠
2
y
In the case of a cross validation, the root mean square error of cross validation (RMSECV)
can be taken as a criterion to judge the quality of the method:
RMSECV = (
× ∑ y i − y( i )
1 M
)
2
(3)
M i= 1
where M is the number of standards in the data set. One method of cross-validation is leave-
one-out cross-validation (LOOCV). Leave-one-out cross-validation is performed by
estimating n calibration models, where each of the n calibration samples is left out one at a
time in turn. The resulting calibration models are then used to estimate the sample left out,
which acts as an independent validation sample and provides an independent prediction of
each yi value, y(i), where the notation i indicates that the ith sample was left out during
model estimation. This process of leaving a sample out is repeated until all of the calibration
samples have been left out. To obtain the root mean square error of prediction (RMSEP), the
validation samples prepared and measured independently from the calibration samples are
used. The number of validation samples, p, should be large, so that the estimated prediction
error accurately reflects all sources of variability in the calibration method. The RMSEP is
computed as:
RMSEP = (
× ∑ y i − y( i )
1 p
)
2
(4)
p i= 1
where p is the number of prediction samples.

Partial least squares (PLS) regression algorithm from Quant2 software of OPUS™ version
6.0 (Bruker Optics) was used to find the best correlation function between the spectral
information and the surface concentration. PLS was used for generating a chemometrics
model for all analyzed standoff distances. Cross validations were made and the root mean
square errors of cross validations (RMSECV) and correlation coefficient (R2) were used as
criteria to judge the quality of the correlations obtained at different standoff distances.
4. OP/FTIR detection of gases and condensable vapors

Griffiths et al. described two general ways in which OP/FTIR can be used for remote
sensing measurements (Griffiths, et al., 2009). When the source and the detector are in line of
sight with each other and they have separate power sources the operational mode is called
bistatic. In this setup, the source is non-modulated by the interferometer and as a result stray
light contributions cannot be minimized. On the other hand, when all the components reside
within the spectrometer, including the MIR source, sharing a common power source, the
operational mode is termed monostatic. This setup has clear advantages as pointed out by
Pacheco-Londoño and collaborators in reducing stray light components, but is limited to
source power and cooling restraints (Pacheco-Londoño, et al., 2009). In the active mode
employed for remote detection, a bistatic setup in which the source is not modulated was
used.
www.intechopen.com
The first task in evaluating the performance of the Bruker Optics EM27 open path
interferometer was to use it in detection of gases and condensable vapors. The spectroscopic
system was configured for bistatic operational mode in active configuration (Fig. 3b) with
source and interferometer in line-of-sight of each other. Measurements were done for
standoff distance of 1-10 m. The spectra of ambient air and ammonia (NH3) at 10 m range at
room temperature are shown in Fig. 4a-d. In the case of NH3, spectra were collected at an
instrumental resolution of 1 cm-1. The presence of ro-vibrational lines in the remote IR
absorption spectrum of ammonia is clearly shown in Figs. 4b and 4c. The intense spectrum
obtained for dichloromethane is illustrated in Fig. 4d. Gas phase standoff IR spectra of some
high vapor pressure liquids are shown in Fig. 5a-d. Spectra are arranged in increasing order
of their room temperature vapor pressure and absorbance of most prominent spectroscopic
features (vibrational signals). The presence of spectral contributions from ambient water
vapor and carbon dioxide ro-vibrational lines can be seen. These persistent lines were not
removed by any of the widely used algorithms since spectral windows for sample
identification were available in all cases.
NH3 0.2
0.185
a Air
b NH3
0.16 Air
Absorbance
Absorbance
0.135
0.12
0.085
0.08
0.035 0.04
-0.015 0
700 1100 1500 1900 2300 2700 3100 3500 3900 700 800 900 1000 1100 1200 1300
Wavenumber / cm-1 Wavenumber / cm-1
0.1
NH3
0.08
c Air d
Absorbance
0.06
0.04
0.02
0
3100 3250 3400 3550 3700 3850
Wavenumber / cm-1
Fig. 4. Active mode OP/FTIR spectra of: (a) air and NH3 complete spectrum; (b), (c) details
of NH3 spectrum; (d) dichloromethane
www.intechopen.com
0.28
METHYL ETHER 0.33
0.23 a b ACETONE
0.28
e 0.18
Absorbance
c 0.23
n
a
b
r 0.13 0.18
o
s 0.13
b 0.08
A
0.08
0.03
0.03
-0.02 -0.02
700 1100 1500 1900 2300 2700 3100 3500 3900 700 1100 1500 1900 2300 2700 3100 3500 3900
Wavenumbers / cm-1 Wavenumber / cm-1
0.04
ISOPROPANOL
0.1 0.035 d
c 0.03
ACETONITRILE
e
c0.06 Absorbance
n 0.025
a
b
r
o 0.02
s0.02
b 0.015
A
-0.02 0.01
0.005
-0.06
0
700 1100 1500 1900 2300 2700 3100 3500 3900
700 1100 1500 1900 2300 2700 3100 3500 3900
Fig. 5. Active mode remote spectra of condensable vapors: (a) methyl ether; (b) acetone, (c)
isopropanol; acetonitrile
The spectral band shapes observed in stand-off detection mode, shown in Figures 4 and 5,
are superimposed on a ramp-shaped background, and the bands themselves exhibit strong
reflective (dispersive) band profiles. Since these measurements were done on a reflective
metal substrate, the distortion of the band profiles is expected; similar effects have been
reported in DRIFT (diffuse reflection infrared Fourier Transform) spectroscopy (Chalmers;
Mackenzie, 1985). and in microscopically acquired infrared spectra of microspheres (Basan,
el at. 2009a). In both cases, the distortion of the absorptive line shapes is due to the fact that
within an absorption peak, the reflective index undergoes anomalous dispersion, as shown
in Figure 6. In spectroscopic experiments carried out in reflectance mode, a mixing of the
absorptive and dispersive line shapes can occur, resulting in bands that have a negative dip
at the high wavenumber side of the peak, cf. Figures 4 and 5. This will shift the peak
maximum by up to 15 cm-1 toward lower values (Basan, el at. 2009b).
For ‘real-life’ stand-off detection, strong reflectance band distortions, such as those shown in
Figures 4 and 5 are not likely, since these are typical for reflectance spectra on metals, and
should be much weaker if explosives are distributed on fabric. However, mixing of
absorptive and reflective line shapes can also be mediated by scattering effects (Basan, et al.
2009a) and could produce significant band distortions.
Unsupervised correction of the spectral distortions will be necessary since the distortions
cause apparent frequency shifts which will confound spectral search and identification
algorithms. Although several methods have been developed to correct the dispersive line
www.intechopen.com
shapes observed in biological systems (Basan, et al. 2009a; Bird, et al. 2010) they are not
applicable here, since they are based on multivariate methods, and require large number of
spectra, as well as undistorted reference spectra. For the spectra reported here, a method
originally published in 2005 (Romeo and Diem, 2005), may be more suitable. This method is
based on phase correction between real and imaginary spectral contributions which can be
obtained by reverse Fourier transform of the contaminated spectra. The original
implementation of this method contained a minor logical error, which since has been
corrected (Bird, et al. 2010).
Wavenumber / cm-1
Fig. 6. Anomalous Dispersion of the refractive index (n) in the vicinity of an absorption band
5. Active standoff IR detection of solids deposited on substrates

Active mode standoff IR measurements of solids, including nitroexplosives and other highly
energetic materials (HEM) smeared on Al plates at various surface loadings were carried out.
Spectra in the fingerprint region were obtained with EM27 spectrometer. In the active mode
used in the current application, a bistatic operation setup in which an IR telescope was used to
steer a MIR source which was not modulated and sat side by side to the reflective IR telescope
samples deposited onto aluminum plates at the highest surface loading used: 400 μg/cm2.
collector, operating in back-reflection mode (Fig. 7). Initial experiments were done with solid
Samples were transferred to the metallic test substrates by dissolving in an appropriate solvent
and then smearing them on the test plates. A Teflon stub was used to assist in sample
smearing. Coated plates were allowed to dry in air at room temperature.
Initial remote IR experiments were designed to optimize experimental considerations,
including sample placement and measurement geometry. For the reflectance IR measurements
at one meter (1 m) standoff distance, as shown in Fig. 8, the angle of incidence of the MIR
radiation (and angle of reflection) was varied from 0° to ~ 30°. TNT signals decreased
drastically at high incident angles, becoming nearly unobservable at 27°. Spectra of selected
www.intechopen.com
standoff distance of 8 m and loading concentration of 400 μg/cm2 was used for the SOIR
solid phase compounds deposited as traces on Al substrates are shown in Fig. 9. A fixed
measurements. The first spectrum of Fig. 9 is that of caffeine deposited on Al plate. Fig. 9b and
9c show the spectra of p-benzoic acid and benzoic acid, respectively at 25°C. For the
acquisition of the data shown in Fig. 9d the sample plate was heated to 28°C to demonstrate
how the emission of vibrational quanta is significantly enhanced by small temperature
differences. The remote IR spectrum of aliphatic nitrate ester PETN is shown in Fig. 9e and the
corresponding spectra aliphatic nitramine RDX are shown as %T and absorbance in Fig. 9f.
Pacheco and co-workers used a modulated home built setup for remote IR measurements of
nitroexplosives from ~ 1 m to ~ 4 m range. At short distances (0.9 m and 1.8 m) the
maximum and minimum signal to noise (SNR) values showed high dispersion. They found
out that with their uncollimated MIR beam, the problem both was sample and transfer
solvent dependent in the low to very low loading concentrations studied. They argued that
part of the problem was the lack of uniformity in surface coverage due to nucleation and
crystallization phenomena. When the IR beam used was 1-2 cm in diameter or less, sample
discontinuities could be detected and this was reflected by the relatively high dispersion in
the values of SNR. At longer target-collector distances (1.8, 2.7 and 3.7 m) the maximum and
minimum SNR values were very close due to higher sample coverage by a beam spot of ~ 5
determined was 2 μg/cm2 for TNT. According to the authors, LOD values determined could
- 11 cm in diameter. At a source-target distance of 0.9 m, limit of detection (LOD) value
have been influenced by several factors, such as: humidity, alignment of detector, source
pointing accuracy, detector efficiency and reflectivity of samples and substrates.
Al plates
Range
MIR source Spectrometer
Fig. 7. Experimental setup used for SOIR detection. In active mode operation, both MIR
source and FT interferometer were coupled to gold coated mirrors MIR reflective telescopes.
In passive mode only the IR spectrometer was used
The reflectivity of sample: substrate and analyte, is mainly determined by how the analyte is
deposited and by the solvent used for deposition. If the explosive exists on the surface as a
thin layer, the backscattering signal is low but the reflection-absorption infrared (RAIRS)
www.intechopen.com
spectra measured is of high signal-to-noise (SNR) value. When the explosive was present on
the surface as discrete particles (crystals), the backscattered SIRS signal improved;
correspondingly, the RAIRS signal measured for surface loading validation got worse.
Sample smearing was done twice using the Teflon applicator: from left to right and then
right to left. When the first pass was done the solvent was allowed to evaporate. Then, the
second pass was carried out to induce more sample roughness and particle formation
(crystallization) on the surface. TNT deposits on the substrate did not result in a thin layer
covering the metallic surface. Instead droplets of a metastable phase were formed on the
surface (Manrique-Bastidas, et al., 2004; Vrcelj, et al., 2001; Manrique-Bastidas, et al. (2)
2004). This metastable phase could be easily turned to its crystalline phase by friction,
abrasion or even by pressing hard with the Teflon applicator for the sample smearing stage.
This effect enhanced the SOIR experiments of TNT because the metastable phase was
formed in the first smearing step. When the methanol evaporated and the second smearing
TNT than for RDX. Standoff IR spectra of 400 μg/cm2 TNT at 1 m and 14.5 m are shown in
stage was performed, crystalline roughness was induced resulting in a lower LOD value for
Fig. 10. A reference spectrum of neat, microcrystalline sample of TNT (1 mg/100 mg KBr)
obtained in the macro sample chamber of a benchtop interferometer (Bruker Optics IFS-
66/v) is included for comparison purposes.
0
b
0˚ 12˚
-0.025
-0.02
-0.04
-0.075
Absorbance
Absorbance
-0.06
-0.125
-0.08
12˚
0˚
-0.1
-0.175
-0.12
-0.225 -0.14
700 1000 1300 1600 1900 2200 2500 700 1000 1300 1600 1900 2200 2500
c d 27˚
-0.105 17˚ -0.18
-0.115 -0.2
-0.22
-0.125
Absorbance
Absorbance
-0.24
-0.135
-0.26
-0.145
17˚ -0.28
-0.155 -0.3 27˚
-0.165 -0.32
-0.34
-0.175
700 1000 1300 1600 1900 2200 2500
700 1000 1300 1600 1900 2200 2500
Wavenumber / cm-1
Wavenumber / cm-1
Fig. 8. Active mode OP/FTIR spectra of 400 μg/cm2 TNT deposited on Al plate at: (a) 0°; (b)
12°; (c) 17°; (d) 27°. Data shown demonstrates the specular reflectance nature of the IR
reflection-absorption (transflection) experiment
www.intechopen.com
0.6
CAFFEINE 0.6
0.55 p-NITROBENZOIC ACID
0.55
0.5 a 0.5
0.45 b
Intensity
Intensity
0.45
0.4
0.4
0.35
0.35
0.3
0.3
0.25
0.25
0.2
0.2
700 800 900 1000 1100 1200 1300
700 800 900 1000 1100 1200 1300 1400
0
0.65 BENZOIC ACID HEATED BENZOIC ACID
0.6 -0.01
0.55 -0.02
0.5 c
Intensity
-0.03
Inensity
0.45
-0.04
d
0.4
0.35 -0.05
0.3 -0.06
0.25
-0.07
0.2
700 800 900 1000 1100 1200 1300 1400 -0.08
700 800 900 1000 1100 1200 1300 1400
0.25 1.07
1.1
0.2 0.97
RDX ABS f 1
RDX %T 0.9
0.87
Absorbance
0.8
0.15
Intensity
0.77 0.7
e
%T
0.1 0.6
PETN 0.67
0.5
0.57
0.05 0.4
0.47
0.3
0
0.37 0.2
700 800 900 1000 1100 1200 1300 1400 700 800 900 1000 1100 1200 1300 1400
-1
Wavenumber / cm Wavenumber / cm-1
temperature (25 °C); (d) benzoic acid heated to 28 °C; (e) aliphatic nitrate ester PETN; (f)
Fig. 9. Remote IR spectra of: (a) caffeine; (b) p-nitrobenzoic acid, (c) benzoic acid at room
nitramine RDX in absorption and in %T

Other remote IR detection experiments were done using only TNT as target. Loading
concentrations ranging from 50 to 400 µg/cm2 of TNT were deposited on Al plates. The
targets were carefully aligned to the source and collector and then the SOIR spectra were
recorded. The analyzed target-collector distances were 4, 8, 12, 16, 20, 25, and 30 m. A total
of 10 spectra were taken for each sample, at 20 scans and 4 cm-1 resolution. Experiments
were carried out at room temperature (25°C). Spectra were collected in remote, bistatic
active mode detection IR at various surface concentrations at a fixed standoff distance of 8
m. Typical results are shown in Fig.11. These traces were not submitted to any pre-
processing routine: offset correction, baseline correction, smoothing, water vapor rotational
lines removal, etc. Thus, there is no common baseline for these spectra and some traces
exhibit positive intensity ramps to higher wavenumber. However, increase of signal
intensity as function of loading surface concentrations is clearly shown without the use of
www.intechopen.com
chemometrics routines. Intense vibrational band about 908 cm-1 was tentatively assigned to
C-N stretching, vibrational band at 938 cm-1 was assigned to C–H out-of plane bend (ring)
and symmetric stretch band of the nitro groups appears at 1350 cm-1. Results agree with
reported values (Pacheco-Londoño, et al., 2009; Clarkson, et al., 2003).
Fig. 10. Comparison of IR spectra of TNT: (a) KBr pellet with 1 mg TNT obtained in a macro
sample compartment of a FTIR; (b) active mode SOIR spectrum of 400 µg/cm2 deposited on
Al plate measured at 1 m range distance; (c) same sample before measured at a source-target
distance of 14.5 m. Prominent spectral features are present
When a MIR source was used for carrying out active mode experiments, the intensity of the
peaks decreased when the distances increase. This is illustrated in Fig. 11a for spectra of Al
plate coated with surface loading of 400 µg/cm2 TNT and measured at standoff distances of
4, 8, 12, 16, 20, 25 and 30 m. At standoff distances higher than 25 m it was not possible to
visualize clearly some of TNT vibrational signatures. At these distance the density of
infrared radiation that gets to the Al plates from the MIR source is low, leading to a smaller
number of excited molecules, so that the detector cannot register the low intensity signals
emitted. Fig. 11b shows SOIR spectra of TNT as function of loading concentration. Spectra
were collected in active mode detection standoff IR at a fixed standoff distance of 8 m.
The statistical treatments with chemometrics using PLS were carried out using the spectral
region 700 to 1400 cm-1, where the nitro symmetric stretch and aromatic C-H vibrations are
present. Data pre-processing is an important stage in performing a calibration. Thus, the
PLS models were built using mean centering as only pre-processing of variables. To ensure
the reproducibility of the calibration samples, several spectra of each sample (fixed loading
concentration and standoff distance) had to be acquired. If spectra of the same sample are
not identical, a data pre-processing procedure must be chosen to bring them in line with
each other. Data pre-processing can eliminate variations in offset or different linear
baselines. Different treatments data were used, including: vector normalization, first
derivative and second derivative, mean-centering, but no other pre-processing routine
was applied, achieving best results for RMSECV and R2. Results indicate that the
experimental setup has good management of external variables, such as humidity,
m and surface loadings of 50 μg/cm2.

temperature changes, homogeneity of samples on Al plates, and others at ranges as far as 30
www.intechopen.com
Fig. 11. Active mode FTIR spectra of TNT: (a) at different standoff distances of Al plate
coated with surface concentration of 400 µg/cm2; (b) at several loading concentrations at
fixed range of 8 m
www.intechopen.com
Fig. 12. Predicted loadings vs. true loadings for TNT on Al plates at various ranges using
Opus 6.0™ Quant2: (a) 20 m; (b) 25 m; (c) 30 m
www.intechopen.com
Fig. 12 illustrates the results obtained of the cross validation at the analyzed standoff
distances: 20, 25 and 30 m and Table 2 summarizes the results of RMSECV and R2 obtained
in the PLS models. In these correlation charts, each point represents ten spectra with a fixed
surface concentration (0, 50, 100, 200, 300, or 400 µg/cm2). All PLS correlation charts of
predicted surface loading value vs. true surface concentration value for 4, 8, 12 and 16 m
were similar to the correlation chart presented for 20 m standoff distance (Fig. 11a). As the
standoff distance increases (> 20 m) some of the spectral information is lost, causing the
spectra for each sample to be slightly different from the others (within experimental error)
thus making it difficult to predict its concentration (see Figs. 11b and 11c). Taking into
account the low values of RMSECV and high values of R2 obtained at maximum distance of
20 m, the correlation (PLS) models are useful tools to determine precisely the surface
concentration of TNT unknown samples using OP/FTIR spectroscopy. More precise
alignment of both transmitter and receiver MIR telescopes would be required to perform
similar correlations for ranges > 20 m.
6. Passive mode standoff IR detection

The setup used for passive mode detection using thermal excitation of the sample is shown
in Fig. 13. The MIR source and transmitter telescope were not used in these experiments.
Temperature differences tested were 1 to 7°C in one degree interval. Aluminum plates (Fig.
13a) were heated by a 500 W tungsten lamp that was placed on back of the Al plates (Fig.
13b). The standoff distances studied were 8, 16 and 30 m. In passive mode experiments it is
not so critical to carefully align the target and detector while recording the spectra. Ten
spectra were taken for each sample, at 10 scans and 4 cm-1 resolution for passive mode
experiment.
The emission from a heated, uncoated with TNT Al plate, used as blank to measure
μg/cm2 TNT is also shown overlapping the blank Al plate spectrum. Both traces were
background contributions is shown in Fig.14. The corresponding blackbody spectrum of 400
measured at a range of 8 m and the plates were maintained at an equilibrium temperature of

32°C by heating with a 500 W tungsten lamp.
Fig. 15 shows the TNT IR vibrational signatures recorded with an EM27 spectrometer using
passive mode standoff IR detection of Al plates heated with tungsten lamp to different
surface temperatures from 25 to 32°C. These spectra were measured at a standoff distance of
16 m and a TNT surface concentration of 400 µg/cm2. Most of the characteristic vibrational
signatures of TNT are well defined. The bands that allow identifying TNT were taken in the
spectral region 700-1400 cm-1. Most of the persistent bands are observed and the standoff
spectrum agrees very well with traditional infrared techniques: sample compartment, KBr
pellet, micro-IR, attenuated total reflectance (ATR, both micro and macro) and grazing angle
reflectance (GA, both micro and macro). Bands tentative assignments are: 910 cm-1 (2,6-NO2
scissors and C–N stretch); 1087 cm-1 (C–H (ring) in-plane bend); 1171 cm-1 (C–C in-plane
ring trigonal bend, 2,4,6-C–N and C–CH3 stretch) and 1350 cm-1 due to the symmetric
stretching vibration of the NO2 (nitro) group bond. TNT vibrational markers change
significantly with surface temperatures. At 25°C (black spectrum, room temperature) TNT
vibrational signatures are present, but they can barely be noticed at 16 m standoff distance.
For the spectrum at 26°C (green spectrum) there is a significant increment in intensity of
TNT vibrational signals.
www.intechopen.com
Fig. 13. Passive mode remote IR setup: (a) Al plate with TNT sample smeared-on; (b) 500 W
tungsten lamp assembly
1
1
0.9
0.95
0.8
0.9
0.7 0.85
Emissivity
0.6 0.8
0.5 0.75 Black_body+TNT…

0.4 0.7
700 750 800 850 900 950 1000
0.3
0.2
Black_body+TNT_Emission
0.1 Black_body_Background
0
700 1000 1300 1600 1900 2200 2500 2800
Wavenumber / cm-1
heated Al plate with 400 μg/cm2 TNT at 8 m range and plate heated to 32°C
Fig. 14. Overlap of emissivity spectra of background air and air with TNT emissions from
www.intechopen.com
Fig. 15. Passive mode remote IR spectra of 400µg/cm2 TNT at several temperatures
The variation in peak areas with temperature for TNT for a surface loading of 200 µg/cm2 at
two different standoff distances is shown in Fig. 15. The vibrational spectra were measured
out to a room temperature (25ºC). For a standoff distance of 8 m (Fig. 11a) peak areas of the
vibrational bands at 793 and 1087 cm-1 were measured. For a standoff distance of 16 m the
bands used for peak areas calculations were 793, 1087 and 1171 cm-1 (Fig. 11b). In both cases
when TNT was heated to higher temperatures more intense bands were observed. This
study shows than the increase of the vibrational signatures has a second order polynomial
behavior in all cases, with excellent correlation coefficients squared. These results are very
useful for real field standoff detection, because when the target is warmer than room
temperature, the vibrational signatures of the explosive are increased significantly.
The results of the effect of standoff distance on the intensity of heated samples are shown in
Fig. 16. The spectra were taken at different distances and a specific surface temperature for
each one. The tested temperature was always higher that the ambient temperature. Spectra
taken at room temperature did not show some of the characteristic TNT vibrational signals
(spectra not shown). Fig. 16 shows that the standoff detection in passive mode using thermal
excitation is a useful tool for recording IR spectra to maximum range distance of 30 m,
under the current experimental conditions.
PLS regression algorithm from Quant2 software for OPUS™, version 6.0 (Bruker Optics,
Billerica, MA) was used to find the best correlation function between the IR spectral
information and the TNT surface concentration. PLS was used for generating a
chemometrics model of analyzed standoff distances at specific temperatures (25 to 32°C).
Cross validations were made and RMSECV and R2 were used as criteria to evaluate the
quality of the correlations obtained. The statistical data treatments prepared using
chemometrics routines (PLS) were carried out using spectral region 700 to 1400 cm-1, where
www.intechopen.com
y = 0.0038x2 - 0.1774x + 2.0785

0.25 a R² = 0.9996
y = 0.0012x2 - 0.0582x + 0.7166
0.2 R² = 0.9916
a0.15
e
r
A 793 cm-1 Band
0.1
1087 cm-1 Band
0.05
0
25 26 27 28 29 30 31 32
Temperature / °C
0.25
y = 1.32E-03x2 - 5.18E-02x + 5.37E-01
R² = 9.99E-01
0.2 y = 7.96E-04x2 - 3.50E-02x + 4.08E-01

R² = 9.95E-01
y = 3.39E-04x2 - 1.31E-02x + 1.30E-01
0.15 R² = 9.99E-01
793 cm-1 Band
Area
1087 cm-1 Band

0.1 1171 cm-1 Band
0.05
Temperature / °C
25 26 27 28 29 30 31 32
Fig. 16. Effect of temperature on intensity of TNT vibrational signals for 200 µg/cm2 at range
of: (a) 8 m; (b) 16 m
the nitro symmetric stretch and aromatic C-H and C-C vibrations are present. As before, PLS
models were made using mean centering as pre-processing of variables. Data pre-processing
included: “vector normalization, first derivative and second derivative but NO pre-
processing” achieving best results for RMSECV and R2. Fig. 14 shows the results obtained
www.intechopen.com
for the cross validations at analyzed standoff distances using different surface temperatures.
Table 3 contains the summary of results for RMSECV and R2 obtained in the PLS models.
Fig. 17. Effect of the standoff distance on intensity of TNT IR signals. Surface concentration
of 400 µg/cm2 and 8-16 m range at sample temperatures: 28-32°C
distances and temperature in passive mode: (a) 8 m range, 32 °C surface temperature; (b) 16
Fig. 18. Predicted vs. true surface concentration for TNT explosives at different standoff
m range, 32 °C surface temperature
www.intechopen.com
Standoff Distance (m) Temperature (°C) R2 RMSECV RMSEP Rank
8 25 0.9461 32.6 28.1 6
8 26 0.9737 22.8 29.2 9
8 27 0.9864 16.4 22.8 3
8 28 0.9692 24.7 24.3 2
8 30 0.9658 26.0 28.1 4
8 32 0.9760 21.8 17.7 6
16 25 0.9555 29.7 24.0 3
16 26 0.9624 27.3 29.0 3
16 27 0.9202 39.7 41.4 3
16 28 0.9567 29.3 27.6 4
16 30 0.9506 31.3 30.5 3
16 32 0.9684 25.0 26.6 3
Table 3. PLS calibration parameters for the different tested standoff distances and
temperatures. Spectral range: 700 – 1400 cm-1; no preprocessing
All graphs of predicted vs. true surface concentration for specific distance and temperature
(Table 3) have similar behavior to that of Fig. 18a for standoff distance of 8 m and Fig. 18b at
a range of 16 m. Each point represents ten spectra with a specific surface concentration (0-
400 µg/cm2). Taking into account the high values of R2 (~ 0.96) and relatively low values of
RMSECV (around 26.0) obtained for all the tested distances and temperatures these models
could be used as a tools to determine the surface concentration of unknown samples of TNT
at remote distances using SOIR.
7. Conclusion
A standoff technique using an Open-Path Fourier transform infrared (OP/FTIR)
spectrometer has been demonstrated to obtain spectral information of TNT samples
deposited on Al plates. The system consisted in an infrared telescope coupled MIR source
and an IR coupled EM27 spectrometer manufactured by Bruker Optics. The remote
www.intechopen.com
detection IR system was first tested for the standoff detection of gases and condensable
vapors, which is the application that was developed for. High quality measurements were
achieved by using a sensitive photoconductive cryo-cooled MCT detector. Standoff
detection both in active and passive modes proved to be useful for recording TNT
vibrational signatures in the range from 700 to 1400 cm-1 of the MIR. Very good results of
RMSECV and R2 were obtained in cross validations for active and passive mode
experiments. The active mode standoff detection in worked very well for distances lower
than 30 m. Is necessary carefully aligning the target with the detector to be able to measure
with high accuracy at ranges higher than 30 m.
For passive mode experiments thermal excitation proved to be a useful tool for enhancing
TNT vibrational signatures for standoff detection. Achieving temperature difference of just
1°C between the sample and the spectrometer was enough to bring out spectral information
to standoff distances of 8, 16 and 30 m. The increase in intensity of TNT signatures as a
function of temperature can be modeled very well with second order polynomials in the
temperature range tested above room temperature. In this experiment alignment of sample
and detector was not critical as in the standoff active modality configuration. Partial least
squares routines of commercial chemometrics statistical routines of spectroscopic analysis
were used to enhance the data in multivariate mode.
8. Acknowledgments
Parts of the work presented in this contribution were supported by the U.S. Department of
Defense, University Research Initiative Multidisciplinary University Research Initiative
(URI)-MURI Program, under grant number DAAD19-02-1-0257. The authors also
acknowledge contributions from Mr. Aaron LaPointe from Night Vision and Electronic
Sensors Directorate, Fort Belvoir, VA, Department of Defense, Dr. Jennifer Becker MURI
Program Manager, Army Research Office, DoD and Dr. Stephen J. Lee Chief Scientist,
Science and Technology, Office of the Director, Army Research Office/Army Research
Laboratory, DoD.
Support from the U.S. Department of Homeland Security under Award Number 2008-ST-
061-ED0001 is also acknowledged. However, the views and conclusions contained in this
document are those of the authors and should not be interpreted as necessarily representing
the official policies, either expressed or implied, of the U.S. Department of Homeland
Security.
9. References
ALERT Department of Homeland Security university based Center of Excellence: at
Northeastern University: http://www.northeastern.edu/alert/index.php/
ALERT Department of Homeland Security university based Center of Excellence: at
University of Rhode Island: http://energetics.chm.uri.edu/
Akhavan, J. (2004). The Chemistry of Explosives. 2nd ed. T J International Ltd, Padstow,
Cornwall, UK.
American Society for Testing and Materials (1997) Standard guide for open-path Fourier-
transform infrared (OP/FTIR) monitoring of gases and vapors in air, E-1865–97. In:
www.intechopen.com
Annual book of ASTM standards, American Society for Testing and Materials, West
Conshohocken, PA, vol 03.06.
Bacci, M.; Fabbri, M.; Picollo, M.; Porcinai, S. (2001). Non invasive fibre optic Fourier
Transform-infrared reflectance spectroscopy on painted layers identification of
materials by means of principal component analysis and Mahalanobis distance.
Anal. Chim. Acta. 446, 15-21.
Bangalore, A.S.; Small, G.W.; Combs, R.J.; Knapp, R.B.; Kroutil, R.T.; Traynor, C.A.; Ko, J.D.
(1997) Anal. Chem. 69, 2, 118.
Blake, T.A.; Kelly, J.F.; Gallagher, N.B.; Gassman, P.L.; Johnson, T.J. (2009) Anal. Bioanal.
Chem. 395, 337.
Beebe, K.; Pell, R.; Beth, M. (1998) Chemometrics: A Practical Guide. John Wiley & Sons. N.Y
Caron, T.; Guillemot, M.; Montméat, P.; Veignal, F.; Perraut, F.; Prené P.; Serein-Spirau, F.
(2010) Talanta 81, 1,543.
Castro-Suarez, J.R.; Pacheco-Londoño, L.C.; Vélez, M.; Diem, M.; Tague, Jr. T.J.; Hernandez-
Rivera, S.P. (2010) submitted, Appl. Spectrosc.
Clarkson, J.; Smith, W.E.; Batchelder, D.N.; Smith, D.A. Coats, A.M. (2003) J. Molec. Structure,
648, 203.
Committee on the Review of Existing and Potential Standoff Explosives Detection
Techniques, Existing and Potential Standoff Explosives Detection Techniques (2004)
National Academies Press: Washington, DC.
Diem, M. (1993). Introduction to Modern Vibrational Spectroscopy, John Wiley, New York, NY.
Griffiths, P. R.; de Haseth, J. A. (2007) Fourier transform infrared spectrometry, 2nd ed. Wiley
Interscience, Hoboken, NJ, pp 466–479.
Griffiths, P.R.; Shao, L.; Leytem, A.B. (2009) Anal. Bioanal. Chem. 393, 45.
Gunzler, H.; Gremlich, H.-U. (2002) IR Spectroscopy: An Introduction. Wiley-VCH, DE.
Hamilton, M. L.; Perston, B. B.; Harland, P. W.; Williamson, B. E.; Thomson, M. A.; Melling,
P. J. Organic Process Research & Development. 2005, 9, 337–343.
Hart, B.K., Griffiths, P.R. (1998) Proc 11th Int. Conf Fourier Transform Spectrosc. Am. Inst.
Phys. Conf. Proc. 430:241–242.
Hart, B. K.; Berry, R. J.; Griffiths, P. R. (2000) Environ. Sci. Technol. 34:1346–1351.
Hilmi, A.; Luong, J. (2000) Environ. Sci. Technol., 34, 14, 3046.
Huberty, C.J. (1994) Applied Discriminant Analysis. Wiley–Interscience, Hoboken, NJ.
Johnson, R.A.; Wichern, D.W. (1992) Applied Multivariate Statistical Analysis. Prentice-Hall:
Englewood Cliffs, NJ.
Kramer, R., (1998). Chemometrics Techniques for Quantitative Analysis, Marcel Dekker, New
York, NY.
Lin-Vien, D.; Colthup, N.B.; Fateley, W.G.; Grasselli, J.G. (1991) The Handbook of Infrared and
Raman Characteristic Frequencies of Organic Molecules. San Diego, CA pp. 179–189.
Manrique-Bastidas, C.A.; Castillo-Chará, J.; Mina, N.; Castro, M.E.; Hernández-Rivera, S.P.
(2004a), Proc. SPIE Int. Soc. Opt. Eng. Eng. 5415: 1345.
Manrique-Bastidas CA, Primera-Pedrozo OM, Pacheco-Londoño LC, Hernández-Rivera,
S.P. (2004b) Proc. SPIE Int. Soc. Opt. Eng. 5617: 429.
Mardia, K.V.; Kent, J.T.; Bibbly, J.M. (1979) Multivariate Analysis. Academic Press: New
York, N.Y.
www.intechopen.com
Marshall, M.; Oxley, J.C. (2009) Aspects of Explosives Detection. Elsevier, Amsterdam, The
Netherlands.
Mehta, N. K.; Goenaga-Polo, J. E.; Hernández-Rivera, S. P.; Hernández, D.; Thomson, M. A;
Melling, P. J. Spectroscopy, April, 2003.
Miller, C. J.; Yoder, T.S. (2010) Sens. Imaging: An International Journal. 11, 2, 77.
OPUS version 4.2 (2003) User Manual. Bruker Optics: Billerica, MA.
OPUS version 6.0 (2006) Spectroscopic software manual Bruker Optik GmbH, DE.
Pacheco-Londoño, L.C.; Ortiz, W.; Primera, O.M.; Hernández-Rivera, S.P. (2009) Anal.
Bioanal. Chem. 395, 323.
Parmeter, J. E. (2004) Proceedings of the 38th Annual 2004 International Carnahan
Conference on Security Technology; 355 IEEE: New York.
Perston, B.B.; Hamilton, M.L.; Williamson, B.E.; Harland, P.W.; Thomson, M.A.; Melling, P.J.
(2007) Anal. Chem., 79, 1231–1236.
Primera-Pedrozo, O.M.; Pacheco Londono, L.C.; De la Torre-Quintana, L.F.; Hernandez-
Rivera, S.P.; Chamberlain, R.T.; Lareau, R.T. (2004) Proc. SPIE Int. Soc. Opt. Eng.
5403, 237–245.
Primera-Pedrozo, O.M.; Pacheco-Londono, L.C.; Ruiz, O.; Ramírez, M.; Soto-Feliciano, Y.M.;
De La Torre-Quintana, L.F.; Hernandez-Rivera, S.P. (2005) Proc. SPIE Int. Soc. Opt.
Eng. 5778, 543-552.
Primera-Pedrozo, O. M.; Rodríguez, N.; Pacheco-Londoño, L. C.; Hernández-Rivera, S. P.
(2007) Proc. SPIE Int. Soc. Opt. Eng., 6542, 65423J.
Primera-Pedrozo, O.M.; Soto-Feliciano, Y.; Pacheco-Londoño, L.C.; Hernandez-Rivera, S.P.
(2008) Sens. Imaging: An International Journal. 9, 27–40.
Primera-Pedrozo, O.M.; Soto-Feliciano, Y.; Pacheco-Londoño, L.C.; Hernandez-Rivera, S.P.
(2009) Sens. Imaging: An International Journal. 10, 1–13.
Soto-Feliciano, Y.; Primera-Pedrozo, O.M.; Pacheco-Londoño, L. C.; Hernandez-Rivera, S.P.
(2006) Proc. SPIE Int. Soc. Opt. Eng. 6201, 62012H.
Russwurm, G. M.; Childers, J.W (1999) FTIR open-path monitoring guidance document, 3rd ed.,
TR-4423–99–03. ManTech Environmental Technology. Research Triangle Park, NC,
USA.
Russwurm, G.M.; Childers, J.W. (2001) Open-path Fourier transform infrared spectroscopy. In:
Chalmers, J.W.; Griffiths, P. R. (eds.) Handbook of vibrational spectroscopy, vol. 2.
Wiley, Chichester, UK.
Yinon, J.; Zitrin, S. (1996) Modern Methods and applications in analysis of explosives. John Wiley
& Sons Ltd., Chichester, UK.
Sharma, S.P.; Lahiri, S.C. (2008) Spectrochim. Acta Part A, 70, 144.
Shao, L.; Griffiths, P.R.; Leytem, A.B. (2010) Anal. Chem. 82, 8027–8033.
Smith, B. (2000) Fundamentals of Fourier Transform Infrared Spectroscopy. CRC Press, Boca
Raton, FL, USA.
Szakal, C.; Brewer, T.M. (2009) Anal. Chem., 81, 13, 5257.
Theriault, J.M.; Puckrin, E.; Hancock, J.; Lecavalier, P.; Lepage, C.J.; Jensen, J.O. (2004) Appl
Opt, 43, 5870.
Urbanski, T. (1964). Chemistry and Technology of Explosives. Vol. 1, Macmillan Co.: New York,
NY.
Van Neste, C.W.; Senesac, L.R.; Thundat, T. (2009) Anal. Chem. 81, 5, 1952.
www.intechopen.com
Yinon, J. (1996) J. Chrom. A, 742, 1, 205.

Vrcelj, R.M.; Gallagher, H.G.; Sherwood, J.N. (2001) J. Am. Chem. Soc. 123:2291.
Weber, K.; Fischer, C.; van Haren, G.; Krause, H.; Bunte, G.; Schweikert, W.; Härdle, T.
(2006) Stand-Off Detection of Explosives of Suicide Bombers by Means of Open-Path FTIR
Spectroscopy in Schubert, H.; Rimski-Korsakov, A. Stand-Off Detection of Suicide-
Bombers and Mobile Subjects, Proceedings of the NATO Advanced Research
Workshop on Stand-Off Detection of Suicide Bombers and Mobile Subjects, NATO
Security through Science Series B: Physics and Biophysics, Pfinztal, Germany,
Springer, DE.
www.intechopen.com
21
Remote Sensing of Atmospheric Trace Gases

by Ground-Based Solar Fourier Transform
Infrared Spectroscopy
Clare Paton-Walsh
University of Wollongong
Australia
1. Introduction
The changing composition of the earth’s atmosphere is a matter of intense scientific research
as we strive to understand details of the physical and chemical mechanisms that control our
climate. Fourier transform spectroscopy has been applied very successfully to the study of
trace gases in the atmosphere by examining terrestrial atmospheric absorption lines in the
infrared spectrum from the Sun. In fact many gases were first discovered in the atmosphere
during the 1940’s from their absorption features in the infrared solar spectrum. These early
optical absorption measurements of the atmosphere using the Sun as a source were made
with grating spectrometers and examples of atmospheric gases first detected this way
include methane and carbon monoxide (Migeotte, 1948; 1949).
Continuous or semi-continuous records of infrared solar atmospheric absorption spectra
have been made from ground-based Fourier transform spectrometers (FTS) since the late
1970s and early 1980s, when the first ground-based solar-tracking FTS systems were
installed at Kitt Peak National observatory in the USA and at the Jungfraujoch Observatory
in Switzerland (Goldman et al., 1979; Murcray et al., 1978; Zander et al., 1977). Initially interest
was focused on the detection and quantification of stratospheric trace gases (Rinsland et al.,
1986; Zander et al., 1986). The discovery of the Antarctic ozone hole (Farman et al., 1985)
intensified interest in stratospheric chemistry and helped support the establishment of the
Network for Detection of Stratospheric Change (NDSC). This global network of instrument
sites became operational in 1991 with ground-based FTS amongst the suite of primary
techniques being used. Photographs of the instrument at the NDSC site at Wollongong,
Australia are shown for illustrative purposes in figure 1 below. Other NDSC instruments are
lidars for ozone, temperature, water and aerosols; microwave instruments for ozone, water
and chlorine monoxide; UV/Visible spectrographs for ozone and nitrogen dioxide;
Dobson/Brewer spectrophotometers for total column ozone and regular ozone sondes. The
establishment of the NDSC resulted in a huge increase in the number of infrared solar
absorption measurements being made around the globe during the next few years.
More recently interest in atmospheric chemistry has been focused on tropospheric pollution
and anthropogenic emissions of greenhouse gases (Barret et al., 2003; Jones et al., 2009; Mahieu
et al., 1995; Nagahama and Suzuki, 2007; Paton-Walsh et al., 2008; Rinsland et al., 2000; Rinsland
et al., 2001; Rinsland et al., 2002; Rinsland et al., 2008; Warneke et al., 2006; Zhao et al., 2000; Zhao
www.intechopen.com
Fig. 1. The ground-based solar Fourier Transform infrared spectrometer at Wollongong,

Australia. The sun’s radiation is captured by the solar tracker (shown above) and sent to the
entrance optics of the high resolution Fourier transform spectrometer in the laboratory
below (shown right). (Photography by R Macatangay)
et al., 2002). As a result, the NDSC has changed its emphasis and name to the Network for
Detection of Atmospheric Composition and Change (NDACC) – see
http://www.ndacc.org/. As well as an ever increasing number of sites in the global
network the new millennium has seen an expansion into the near infrared spectra region in
www.intechopen.com
Remote Sensing of Atmospheric Trace Gases by
Ground-Based Solar Fourier Transform Infrared Spectroscopy 461
an effort to provide extremely accurate and precise measurements of carbon dioxide and
other greenhouse gases. The Total Column Carbon Observing Network (TCCON) was
established to help characterise biogenic and oceanic sources and sinks of greenhouse gases
to and from the atmosphere and to validate current and future satellite based measurements
(http://www.tccon.caltech.edu/ ).
2. Atmospheric solar infrared Fourier transform transmission spectra

Fourier transform infrared spectroscopy is a powerful tool for monitoring the changing
composition of the atmosphere because it can measure so many different gases
simultaneously. The ability to measure atmospheric trace gases in the infrared is enhanced
by the fact that the major components of the atmosphere (nitrogen, oxygen and argon) have
no dipole moment and thus are infrared inactive.
Fig. 2. An example spectrum recorded by a ground-based solar Fourier transform

spectrometer
In ground-based solar Fourier transform spectroscopy, the Sun acts as the radiation source
and the sample is the atmosphere. The shape of the resulting spectrum depends upon the
solar radiation reaching the top of the Earth’s atmosphere, absorption by the atmosphere
and the optical properties of instrument recording the radiation. The solar radiation
reaching the top of the Earth’s atmosphere is in essence a blackbody curve at 5800K with
emission and absorption lines of gases in the solar atmosphere imposed. Terrestrial
atmospheric absorption lines contain information about the species of trace gases present in
the atmosphere (line positions), the amounts of each gas present (line depths/areas) and
some information about the altitude distribution of each gas (line shapes). In reality there is
also a component of radiation as a result of atmospheric emission, but this is so small in
comparison to the radiation from the sun that its effects are negligible.
www.intechopen.com
3. Calculating synthetic solar atmospheric transmission spectra

Clearly it is not possible to measure a background spectrum or a set of calibration spectra
when the atmosphere is the sample. For this reason the analysis of these spectra requires the
calculation of a synthetic spectrum using a database of absorption line parameters such as
the HITRAN database (Rothman et al., 1998; Rothman et al., 2003; Rothman et al., 2005) and a
model of the atmospheric conditions such as pressure, temperature and gas concentrations
that all vary with altitude. The calculation must also take into account instrumental effects
such as line shape and resolution.
The HITRAN (HIgh resolution TRANsmission) database contains calculated quantum
mechanical parameters that describe the vibrational-rotational transitions of the most
common atmospheric molecules. Parameters include the line positions (frequencies of
absorption lines), line strengths at a reference temperature and pressure, lower state energy
and pressure broadening and shift parameters.
3.1 Atmospheric absorption line positions

The characteristic absorption features seen in Figure 2 are caused by molecules absorbing
radiation at frequencies that correspond to the allowed transitions between different
vibrational and rotational states. The frequencies at which molecules absorb are determined
by the allowed transitions between energy levels of the molecule and provide a unique
identifier of trace gases in the atmosphere. Vibrational energy states are much more widely
spaced than rotational energy states and so the line position is determined mainly by the
change in vibrational state with small frequency differences depending on the
accompanying change in rotational energy.
The energy of allowed states of a simple diatomic molecule can be approximated by
Equation 1:
E=¥ ( v + 1 ) + BJ ( J + 1)
μ
k
(1)
2
(if it is assumed that molecule behaves like a simple harmonic oscillator and a rigid rotor)
•
and where:
¥ is Planck’s constant divided by 2π,
• k and μ are the force constant and the reduced mass of the molecule respectively,
• B is the rotational constant (B= ¥ /2I, where I is the moment of inertia, the product of
•
the reduced mass and the square of the radius of rotation)
and v and J are the vibrational quantum number and rotational quantum number of the
state respectively.
For diatomics the allowed transitions between states are Δv = ±1,±2,±3, . . and ΔJ = ±1.
A vibration-rotation transition with Δv= -1 results in an emission spectrum and a transition
with Δv= +1 an absorption spectrum, and both types of spectra may be accompanied by
either a gain in rotational energy ΔJ=+1 or a loss in rotational energy ΔJ= -1. This produces a
spectrum with two branches, one either side of the pure vibrational frequency, v0. The R
branch corresponds to a vibrational transition accompanied by a gain in rotational energy
ΔJ= +1 and consists of a set of lines spaced approximately 2B apart to the high frequency
side of the pure vibrational frequency, v0, becoming more closely packed as the rotational
energy increases further from the band centre, (assuming B1 < B0). The P branch corresponds
www.intechopen.com
to a vibrational transition accompanied by a loss in rotational energy ΔJ= -1 and consists of a

set of lines spaced approximately 2B apart to the low frequency side of the pure vibrational
frequency, v0, becoming more widely spaced as the rotational energy increases away from
the band centre. Linear polyatomic molecules can also vibrate in a manner such that the
dipole moment is changed perpendicular to the principal axis of rotational symmetry and in
this case the selection rules also allow ΔJ = 0, i.e. the vibrational energy change can occur
without an accompanying change in rotational energy. In such spectra the Q-branch appears
at the band centre between the P and R branches (at the pure vibrational frequency, v0). The
Q-branch is a relatively intense feature because the vibrational transitions occur from all
existing rotational states with approximately the same energy change and hence the same
frequency.
3.2 Line broadening and line shapes

As a consequence of Heisenberg’s uncertainty principle, absorption lines are never infinitely
narrow. The uncertainty in the energy of a state multiplied by the uncertainty in time (the
lifetime of a state) must be greater or equal to ħ, (ΔEΔt > ħ). So the shorter the lifetime, the
larger the uncertainty in a state’s energy and the broader the absorption or emission line (as
the energy uncertainty manifests itself as an uncertainty in the frequency of the line).
Uncertainty broadening due to the radiative or intra-molecular lifetime of the state in
isolation is always present but for vibrational-rotational states is usually very small (< 10-6
cm-1). Another form of uncertainty broadening, which dominates at atmospheric pressures,
is collisional broadening (also called pressure broadening) and occurs when the collisions of
atoms, ions or gas molecules shorten the lifetime of states. In gases it is proportional to
pressure and this means that absorption lines from spectra taken through the whole
atmosphere will have different shapes depending upon the distribution of the absorbing gas
in the atmosphere. A gas that is mainly located in the troposphere (such as methanol) will
display broad absorption lines because of the high pressure whilst predominantly
stratospheric gases (like ozone) will produce much narrower lines since the pressure is low.
Uncertainty broadening leads to a Lorentzian line shape contribution at a given
wavenumber :
αL /π
f L ( v) =
( v − v0 ) 2 + α L 2
(2)
•
where:
•
v0 is the absorption line frequency in wavenumbers, and
αL is the Lorentzian half-width at half height, which is proportional to the total
pressure.
Absorption lines due to atmospheric gases are also subject to Doppler broadening. Doppler
broadening occurs because molecules that travel with different velocities with respect to the
light source, absorb at different wavelengths, just as light from stars accelerating away from
the Earth is red-shifted. Doppler broadening produces a Gaussian line shape due to the
Gaussian distribution of molecular velocities:
⎛ ( v − v )2 ⎞
fG ( v) = exp ⎜ − ⎟
⎜ αG 2 ⎟
1
αG π
0
⎝ ⎠
(3)
www.intechopen.com
•
where
αG is the Gaussian half-width at half height and is given by:
αG =
vl 2 kT
(4)
c m
•
where
•
k is the Boltzman constant,
•
T is the Kelvin temperature and
m is the molecular mass.
Pressure broadening (with Lorentzian lineshape) dominates in the troposphere1, but its
effects drop off rapidly with altitude as the pressure drops. Doppler broadening (with
Gausssian line shape) is temperature dependent but its variation through the atmosphere is
much smaller than pressure broadening. Stratospheric gas lines are primarily Doppler
broadened and the two types of broadening become equally significant at around 25 km.
The convolution of Lorentzian and Gaussian line shapes produces a Voigt line shape. This
variation of the shape and width of absorption lines with the pressure of the absorbing gas
means that spectra of atmospheric gases contain information about the altitude of the
absorbing gas as well as the total number of absorbing molecules in the path.
3.3 Atmospheric absorption line intensities

The integrated line strength of each absorption line of each molecule is determined by the
transition probability, the population and degeneracy of the initial state and the number of
absorbing molecules in the path. In local thermodynamic equilibrium the population of
states is determined by the Boltzmann distribution, which is dependent upon the
temperature of the absorbing molecules and for this reason line strengths are temperature
dependent.
At a given wavenumber (v), the intensity of radiation (the radiative power) that reaches the
ground I(v), is related to the radiative power at the top of the atmosphere I0(v) by Equation 5:
I ( v ) = I 0 ( v )e − mτ ( v ) (5)
•
where
•
(v) is the optical depth of the atmosphere and
m is the airmass factor - a geometrical factor accounting for the slant path through the
atmosphere.
In order to calculate a synthetic spectrum all molecules that absorb in the spectral region
being simulated must be considered together. At any given wavenumber (v), the
contribution to the optical depth (τ) for every absorption line k of each molecule i, is given
by
τ ik (ν ) = σ ik (ν ).ai (6)
σ ik ( ) is the absorption coefficient or cross section at

where
•
1 Typical values for a mid-size molecule are αL ~ 0.15 cm-1 atm-1 and αG ~ 0.004 cm-1.
www.intechopen.com
• , for each absorption line k of each molecule i, (typically in units cm2 molec-1)
•
The absorption coefficient σ ik ( ) is the convolution ( ⊗ ) of the integrated line strength, Sik
ai is the amount of molecule i, in units of molec cm-2.
and the pressure broadening and Doppler broadening line-shapes:
σ ik (ν ) = Sik ⊗ ⎡⎣ f L ( v )⎤⎦ i ⊗ ⎡⎣ f G ( v )⎤⎦ i

k k
(7)
The HITRAN 2004 database contains line parameters for 39 atmospheric gases including the
wavenumber of each absorption line, the integrated line strengths at 296K, the lower state
energy E0, the air-broadened Lorentzian half-width at atmospheric pressure and its
temperature dependence.
The integrated line strength is temperature dependent because of the temperature
dependence of the population of the lower-state energy and a small contribution from the
spontaneous emission. The integrated line strength at any given temperature (ST) may be
calculated using the integrated line strengths at 296K and other parameters given in
HITRAN with Equation 8:
exp( − 2 0 ) (1 − exp( − 2 0 ))
cE c v
S(T ) = S(296)
Q(296) T T
Q(T ) exp( − c 2 E0 ) (1 − exp( − c 2 v0 ))
. . (8)
296 296
•
where
•
Q(296) and Q(T) are the partition functions at these temperatures and
c2 is the second radiation constant, (c2=hc/k = 1.439 cm K) (Griffith, 1996).
3.4 Calculating synthetic solar atmospheric transmission spectra

The dry atmosphere is composed of mainly nitrogen (78%) and oxygen (21%) with minor
and trace gases making up the other 1%. The normal modes of vibration of nitrogen and
oxygen are infrared inactive because the symmetry of the molecules is such that vibrations
do not cause a change in the dipole moment, so there is nothing for incoming infra-red
radiation to interact with. Thus the main constituents of the atmosphere transmit infra-red
radiation and the infrared spectrum of the atmosphere as shown in Figure 2 is dominated by
minor constituents and trace gases such as water, carbon dioxide and methane. This is the
reason that these species are important greenhouse gases.
The temperature, pressure and concentrations of trace gases in the atmosphere change
continuously with altitude. In order to calculate a synthetic spectrum the atmosphere is
modelled as a series of homogeneous layers each with a defined temperature, pressure and
gas composition2. To model the radiative transfer through the atmosphere the path length of
radiation from the sun through each layer of the modelled atmosphere must be calculated
using a ray-tracing algorithm that takes into account the solar zenith angle at the time of the
2 The amounts of atmospheric trace gases are often measured in mole ratios, defined as the number of
moles of the substance of interest per mole of air. For trace gases the usual units are micro moles per
mole (μmol.mol-1) – sometimes expressed as parts per million (ppm), e.g. 5 ppm CO means 5 molecules
of carbon monoxide per million molecules of air). For ultra-trace gases the units often used are nano
moles per mole (nmol.mol-1) or picomoles per mole (pmol.mol-1).
www.intechopen.com
measurement and the effects of atmospheric curvature and refraction. At least thirty
different atmospheric layers are required in order to do the radiative transfer calculation
without errors in this calculation becoming a dominant uncertainty (Meier et al., 2003).
Fig. 3. The infrared transmission spectrum of water (top panel), carbon dioxide (middle
panel) and methane (bottom panel) in the atmosphere from 500 cm-1 to 4400 cm-1 as
simulated using the HITRAN 2000 database for a solar zenith angle of 70° and concentration
profiles taken from US standard atmosphere (Meier et al, 2004)
www.intechopen.com
Following on from Equation 6, when all the absorption lines of all molecules are considered,
the total optical depth at any given frequency, (v), is the sum of the contribution of all the
absorption lines of all molecules, for all homogeneous layers:
τ ( v) = ∑ ∑∑τ ik ( v ) (9)
layers i k
The true atmospheric transmittance spectrum, T(v) is the ratio of the intensity at the ground,
I(v), and the intensity at the top of the atmosphere, I0(v):
T ( v) = = exp( −τ ( v ))
I( v)
(10)
I0 ( v)
The calculation of the synthetic spectrum must also include the effects of instrumental
parameters such as the resolution and instrumental line-shape including phase error and
wavelength shift. The spectrum that will actually be measured is a convolution of the
intensity at the ground, I(v) and the instrumental line-shape. The instrumental line-shape
(ILS) may be derived using measurements of low-pressure gas cells (Hase et al., 1999), or a
theoretical ideal ILS can be calculated from the instrument’s field of view, resolution and the
apodisation function3.
Example simulations calculated in this manner are shown in Figure 3 (Meier et al., 2004). In
these cases the infrared transmission spectra that result from just one component
atmospheric gas at a time are given, namely water, carbon dioxide and methane. The spectra
are simulated from 500 cm-1 to 4400 cm-1 using the HITRAN 2000 database for a solar zenith
angle of 70° and concentration profiles taken from US standard atmosphere.
4. Theoretical basis for the retrieval of trace gas amounts from atmospheric
solar infrared Fourier transform transmission spectra
4.1 Development of analysis algorithms
Over the years several different analysis algorithms have been developed to perform the
retrieval of trace gas amounts from solar FTIR spectra. All the techniques calculate a
synthetic spectrum using a “forward model” as described above, that includes a model of
the instrumental effects and a layered model of the atmosphere with assumptions about
environmental parameters such as the pressure, temperature and composition of each layer.
The synthetic spectrum is compared to the measured spectrum and adjustments made to the
forward model until a best fit with the measured spectrum is obtained.
Different analysis algorithms allow different adjustments to the forward model in order to
•
achieve the best fit to the measurement.
“SFIT1” (Rinsland et al., 1982; Rinsland et al., 1984) and “GFIT” (Washenfelder et al., 2006)
allow only a scaling of the a priori concentration profile of the absorbing gases to
achieve best fit. This means that the distribution of the absorbing gas is defined entirely
3 An apodisation function is a mathematical function that is applied to the interferogram that may give
greater weight to the information around zero path difference compared to information at greater
optical path differences. In the analyses presented in this thesis a boxcar apodisation function was used
that gives equal weight to the information throughout the interferogram.
www.intechopen.com
by the a priori assumptions – the fitting process for example cannot put more gas in the
troposphere and less in the stratosphere but must multiply the concentration in each of
•
the modelled layers by the same multiplicative factor.
Later development “SFIT2” (Pougatchev et al., 1995) using optimal estimation techniques
(Rodgers, 1990) aimed to extract the limited spectral information on the vertical
distribution of the target gas, which is contained in the shape of the absorption features
as a result of pressure broadening by the surrounding atmospheric gases (Hase et al.,
•
2004; Rinsland et al., 1998).
More recently another analysis code (“PROFITT”) has been developed using optimal
estimation techniques that allows the temperature profile and concentration profiles to
be adjusted (Hase et al., 2004).
4.2 Inverse modelling and optimal estimation

In contrast to GFIT, SFIT2 allows the volume mixing ratio profile of the absorbing gas in the
simulated spectrum to be adjusted so that the shape of the absorption line can best fit the
measured spectrum. The principle difficulty in this technique is that the radiative transfer
model requires at least thirty atmospheric layers to achieve a reasonable model of the
transmission of solar radiation through the atmosphere, but the shape of an absorption
feature typically contains between one and five independent pieces of information. Thus the
problem is mathematically underdetermined and much of the information must still be
provided by the a priori concentration profile in the forward model.
SFIT2 uses an inverse modeling technique (Rodgers, 1990; Rodgers, 2000) to extract the
volume mixing ratio profile of the gases of interest from the measured spectrum. The
volume mixing ratios of the gas of interest at each of the modeled layers are the variables of
interest (called state variables or together the state vector x, with n elements). The measured
spectrum (a series of observed radiances at different frequencies) is represented by the
observation vector y (with m elements), and the forward model, F, describes the relationship
between the observation vector y (the spectrum) and the state vector x, (the volume mixing
ratios of the gases of interest).
y = F( x , b ) + ε (11)
where b is a parameter vector including all model variables that are not to be optimized (also
called the model parameters), and ε is the error vector including errors in the observations, in
the forward model, and in the model parameters.
Inverting Equation 11, x may be obtained given y, but due to the error term ε the best that
can be achieved is a statistical estimate. As stated before the problem is mathematically
underdetermined and in optimal estimation x is weighted by our prior (a priori) knowledge
of the state vector xa (the a priori volume mixing ratio profile). The optimal solution of x
including this a priori knowledge is called the “optimal estimate” or the ‘retrieval’ (also
sometimes called the “a posteriori solution”).
The a priori estimate has its own error: xa=x + εa and the key to solving the optimal
estimation problem is weighting the error statistics of ε and εa (Rodgers, 2000).
The inverse problem lends itself to the use of matrix algebra, and the Jacobian matrix, K, is a
linearization of the forward model that represents the sensitivity of the observation
variables y to the state variables x, assembled in matrix form (Jacob, 2007):
www.intechopen.com
∂y
K = ∇xF =
∂x
(12)
K changes with x and so it is calculated initially for the a priori value xa and then re-
calculated iteratively until the solution converges. The algorithm iterates until the cost
function J(x) is minimised by solving for ∇ x J ( x ) = 0 as shown in Equation 13:
∇ x J ( x ) = 2Sa−1 ( x − xa ) + 2K T Sε−1 (Kx − y ) = 0 (13)
where Sa and Sε are the a priori error and observational error covariance matrices
respectively (the matrix equivalents of εa and ε ). Note that observational error includes
errors in the forward model as well as spectral noise and that in many instances it is the
forward model error that dominates the Sε matrix.
The solution to Equation 13 yields the optimal estimate or retrieval x̂ and is given by Equation 14:
xˆ = x a + G( y − Kxa ) (14)
where G is known as the gain matrix and describes the sensitivity of the retrieval to the
∂x̂
observations, i.e. G =
∂y
, and is given by Equation 15:
G = (K T Sε−1K + Sa−1 )−1 K T Sε−1 (15)
4.3 Uncertainties in retrieved trace gas amounts and the “averaging kernel”
∂x̂
The ability of the measured spectrum to constrain the volume mixing ratio profile of the gas
of interest, (the state vector, x) is given by the averaging kernel matrix A =
∂x
, which
represents the sensitivity of the retrieval x̂ to the true state vector x.

∂x̂
The averaging kernel matrix, A=GK, i.e. it is the product of the gain matrix G =
∂y
and the
∂y
Jacobian matrix K =
∂x
.
The information content may be defined in terms of the degrees of freedom for signal which
is the trace of the averaging kernel. Mathematically this is the sum of the diagonal elements
of the averaging kernel matrix (Rodgers, 2000).
Using the averaging kernel matrix leads to an alternative expression for the optimal estimate
or retrieval x̂ :
xˆ = Ax + ( I n − A)x a + Gε (16)
where In is the identity matrix of dimension n. There are three terms on the right hand side
of Equation 16. The first term, Ax represents the contribution of the true state x to the
solution. The second term (In – A)xa represents the contribution of the a priori assumptions.
The third term Gε is the contribution from random observational error. An ideal
measurement would have an averaging kernel matrix that was an n dimensional identity
matrix, A = In in which case the 2nd term is zero (Jacob, 2007; Rodgers, 2000).
www.intechopen.com
The second term in Equation 16, [(In – A)xa] is called the smoothing uncertainty (since it
results in a smoothing of the retrieval towards the a priori values). The third term, Gε, is
known as the retrieval uncertainty (or signal-to-noise uncertainty) but again it should be noted
that this includes not only measurement noise but also errors in the forward model such as
errors in the HITRAN database and that these factors often dominate the spectral noise.
There may also be errors in the forward model that are correlated with one another and are
therefore not random, potentially leading to uncharacterised biases in the retrievals. One
other error that can be estimated is the temperature uncertainty, due to errors in the assumed
temperature profile in the forward model. These lead to errors in the line strengths and
resulting retrieved trace gas amounts that may be calculate using Equation 8.
5. A summary of past successes for ground-based solar Fourier transform

infrared spectroscopy of atmospheric composition
In the late 1970’s there were two groups of researchers actively engaged in making solar
infrared measurements of atmospheric composition from the ground. The scientific focus at
the time was on understanding the chemistry of the stratosphere and many measurements
were made using balloon-borne grating spectrometers (Murcray et al., 1975; Zander, 1976).
Bradford et al (1976) outlined the feasibility of monitoring atmospheric trace gases using
ground-based high resolution infrared spectroscopy. Within a couple of years the first
scientific results from ground-based stations had been published, confirming the presence of
hydrogen fluoride in the stratosphere from ground-based measurements from the
Jungfraujoch in Switzerland, (47°N, 8.0°E) (Zander et al., 1977) and identifying atmospheric
absorption features of ammonia in spectra from Kitt peak in the USA, (32°N, 112°W)
(Murcray et al., 1978). In 1979 Goldman et al, published their “New Atlas of Infrared Solar
Spectra” which included telluric absorption features of carbon dioxide, water, methane,
carbon monoxide, nitrous oxide, ozone, carbonyl fluoride, chlorine nitrate,
difluorochloromethane (CFC-22), dichlorodifluoromethane (CFC-12), sulphur hexafluoride
and nitric acid as well as a number of absorption features from carbon monoxide and
hydroxyl radicals in the solar atmosphere.
As the capabilities increased more trace gases were added to the list of measureable species
including hydrogen cyanide (Rinsland et al., 1982), nitric oxide (Rinsland et al., 1984) and
chlorine nitrate (Zander et al., 1986). After the discovery of the Antarctic ozone hole (Farman
et al., 1985) and the establishment of the Network for Detection of Stratospheric Change
(NDSC), the number of ground-based Fourier transform spectrometers making solar
atmospheric absorption measurements increased dramatically. During the first half of the
1990’s Fourier transform solar remote sensing spectrometers were installed at a number of
new sites including Lauder, New Zealand (45°S, 170°E) in 1990 (Jones et al., 1994); Mauna
Loa, Hawaii (20°N, 156°W)(David et al., 1993) & Arrival Heights, Antarctica (78°S,
167°E)(Kreher et al., 1996; Wood et al., 2004) in 1991, Ny Alesund, Spitzbergen (79°N, 12°E) in
1992 (Notholt and Schrems, 1994) and Harestua, Norway (60°N, 11°E) in 1994 (Mellqvist et al.,
2002). In 1994 the first measurements using the moon as a light source were reported from
Spitzbergen, enabling measurements of the Arctic atmosphere during polar night (Notholt,
1994).
In 1995 three additional sites were added to the network at Zugspitze, Germany (47°N,
11°E) (Sussmann and Schafer, 1997), Rikubetsu, Japan (44°N, 144°E)(Zhao et al., 2000) and
Wollongong, Australia (34°S, 151°E) (Rinsland et al., 2001). The following year saw a further
www.intechopen.com
three sites established at Kiruna, Sweden, (68°N, 20°E) (Blumenstock et al., 1997), Eureka,
Nunavit (80°N, 86°W) (Batchelor et al., 2009; Donovan et al., 1997) and Moshiri, Japan (44°N,
142°E) (Zhao et al., 2000). Measurements at Tsukuba, Japan (36°N, 140°E) started in 1998,
followed by Thule, Greenland (77°N, 69°W), Poker Flat, Alaska (65°N, 147°W) and Izana,
Tenerife Island (28°N, 16°W) all in 1999.
In the meanwhile there were a number of mobile instruments operating. The Jet Propulsion
Laboratory’s “Mark IV” instrument was making regular balloon flights from Alaska
interspersed with ground-based measurements from Esrange (68°N, 21°E), Fairbanks (65°N,
148°W), Mount Barcroft (38°N, 118°W) or Table Mountain (34°N, 118°W) (Toon et al., 1999b).
The Alfred Wegner Institute had a ship-borne spectrometer onboard the Polarstern (Notholt
et al., 1995; Notholt et al., 2000), and the National Physical Laboratory had a mobile
instrument that made a series of side-by-side instrument intercomparisons with a number
of other spectrometers in the network as well as a number of campaign measurements at
Are, Sweden, Aberdeen, Scotland and Calar Alto, Spain (Bell et al., 1994; Bell et al., 1998;
Paton-Walsh et al., 1997). In addition to instrument intercomparisons, there were a number of
algorithm intercomparison exercises and a standard procedure for characterising the
instrument line-shape was developed (Hase et al., 1999; Hase et al., 2004). Apriori profiles
were most commonly based upon measurements from the balloon-based MarkIV
instrument (Toon et al., 1999a) or from the ATMOS instrument that was flown on the Space
Shuttle (Gunson et al., 1996).
Early in the new millennium the NDSC changed its name to the Network for Detection of
Atmospheric Composition Change (NDACC) to highlight the change in scientific focus from
stratospheric chemistry to changing tropospheric composition and greenhouse gases. The
NDACC database (see http://www.ndacc.org/) contains standard gases for most stations
equipped with a remote sensing FTIR spectrometer including ozone, nitric acid, hydrogen
chloride, chlorine nitrate, hydrogen fluoride, nitrous oxide, carbon dioxide, carbon
monoxide, methane, ethane and hydrogen cyanide. Many stations also provide other gases
such as CFCs, nitrogen dioxide, nitrogen oxide and acetylene. Further stations were added
including Toronto, Canada (44°N, 80°W) in 2002 and Bremen, Germany (53°N, 9°E) in 2004,
whilst campaign measurements have been made at sites including Reunion Island, (22° S,
56°E) (Senten et al., 2008), Paramaribo, Suriname (6°N, 55°W)(Petersen et al., 2010) and Addis
Ababa (9°N, 39°E). Further gases have also been identified in spectra from solar remote
sensing Fourier transform spectrometers included chlorine monoxide (Bell et al., 1996),
formic acid, (Rinsland et al., 2004), ethylene(Rinsland et al., 2005) and methanol (Paton-Walsh
et al., 2008). In addition there have been a number of studies that examine different isotopes
(Frankenberg et al., 2009; Goldman et al., 2000; Goldman et al., 2002; Haverd et al., 2005; Irion et
al., 1996; Meier and Notholt, 1996; Notholt et al., 2010). Trends in both stratospheric gases eg
(Rinsland et al., 2003) and tropospheric gases eg (Jones et al., 2009) have been established from
NDACC remote sensing FTIR spectrometers. Recently there has been significant interest in
the ability of this technique to characterise free tropospheric water vapour and its isotopes
(Palm et al., 2010; Schneider et al., 2010). The NDACC continues to provide ground-validation
for a number of retrieved products from several different satellite-based sensors (Dils et al.,
2006; Dupuy et al., 2009; Mahieu et al., 2008; Payan et al., 2009; Strong et al., 2008; Vigouroux et
al., 2007; Wolff et al., 2008; Yurganov et al., 2008).
In 2004 the Total Carbon Column Observing Network (TCCON) was established with the
aim of making very accurate measurements of greenhouse gases in the near-infrared
spectral region (see https://tccon-wiki.caltech.edu/ ). The primary function of TCCON was
www.intechopen.com
to act as a network of ground-thruthing stations for the Orbiting Carbon Observatory (Crisp
et al., 2004), however TCCON also had a remit to provide independent constraints to models
of the carbon cycle and provide validation data to other satellite-based sensors. New sites
were established at Park Falls, Wisconsin, USA (46°N, 90°W)(Washenfelder et al., 2006) and
Darwin, Australia (12°S, 131°E) (Deutscher et al., 2009) whilst in a number of existing
NDACC sites have been upgraded to allow for the extended spectral coverage and
improved precision demanded for TCCON. The failure of the launch of the Orbiting
Carbon Observatory has meant that TCCON has yet to fulfil its primary function however a
re-launch is now planned and in the meanwhile TCCON has started to elucidate details of
the carbon cycle itself (Yang et al., 2007). More TCCON stations are being set-up and the data
is available for the validation of other satellite instruments such as SCIAMACHY, AIRS,
IASI and GOSAT(Bosch et al., 2006).
6. Concluding remarks
Remote sensing of atmospheric trace gases by ground-based solar Fourier transform
infrared spectroscopy has developed rapidly since its origins in the 1970s. It has had major
successes in characterising the composition of the atmosphere and trends in both
stratospheric and tropospheric gases. Whilst the technique has greater uncertainties than
ground level in situ measurements of atmospheric composition, it characterises the total
atmospheric column amount. Interpretation of ground level measurements is often
complicated by the effects of vertical transport and changing boundary layer height, but
total column measurements are less sensitive to this problem because the measurement is
integrated over the whole atmosphere. A significant drawback to these remote sensing
techniques is that they are numerically ill-posed and thus some apriori information is
required to derive total column amounts from the spectra. In future atmospheric models are
likely to be constrained by use of combined in situ and remotely sensed data such that the
total uncertainties are minimised.
Remote sensing Fourier transform spectrometers play a vital role in our efforts to understand
the changing composition of the atmosphere. The networks of ground-based instruments are
continuing to expand, with improving precision & accuracy. These measurements will
continue to improve our understanding of atmospheric composition and chemistry and
provide ground validation for a new generation of satellite-based instruments.
7. References
Barret, B., M. De Maziere, and E. Mahieu (2003), Ground-based FTIR measurements of CO
from the Jungfraujoch: characterisation and comparison with in situ surface and
MOPITT data, Atmospheric Chemistry And Physics, 3, 2217-2223.
Batchelor, R. L., K. Strong, R. Lindenmaier, et al. (2009), A New Bruker IFS 125HR FTIR
Spectrometer for the Polar Environment Atmospheric Research Laboratory at
Eureka, Nunavut, Canada: Measurements and Comparison with the Existing
Bomem DA8 Spectrometer, Journal of Atmospheric and Oceanic Technology, 26(7),
1328-1340.
Bell, W., N. A. Martin, T. D. Gardiner, et al. (1994), Column Measurements of Stratospheric
Trace Species over Are, Sweden in the Winter of 1991-1992, Geophysical Research
Letters, 21(13), 1347-1350.
www.intechopen.com
Bell, W., C. PatonWalsh, T. D. Gardiner, et al. (1996), Measurements of stratospheric chlorine

monoxide (ClO) from groundbased FTIR observations, Journal of Atmospheric
Chemistry, 24(3), 285-297.
Bell, W., C. Paton-Walsh, T. D. Gardiner, et al. (1998), Ground-based FTIR measurements of
stratospheric trace species from Aberdeen during winter and spring 1993/94 and
1994/95 and comparison with a 3D model, Journal of Atmospheric Chemistry, 30(1),
119-130.
Blumenstock, T., H. Fischer, A. Friedle, et al. (1997), Column amounts of ClONO2, HCl,
HNO3, and HF from ground-based FTIR measurements made near Kiruna,
Sweden, in late winter 1994, Journal of Atmospheric Chemistry, 26(3), 311-321.
Bosch, H., G. C. Toon, B. Sen, et al. (2006), Space-based near-infrared CO2 measurements:
Testing the Orbiting Carbon Observatory retrieval algorithm and validation
concept using SCIAMACHY observations over Park Falls, Wisconsin, Journal Of
Geophysical Research-Atmospheres, 111(D23).
Bradford, C. M., F. H. Murcray, J. W. Vanallen, et al. (1976), GROUND LEVEL DETECTION
AND FEASIBILITY FOR MONITORING OF SEVERAL TRACE ATMOSPHERIC
CONSTITUENTS BY HIGH-RESOLUTION INFRARED SPECTROSCOPY,
Geophysical Research Letters, 3(7), 387-390.
Crisp, D., R. M. Atlas, F. M. Breon, et al. (2004), The orbiting carbon observatory (OCO)
mission, in Trace Constituents in the Troposphere and Lower Stratosphere, edited by J. P.
Burrows and A. M. Thompson, pp. 700-709, Pergamon-Elsevier Science Ltd,
Kidlington.
David, S. J., S. A. Beaton, M. H. Anderberg, et al. (1993), Determination of Total Ozone over
Mauna-Loa Using Very High-Resolution Infrared Solar Spectra, Geophysical
Research Letters, 20(19), 2055-2058.
Deutscher, N. M., D. W. T. Griffith, G. W. Bryant, et al. (2009), Total column CO2
measurements at Darwin, Australia - Site description and calibration against in situ
aircraft profiles, Journal of Geophysical Research.
Dils, B., M. De Maziere, J. F. Muller, et al. (2006), Comparisons between SCIAMACHY and
ground-based FTIR data for total columns of CO, CH4, CO2 and N2O, Atmospheric
Chemistry And Physics, 6, 1953-1976.
Donovan, D. P., H. Fast, Y. Makino, et al. (1997), Ozone, column ClO, and PSC
measurements made at the NDSC Eureka observatory (80 degrees N, 86 degrees W)
during the spring of 1997, Geophysical Research Letters, 24(22), 2709-2712.
Dupuy, E., K. A. Walker, J. Kar, et al. (2009), Validation of ozone measurements from the
Atmospheric Chemistry Experiment (ACE), Atmospheric Chemistry And Physics, 9(2),
287-343.
Farman, J. C., B. G. Gardiner, and J. D. Shanklin (1985), Large Losses of Total Ozone in
Antarctica Reveal Seasonal Clox/Nox Interaction, Nature, 315(6016), 207-210.
Frankenberg, C., K. Yoshimura, T. Warneke, et al. (2009), Dynamic Processes Governing
Lower-Tropospheric HDO/H2O Ratios as Observed from Space and Ground,
Science, 325(5946), 1374-1377.
Goldman, A., R. D. Blatherwick, F. H. Murcray, et al. (1979), NEW ATLAS OF IR SOLAR
SPECTRA, Applied Optics, 18(5), 604-605.
www.intechopen.com
Goldman, A., M. T. Coffey, T. M. Stephen, et al. (2000), Isotopic OCS from high-resolution
balloon-borne and ground-based infrared solar absorption spectra, Journal of
Quantitative Spectroscopy & Radiative Transfer, 67(6), 447-455.
Goldman, A., C. P. Rinsland, A. Perrin, et al. (2002), Weak ozone isotopic absorption in the 5
mu m region from high resolution FTIR solar spectra, Journal of Quantitative
Spectroscopy & Radiative Transfer, 74(1), 133-138.
Griffith, D. W. T. (1996), Synthetic calibration and quantitative analysis of gas phase infrared
spectra, Applied Spectroscopy, 50(1), 59-70.
Gunson, M. R., M. M. Abbas, M. C. Abrams, et al. (1996), The Atmospheric Trace Molecule
Spectroscopy (ATMOS) experiment: Deployment on the ATLAS Space Shuttle
missions, Geophysical Research Letters, 23(17), 2333-2336.
Hase, F., T. Blumenstock, and C. Paton-Walsh (1999), Analysis of the instrumental line shape
of high-resolution Fourier transform IR spectrometers with gas cell measurements
and new retrieval software, Applied Optics, 38(15), 3417-3422.
Hase, F., J. W. Hannigan, M. T. Coffey, et al. (2004), Intercomparison of retrieval codes used
for the analysis of high-resolution, ground-based FTIR measurements, Journal of
Haverd, V., G. C. Toon, and D. W. T. Griffith (2005), Evidence for altitude-dependent
photolysis-induced O-18 isotopic fractionation in stratospheric ozone, Geophysical
Research Letters, 32(22), 4.
Irion, F. W., M. R. Gunson, C. P. Rinsland, et al. (1996), Heavy ozone enrichments from
ATMOS infrared solar spectra, Geophysical Research Letters, 23(17), 2377-2380.
Jacob, D. J. (2007), Lectures on Inverse Modeling, edited.
Jones, N. B., M. Koike, W. A. MAtthews, et al. (1994), Southern Hemisphere Seasonal Cycle
in total column Nitric Acid, Geophys. Res. Lett., 21(7), 593-596.
Jones, N. B., K. Riedel, W. Allan, et al. (2009), Long-term tropospheric formaldehyde
concentrations deduced from ground-based fourier transform solar infrared
measurements, Atmospheric Chemistry And Physics, 9(18), 7131-7142.
Kreher, K., J. G. Keys, P. V. Johnston, et al. (1996), Ground-based measurements of OClO
and HCl in austral spring 1993 at Arrival Heights, Antarctica, Geophysical Research
Letters, 23(12), 1545-1548.
Mahieu, E., C. P. Rinsland, R. Zander, et al. (1995), Vertical Column Abundances of HCN
Deduced from Ground-Based Infrared Solar Spectra - Long-Term Trend and
Variability, Journal of Atmospheric Chemistry, 20(3), 299-310.
Mahieu, E., P. Duchatelet, P. Demoulin, et al. (2008), Validation of ACE-FTS v2.2
measurements of HCl, HF, CCl3F and CCl2F2 using space-, balloon- and ground-
based instrument observations, Atmospheric Chemistry And Physics, 8(20), 6199-6221.
Meier, A., and J. Notholt (1996), Determination of the isotopic abundances of heavy O-3 as
observed in arctic ground-based FTIR-spectra, Geophysical Research Letters, 23(5),
551-554.
Meier, A., A. Goldman, P. Manning, et al. (2003), Improvements to Air Mass Calculations for
Ground-Based Infrared Measurements, J. Quant. Spectrosc. Radiat. Transfer,
doi:10.1016/S0022-4073(02)00018-3.
Meier, A., G. Toon, C. Rinsland, et al. (2004), Spectroscopic Atlas of Atmospheric
Microwindows in the Middle Infra-red, Institutet for Rymdfysik, Kiruna, Sweden.
www.intechopen.com
Mellqvist, J., B. Galle, T. Blumenstock, et al. (2002), Ground-based FTIR observations of

chlorine activation and ozone depletion inside the Arctic vortex during the winter
of 1999/2000, Journal Of Geophysical Research-Atmospheres, 107(D20).
Migeotte, M. V. (1948), Spectroscopic evidence of methane in the earth's atmosphere, Phys
Rev, 73, 519-520.
Migeotte, M. V. (1949), The fundamental band of carbon monoxide at 4.7 microns in the
solar spectrum, Phys Rev, 75, 1108-1109.
Murcray, D. G., F. S. Bonomo, J. N. Brooks, et al. (1975), DETECTION OF
FLUOROCARBONS IN STRATOSPHERE, Geophysical Research Letters, 2(3), 109-
112.
Murcray, D. G., A. Goldman, C. M. Bradford, et al. (1978), IDENTIFICATION OF NU-2
VIBRATION-ROTATION BAND OF AMMONIA IN GROUND LEVEL SOLAR
SPECTRA, Geophysical Research Letters, 5(6), 527-530.
Nagahama, Y., and K. Suzuki (2007), The influence of forest fires on CO, HCN, C2H6, and
C2H2 over northern Japan measured by infrared solar spectroscopy, Atmospheric
Environment, 41(40), 9570-9579.
Notholt, J. (1994), The Moon as a Light-Source for Ftir Measurements of Stratospheric Trace
Gases During the Polar Night - Application for Hno3 in the Arctic, Journal Of
Geophysical Research-Atmospheres, 99(D2), 3607-3614.
Notholt, J., and O. Schrems (1994), GROUND-BASED FTIR MEASUREMENTS OF
VERTICAL COLUMN DENSITIES OF SEVERAL TRACE GASES ABOVE
SPITSBERGEN, Geophysical Research Letters, 21(13), 1355-1358.
Notholt, J., I. Beninga, and O. Schrems (1995), SHIP-BORNE FT-IR MEASUREMENTS OF
ATMOSPHERIC TRACE GASES ON A SOUTH (33-DEGREES-S) TO NORTH (53-
DEGREES-N) ATLANTIC TRAVERSE, Applied Spectroscopy, 49(10), 1525-1527.
Notholt, J., G. C. Toon, C. P. Rinsland, et al. (2000), Latitudinal variations of trace gas
concentrations in the free troposphere measured by solar absorption spectroscopy
during a ship cruise, Journal Of Geophysical Research-Atmospheres, 105(D1), 1337-
1349.
Notholt, J., G. C. Toon, S. Fueglistaler, et al. (2010), Trend in ice moistening the stratosphere -
constraints from isotope data of water and methane, Atmospheric Chemistry And
Physics, 10(1), 201-207.
Palm, M., C. Melsheimer, S. Noel, et al. (2010), Integrated water vapor above Ny Alesund,
Spitsbergen: a multi-sensor intercomparison, Atmospheric Chemistry And Physics,
10(3), 1215-1226.
Paton-Walsh, C., W. Bell, T. Gardiner, et al. (1997), An uncertainty budget for ground-based
Fourier transform infrared column measurements of HCl, HF, N2O, and HNO3,
deduced from results of side-by-side instrument intercomparisons, Journal Of
Geophysical Research-Atmospheres, 102(D7), 8867-8873.
Paton-Walsh, C., S. R. Wilson, N. B. Jones, et al. (2008), Measurement of methanol emissions
from Australian wildfires by ground-based solar Fourier transform spectroscopy,
Geophysical Research Letters, 35(8).
Payan, S., C. Camy-Peyret, H. Oelhaf, et al. (2009), Validation of version-4.61 methane and
nitrous oxide observed by MIPAS, Atmospheric Chemistry And Physics, 9(2), 413-442.
www.intechopen.com
Petersen, A. K., T. Warneke, C. Frankenberg, et al. (2010), First ground-based FTIR

observations of methane in the inner tropics over several years, Atmospheric
Chemistry And Physics, 10(15), 7231-7239.
Pougatchev, N. S., B. J. Connor, and C. P. Rinsland (1995), Infrared measurements of the
ozone vertical-distribution above Kitt Peak, Journal Of Geophysical Research-
Atmospheres, 100(D8), 16689-16697.
Rinsland, C. P., M. A. H. Smith, P. L. Rinsland, et al. (1982), Ground-based infrared
spectroscopic measurements of atmospheric hydrogen-cyanide, Journal Of
Geophysical Research-Oceans and Atmospheres, 87(NC13), 1119-1125.
Rinsland, C. P., R. E. Boughner, J. C. Larsen, et al. (1984), Diurnal-variations of atmospheric
nitric-oxide - ground-based infrared spectroscopic measurements and their
interpretation with time-dependent photochemical model-calculations, Journal Of
Geophysical Research-Atmospheres, 89(ND6), 9613-9622.
Rinsland, C. P., R. Zander, L. R. Brown, et al. (1986), Detection of Carbonyl Fluoride in the
Stratosphere, Geophysical Research Letters, 13(8), 769-772.
Rinsland, C. P., N. B. Jones, B. J. Connor, et al. (1998), Northern and Southern Hemisphere
Ground-Based Infrared Spectroscopic Measurements of Tropospheric Carbon
Monoxide and Ethane, Journal of Geophysical Research, 103, 28,197-128,218.
Rinsland, C. P., E. Mahieu, R. Zander, et al. (2000), Free tropospheric CO, C2H6, and HCN
above central Europe: Recent measurements from the Jungfraujoch station
including the detection of elevated columns during 1998, Journal Of Geophysical
Research-Atmospheres, 105(D19), 24235-24249.
Rinsland, C. P., A. Meier, D. W. T. Griffith, et al. (2001), Ground-based measurements of
tropospheric CO, C2H6, and HCN from Australia at 34 degrees S latitude during
1997-1998, Journal Of Geophysical Research-Atmospheres, 106(D18), 20913-20924.
Rinsland, C. P., A. Goldman, E. Mahieu, et al. (2002), Ground-based infrared spectroscopic
measurements of carbonyl sulfide: Free tropospheric trends from a 24-year time
series of solar absorption measurements, Journal Of Geophysical Research-
Atmospheres, 107(D22), -.
Rinsland, C. P., E. Mathieu, R. Zander, et al. (2003), Long-term trends of inorganic chlorine
from ground-based infrared solar spectra: Past increases and evidence for
stabilization, Journal Of Geophysical Research-Atmospheres, 108(D8), 21.
Rinsland, C. P., E. Mahieu, R. Zander, et al. (2004), Free tropospheric measurements of
formic acid (HCOOH) from infrared ground-based solar absorption spectra:
Retrieval approach, evidence for a seasonal cycle, and comparison with model
calculations, Journal Of Geophysical Research-Atmospheres, 109(D18).
Rinsland, C. P., C. Paton-Walsh, N. B. Jones, et al. (2005), High spectral resolution solar
absorption measurements of ethylene (C2H4) in a forest fire smoke plume using
HITRAN parameters: Tropospheric vertical profile retrieval, Journal Of Quantitative
Spectroscopy and Radiative Transfer, 96(2), 301.
Rinsland, C. P., L. Chiou, E. Mahieu, et al. (2008), Measurements of long-term changes in
atmospheric OCS (carbonyl sulfide) from infrared solar observations, Journal of
Rodgers, C. D. (1990), Characterization and error analysis of profiles retrieved from remote
sounding measurements, Journal of Geophysical Research, 95, 5587-5595.
www.intechopen.com
Rodgers, C. D. (2000), Inverse Methods for Atmospheric Sounding: Theory and Practice, World
Scientific Publishing Co. Pte. Ltd, New Jersey.
Rothman, L. S., C. P. Rinsland, A. Goldman, et al. (1998), The HITRAN molecular
spectroscopic database and HAWKS (HITRAN Atmospheric Workstation): 1996
edition, Journal of Quantitative Spectroscopy & Radiative Transfer, 60(5), 665-710.
Rothman, L. S., A. Barbe, D. C. Benner, et al. (2003), The HITRAN molecular spectroscopic
database: edition of 2000 including updates through 2001, Journal of Quantitative
Spectroscopy & Radiative Transfer, 82(1-4), 5-44.
Rothman, L. S., D. Jacquemart, A. Barbe, et al. (2005), The HITRAN 2004 molecular
spectroscopic database, Journal of Quantitative Spectroscopy & Radiative Transfer,
96(2), 139-204.
Schneider, M., K. Yoshimura, F. Hase, et al. (2010), The ground-based FTIR network's
potential for investigating the atmospheric water cycle, Atmospheric Chemistry And
Physics, 10(7), 3427-3442.
Senten, C., M. De Maziere, B. Dils, et al. (2008), Technical note: New ground-based FTIR
measurements at Ile de La Reunion: observations, error analysis, and comparisons
with independent data, Atmospheric Chemistry And Physics, 8(13), 3483-3508.
Strong, K., M. A. Wolff, T. E. Kerzenmacher, et al. (2008), Validation of ACE-FTS N2O
measurements, Atmospheric Chemistry And Physics, 8(16), 4759-4786.
Sussmann, R., and K. Schafer (1997), Infrared spectroscopy of tropospheric trace gases:
Combined analysis of horizontal and vertical column abundances, Applied Optics,
36(3), 735-741.
Toon, G. C., J. F. Blavier, B. Sen, et al. (1999a), Comparison of MkIV balloon and ER-2
aircraft measurements of atmospheric trace gases, Journal Of Geophysical Research-
Atmospheres, 104(D21), 26779-26790.
Toon, G. C., J. F. Blavier, B. Sen, et al. (1999b), Ground-based observations of Arctic O3 loss
during spring and summer 1997, Journal Of Geophysical Research-Atmospheres,
104(D21), 26497-26510.
Vigouroux, C., M. De Maziere, Q. Errera, et al. (2007), Comparisons between ground-based
FTIR and MIPAS N2O and HNO3 profiles before and after assimilation in
BASCOE, Atmospheric Chemistry And Physics, 7, 377-396.
Warneke, T., J. F. Meirink, P. Bergamaschi, et al. (2006), Seasonal and latitudinal variation of
atmospheric methane: A ground-based and ship-borne solar IR spectroscopic
study, Geophysical Research Letters, 33(14).
Washenfelder, R. A., G. C. Toon, J. F. Blavier, et al. (2006), Carbon dioxide column
abundances at the Wisconsin Tall Tower site, Journal Of Geophysical Research-
Atmospheres, 111(D22).
Wolff, M. A., T. Kerzenmacher, K. Strong, et al. (2008), Validation of HNO3, ClONO2, and
N2O5 from the Atmospheric Chemistry Experiment Fourier Transform
Spectrometer (ACE-FTS), Atmospheric Chemistry And Physics, 8(13), 3529-3562.
Wood, S. W., R. L. Batchelor, A. Goldman, et al. (2004), Ground-based nitric acid
measurements at Arrival Heights, Antarctica, using solar and lunar Fourier
transform infrared observations, Journal Of Geophysical Research-Atmospheres,
109(D18).
Yang, Z., R. A. Washenfelder, G. Keppel-Aleks, et al. (2007), New constraints on Northern
Hemisphere growing season net flux, Geophysical Research Letters, 34(12), 6.
www.intechopen.com
Yurganov, L. N., W. W. McMillan, A. V. Dzhola, et al. (2008), Global AIRS and MOPITT CO
measurements: Validation, comparison, and links to biomass burning variations
and carbon cycle, Journal Of Geophysical Research-Atmospheres, 113(D9).
Zander, R. (1976), HIGH-RESOLUTION INFRARED SOLAR OBSERVATIONS BY
BALLOON, Infrared Physics, 16(1-2), 125-127.
Zander, R., G. Roland, and L. Delbouille (1977), CONFIRMING PRESENCE OF
HYDROFLUORIC-ACID IN UPPER STRATOSPHERE, Geophysical Research Letters,
4(3), 117-120.
Zander, R., C. P. Rinsland, C. B. Farmer, et al. (1986), Observation of Several Chlorine
Nitrate (ClONO2) Bands in Stratospheric Infrared-Spectra, Geophysical Research
Letters, 13(8), 757-760.
Zhao, Y., Y. Kondo, F. J. Murcray, et al. (2000), Seasonal variations of HCN over northern
Japan measured by ground-based infrared solar spectroscopy, Geophysical Research
Letters, 27(14), 2085-2088.
Zhao, Y., K. Strong, Y. Kondo, et al. (2002), Spectroscopic measurements of tropospheric CO,
C2H6, C2H2,and HCN in northern Japan, Journal Of Geophysical Research-Atmospheres,
107(D18), -.
www.intechopen.com
22
Earth Scientist’s Guide to Discrete-Time

Power Spectrum Analysis
Bjørn-Gustaf J. Brooks
Center for Climatic Research, University of Wisconsin-Madison
U.S.A.
1. Introduction
Fourier methods are commonplace in the Earth Sciences and have greatly enhanced our
understanding and forecast capabilities for cyclical phenomena that recur on interannual (e.g.
El Niño, Pacific Decadal Oscillation) to millennial scales (e.g. Milankovitch cycles). Nowadays
most low level programming languages (C, Fortran) have math libraries that include the fast
Fourier transform algorithm (FFT) and nearly all abstract programs (Python, Octave/Matlab,
IDL, R) provide an array of Fourier functions for scripting sophisticated signal processing
routines. Whether your interest as a practicing Earth scientist is in Fourier transformation
for efficient data manipulation, or for problems where the Fourier transform or its power
spectrum is needed for direct analysis, you have probably found no shortage of relevant
literature. Nonetheless, you may also have found some difficulty in making sense of which
Fourier methods to implement for your particular analysis idea, and how to appropriately
apply them. This chapter will serve you as a basic guide for unraveling some of the
complicated implementations of discrete-time power spectrum analysis using direct language
and supplementary Matlab/Octave routines using both observed and modeled data.
This chapter assumes that you have a certain task to accomplish, and therefore it is designed
to teach you how to set up an approach appropriate for Fourier analysis, and also to advise
you of potential pitfalls and limitations in Fourier analysis. The reader need not have prior
exposure to signal processing methodologies, but should have a solid base in mathematics,
probability theory and more importantly the issues related to your analysis data so that the
significances of cycles within your data can be rationally interpreted. If you find yourself
lost by the terminology I recommend you familiarize yourself with basic treatments of
discrete-time systems, for example Press et al. (1992) or Cadzow (1973).
The body of this chapter is split into three sections, Preprocessing data, Single Series
Spectrum Analysis, and Multiseries Spectral Analysis. Step-by-step examples are given on the
analysis of a variety of freely accessible earth science datasets covering atmospheric science,
biosphere-atmosphere carbon cycling, climate modeling, and paleodiversity as well as some
example implementations of Markov chain Monte Carlo routines for computing statistical
significances. Each section contains direct explanations with ready to deploy example code
that you are free to use for your own investigations. Supplementary code can be accessed
online from ftp://ftp.climatemodeling.org/pub/esg/.
www.intechopen.com
2. Preprocessing data: Identifying the pertinent information

An often overlooked aspect in the early stages of data analysis and exploration is the
preliminary processing of time series. No matter what your particular topic is it is often
desirable to filter measurements that confound your analysis or do not contribute useful
information. A thorough exploration of the many methods for reducing noise and removing
trends prior to computing the Fourier power spectrum may seem like a lot of work, but it
will be effort well-rewarded because it will increase confidence in your results. Even data
that has already been heavily processed, for example satellite products like leaf area index
(LAI), may still require treatment beyond, or in stead of, monthly averaging, seasonalizing, or
annualizing. This section discusses filtering, subsetting, and detrending topics and provides
example code for deploying this in your own data.
Within any dataset there are data that contribute useful information as well as data that do not,
or even confound your analysis. One example comes from the study of atmospheric tracers of
photosynthesis such as carbon dioxide or carbonyl sulfide. Atmospheric CO2 and OCS have
diurnal and annual cycles that are strongly driven by biotic processes, primarily reflecting the
uptake of atmospheric CO2 by plant photosynthesis and CO2 release through heterotrophic
and autotrophic respiration. Measurements of these tracers should show stronger annual
or diurnal cycles during years when precipitation is substantially influenced by El Niño.
But strong cycles may be difficult to observe unless the observations are preprocessed or
filtered. This is because a series of tower measurements of atmospheric trace gasses will
include both locally representative observations made when turbulent mixing is low and
winds are calm, and regionally representative observations made when the boundary layer
is well-mixed. If your goal were to observe smaller influences caused by interannual ENSO
cycles among much larger biotic influences you may have to detrend from a fitted polynomial
(section 2.1) filter the noise (section 2.2), or subset the data to remove locally representative
measurements (section 2.3), which would result in non-uniform series that can be analyzed
using Lomb-Scargle (section 3.2).
2.1 Detrending
Within geological time series, from ice cores to cyclostratigraphy, it is not uncommon to find
long term and systematic trends. If your objective is to understand something other than these
trends it will be necessary to detrend the data. An example of the importance of detrending
was observed by Cornette (2007) who showed that the prominence of a 62-million-year cycle
in extinctions recorded in the geologic rock record was strongly affected by detrending the
original diversity time series. This is because the most significant variability was caused
by a large non-linear biodiversification trend (Figure 1A) representing three major phases of
biodiversification. Extinction cycles were only apparent in the Fourier power spectrum after
detrending the time series. Likewise if a very long lived and significant harmonic or trend
exists in your data detrending should be considered. Pseudo-code showing how this fossil
diversity time series was detrended appears in Procedure 1 (and also in supplementary code
dftps.m).
2.2 Noise and smoothing filters

Noise adds considerable clutter that can confound Fourier power spectrum (PS) analysis and
reduce the strength of important cycles. Noise is given a broad definition here to refer to
variability in any series that cannot be resolved or directly tied to a predictable physical cause.
www.intechopen.com
Earth Scientist’s Guide to Discrete-Time Power Spectrum Analysis 483
4
0.8 140 Myr 62 Myr
B
0.5 C
Spectral Power (normalized)
Diversity (thousands of genera)
Diversity (thousands of genera)

0.6
A
0.4 0
3 0.2
0
0.01 0.02 0.03 0.04 0.05
-0.5 0.8 140 Myr 62 Myr

2 Frequency (cycles/million years)
0.6
-1 0.4
1 0.2
-1.5 0
0.01 0.02 0.03 0.04 0.05
Frequency (cycles/million years)
0
Cm O S D C P Tr J K Pg Ng Cm O S D C P Tr J K Pg Ng
500 400 300 200 100 0 500 400 300 200 100 0
Time (millions of years ago) Time (millions of years ago)
Fig. 1. Detrending effects. Subplots A and B show diversity without detrending– C and D are
detrended. A) Time series of Sepkoski’s marine fossil diversity covering the past 542 million
years, as reproduced from Rohde & Muller (2005). B) The Fourier power spectrum (PS) of
that series. C) Time series of diversity residuals obtained from A by detrending by a cubic
polynomial (cf. Cornette, 2007). D) The PS of residual diversity showing significant cycles
corresponding to the 62 and 140 million year cycles. These show that in non-detrended fossil
diversity nearly all the power of the 62 and 140 million year cycles is subsumed by the larger
trend in A.
Procedure 1 General procedure for detrending a series by a polynomial curve of order N. The
diversity data in Figure 1A was detrended by a polynomial of N = 3 to produce Figure 1C.
Series x is comprised of nterms measured at increments of t.
input: data series, {(tk , xk )}nterms
k =1
output: data series of residuals, {rk }nterms
k =1
p = poly_fit(t,x,N)
pv = poly_val(p,t)
r = x - pv
It is also important to keep in mind that noise may simply reflect events that are not well
represented by the particular measurement system or point of observation being used. Noise
generally has a limited systematic effect and should be filtered if at all possible.
A consideration when filtering noise is to choose a filter appropriate to the type of noise within
your data (i.e. white, red, blue, gray). White noise can give the Fourier power spectrum plots
(like those in Figure 1B and 1D) broad spectrum noise, which appears as peaks (all of similar
magnitude) dispersed across a range of frequencies in the power spectrum. Least squares
filters including Savitzky-Golay are often used to reduce this kind of broad spectrum noise.
In Matlab/Octave such filtering is achieved by calling the sgolayfilt function. In IDL
calling savgol in conjunction with convol achieves the same purpose.
Confounding variability can also come from red noise, which is not uncommon in
paleoclimate datasets and some long term atmospheric and ecosystem records. Power
spectrum plots of series containing red noise have characteristic slopes that diminish with
increasing frequency (toward the right of the spectral plot), resulting in power spectrum
plots that appear cluttered on left near longer cycles (i.e. where red wavelengths of the
www.intechopen.com
visible spectrum would appear). Red noise can be caused by quasi-stationary processes, for
example a precipitation record with repeated and prolonged multi-year droughts that cause
precipitation to fall below the mean. Red noise can be smoothed by filters that use lower order
polynomials, for example in IDL using savgol with a degree of 2.
Blue noise causes clutter around shorter cycles toward the right of the power spectrum, and
can be filtered using higher order polynomials. Grey noise is identified by its bimodal noise
peaks at both the ‘red’ and the ‘blue’ end of the power spectrum. Several test cases are
provided in the supplementary Matlab/Octave code noise_filt.m, and you should use them
to test the effectiveness of the Savitzky-Golay filter using various filter configurations.
If however you find that it is not possible to effectively filter noise without reducing the
information content of your time series, you may instead evaluate the power spectrum of
the unfiltered data against a stringent statistical test such as a Markov chain Monte Carlo test
(see the random walk test for red noise in section 3.3).
2.3 Subsetting and subsetting filters

Where noise filtering is not tenable or where it is desirable to search for the sources of cycles
you should consider partitioning your time series based on the relative contributions of
various groups (subsetting) or applying a subsetting filter to reject entire observations. Here
I use the term subsetting to indicate partitioning of each value, as opposed to subset filtering,
which I use to indicate rejecting entire observations that exceed some cutoff. Subset filtering
would be relevant for analyses such as in atmospheric observation where it is not appropriate
to smooth a biased measurement because ties to the actual observations are required.
A common issue with observational data is that not all measurements communicate useful
information about the processes you are interested in, therefore it may be necessary to
partition the complete set of observations in some way and to compare the significances of
cycles between subsets. One example subsetting strategy is to partition each value in the
series based on the relative contributions from different groups. Given that you also possessed
meta-data, i.e. the annotations describing each specimen, you could begin to address some
very interesting questions about the nature of cycle. As it happens, the diversity values used
to construct Figure 1A can be partitioned by groups such as deep water fossils vs. shallow
water fossils, or hard-shelled fossils vs. all others. Now we might ask, ‘Is any one group of
organisms responsible for the cycles in fossil diversity seen in Figure 1D?’
Figure 2 uses a similar but more comprehensive dataset of fossil diversity downloaded
from the Paleobiology Database (PD, 2008) to subset the total diversity into multiple curves
representing the contributions from each phylogenetic group, such as mammals, birds, and
reptiles. Figure 2A compares the diversity curve of several dominant shelled fossil animals
(gastropods, bivalves, and articulate brachiopods or GBA) against all other phyla. Figure 2B
shows the Fourier power spectra of those same subsets and suggests that the 62 million year
cycle in this dataset is largely driven by the GBA group.
On the other hand meta-data annotating each observation may not be available to you, and
you may have to create a subsetting filter. Here you would make a decision, for example using
a statistical basis to reject observations from a subset. For example, given a dataset of wind
speed measurements from a meteorological station we might use a subset filter to reject all
measurements of wind speeds less than 4 meters per second, leaving us with a new subset
of unevenly spaced measurements. Non-uniformly sampled data cannot be analyzed using
www.intechopen.com
A B 0.8 140 Myr 62 Myr

Diversity (hundreds of genera)

8
0.6
0.4
4
0.2
0
Cm O S D C P Tr J K Pg Ng 0
500 400 300 200 100 0 0.01 0.02 0.03 0.04 0.05
Time (millions of years ago) Frequency (cycles/million years)
Fig. 2. Using metadata to subset by groups. A) shows two time series of fossil abundances
derived from the Paleobiology Database (paleodb.org) and ranging from 542 million years
ago to present. The first subset in green is comprised of gastropod, bivalve, and articulate
brachiopod fossils (GBA). Their complement (non-members of that group) appear in black.
B) shows the Fourier power spectrum of those same curves when detrended. Notice that a
rather significant peak appears in the GBA group near the 62-million-year cycle frequency,
which is not present in the non-GBA group. This shows that by using the annotations of your
data it is possible to investigate the sources of cycles.
standard Fourier decomposition, but can be processed using the Lomb-Scargle method, which
will be addressed later in Section 3.2).
Subsetting filters utilize a priori knowledge to reject outliers, which means you decide based on
your analysis of the dataset what an acceptable cutoff should be. One way to avoid criticisms
on the subjectivity of your cutoff would be to process your data using a series of cutoffs that
range from very inclusive to very specific, and to do spectral analysis on each differently
filtered subset, and compare.
2.4 Normalizing
A final but critical note on preprocessing is about the importance of normalizing your time
series to a standard deviation of 1 prior to Fourier decomposition. Whether you apply filters
to your data or not it is always necessary to normalize the variance of your original time series
to make it comparable to other series. This is simple to do. Divide each observation by the
standard deviation of the time series. See lines 21–22 in supplementary code dftps.m.
3. Single series spectrum analysis

Once your series of uniformly sampled values has been normalized (perhaps even filtered
or subset) Discrete-time Fourier transform power spectrum analysis (DFTPS) can be used as
a powerful tool for determining the relative strength of cycles within your time (or spatial)
series. You should be aware that Fourier transformation works by decomposing a series
into its complex conjugate of real (ℜ) and imaginary (ℑ) signal parts. The spectral power is
computed as the square of the signal power from ℜ, which produces a series of amplitudes
across a range of frequencies that describe the strength of cycles. ℑ is used to determine the
phase of the cycles, which can tell you when in your time series the peaks of cycles of different
frequencies should occur. These methods can be applied to any series of uniformly sampled
www.intechopen.com
data. If your interest lies in the theory behind spectral analysis of discrete time systems a good
place to start is Cadzow (1973) or Bloomfield (1976). More sophisticated implementations
of signal processing designed for specific purposes (e.g. Brooks, 2009) are often useful when
computational power is limited. But if you have access to a high performance machine or
cluster then you will probably be able to solve big data problems easier using distributed
memory parallelism.
3.1 Computing the power spectrum

As opposed to continuous analog signals, discrete time series are segmented and require a
special step in treatment before transformation by the fast Fourier transform algorithm (FFT).
You must pad your series of normalized values with zeros so that the transformed signal
of your discretely sampled values is sufficiently long to allow for measurement of its peaks
across a range of frequencies (see lines 18-19 of dftps.m).
Zero padding is particularly important when examining cycles with longer periods. Consider
that you have 100 years of monthly means from a model and you want to know how
well the model is able to reproduce interannual variability, caused for example by El
Niño–Southern Oscillation (ENSO, 2-7 yr. period) or the Pacific Decadal Oscillation (PDO,
20-30 yr. period). Despite having cycle lengths that are roughly 20 years apart (ENSO:
∼5 years, PDO:∼25 years), these cycles will show up very near each other in frequency on the
1
power spectrum plot. Since frequency is the inverse of time f ENSO = 12 (months) ×5 (years)
1
= 0.0167, which is near f PDO = 12 × 25 = 0.0033. In order to resolve such small frequency
differences between longer cycles it is necessary to pad the signal with zeros. To better
understand this you might try commenting-out line 18 of dftps.m and substituting it with:
lps=2ˆ (8); (you will also have to adjust the y-axis range). The subsequent plot will be very
coarsely sampled on the left toward longer frequencies because you reduced the padding.
Zero padding will not change the amplitude of your peaks but it will help you better resolve
the frequency and timing of cycles. Fortunately Matlab, Octave, and Python provide handy
methods for padding the series and computing the Fourier transform in one step using the
fft function (Procedure 2). The computation of uniform increments of f to plot your power
spectrum peaks against is also straight-forward (see line 35 of dftps.m).
Procedure 2 General FFT procedure for Matlab, Octave, Python. Given the series x of discrete
values xk the function fft is used to compute its complex conjugate meanwhile ‘stretching’
the signal to length ne, provided that ne > nterms.
input: time series, { xk }nterms
k =1
output: x_ft (complex conjugate with ℜ & ℑ parts), sp (spectral power), f (frequency), p
(period)
x_n = x/std_dev(x) # Normalize to variance of 1
x_ft = fft(x_n,ne) # Transform and pad to length ne
sp = x_ft .* conj(x_ft) # Compute spectral density curve. sp is
# complex and contains real and imag parts
f = flt_intgen(ne)/ne # Generate sampling frequencies for sp
p = 1/f # Cycle periods in original units
www.intechopen.com
Remember that the FFT algorithm will give you complex output: a real part (ℜ), which you
will use to make your power spectrum plot, and an imaginary part (ℑ), which you might use
to determine the phase-synchronization. The imaginary part is not to be overlooked. In order
to determine causality relationships for example, the phase data (ℑ) of the independent and
dependent cycles should coincide in a logical way. Computation of the phase is discussed in
Section 3.4 and Procedure 3.
3.2 Non-uniformly sampled data

The FFT algorithm requires uniform time steps, and cannot be used to analyze non-uniform
series such as in Figure 3A. Fortunately there are alternative ways for dealing with
unevenly sampled data including the Lomb-Scargle (LS) method. The LS approach is a
common implementation, but you should be aware that estimating the period of cycles from
non-uniformly sampled data is not a trivial issue, and precision can vary depending on the
frequency estimators used. An in-depth treatment of spectral analysis for unevenly sampled
data appears in Press et al. (1992), which should be referred to for a detailed explanation.
A B
Net Ecosystem Exchange (µmol m-2 s-1)
20 200
1 Yr 24 Hr
150
0
100
-20
50
-40
Jul-08 Oct-08 Jan-09 Apr-09 Jul-09 0 0.01 0.02 0.03 0.04 0.05
Month-Year Frequency (cycles/hour)
Fig. 3. Series that are not uniformly sampled can be transformed using Lomb-Scargle. A)
Non-uniformly sampled time series of CO2 exchange between the atmosphere and biosphere
(Net Ecosystem Exchange) measured at the Park Falls, Wisconsin tall for June, 2008-Oct.,
2009 (UW-Madison, 2010). B) The spectral power of that data when the LS method is used.
LS implementations (such as the example in supplementary code lsps.m) work by computing

the spectral power across an increasing set of frequencies (usually controlled by a variable
called ofac), up to a frequency limit (hifac). The resulting power spectrum represents an
oversampling of the data (suitable results can be obtained by oversampling rates where ofac
≥ 4). hifac, which is related to the Nyquist frequency, sets the limit of frequencies to be
explored. If, for example, you are not interested in frequencies larger than f = 0.05, then you
should conserve computational time by setting hifac = 0.1.
Some considerations should be kept in mind when specifying your ofac resampling rate. If
you are investigating two narrowly spaced cycles and your series has relatively few broad
gaps, then you would probably be justified in using higher values for ofac that will allow
you to resolve closely spaced cycles. On the other hand if your series is subject to substantial
gaps, large values for ofac are probably not a good idea.
Figure 3B is a power spectrum plot, the same as in Figures 1B, 1D and 2B, except that this
one was computed using the LS method. Figure 3B illustrates an important issue common to
www.intechopen.com
all power spectra: the dispersion of peaks reflecting a cycle with a period that slightly varies
throughout the time series. Minor variations in shorter frequencies, such as those to the right
( f > 0.03) in Figure 3B, can result in either a broader peak or multiple peaks near the frequency
of interest. The primary diurnal cycle should occur at f = 0.0416 in Figure 3B. Instead there
are actually three peaks of nearly equal amplitude, which indicates that the timing of the
diurnal NEE cycle pictured in Figure 3A varies throughout the time series, probably due to
changes in daylight length throughout the year as well as synoptic frontal passages and other
variability that can affect the exchange of carbon (NEE). For example, small shifts in timing of
peak NEE by 1 hour in either direction (e.g. 23-hour/25-hour spacing) would result in large
spreads in frequency ( f = 0.04 to 0.045).
In spectral plots longer cycles have fewer opportunities to occur in a limited dataset and
therefore the certainty about their significance is generally less than a shorter cycle of similar
amplitude. The next section will discuss a method for dealing with uncertainty in the
significance of longer-term cycles using Monte Carlo methods.
3.3 Estimating uncertainty using Markov chain Monte Carlo trials

By this point in the chapter you already know how to identify whether or not a cyclical
pattern exists within your series. If you have already used a script to produce a power
spectrum plot (e.g. dftps.m, lsps.m), you probably asked yourself: ‘How do I know if these
cycles are significant?’ There are a variety of ways to estimate the statistical significance
of cycles within your data and many of them involve Monte Carlo trials. Markov chain
Monte Carlo methods (MCMC) are a widely used class of algorithms that iteratively and
randomly resample (permute) time series, and can be used to test the significances of cycles by
comparing them to randomly derived cycles. MCMC tests allow you to compare the original
time series to many randomized versions of that same data, and to examine whether or not
cycles of equal or greater magnitude exist in the randomized versions. It is important to
understand, however, that Monte Carlo significances do not rule-out bias in your sampling
protocol. Significance tests of this kind can only inform you about how likely a cycle is to
occur given a particular collection of values.
As a starting point for understanding Markov chain Monte Carlo methods I present an
example called random walk trials, sometimes referred to as random step or drunkard’s walk.
Random walk trials can be used to determine statistical significances and are particularly
useful for evaluating low frequency cycles. Many random walk implementations exist,
and here we will keep things simple by using just one implementation, which appears in
supplementary code dftps_mcmc.m.
If you have experimented with the dftps_mcmc.m routine you may have wondered how it
computes significances. The significance calculation used in dftps_mcmc.m is described as
follows: For a given peak of height h at a frequency f , the significance of a peak can be
computed as the fraction, p, of N randomly generated sequences of the original series, from
which the spectrum at f exceeds h. Here you might report the significance of a cycle as the
fraction of trials in which the height of the peak in the original series exceeded the height from
the randomly generated series. Since random walk trials work by randomly permuting the
original series before Fourier decomposition, 10,000 MCMC trials would represent the null
hypothesis that cycles within your data are merely coincidental and can be reproduced given
sufficient randomization.
www.intechopen.com
Uncertainty can be represented not only as a percentage but also in your power spectrum
plot by a line, for example representing the mean spectral power of the MCMC iterations.
Supplementary code dftps_mcmc.m outputs both the significance estimates and a power
spectrum plot with the MCMC random walk significance curve. Note that all peaks that
occur at or below the random walk significance curve should be interpreted as insignificant,
whereas peaks with amplitudes greater than the curve are significant to the degree indicated
by the statistics of the random walk trials.
3.4 Phase synchronization and phase shifted cycles

Two cycles (from different time series) with exactly the same period might seem to be
related. However, their phases could indicate that their peaks in the original time series
are asynchronous, which could indicate a lag in the cause-and-effect relationship, or they
may not be related at all. Let’s say that you have two time series of daily means, one for air
temperature and another for the temperature of a nearby lake. Both records will have strong
1
annual cycles, near f = 365 = 0.0027, but because of the specific heat capacity of water, the
peaks in lake temperature will lag behind air temperature. If you examined the imaginary
part (ℑ) of the Fourier transform at f = 0.0027 you would see the phase shift of the cycle in
the range −π : π radians. (For example find the index number of the element in f on line 29 of
dftps.m that is closest to the frequency 0.0027 and then find the corresponding element by its
index in the imaginary part of xn_fft on line 19 of dftps.m) A good way of visualizing this
is to plot the original series along with a sine wave corresponding to the cycle of interest with
the appropriate phase shift as determined from the imaginary part of the Fourier transform.
This is readily done in Matlab/Octave using angle, and Python using phase, and IDL using
atan, which compute the phase angle shift in radians ([−π : π ]). Pseudocode describing the
construction of the phase shifted sine wave appears in Procedure 3. Note also that this can be
used to extrapolate the continuation of a cycle beyond the observational data, which might be
useful when predicting future cycle peaks.
An example that should give us no trouble comes from modeled temperature data from the
Parallel Climate Model (Washington et al., 2000). Because the complex transform corresponds
perfectly to the magnitude and phase, we need only to convert the real and imaginary parts
back into physical coordinates according to Procedure 3 in order to produce Figure 4.
Procedure 3 General procedure for determining the phase shift and computing the sine wave
corresponding to the period and phase of the cycle of interest. This procedure uses the
imaginary data at index k to compute the phase shift (ps) for constructing the sine wave.
input: { x_ f t}interms
=1 (complex conjugate); p (period); t (time increments in physical units); a
(sine wave amplitude in physical units)
output: ps_k, p_k
ps = atan(x_ft[k])
sw = a * sin(2 * (pi/p) * t + ps) # the phase shifted sine wave
4. Multiseries spectrum analysis

Cause and effect relationships are common, but sometimes difficult to verify. One way
to test whether cycles with the same period from two different time series are related is
www.intechopen.com
A B 250
10 1 Yr
Cospectral Power (normalized)
Monthly Mean Air Temp. (K)

8 240
6
230
4
220
2
0 210
0 0.02 0.04 0.06 0.08 0.1 2010 2012 2014 2016 2018
Frequency (cycles/month) Year
Fig. 4. Power spectrum plot (A) and time series with phase shifted wave (B). The phase shift
of the annual cycle in A is applied to the green sine wave in B, which shows where the peaks
and troughs of the annual cycle should occur according to their average occurrence over 100
years of model data.
through multiseries spectrum analysis. Cross-spectrum analysis is a common implementation

used to compare how well cycles from two different series covary in period and phase
synchronization. Think of cross-spectrum analysis as being the Fourier transform equivalent
of cross correlation. For example, you would expect that net radiation measured at a
meteorological station would correlate well with air temperature. Cross-spectrum analysis
should reveal diurnal and annual cycles that are strong and nearly synchronized.
As opposed to the results of single series spectrum analysis the cross-spectrum density
represents the covariance between two series. Cross-spectrum functions in abstract languages
like Matlab typically compute spectrum density using the Fourier method or Welch’s method,
and their power spectra can be interpreted in the same way, except that the phase data takes
on a heightened importance. Figure 5 was produced using supplementary code csps.m again
using model data for years 2000-2099 from the Parallel Climate Model (Washington et al.,
2000). Figure 5A shows the cross-spectrum density of soil moisture and precipitation for one
arctic land surface grid cell, which signals a very strong correlation in both series of the annual
cycle. Figure 5B represents the phase shift across all frequencies. The frequency of the annual
cycle is located by the vertical gray band, which indicates that the two annual cycles are nearly
π radians (180◦ ) out of phase, meaning that precipitation is high when soil moisture is low.
This seems surprising at first, but remember that in the arctic a majority of the precipitation
comes as snow, which has a delayed release into soils or the soils may be frozen (permafrost)
for most of the year.
5. Summary
There are many worthwhile uses for discrete-time Fourier power spectrum analysis methods,
as shown in this chapter including: 1) correlating the amplitude and phase of similar
cycles from different time series, 2) estimating the statistical significance of cycles, and 3)
investigating the sources of cycles using subsets of the complete set of data. However,
by now you have probably also realized a few of the limitations of power spectrum
analysis such as: 1) very long cycles are difficult to detect given time series of limited
length, 2) inappropriately preprocessing/filtering your data prior to Fourier decomposition
www.intechopen.com
A 3 B
1 Yr 3 1 Yr
Cospectral Power (normalized)
Phase Synchronization (-π:π)

2
2 1
1 -1
-2
-3
0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Frequency (cycles/month) Frequency (cycles/month)
Fig. 5. Cospectrum and phase synchronization. A) shows the total power of the cospectrum
representing a strong cycle in both soil moisture and precipitation. The gray band locates the
annual cycle. B) shows that the phases of the annual cycles are nearly 180◦ (π radians) out of
phase, which underscores the importance of considering the phase data in cross-spectrum
analysis.
can lead to erroneous conclusions, 3) choosing a noise filter is subjective and can lead to
different results, and 4) failing to specify the correct null hypothesis to test against possible
cause-and-effect relationships between cycles (i.e. type iii error) can lead to false conclusions.
Understanding the strengths and weaknesses of Fourier power spectrum analysis as you have
been introduced to here will help you to place the correct emphasis on signal processing
results in your study. If you do find room for such analysis in your work feel free to implement
the supplementary code presented here without reservation and in any way you like.
6. References
Bloomfield, P. (1976). Fourier analysis of time series: An introduction, John Wiley & Sons Inc.
Brooks, B.-G. J. (2009). Applying Wavelet and Fourier Transform Analysis to Large
Geophysical Datasets, Vol. 5545/2009, Springer Berlin, Heidelberg, pp. 426–434. doi:
10.1007/978-3-642-01973-9_47.
Cadzow, J. A. (1973). Discrete–Time Systems: An Introduction with Interdisciplinary Applications,
Prentice-Hall, Inc.
Cornette, J. L. (2007). Gauss-vaníček and fourier transform spectral analyses of
marine diversity, Computing in Science and Engineering 9(4): 61–63. doi:
10.1109/MCSE.2007.76.
PD (2008). The Paleobiology Database, http://www.paleodb.org/. Data were downloaded
on 8 August, 2008. Ichnotaxa and open nomenclature genera were excluded. The
collections reflect 8,821 publications. Lithological categories can be found online at
http://paleodb.org/public/tips/lithtips.html.
Press, W. H., Flannery, B. P., Teukolsky, S. A. & Vetterling, W. T. (eds) (1992). Spectral Analysis
of Unevenly Sampled Data, Cambridge University Press, 2 edition, Cambridge, p. 992.
Rohde, R. A. & Muller, R. A. (2005). Cycles in fossil diversity, Nature 434: supplementary
material 1. http://www.nature.com/ nature/journal /v434/n7030/extref /nature
03339-s3.xls.
www.intechopen.com
UW-Madison (2010). University of Wisconsin Park Falls tall tower NEE data, Website.
Data were downloaded on 24 January, 2011 from http://flux.aos.wisc.edu/
data/wlef/flux/2008/prefnee_2008.txt and http://flux.aos.wisc.edu/data/wlef/
flux/2009/prefnee_2009.txt.
Washington, W. M., Weatherly, J. W., Meehl, G. A., Semtner, Jr., A. J., Bettge, T. W., Craig,
A. P., Strand, Jr., W. G., Arblaster, J., Wayland, V. B., James, R. & Zhang, Y. (2000).
Parallel climate model (PCM) control and transient simulations, Climate Dynamics
16: 755–774. doi: 10.1007/s003820000079.
www.intechopen.com
23
Imaging Fourier Transform Spectroscopy

for Astronomy
Laurent Drissen1, Anne-Pier Bernier1, Maxime Charlebois1,
Alexandre Alarie1, Frédéric Grandmont2 and Julie Mandar1,2
1Université Laval, Québec
2ABB Inc., Québec
Canada
1. Introduction
Unlike most other scientists, astronomers do not have direct access to the objects they are
studying. As a matter of fact, except for a few cases (solar wind and neutrinos, lunar
samples, meteorites, cosmic rays), all information originating from the Universe is
transmitted to us by light. Because it has the ability to interact with matter, it keeps a lasting
impression of the environment where it was born or has had interaction with. One of the
greatest challenges in astronomy is thus to extract, using methods ever more clever, the
maximum information from photons that crossed through space over thousands, if not
billions, of years. A giant step forward was taken nearly 400 years ago when Galileo Galilei
pointed a modest telescope towards the sky. Technological developments have since
considerably increased the dimension and visual acuity of telescopes (segmented mirrors,
adaptive optics), the quantum efficiency of detectors (often close to 90%), the detectable
wavelength range (from radio waves to gamma rays), as well as all the specific
measurement techniques such as photometry, spectroscopy and polarimetry.
There are basically two traditional approaches to obtaining spectral information on
extended astrophysical objects: narrow-band imagery and integral field dispersive
spectroscopy. Imagery with filters allows the observer to map a target in selected
wavelength ranges and to extract the required physical information by comparing the
relative flux of the sources in these bands. This technique is used to obtain color-magnitude
diagrams of star clusters or resolved galaxies (SLOAN ugriz broad-band filters for example),
or to map abundance gradients in nebulae or gas-rich galaxies (using narrow-band
interference filters centered on specific emission lines such as Hα 656.3 nm, [NII] 658.4 nm
or [OIII] 500.7 nm). Images of the targets in the different band passes must be obtained one
after the other with a CCD detector, rejecting each time all photons excluded by the selected
filters (up to 99.8%). Moreover, narrow-band imagery does not provide a high enough
Dispersive spectroscopy with slits allows a much finer spectral resolution (R = λ/Δλ ~ 103 –
spectral resolution to determine the gas velocity.
105) at the expense of spatial information on the targets. Extensively used since the mid-19th
century to obtain the spectrum of individual stars or small slices of extended objects,
dispersive spectroscopy has been transformed by the advent of multi-object spectrographs
www.intechopen.com
(MOS) in the 1990’s: multiple slitlets or optical fibers are positioned at the location of the
targets in a wide field of view, the light of which is then sent to a disperser and recorded on
a CCD. Major breakthroughs have also been obtained in astronomical instrumentation over
the last decade by combining imagery and spectroscopy into a single experimental
observation technique that produces cubes of data. These are typically referred to as Integral
Field Unit (IFU) instruments. Given the limitations of modern array detectors, different
instrument concepts convey different trades that each enhances the possibility of discovery
for a given science program category. The ability to cover a greater area than the classical
spectrometer slit has often been the driving motivation behind these 3-D instrument
developments (Monnet 2009). The use of MOS [Sloan digital sky survey (Stoughton et al.
2002) or 2dF (Colles et al. 2001)] and integral field spectrographs [GMOS-IFU on Gemini
(Allington-Smith et al. 2002) , or VIMOS-IFU (Sanchez et al. 2004)] on large telescopes has
revolutionized data collection by allowing respectively to obtain spectra of a large number
(up to a few hundred) of objects dispersed in a large field or to spatially sample relatively
small (of the order of 10 arcseconds) objects. An integral field spectrograph allowing
observations across a relatively large field field of view (41 x 33 arcseconds at a spectral
resolution R ~ 1000), SAURON (Bacon et al. 2001), has revolutionized the study of late-type
galaxies, and a similar, but much more complex, instrument, MUSE, is being built for the
VLT (Bacon et al. 2010).
The vast majority of imaging spectrometers used on telescope to date however build on
dispersive approaches which must “sacrifice” detector pixels to retrieve the spectral content
instead of scene elements. Typical ratio of distinct scene elements (pixels) to available
detector pixels is on the order of 1/1000. A pure imager would have a ratio of 1 but its
spectral capability is limited by the width of the filters used to select specific spectral
wavebands.
A variety of concepts now propose different balances between field and spectral elements in
terms of both coverage and resolution in order to make the best use of their detector pixels.
By using non-dispersive approaches such as interferometric ones, one can hope to use all
detector pixels for imagery thus prioritizing spatial coverage, resolution or both. The
compromise then usually shifts to the spectral or temporal side as spectra for scene elements
must be acquired in the time domain using multiple exposures. The most familiar
instrument of this kind is probably the Faby-Perot interferometer in which spectral slices of
the final data cube are acquired one by one while mechanically changing the central
wavelength transmitted by the etalon. Although unrivaled for field coverage in the IFU
group, this instrument faces an important waveband width limitation imposed by the
multiple orders transmitted by the etalon, which must be filtered out optically to permit
unambiguous retrieval of the spectral information. Fabry-Perot are mostly used to obtain
high resolution (R ~ 20 000) spectra of individual lines such as Hα, [OIII] 500.7 nm or the
[SII] doublet at 671.7, 673.1 nm (Hernandez et al. 2008, Lagrois & Joncas 2010).
We present in this paper a very brief historical review, as well as the most recent
developments, of another approach , imaging Fourier transform spectroscopy (FTS), which
has been given a strong boost during the past decade, mostly because of enormous
improvements in digital imaging capabilities, computer power and servo control systems. A
large number of research programs would benefit from an instrument capable of
simultaneously obtaining spatially resolved, high quality spectra on extended areas (of the
order of 10 arcminutes) and with a resolution up to R ~ 104. Imaging Fourier transform
www.intechopen.com
Imaging Fourier Transform Spectroscopy for Astronomy 495
spectroscopy is very promising in that regard. Based on the principle of the Michelson
interferometer, Fourier Transform Spectrometers (FTS) are extremely efficient because all
photons are collected and analyzed. Moreover, by using appropriate optical configurations,
it is possible to transform the traditional one-pixel FTS into a truly integral field
spectrometer. In this chapter, we will :
a. Present a brief historical review of the use of FT spectrographs in astronomy;
b. Introduce the concept of an imaging FTS aimed at observing astronomical sources;
c. Use our experience with our instrument, SpIOMM, to illustrate the technical challenges
that must be overcome to ensure the efficiency of such an instrument;
d. Present some of the most interesting scientific results obtained with SpIOMM ;
e. Discuss future developments of wide-field imaging FTS on large ground-based and
space-based telescopes.
2. A very brief history of the imaging FTS in astronomy

Although FTS are most widely used for military and chemical applications, they have also
been very successful in planetary exploration (on board the Mariner, Voyager and more
recently Cassini spacecrafts; Flasar et al. 2004) and in the analysis of the Earth’s atmosphere
(a recent example being the ACE-FTS instrument on board the SCISAT-1 remote sensing
Canadian satellite; Bernath et al. 2005). The use of FTS in astronomy is not widespread,
mostly because of the technical difficulties in building such instruments, but some examples
need to be mentioned. The FTS at Kitt Peak’s Mayall telescope was used the 1970’s and
1980’s to provide exquisite spectra of late-type stars (Scoville et al. 1979, Ridgway et al.
1984). At the Canada-France-Hawaii Telescope (CFHT), the high-resolution FTS was widely
used on a large variety of planetary and stellar programs (Chalabaev & Maillard 1985,
Maillard et al. 1987). Made able to work on an imaging mode in the early 1990’s, the CFHT
FTS was renamed BEAR (Maillard & Simons 1992); it provided integral field spectra of a
variety of objects such as planetary nebulae, massive star clusters and star-forming regions
in a 24 arcsecond field of view (Paumard et al. 2004). Other examples include the FTS built
by D. Naylor (University of Lethbridge) on the James Clerk Maxwell submillimeter
telescope (Naylor et al. 2004, Friesen et al. 2005), an FTS for SPIRE, one of three instruments
to fly on ESA's Herschel Space Observatory (Naylor et al. 2010, White et al. 2010), a far-
infrared FTS on the Japanese satellite AKARI, a mid-IR FTS (CIRS) on the Cassini spacecraft
and for the nar-IR, PFS on Mars Express with a copy on Venus Express. We would also like
to mention another imaging FTS prototype, working in the visible part of the spectrum that
was built at the Laurence Livermore Lab and tested at the 3.5-m Apache Point Observatory
telescope (Wurtz et al. 2002a, 2002b), in which one of us (FG) was involved, but which
development ceased a few years ago. The development of this instrument was a major step
forward to demonstrate the ability of an imaging FTS to acquire hyperspectral images in the
visible band. The advantages and disadvantages of the imaging FTS technique, as well as
the relative merit of different approaches to 3-D imagery are discussed by Ridgway & Brault
(1984) and, more recently, by Bennett (2000).
The development of imaging FTS in astronomy was given a strong incentive during the
early definition phases of the NGST (now known as the James Webb Space Telescope, a 6.5-
m segmented mirror infrared telescope to be launched at the Sun-Earth L2 point in 2014):
astronomers supported by the three participating space agencies (NASA, ESA and the
Canadian Space Agency) presented studies of imaging FTS at the NGST Instrumentation
www.intechopen.com
meeting in Hyannis in 1999 september (Graham 2000, Morris et al. 2000, Posselt et al. 2000) .
None of these concepts however were included in the final instrument suite of the telescope.
More recently, Boulanger et al (2008) proposed the design of a 1.2-m space telescope, H2EX,
equipped with a wide-field imaging FTS specifically aimed at studying molecular hydrogen
in the universe.
A recent review of the imaging FTS concept, with some historical perspective and technical
details not discussed in the present paper, is presented by Maillard et al (2011).
3. The IFTS concept

An astronomical imaging Fourier transform spectrometer (IFTS) is basically a Michelson
interferometer inserted into the collimated beam of an astronomical camera system, equiped
with two detectors. Contrary to conventional integral field dispersive spectrographs, all
detector pixels are used for imagery as with the Fabry-Perot, but instead of acquiring
spectral slices one by one (thus rejecting every other wavelength) to cover the entire
waveband of interest, an "all in one" approach is used. Moreover, while the Fabry-Perot is
limited to a very narrow wavelength range (typically 1 or 2 nm), an IFTS has no waveband
limitations other than the sensitivity of the detectors and the transmitance and reflectance
properties of its optics. Schematically, the core of an IFTS is a Michelson interferometer
consisting of a beamsplitter used to separate the incoming beam into two equal parts; two
mirrors on which the two halves of the original beam are reflected back; a moving
mechanism to adjust the position and orientation of one of the mirrors (the other mirror is
fixed); and a metrology system to monitor the mirror alignement. All wavelengths from the
field are simultaneously transmitted to either one or both of the interferometer outputs in
which the array detector sits. The interferometer is configured to modulate the scene
intensity between the two outputs instead of spectrally filtering it. This configuration results
in a tremendous light gathering power since no light is lost except through items common to
any optical design (substrate transmission, coatings efficiency, quantum efficiency of
detector). All photons from the scene can hence be recorded at each exposure provided that
both complementary outputs of the interferometer are recorded (see below).
Hence, instead of corresponding to a particular slice of the expected spectral data cube, each
exposure populates what is called the "interferogram cube": a series of broadband images of
different intensities. The key to an unambiguous spectral information recovery lies in the
calculation of a Discrete Fourier Transform (DFT or FFT) on each pixel recordings through
the interferogram cube. The vector composed of such a pixel recording is called an
interferogram and is uniquely determined by the spectral content of the light shined on the
pixel. The interferogram cube can at any time during acquisition be turned into a spectral
cube since each acquired broadband image contains information covering the whole
waveband. The inclusion of additional exposures to an interferogram cube simply refines
the meshing of the output spectra (spectral resolution, see section 3.2) and has no effect on
the waveband which is determined by scanning parameters and optics transmission
(including a filter introduced in the optical path to reduce the width of the waveband, if
necessary).
In order to properly generate the interferogram cube, a moving component of the
interferometer must be precisely positioned at predetermined sequential interference
position before each exposure can be recorded. This operation is referred to as scanning
through the interference patterns. For the IFTS, the scanning parameter is the Optical Path
www.intechopen.com
Difference (OPD) existing between both arms of the interferometer and the moving
component a mirror. The DFT calculation assumes that all data points of the interferogram
vector are acquired at equidistant OPD intervals. Deviations from this assumption result in
an increased noise level or artifacts in the resulting spectra. The performance of an IFTS
instrument is thus tightly linked with the performance of its OPD scanning system, which
can be very challenging in the visible band (350 - 850 nm). The technical challenges
associated with building an efficient Michelson interferometer in this wavelength range are
partly responsible for the absence of a widespread use of IFTS in astronomical observatories
today. In practice however, a lot of the recent IFUs require increasingly complex data
processing software such that this apparent distance to the final data tends to even out
among 3D capable instruments. The main IFTS limitation that remains in the classic IFU
selection trade-off is the balance between spectral resolution desired and acquisition time (or
number of exposures required). Another potential hurdle is, we think, conceptual. While
dispersive spectroscopy is very intuitive, to a point where most highschool students have
experienced the use of a prism or a dispersive grating, and understood the process giving
rise to a rainbow, Fourier transforms are non-trivial mathematical concepts standing
between the data acquisition and the desired spectrogram. The fact that a given pixel
recording cannot be directly related to a given spectral point typically leaves observers with
a somewhat less tangible feeling for the data which must be addressed by the careful design
of a comprehensive user interface.
3.1 Spectroscopy with a Fourier transform spectrometer

To explain how a Fourier transform spectrometer works, we first assume that the
interferometer at the core is a classical Michelson, that the light coming to the telescope is
monochromatic, such as a laser beam , and that the signal is recorded on a single-pixel
detector (see Fig. 1). The incoming beam is first split into two equal parts by a beamsplitter.
Fig. 1. Optical configuration of a classical Michelson interferometer at the core of a Fourier

transform spectrometer
www.intechopen.com
Half the light is transmitted through the beamsplitter, bounces back on a moving mirror and
interferes, in the beamsplitter, with the other half-beam which has, in the meantime, been
reflected by the beamsplitter to a fixed mirror and bounced back. Initially, the optical path
travelled by the two beams are the same; we are at the Zero Path Difference (ZPD) position.
The two beams are in perfect phase when they combine and the interference is completely
constructive: the detector receives the sum of the two beams while none of the light goes
back to the source. The moving mirror is then slightly displaced (typical displacements vary
between 175 nm and a few micrometers, depending on the wavelength range and the
desired spectral resolution), creating a small offset in optical paths between the beams.
These not being in phase anymore, the interference is not entirely constructive and the
detector receives a bit less light while the difference goes back to the source.
Fig. 2. Simulations of interferograms from different light sources, where step 0 corresponds
to the ZPD. Upper left: single He-Ne (632 nm) laser, sampled with mirror displacement
steps of 175 nm; the sinusoidal pattern is clearly seen. Upper right: mercury doublet (577 nm
and 579 nm); the beating between the two frequencies is obvious. Lower graph: a continuum
5000 K blackbody source; most of the action occurs near the ZPD
Eventually, after a sufficiently large mirror displacement, the interference will be totally
destructive: no light will be recorded on the detector, all the beam will go back to the source.
In practice, an interferogram is obtained by first moving the mirror to a predetermined
position away on one side of the ZPD, moved by equidistant steps to the same position on
the other side of the ZPD. Since the incoming light source is monochromatic, the temporal
www.intechopen.com
signal pattern received by the detector after a large number of mirror displacements,
called an interferogram, will be sinusoidal. The original frequency, or wavelength, of the
laser light beam is thus recovered by calculating the Fourier transform of the time-
dependent signal recorded by the detector. An incoming beam which includes two or
more emission lines will produce a more complex interferogram, while the interferogram
pattern of a continuum source will be mostly concentrated near the ZPD. Astrophysical
objects, such as planetary nebula, ionized gaseous nebulae or galaxies present complex
spectroscopic features (many emission lines produced by ionized gas superimposed on a
stellar continuum) which produce very diverse types of interferograms. Again, a Fourier
transform of these interferograms will recover the original spectra, in terms of both
frequency and intensity. Fig. 2 illustrates the shape of typical interferograms from
different artificial sources.
3.2 Instrument line shape

What does a spectrum obtained in such a way look like? In principle, the Fourier transform
of a continuous, infinite interferogram will provide exactly the same frequencies and
amplitude as the incoming beam. If the light beam to be analyzed is monochromatic, the
interferogram will be a sine wave. Assuming that this interferogram is continuous and
infinite, its Fourier transform will be a delta function. However, in real life, the
interferogram is sampled at discrete, and in principle equal, intervals of optical path
difference: the intensity (or, in the case of an imaging FTS, an image) of the source is
obtained, then the mirror is moved, another image is obtained, and so on. Moreover, the
signal is observed for a limited time and thus the interferogram is sampled with a finite
number of data points. For example and as we shall see later, for a typical astronomical
scene observed in the 650 – 680 nm range, the interferogram is sampled with a mirror step of
about 5 micrometers and 300 steps for a total optical path difference of ~ 3 mm (the optical
path corresponds to twice the mirror translation). Therefore, the recovered line shape will
be the “ideal” incoming line shape, degraded by a function corresponding to the
“imperfect” instrument. In mathematical terms, the limited spectral resolution of the final
spectrum is caused by the convolution of the original line shape with a function that
depends on the properties of the instrument and the sampling technique. Using the
definition of the Fourier transform and its properties (seen elsewhere in this book), we see
that:
f ( x ) : perfect, infinite interferogram

g(x): instrumental function
∫
∞
FT ⎡⎣ f ( x ) ⋅ g( x )⎤⎦ = f ( x ) ⋅ g( x ) e − iω x dx = F(ω ) ⊗ G(ω )
−∞
where F(ω ) = FT ⎡⎣ f ( x )⎤⎦ and G(ω ) = FT ⎡⎣ g( x )⎤⎦
Let’s first have a look at the effect of a finite interferogram on the instrument line shape. In
this case, the observed interferogram can be seen as the product of an infinite interferogram
(a sine wave for a monochromatic light source) with a square box function, g(x) = sb(x),
having an intensity of 1 between –d and +d (the maximum optical path difference on both
sides of the ZPD) and 0 everywhere else. The instrument line shape (ILS) will thus be
determined by the Fourier transform of the square box:
www.intechopen.com
ILS = FT ⎡⎣sb ( x ) ⎤⎦ = ∫ sb ( x )e iω x dx = ∫
∞ ∞
e iω x
e iω x dx =
d
−∞ −∞ iω
{ } { }
−d
cos (ω d ) + i sin (ω d ) − cos (ω d ) − i sin (ω d ) ⎤

1 ⎡
= ⎡ e − e − iω d ⎤ =
1 iω d
iω ⎣ ⎦ iω ⎣ ⎦
2i sin (ω d ) sin (ω d )
= = 2d
iω ωd
Fig. 3. FTS instrument line shape of a monochromatic source for which the interferogram
was truncated by a limited observing time. The recovered spectrum is the convolution of a
delta function with the sinc function, which is the Fourier transform of a square box
The instrument line shape generated by the fact that the interferogram is bounded is thus a
the convolution of a delta function F(ω) with the sinc function G(ω) (see Fig. 3): a sinc
sinc function. If the observed light source is monochromatic, the recovered spectrum will be
function centered at the wavelength of the incoming light source. Let’s now determine the
width of this function, which dictates the spectral resolution attainable with an FTS, by
calculating the half width at half maximum of the ILS:
sin (ωhwhm d ) sin (ωmax d )

= 0.5 × 2 d
ωhfhm d ωmax d
2d
;1 − + − ... , we see that the maximum of this function, 1,

sin x x2 x4
Using the fact that
occurs at ωmax = 0 . Therefore,

x 6 120
sin (ωhwhm d ) = 0.5 ⇒ ωhwhm =

1.9
d
www.intechopen.com
The larger d is, the thinner the ILS becomes. Spectral resolution of an FTS is thus directly
proportional to the total optical path difference sampled by the interferometer. The
maximum spectral resolution attainable with an FTS is then set by the maximum
displacement of the moving mirror within the interferometer. In an imaging FTS, other
factors, such as image quality (defined by the instrument optics or, more likely, the seeing
disk blurred by convection in the atmosphere) or image sampling can decrease the
theoretical spectral resolution of the instrument. Figure 4 illustrates the typical line shape of
a monochromatic source obtained with an imaging FTS. It is possible to modify the shape of
the ILS a posteriori by multiplying the interferogram by a function that would attenuate the
presence of the “side lobes” associated with the sinc function; one then talks about
apodization. One of the most widely used apodization functions is the Gaussian, since it is
the only function whose frequency content is the same as the function itself; the Fourier
transform of a Gaussian is a Gaussian. The counterpart of apodizing the interferogram
however is a loss of resolution. In our typical astronomical applications, the ILS sidelobes
disappear in the noise after an apodization with a Gaussian that decreases the spectral
resolution by about 25%.
Fig. 4. Spectrum of a He-Ne laser (632 nm) obtained with the imaging FTS SpIOMM,
showing the typical instrument line shape
3.3 Modulation efficiency

Even if a good transmission is achieved in the optical design, contributing positively to the
image’s signal-to-noise ratio, it does not necessarily translate in a good performance for
spectroscopy. In order to perform well on this aspect, a good modulation efficiency is also
required. The performance of an FTS is thus characterized by its modulation efficiency (ME),
i.e. the capability of the interferometer to modulate the incident light:
I ( modulated light )
ME =
I ( incoming beam )
www.intechopen.com
This parameter can be viewed as an analog to the grating efficiency in dispersive

spectrographs. In the worst case scenario, where the modulation efficiency is zero, the light
from the source is recorded on the detector but the interferogram is a straight line and no
spectral information can be extracted from it.
This efficiency depends on a multitude of factors, the most important being the following:
intensity over the whole wavelength range (ME ∝ 4RT, where R and T refer to the
1. The capability of the beamsplitter to separate the incident beam into two beams of equal
reflectance and transmission of the beamsplitter). The modulation efficiency is however

relatively permissive to deviation from the 50% perfect case. For example, a 60% - 40%
R-T beamsplitter can still generate modulation efficiency near 96% while a 70% - 30%
one would limit the performance to 84%.This is usually not a big problem over a small
wavelength range, but it becomes a challenge if the FTS covers, for example, the entire
visible range (from 350 to 900 nm).
2. The surface quality of the optical components in the interferometer (mirrors and
beamsplitter). Three conclusions are obvious from Fig. 5, where we present the
influence of the surface quality on the ME for different wavelengths and optical
configurations. At a given wavelength, ME is lowered by a decreased surface quality; it
is more and more difficult to obtain a good ME as we move from the infrared to the
ultraviolet (most FTS commercially available indeed work in the infrared); and the
Mirrors with a surface quality of λ/20 (peak-to-valley) are commercially available for a
number of reflections within the interferometer plays a major role in the global ME.
reasonable price, but large λ/30 mirrors must be custom made and are therefore much
more expensive. Moreover, even if the mirror substrate is of high enough quality, any
error in the coating deposit or any tension caused by the mechanical parts used to
maintain the mirror within the interferometer can ruin the initial surface figure and
dramatically reduce the modulation efficiency, especially in the blue part of the spectral
range.
3. Homogeneity of the refraction index within the interferometer cavity. Any convection
or temperature inhomogeneity in one or both arms can alter the OPD and therefore
lower the ME, since the velocity of light depends on the refraction index of the material
in which it travels.
4. The mirror alignment. In order for the beams from the two arms to interfere properly,
the two mirrors need to be very well aligned. The smallest deviation, in any direction,
from a right angle between the two mirrors reduces the spatial coherence (interference)
of the two beams as they recombine. Again, this effect is more obvious at small
wavelengths. A deviation of only 1.5 microradian from perfect alignement can decrease
the ME by up to 25% at 350 nm.
5. Stability of the OPD during an exposure. The optical distance between the two mirrors
must be kept constant during an exposure at a given step. An OPD jitter with a
standard deviation of 10 nm typically reduces ME by 1 to 2%.
The metrology and servo system play a crucial role with regard to the last two points, since
the mirrors must be aligned with a precision of less than a microradian and their distance
kept constant to within a few nanometers during an exposure. As we shall see later,
monitoring the distance between the two mirrors as well as their alignment many thousand
times per second, and a fast correction of any deviations, are required to ensure a constant,
high modulation efficiency.
www.intechopen.com
configurations and mirror quality. Peak-to-peak asperities of λ/20 and λ/30 are considered,
Fig. 5. Wavelength-dependence of an FTS’s modulation efficiency for different
as well as three mirror configurations: flat mirrors (FM; one reflection on each arm), roof-top
mirrors (RT; two reflections), and cube corners (CC; three reflections)
3.4 Four ports design

In a traditional Michelson interferometer, half the light goes back to the source after it
interferes with the other beam. But in astronomy, there is so little flux coming from distant
nebulae and galaxies that every photon counts; moreover, returning back half of the photons
coming from a galaxy after it had crossed millions of light-years seems absurd. A third
factor also needs to be considered in astronomical observations: the sky transparency. Any
change in the sky transparency during the observations, caused by thin clouds or a change
of airmass, will affect the interferogram, introducing noise or even artefacts in the recovered
spectrum. But since the sum of both outputs from the interferometer is, in principle,
constant during he whole data acquisition, any fluctuation in the total signal can only be
interpreted as a sky transparency variation and be corrected for.
Therefore, the Michelson interferometer used in an astronomical IFTS must be designed in a
way that allows the two half-beams to be recovered after having interfered with each other.
Such configurations, with two output ports, are usually achieved using cube corner
retroreflectors or cat's eye mirrors (Maillard & Simmons 1992, Boulanger et al. 2008). If one
wants to produce an interferometer that strongly modulates the visible light down to the
near UV, the optical surfaces located between the two beamsplitter passes must exhibit
excellent surface figures (< /20) to allow a good interference of the two recombining
wavefronts, as seen in the previous section. The two mirror options mentioned above
complicate this task because of the additive errors of all three reflections encountered (see
Fig. 5) in addition to optical shape consideration in the case of cat's eyes or orthogonality
issues in the case of cube corners. The flat mirror interferometer, besides providing a better
throughput than the others and improved wavefront accuracy, can also provide a two
output-ports configuration when used off axis as shown on Fig. 6. There are however two
disadvantages of using flat mirrors. First, the alignment of the two mirrors must be very
carefully controled by a servo control system (cube corners are insensitive to slight
misalignements); this puts serious constraints on the metrology (which measures the
www.intechopen.com
distance between the two mirrors and their alignement) and piezo actuators which must
respond very rapidly and accurately to any misalignement. Second, since the center of the
interferometric pattern is not aligned with the detector’s center, the ultimate spectral
resolution attainable by the FTS is lower than that obtained with cube corners. As we shall
see, neither of those constraints is a show stopper for practical purposes.
Fig. 6. Interferometer design of an of-axis imaging FTS
3.5 An imaging FTS in the visible band

By adding input and output optics (a group of collimating lenses before the interferometer,
and one group of imaging lenses for each output port) to image a wide field of view and a
multi-pixel detector such as a CCD at each output port, the standard Fourier transform
spectrometer becomes an imaging FTS. Since most Integral Field Spectrographs used in
astronomy today cover a small field of view (typically a few arcseconds on a side), an
enormous advantage of the imaging FTS is its ability to cover a much wider (arcminutes)
field. As we shall see in the next section, a large number of scientific programs benefit from
an instrument capable of simultaneously obtaining spatially resolved, high quality spectra
of extended (of the order of 10 arcminutes) sources.
The basic design of the optical system surrounding the interferometer is very similar to that
of conventional astronomical cameras and focal reducers, but with an additional constraint
on the optics. Astronomical imagery is performed with filters which select specific
the SDSS u’g’r’i’z’ system, with a typical bandwidth Δ λ ~ 150 nm) or on specific emission
bandpasses aimed at extracting information on the color of objects (broadband filters such as
www.intechopen.com
lines (narrowband filters centered, for example, on the Hα line, with a typical bandwidth Δ
λ ~ 2 nm) to highlight the ionized gas in galaxies. These cameras usually observe one band
at a time allowing for refocus between each of them, thereby relaxing the constraints on the
optical design. The challenge for the optical design of an imaging FTS, which can in
principle accept a very wide wavelength range at every single step is to satisfy the image
quality requirement over the whole waveband throughout the field of view.
Two more constraints on detector technology are imposed by the imaging FTS concept and
need to be discussed because they were partly responsible for the absence of wide FOV
imaging FTS up to recently and still drive the choice of the detector today. In order to build
an interferogram, an exposure of the scene must be obtained at every mirror step and read
out from the detector. In the visible range, the most efficient detector is without contest the
charge-coupled device, or CCD. An electronic readout noise is added to the shot noise from
the scene every time the CCD is read out. Moreover, it takes time to read an entire CCD and
register its content on a hard disk, from a few seconds up to one minute, depending on the
readout rate and the number of pixels. In order to minimize the « dead time » due to this
transfer of information, the CCD readout rate ought to be very high. But the readout noise is
proportional to the readout rate, so the two problems are linked. In the 1990’s, the largest
CCDs had 1024 x 1024 pixels ; their readout time was around one minute and the readout
noise was about 10 electrons. Nowadays, 2048 x 2048 pixel CCDs are common, and many of
them are equipped with multiple amplifiers allowing the simultaneous reading of its four
quadrants at speeds up to 1 MHz : the readout time of the entire detector is reduced to a
couple of seconds and the readout noise is down 2 to 5 electrons. In the near future, zero-
noise EMCCDs (Daigle et al. 2009) will replace conventional CCDs for this type of
application and will enable speed-scanning of the OPD instead of the step-scan approach;
this will at the same time increase the signal-to-noise ratio at very low flux level and lower
the dead time due to the CCD readout.
4. SpIOMM, a wide-field imaging FTS for the Mont Mégantic Observatory

In 1999, the NGST science and technology exhibition meeting was held in Hyannis,
Massachusetts, to present and discuss, among other things, the best possible suite of
instruments to be attached to the giant space telescope now known as the James Webb Space
Telescope (JWST: www.jwst.nasa.gov). Three teams then proposed concepts of IFTS that
would allow the James Webb to acquire hyperspectral images over large field of view
(Graham 2000, Morris et al. 2000, Posselt et al. 2000). The scientific rationale and a
preliminary technical performance study of a space-based IFTS had been published a year
earlier (Graham et al. 1998). Although the enormous qualities of such an instrument were
recognized by the panel of scientists and engineers who had to select the JWST instrument
suite, the IFTS solution was not selected. One of the reasons invoqued was that an IFTS is
competitive with a conventional spectrograph only if the number of sources in the field of
view is large. To quote the report of the NGST ad hoc working group on the near-infrared
spectrograph study (Huchra et al. 1999) about IFTS:
In addition to preserving the diffraction limit of the telescope, this technique allows unbiased
spectroscopy since there is no preselection of objects. FTS instruments trade their spatial
multiplex gain against the spectral multiplex advantage of grating instruments. Thus the
richness of the observation scene largely dictates the advantage of an FTS against dispersive
designs. The FTS is most efficient for low spectral resolution of extremely crowded fields.
www.intechopen.com
We would add to the last sentence "or at moderate to high resolution, over selected spectral
wavebands, of extended sources". Indeed, a dispersive spectrograph is more efficient than
an FTS for the observation of single sources. Multi-object dispersive spectrographs are also
more efficient than IFTS if only a limited number of sources (say, a hundred) are spread
over a large area of the sky. This was indeed one of the arguments mentioned to reject IFTS
for the JWST: since one of the most important goals of this telescope is to observe the faint,
distant Universe for cosmological studies, the number density of sources (galaxies) bright
enough for spectroscopic studies was not expected to be large enough to justify the use of
IFTS. However, an IFTS becomes the instrument of choice if the target spans a fair fraction
of the entire field of view, which is the case for Galactic nebulae and nearby galaxies. This
led us to our desire to design and build SpIOMM.
4.1 Science case and science-based requirements

In order to demonstrate the capabilities of a wide-field imaging FTS working in the visible
band, our group has designed and built SpIOMM (Spectromètre Imageur de l’Observatoire
du Mont Mégantic) at Université Laval in close collaboration with ABB-Bomem, a Québec
City-based company specialized in the development and manufacturing of commercial FTS.
The 1.6-m telescope of the Mont Mégantic Observatory (OMM), jointly operated by
Université Laval and Université de Montréal since 1978, is perfectly suited to train highly
qualified personnel such as M. Sc. and Ph. D. students, postdoctoral researchers and
engineers, but is also a perfect test bed to design, build and use innovative astronomical
instruments such as polarimeters (Manset & Bastien 2002), infrared cameras and
spectrographs (Artigau et al. 2009), low-noise detectors for F-P systems (Daigle et al. 2009) as
well as imaging FTS such as SpIOMM.
The primary objective of any astronomical instrument development being to address a
science case, the development of the IFTS at OMM was fed by the desire to acquire complete
data sets to feed existing science projects already ongoing at Université Laval on the
interstellar medium, late stages of stellar evolution, star formation and galaxy evolution and
which could not be obtained with the current instrument suite of the observatory. The
general theme of “cosmic recycling” was our motivation since the beginning: how do
chemical elements such as oxygen, carbon or nitrogen, which are produced by
nucleosynthesis in stellar cores, get transferred via stellar winds and supernova explosions
to the interstellar medium where they “pollute” giant molecular clouds which eventually
contract under their own gravity and form new generations of stars, surrounded by planets
where life can eventually appear and evolve? How does this process occur in our own
galaxy, the Milky way, and in other galaxies? Can we trace the past evolution of galaxies by
measuring the spatial distribution of their heavy elements? Obtaining spatially resolved
spectra of extended objects such as ionized nebulae, supernova remnants, planetary nebulae
and nearby galaxies is a prerequisite to answer these questions and an instrument such as a
wide-field imaging Fourier transform spectrometer could provide some answers.
The work started by first identifying the measurement aspect that could benefit the most
from the IFTS concept. In this respect the instrument development deviated early on from
the JWST instrument concepts and other past IFTS development (such as BEAR) due to the
waveband of interest for our particular research: the visible range (from 350 nm to 850 nm)
is very rich in diagnostic emission lines allowing us to determine the physical characteristics
of ionized nebulae, such as their chemical composition, density or temperature. Theoretical
models abound which relate emission line ratios in this wavelength range to the properties
www.intechopen.com
of nebulae and their ionizing stars (Kewley & Dopita 2002). Moreover, spectra of small
sections of extended objects are available for a direct comparison with those obtained with
an IFTS. But choosing this wavelength range has its drawbacks from an instrumental point
of view, as we have seen in section 3.3. It can easily be understood that controlling
interference of light mechanically is better achieved at longer wavelengths at which a given
fraction of a wave corresponds to a larger displacement hence easier to detect and correct.
As such, FTS users have traditionally had more success when operating in the IR and far-IR
although usage down to far-UV are documented in the literature in other fields of science.
Considering the advances in nanometer level precision actuators, high stability metrology
lasers and electronics speed, new tools are now available to tackle challenge of the past.
Most importantly, the expertise with laboratory demonstration instruments at ABB Bomem
and the NGST breadboard (Wurtz et al. 2002a) on which members of our team had closely
worked gave us a sufficient level of confidence that we could succeed in the visible range.
From our point of view, the niche of an astronomical IFTS clearly sits in the wide field
coverage, moderate spectral resolution (R ~ 2000 -10 000) and the large waveband, although
it is interesting to note that the FTS was classically known for its ability to obtain very high
spectral resolution. Using a FTS in an imaging configuration typically does not affect its
spectral potential, but the faintness of the target diffuse nebulae make the high resolution
measurement difficult to reach in a reasonable time. However, one can still exploit the high
spectral resolution capability of an IFTS by reducing its waveband in order to proportionally
reduce the number of exposures required to obtain a given spectral resolution.
A set of science-based requirements were established as a starting point to design the
•
instrument:
Field of view - Wide field, of the order 10 arcminutes or more; this corresponds to the
size of some of the largest Galactic planetary nebulae and to a fair number of nearby
•
galaxies of interest, up to the Virgo cluster.
Wavelength range - Two important diagnostic emission lines in the spectra of ionized
nebula define the useful spectral range: [OII] 372.7 nm in the near ultraviolet and [SII]
673.1 nm in the red. Allowing for some redshift due to the expansion of the Universe,
which affect galaxies or group of galaxies within reach of the OMM, a lower limit to the
•
red was set to 750 nm.
Minimal spectral resolution – Should be high enough to separate Hα (656.3 nm) from
doublet: R = λ/Δλ > 500. Kinematical study of expanding nebulae and galaxies require
the [NII] (654.8, 658.4 nm) doublet, as well as both members of the [SII] 671.7, 673.1 nm
a higher value of R ~ 2000.

A trade study was then performed to define the best design that would meet these
requirements (Grandmont 2007), which led to the current design, described in the next
section, and construction of SpIOMM in 2002 – 2004. Details about the design and tests of
SpIOMM can be found in a series of SPIE papers (Grandmont et al. 2003, Bernier et al. 2006,
Bernier et al. 2008).
4.2 Description of the instrument

SpIOMM is capable of obtaining spectra in selected bands of the visible spectrum (from 350
to 850 nm) of every light source in a 12 arcminute (circular) field of view of the OMM
telescope; a 12’ x 12’ FOV is actually recorded on the detector but optical abberations are
important in the corners of the field. The spectral resolution is variable, depending on the
www.intechopen.com
need of the observer, from R = 1 (broad-band image) to R = 25 000. The spatial resolution is
limited by the seeing disk (atmosphere blur), which is typically 1 – 1.5 arcsecond. The dual
output design of SpIOMM (see below and Fig. 6) ensures, in principle, that virtually every
photon collected by the telescope reaches the detectors and is analyzed. We have however
CCD camera with an array of 1340 x 1300 20 μm-pixels (corresponding to 0.55 arcsecond on
worked until now, for budgetary reasons, with only one detector, a Princeton Instrument
CCD, an Apogee Alta 2048 x 2048, 15 μm-pixels that will collect the flux from the second
the sky). A data cube thus results in 1.7 million spectra. But we recently acquired a second
output port. SpIOMM is unique in the world, offering the largest field of view of any
integral field spectrograph. After five years of research and development, we have
demonstrated that the concept behind SpIOMM is sound and viable, and that such an
instrument is capable of producing high quality hyperspectral data cubes over a very
extended field of view.
The design (shown on Fig. 6) is based on the Michelson interferometer with a 30 degree
incidence angle on the beamsplitter-compensator assembly, which minimizes their circular
size. The use of two plane mirrors reduces the number of optical surfaces encountered by
the science beam from the telescope output to the camera and therefore grants better
throughput. Also, the instrument configuration places the incoming beam 8 degrees off–axis
perpendicular to the optical interferometer plane, which allows access to the two output
ports. The interferometer uses a dynamic alignment control (metrology laser and
piezoelectric actuators) and step-scan operation. The robustness of an interferometer
operating in the visible range is of paramount importance. FTS are known to serve as good
microphones, seismographs or even thermometer, since they are extremely sensitive to
vibrations and temperature changes. The stabilization of a moving mirror at a small fraction
of the shortest wavelength accepted translates in this case to nanometer level precision. Any
change in temperature, gravity orientation or vibration is at this scale bound to have a
noticeable effect on a macroscopic scan mechanism. Hence much care must be taken in
either passively reject some of these perturbations through a stiff and athermal design or
actively through a well tuned servo control system. SpIOMM’s moving mirror is mounted
on a parallelogram porchswing-type mechanism which offers frictionless displacement. It is
actuated by a piezo-based stepper motor mounted in series with a small range high-
frequency response piezo (the OPD piezo). The stepper actuator can travel a Maximum Path
Difference of 1 cm, corresponding to an average spectral resolution of R ~ 25 000. In practice
however, we have never pushed SpIOMM to this resolution limit. The ensemble allows a
high stiffness (10 N/ m) in order to ensure good passive stability of the Optical Path
Difference (OPD). A dynamic alignment system using piezoelectric (DA piezo) combined
with the mirror displacement mechanism is required for precise positioning and to evenly
sample the interferogram. Also, this maintains the mirror position and alignment subject to
gravity, vibration and thermal perturbation during data acquisition. The interferogram is
sampled at a step size determined by the Nyquist criterion. In order to measure the
wavelength modulation at 350nm, the scan step is 175 nm. When it is set to a value larger
than 175nm, it has to be used with an appropriate filter for the desired band.
The moving mirror position and alignment are detected by a metrology system and
acquired by a servo-computer at a rate of 8 MHz. This feedback is given by an infrared 1550
nm laser beam expanded before passing through the interferometer. We take advantage of
the fact that the telescope’s primary mirror has a hole in its center : the image of the science
www.intechopen.com
beam in the interferometer therefore shows a hole where the metrology beam is sent. The
metrology and science beams are therefore completely separated, both physically and in the
frequency domain (since the CCD is completely insensitive to the laser’s wavelength). The
metrology beam passes through the same optical path as the science beam and strikes the
center of the fixed mirror, where a thin circular layer of glass has been deposited. This layer
acts as a wave-retarder which causes a retardation of /8 at the metrology source
wavelength (1550nm) so that after the reflection on the mirror, the metrology beam is
retarded by /4 since it passes twice in the layer. The diameter of the wave-retarder layer is
smaller than that of the metrology beam, so that a small misalignement of the two mirrors
can be measured (Fig. 7). The metrology beam is then recorded orthogonally by two 2x16
pixel detector arrays. Once the correction values are calculated, commands are sent back to
the dynamic alignment piezos which correct the misalignement.
Fig. 7. Design of SpIOMM's metrology system consisting in the detection of a flat fringe
pattern (of the IR source) comprising a smaller circular region of /4 retardation created by
a wave-retarder layer placed at the center of the fixed mirror (a). The flat fringe pattern is
pixel located in the inner region of retardation of the metrology beam will be delayed by π/2
recorded by two detectors orthogonally placed at interferometer output (b). The signal of a
compared to the signal of a pixel outside (c). For a specific OPD, (d) shows the shape of the
aligned intensity pattern read by one of the detectors. Any variation from the "perfect"
shape shown in (d) causes the piezoelectric actuators to realign the mirror
4.3 Tests and problems encountered

SpIOMM was assembled in the laboratory in early 2004 and the first tests were extremely
encouraging: high resolution spectra of laser sources were obtained on a regular basis, and the
instrument line shape was exactly like it should (Fig. 8). The laboratory environment is ideal to
www.intechopen.com
test an instrument such as SpIOMM: the temperature is constant to within a degree, there is no
wind nor outside vibrations and the gravity vector is constant during data acquisition.
Fig. 8. Example of data obtained in the laboratory. Two light sources (a HeNe laser at 632 nm
and a LED at 375 nm) are sent to an integration sphere which is observed with SpIOMM using
exactly the same technique as for any astronomical source. The upper left image shows the
fringe pattern recorded at a given OPD by the CCD. Notice that the interference pattern is not
centered on the CCD; this is caused by the off-axis approach we have chosen and which is
shown in Fig. 6. The central part of the interferogram of one pixel, near the ZPD, is shown on
the upper right image (the entire interferogram was sampled with 4750 points). A Fourier
transform of this interferogram then recovers the spectrum of the two sources
Happy with the lab results, we installed SpIOMM at the telescope in early 2004. But a
telescope environment is very harsh for an interferometer. Yearly temperatures vary
between +25oC and -35oC, but more importantly excursions of 5 degrees are common during
a single data cube observation. Windy nights are frequent and gusts can move the telescope
abruptly. The Earth rotates, so does the telescope in the opposite direction to follow the
apparent movement of the targets; the orientation of the gravity vector therefore varies
constantly with respect to all components of the instrument, including the interferometer’s
mirrors. The telescope’s motors and the air pump that supports the telescope’s primary
mirror also create vibrations that are transmitted to the instrument. All of these hostile
environmental factors impose very severe and stringent constraints on the servo system to
maintain the mirror stability. The original servo loop ran at 2.5 kHz, which was sufficient in
www.intechopen.com
the lab to correct any misalignment. It became however obvious that it was not enough to
maintain a proper stability at the telescope. The modulation efficiency was not optimal (70 -
75% at 632 nm) and highly variable (excursions of 10% were frequent) during data
acquisition. This introduced noise and spurious artifacts in the spectra. The original
metrology and servo system (described by Bernier et al. 2006, 2008) was then replaced by the
one described above with great success. Science data cubes have been obtained on a regular
basis since 2007, with a very stable modulation efficiency very close to the theoretical limit
of 85% at 632 nm. Fig. 9 schematically describes the data acquisition of a typical science
interferogram and its transformation to a useable data cube. Data processing is an important
part of the entire process and is described in more details in the next section.
Fig. 9. Data acquisition with SpIOMM (a) By scanning the Optical Path Difference (OPD) of
the interferometer and taking images at every step, one gets a datacube composed of one
interferogram for every pixel. (b) For a given pixel, the recorded intensity varies as a
function of the OPD with a pattern that depends on the spectral content of the source; for
example, a monochromatic laser beam would produce a sinusoidal pattern. A Fourier
transform of the signal produces a spectrum for every pixel in the image. (c) After Fourier
transforming every interferogram, one gets a spectral datacube from which monochromatic
images corresponding to the emission lines of interest are extracted. The data shown are
extracted from a data cube of the supernova remnant NGC 6992
www.intechopen.com
4.4 Data processing

As with every imaging system using CCD detectors, data processing starts with a series of
basic corrections such as bias removal (an electronic signal inherent to the CCD) and
flatfield corrections (to correct for the non-flatness of the optical transmission or pixel-to-
pixel sensitivity variations). A datacube acquisistion usually takes between one and four
hours depending of the spectral parameters and the object's surface brightness. Sky
transmission sometimes fluctuates over that period and affects the signal of the
interferogram; this is due to the passage of thin clouds or simply the change of airmass. It
can add low and high frequency contributions to the spectra. To avoid such artifacts, we
want to normalize the baseline signal of the interferogram. To do so, we measure the
photometric variations of the at least a dozen stars in the FOV. The interferogram signal of a
polychromatic source like a star (almost a perfect blackbody!) modulates only at the ZPD.
On either side of the ZPD, the signal intensity is supposed to be flat unless the sky
transmission fluctuates (clouds, airmass, etc). Therefore we determine the mean sky
transmission signal from the interferograms of a list of stars. Using only one CCD as we
have done so far is not an ideal situation because sky transparency fluctuations cannot be
completele corrected for, especially around the ZPD. The implemetation of the second
output port in 2011 should solve this problem and significantly improve the signal-to-noise
ratio of the spectra. Because of flexure within the instrument or between the telescope's
guiding system and the instrument during the entire acquisition (the source can be at the
zenith at the beginning and 45 degrees from the horizon at the end), the images are not
perfectly aligned with each other. Typical shifts of 1 to 3 pixels in both directions are
common between the first and last image of an interferogram. The centroid of a dozen stars
scattered across the field is then measured in all images, which are then realigned
accordingly with a precision of a tenth of a pixel. Finaly, CCDs are very sensitive to cosmic
ray hits, which usually affect a few pixels in each image; they are corrected for by comparing
Before transforming the interferogram cube into a hyperspectral cube (x, y, λ) we must
signals from adjacent images.
apply some operations on the interferogram signal. First, the pre-processed

interferogram should have a null mean in order to have its extremities values
approaching zero. Therefore, we remove the mean value to each interferogram. We then
multiply the interferogram with an apodization function in order to minimize the side
lobes in the spectral lines created by the truncation of the signal (finite number of points;
see section 3.2). We have tested some apodization windows with our data to find the best
function to minimize the side lobes in the spectrum without degrading the spectral
resolution too much. The gaussian function provides the best results. Finally, a zero-
filling operation is performed on the interferogram in order to increase the number of
points up to the next higher power of 2. This is useful for further spectral interpolation
and for a faster processing discrete Fourier transforms. We then apply the discrete
Fourier transform to each processed interferogram of the datacube (up to 1.7 million) in
order to obtain a datacube of spectra. In parallel, we compute the spectral axis scale from
the datacube sampling parameters and we interpolate the total number of points on a
graduation in nanometers. This rescaling and interpolating operation takes also in
account a calibration dataset that corrects for off-axis contributions on the whole FOV: an
interferogram of a He-Ne laser obtained at high resolution during the same observing
run.
www.intechopen.com
5. Science results with SpIOMM

While the number of scientific applications of an imaging FTS is potentially very large, our
group has focused on the interstellar medium of our own galaxy, the Milky way, and nearby
spiral galaxies. A meaningful link between local heavy element enrichment and the global
chemical evolution of galaxies can only be established by detailed studies of individual
wind-blown bubbles in our own galaxy. Winds of evolved stars, and their surrounding
bubbles (planetary nebulae, Luminous Blue Variables and Wolf-Rayet ejecta, supernova
remnants) are known to be globally enriched with products of nucleosynthesis. While
planetary nebulae are ejected by low-mass stars, with slow (20-30 km/s) winds, LBVs and
WRs are the late evolutionary stages of the most massive stars with wind velocities of
hundreds to thousands of km/s. A complete survey of abundance, density, temperature and
kinematic measurements in nebulae surrounding individual evolved stars and ionizing
clusters, looking for inhomogeneities in the distribution of processed material (primarily
nitrogen and oxygen), which has never been undertaken because it requires wide-field
spectroscopic mapping, will provide firm grounds for the interpretation of global galactic
abundance studies.
In particular, we have obtained data cubes of planetary nebulae (the envelope of low-mass
stars ejected at the end of their life), Wolf-Rayet bubbles (cavities created by the interaction
of very massive stars with their surrounding interstellar medium) and supernova remnants,
which are the result of the explosion of the most massive stars. We have also targeted
nearby spiral galaxies. While the observations could be done without any filter, thereby
covering the entire visible range (from 350 nm to 850 nm), we take advantage of the fact that
our prefered targets emit most of their flux in a series of emission lines clustered around 500
nm and 660 nm (rest wavelength). The observations are therefore performed in two steps,
with a blue (450 – 520 nm) and a red (650 – 680 nm) filter to cover the most intense and
diagnostic-rich emission lines. The use of filters also significantly increases the contrast
between the targets (nebulae) and the underlying continuum sources (stars) and
dramatically reduces the background sky intensity, especially when the Moon is up in the
sky. The most important criterium to define the spectral resolution is the capacity to
unambiguously separate the lines from the [SII] doublet at 671.7 nm and 673.1 nm (whose
ratio is a good indicator of the gas electron density). Such a resolution also ensures a clear
separation of the [NII] doublet (654.8, 658.4 nm) from the strong Hα line at 656.3 nm. Such a
resolution is obtained in the red filter with 325 steps. The lines full width at half maximum
corresponds to a velocity of ~ 130 km/s, but the centroid of each line can be determined
with a precision of about 10 – 20 km/s, depending on the line strength, therefore allowing
kinematical analysis of nebulae and rotation curves of galaxies. Exposure times vary from 15
seconds per step for the observations of bright nebulae in the red filter to 90 seconds per
step for the observations of galaxies in the blue filter. Indeed, a few factors contribute to the
difficulty of acquiring blue data : emission lines are generally more intense in the red ;
absorption by interstellar dust and the Earth’s atmosphere is much more efficient in the
blue ; the global throughput of the instrument but especially its modulation efficiency
dramatically decreases from the red to the blue end. To the exposure time, one must add the
« dead time » caused by the CCD readout and the mirror displacement and stabilization
time (a total of 7 seconds per step). During observing runs in 2007 – 2010, we have obtained
more than 60 datacubes of galactic HII regions, planetary nebulae, Wolf-Rayet ring nebulae,
supernova remnants, nearby spiral galaxies and groups of interacting galaxies.
www.intechopen.com
5.1 Nebulae around evolved stars

Planetary nebulae are the remains of the outer envelope of low-mass stars like the Sun
ejected during the late stages of their evolution. They are chemically enriched with the
products of stellar nucleosynthesis and expand at relatively low velocities (~ 50 – 100 km/s).
The planetary nebula NGC 6853 (M27) was one of the first astronomical targets of SpIOMM,
because of its large size, high surface brightness brightness, and because it has been the
object of numerous studies, thereby allowing a direct comparison with previous
observations with classical techniques. Figure 10 highlights some of the data extracted from
the blue and red cubes of M27. While images of this objects had been obtained before at
different wavelengths, and long-slit spectra of a tiny fraction of the nebula been analyzed,
SpIOMM's data are the first to show spatialy resolved spectra of the entire nebula. In
particular, the very high contrast possible by a careful extraction of the images
corresponding to particular wavelengths in the data cube allow us to detect very faint
structures in the outskirts of the nebula, particularly in the [NII] emission line. This line is
particularly difficult to isolate with interference filter imagery because it is very close to the
bright Hα line. Moreover, since one spectrum was obtained for every pixel on the scene, we
were able to obtain a very detailed diagnostic diagram based on two line ratios. A detailed
physical analysis of these data is under way.
Fig. 10. (left) Diagnostic diagram of the planetary nebula M27; each point represents line
ratios for a single 1.1 arcsecond pixel from the data cube. (right) Images of M27 in different
emission lines, from the same data cube, showing very different morphologies. The [NII]
658.4 / Hα 656.3 line ratio map (lower right panel) displays unusually large values,
characteristic of shocks, at the outskirts of the inner bubble, as well as the periphery. The
identification of the individual points on the diagnostic diagram with precise location on the
image provides important constraints for the modelisation of the nebula
5.2 Supernova remnants

Extended galactic supernova remnants were also prime targets for SpIOMM. The most
massive stars in the universe end their life as catastrophic events. When the nuclear fuel has
been entirely consumed by thermonuclear reactions, the core of the star is composed almost
exclusively of iron and nickel. The outer envelope of the star collapses, bounces back on the
dense nucleus and is then expelled at velocities of ~ 15 000 km/s ; this is called a supernova.
www.intechopen.com
The rapidly expanding bubble of chemicaly enriched material has a profound impact on its
surrounding by injecting huge amounts of kinetic energy as well as heavy elements in the
interstellar medium. A classic example of such an interaction is the Cygnus Loop, a 15 000
year-old supernova remnant spanning many degrees in the sky. A tiny fraction of this object
has been mapped with the camera WFPC2 on the Hubble Space Telescope to characterize the
motion, structure and dynamical scale of the blast wave currently encountering the
surrounding medium, in the northeastern part of the nebula (Blair et al. 2005). We have
begun a complete mapping of the Cygnus Loop with SpIOMM is order to characterize this
important object in its entirety. Fig. 11 depicts some characteristics of a single red datacube,
showing the unusually strong [SII] lines. This cube illustrates that SpIOMM can be used at
Fig. 11. Example of science results obtained with a single data cube of a section of NGC
6992, an old supernova remnant in the Milky Way, obtained with SpIOMM. The frames are
12 arcminutes on a side and display 435 000 pixels. (Upper left) - Doppler map from the
[NII] 654.8 nm and 658.4 nm, Hα 656.3 nm, [SII] 671.7 nm and 673.1 nm emission lines;
velocities vary between -20 km/s and + 30 km/s, the blue filaments approaching us and the
red ones receiding from us. (Upper right) [NII]/Hα line ratio; Hα is orange and [NII] blue.
(Lower left) - Electron density of the gas for the same region, based on the ratio of the
[SII] 671.6 and [SII] 673.1 nm lines, assuming an average temperature of 8 500K. (Lower
right) Spectra of two filaments (5 x 5 pixel area each) with different Doppler shifts
www.intechopen.com
the same time as an imager with a set of “perfect” narrow-band filters (note in the lower
right panel of Fig. 11 how well separated all the emission lines are from their neighbors) as
well as a medium-resolution, very-wide field spectrograph.
The case of another, much younger supernova remnant, M1 (also known as the Crab
nebula), is particularly interesting as it illustrates the full power of SpIOMM. M1 is a very
young objects in Galactic terms, since it is the result of an explosion visible from Earth in
1054 AD. The initial explosion propelled the star's outer envelope at more than 10 000 k/s,
but this movement was slowed down by the material surrounding the star. Nevertheless,
the gas is still globally expanding today at velocities of up to ~ 1400 km/s, causing Doppler
shifts of up to 3 nm. Moreover, because of the presence of shocks, the forbidden lines of
[NII] 654.8 and 658.4, as well as the [SII] 671.7, 3.1 doublet are almost as strong as the
(usually strongest) H. Therefore, up to 10 emission lines can be seen in regions where an
approaching and a receding filament are superimposed on the line of sight. Charlebois et al.
(2010) presented a detailed analysis of the M1 data cubes, but we reproduce in Fig. 12 some
of the results. M1 has been partially mapped before with a classical, long-slit spectrograph;
only a small fraction of the nebula could be mapped this way, but we have used these data
to demonstrate that the spectra obtained with SpIOMM showed exacty the same features as
those obtained with classical spectroscopy, but with a 100% filling factor over the entire
nebula. Our data allowed us to obtain a comple tridimensional view of this intriguing object,
as well as to identify a notable asymetric evolution of the oposite lobes of the nebula.
Fig. 12. A Doppler image of the Crab nebula obtained with SpIOMM as determined from the
[OIII] 500.7 nm emission line, showing the rapid expansion of the filaments. On the right,
the spectrum of a pixel where two filaments are superimposed. We see here two sets of
lines, shifted by the Doppler effect
5.3 Galaxies
Powerful constraints on models of galactic chemical evolution, on the star formation
histories of galaxies and on the dynamical processes that transform them can be derived
from accurate and homogeneous determinations of chemical abundances in individual
gaseous nebulae, the distribution of their stellar populations in terms of age and metallicity,
and the gaseous and stellar kinematics. So far, systematic studies between chemical
properties (central abundance or radial abundance gradient), and other parameters have
been conducted for a small sample of spiral galaxies. The effects of the morphological type,
www.intechopen.com
the presence of bars, and environment have all been studied to some extent. Moreover, most
studies so far, performed with slit spectrographs, have concentrated on global properties of
stellar ejecta or abundance gradients in galaxies, thereby neglecting possible small-scale
variations caused by multi-phase stellar wind (individual stars) or localized enrichment by
starburst clusters in peculiar evolutionary stages. SpIOMM is an ideal instrument to conduct
a systematic study of abundances in nearby galaxies and thus easily detect evidence for
small-scale enrichments and establish conditions under which they take place. The
possibility to study the multiple emission line ratios and kinematics for hundreds of HII
regions (nebulae ionized by massive stars) simultaneously in each individual galaxy is an
excellent project for an IFTS. We have targetted a dozen galaxies so far, and show in Fig. 13
Fig. 13. (Upper left) - Velocity diagram of the spiral galaxy M51, based on the centroid of
the Hα 656.3 nm emission line from an SpIOMM data cube. (Upper right) - Image of the
[NII] 658.4 / Hα 656.3 line ratio from the same cube. Notice the large ratio in the core of the
galaxy, characteristic of shocks driven by the central Active Galactic Nucleus. (Lower left) A
comparison between the distributions of the Hα flux (characteristic of ~ 10 Myr old stellar
populations, obtained from the SpIOMM data cube, in orange), and the ultraviolet flux
(characteristic of ~ 100 Myr old stars, obtained with the Galex space telescope, in blue). The
lag between the two stellar populations is indicative of the rotation velocity of the spiral
wave pattern in the galaxy. (Lower right) Spectrum of one HII region from the SpIOMM
data cube
www.intechopen.com
some data on the nearby interacting spiral galaxy M51. As can be seen in the upper left
panel of this figure, the galaxy is not only moving away from us (average recession velocity
of 500 km/s) due to the expansion of the Universe, but it is also rotating. M51 is seen almost
face-on, but a large inclination increases the velocity difference from one side to the other
relative to us. Obtaining line intensities and especially line ratios with narrowband filters
would almost require a set of filters per galaxy, and it would be impossible to obtain precise
line ratios for edge-on galaxies. Most spectra of star-forming regions in galaxies have been
obtained with single slit or multi-object spectrographs, which can only sample a small
fraction of each galaxy. An IFTS is thus the instrument of choice for this kind of work.
6. Conclusions and the future

Despite lots of technical hurdles, we have clearly demonstrated that an astronomical
imaging Fourier transform spectrograph working in the visible band is not only viable, but
also shows an immense potential to spectrally map extended objects. We have so far
discussed and presented examples of emission line spectra. Is an imaging FTS capable of
obtaining spectra of continuum sources with absorption lines, such as stars? The answer is
yes; indeed, the vast majority of commercial applications of IFTS are to detect absorption
lines superimposed on a continuum source, and exquisite infrared spectra of stars and
planets have been obtained with FTS (Ridgway & Brault 1984, Mosser et al. 1993). We have
however observed only emission-line objects with SpIOMM, mostly because we used it,
until recently, with only one detector and used the photometry of stars to correct for sky
transparency variations. The use of the second output port will allow us to work on
absorption line objects. But there is also another reason for us to favor emission-line objects,
and this reason is intrinsic to the FTS concept itself. Since at every step an image of the scene
in the entire waveband is obtained, contrary to Fabry-Perot systems, one of the great
advantages of an IFTS is that a by-product of an interferogram is a deep panchromatic
image of the scene. However, this property has also its downside: the photon noise from
each wavelength is distributed among every spectral resolution element. The spectrum of an
ionized nebula is completely dominated by emission lines and shows very little continuum.
Photon noise therefore comes from the emission lines themselves. Moreover, the intensities
of the lines we are aiming for are all within a factor of ten from each other. The FTS
distributed noise disadvantage is therefore completely negligeable in this case. However,
some lines of interest, such as [OIII] 436.3 nm or [NII] 575.5 nm, are extremely weak and the
inclusion of much stronger lines, such as [OIII] 500.7 or Hα, in the same bandbass would
lower the signal-to-noise of the faint lines and make them very difficult to detect. Once the
second detector becomes available for SpIOMM, we will aim for both absorption line targets
(star clusters, elliptical galaxies) and very faint lines in extended nebulae.
We have recently obtained funding to design and build an improved version of SpIOMM
for the Canada-France-Hawaii telescope. SITELLE, another collaboration between
Université Laval and ABB-Bomem, will have the same field of view, 12' x 12', with 0.35''
pixels, and will see its first light, if all goes according to plans, in early 2013. Because the
CFHT has a larger primary mirror than the OMM telescope (3.6m vs. 1.6m) and SITELLE
will from the start be equipped with two detectors, because of a better sky transparency
(especially in the near UV; the CFHT is located at an altitude of 4200 m) and a better optics
quality, we expect that SITELLE will offer a factor of 15 improvement over SpIOMM in
terms of global efficiency. As such, it will be an ideal instrument to target very faint objects
www.intechopen.com
or very faint diagnostic lines in extended objects. We expect that a fair fraction of its
observing time will be devoted to the study of the star formation at cosmological distances.
As we have seen, technology has evolved enough during the past ten years on different
fronts to allow us to design and build very efficient, wide-field imaging FTS in the visible
range: servo systems and piezoelectric actuators for an accurate and fast correction of the
mirror alignment to enhance the modulation efficiency, high quantum efficiency (~ 90%)
large detectors with fast readout rate and low noise to increase the field of view and the
spectral resolution (which is proportional to the maximum OPD and thus the number of
CCD readouts), and of course computer power and random access memory to allow the
data reduction and analysis of huge data cubes.
7. Ackowledgements
We would like to acknowledge financial contributions from the Canadian Foundation for
Innovation, the Canadian Space Agency, the Natural Sciences and Engineering Council of
Canada, the Fonds Québécois de la Recherche sur la Nature et les Technologies, and
Université Laval. We also thank Ghislain Turcotte, Bernard Malenfant and Pierre-Luc
Lévesque for their help at the telescope.
8. References
Allington-Smith, J., et al. 2002, PASP, 114, 892
Artigau, É., Bouchard, S., Doyon, R., & Lafrenière, D. 2009, ApJ, 701, 1534
Bacon, R., et al. 2001, MNRAS, 326, 23
Bacon, R., et al. 2010, Proc SPIE, 7335, 7B
Bennett, C. L. 2000, in Imaging the Universe in Three Dimensions, ASP Conf. Ser. Vol 195, p.
58
Bernath, P. F., et al. 2005, Geoph. Res. Let. 32, L15S01
Bernier, A.-P., Grandmont, F., Rochon, J.-F.. Charlebois, M., & Drissen, L. 2006, Proc. SPIE,
6269, 135
Bernier, A. -P., Charlebois, M., Drissen, L., & Grandmont, F. 2008, Proc. SPIE, 7014, 70147J
Blair, W. P., Sankrit, R., & Raymond, J. C. 2005, AJ, 129, 2268
Boulanger, F., et al. 2008, Experimental Astronomy, 23, 277
Chalabaev, A. A. & Maillard, J.-P. 1985, ApJ, 294, 640
Charlebois, M., Drissen, L., Bernier, A.-P., Grandmont, F., & Binette, L. 2010, AJ, 139, 2083
Colless, M. M., et al. 2001, MNRAS, 328, 1039
Daigle, O., Carignan, C., Gach, J.-L., Guillaume, C., Lessard, S., Fortin, C.-A., Blais-
Ouellette, S. 2009, PASP, 121, 866
Flasar, F. M., et al. 2004, SSRv 115, 169
Friesen, R. K. et al. 2005, MNRAS, 361, 460
Graham, J. R. 2000, in Next Generation Space Telescope Science and Technology, ASP Conf.
Ser., vol. 207, p. 240
Graham, J. R., et al. 1998, PASP, 110, 1205
Grandmont, F. 2007, Ph. D. Thesis, Université Laval
Grandmont, F., Drissen, L., & Joncas, G. 2003, Proc. SPIE, 4842, 392
Hernandez, O., et al. 2008, PASP, 120, 665
Huchra, J., et al. 1999, NGST NIR Spectrograph Study
www.intechopen.com
Kewley, L. J., & Dopita, M. A. 2002, ApJS, 142, 35

Lagrois, D., & Joncas, G. 2010, ApJ, 700, 1847
Maillard, J. P., Crovisier, J., Encrenaz, T., Combes, M. 1987, A&A, 187, 398
Maillard, J.-P. & Simons, D. 1992, in Progress in Telescope and Instrumentation
Technologies, ESO Conference and Workshop Proceedings, ESO Conference on Progress
in Telescope and Instrumentation Technologies, ESO, p.733
Maillard, J.-P., Drissen, L., Grandmont, F., & Thibault, S. 2011, submitted to Experimental
Astronomy.
Manset, N., & Bastien, P. 2002, AJ, 124, 1089
Monnet, G. 2009, Experimental Astronomy, Volume 25, p. 91
Morris, S. L., et al. 2000, in NGST Science and Technology, ASP Conf. Ser., vol. 207, p. 276
Mosser, B., Mekarnia, D., Maillard, J. P., Gay, J., Gautier, D., & Delache, P. 1993, A&A, 267,
604
Naylor, D. A., Gom, B. G., Tahic, M. K. & Davis, G. R. 2004, SPIE, 5498, 685
Naylor, D., et al. 2010, SPIE, 7731, 29
Paumard, T., Maillard, J.-P. & Morris, M. 2004, A&A, 426, 81
Posselt, W., et al. 2000, in NGST Science and Technology, ASP Conf. Ser., vol. 207, p. 303
Ridgway, S. T. & Brault, J. W. 1984, Ann. Rev. Astron. Astrop. 22, 291
Ridgway, S. T., Carbon, D. F., Hall, D. N. B., & Jewell, J. 1984, ApJS, 54, 177
Sanchez, S. F., Becker, T., & Kelz, A. 2004, Astron. Nachr., 325, 171
Scoville, N., Hall, D. N. B., Kleinmann, S. G., & Ridgway, S. F. 1979, ApJ, 232, L121
Stoughton, C. et al. 2002, AJ, 123, 485
White, G. J., et al. 2010, A&A, 518, 114
Wurtz, R., Wishnow, E., Blais-Ouellette, S., Cook, K., Carr, D., Lewis, I., Grandmont, F., &
Stubbs, C. 2002a, in Next Generation Wide-Filed Multi-Object Spectroscopy, ASP.
Conf. Ser., Vol. 280, p. 139
Wurtz, R., Wishnow, E., Blais-Ouellette, S., Cook, K., Carr, D., Grandmont, F., Lewis, I., &
Stubbs, C. 2002b, in Galaxies: The Third Dimension, ASP. Conf. Ser., Vol. 282, p. 445
www.intechopen.com

Fourier Transforms New Analytical Approaches and FTIR Strategies

Uploaded by

Copyright:

Available Formats

Fourier Transforms New Analytical Approaches and FTIR Strategies

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fourier Transforms New Analytical Approaches and FTIR Strategies

Uploaded by

Copyright:

Available Formats

Fourier Transforms - New Analytical Approaches and FTIR

InTech Europe InTech China

Fourier Transform Infrared Imaging

2. Planning a new experiment

2.1 Sample preparation

2.1.1 Chemically fixed histological samples

2.1.3 The effect of the variable section thickness

2.2 Data collection

2.2.1 Spatial resolution

2.2.2 Spectral resolution

2nd derivative absorbance

Fig. 1. An IR absorbance spectrum of articular cartilage measured with 4 cm-1 spectral

2.2.3 Standardized measurement conditions

3.1 Spectral pre-processing

3.2 Univariate methods

3.3 Multivariate methods

4. How to select appropriate analysis method for different research

4.1 Quantitative measurement of tissue constituents

4.2 Qualitative visualization of tissue morphology

4.3 Classification studies

Analysis of Bioactive Olygosaccharide-Metal

importance. For commercial reasons raw polysaccharides were depolymerized to the

2. Character and spectroscopy of bioinorganic compound

3. Copper(II) ion and its significance

Copper(II) ion is an essential component of several enzymes such as ceruloplasmin,

4. Bioactive copper-pullulan complex

polymer is a α-glucan in which α-(1→4) linkages predominate (Bender et al., 1959).

frequently described as a polymer of α-(1→6) linked maltotriose subunits (Fig. 1).

Fig. 1. Molecular structure of a representative portion of pullulan, illustrating the primary

Fig. 2. Molecular structure of the secondary (minor) repeating structure of pullulan,

a 15x objective and a 250 μm liquid-nitrogen cooled, narrow-band mercury–cadmium–

Nikolic et al., 2008).

δ(OH) on the 1384 cm-1 (data from LNT-FTIR) (Fig. 4c).

only in synthesized Cu(II)–RLMP complexes. An approximate effect exists in the stretching

Fig. 8. ATR–FTIR spectra of Cu(II)–RLMP complex synthesized at pH 7.5 (type I) from

cm-1, show blue shift on cooling.

5. Bioactive copper-dextran complex

narrow-band mercury–cadmium–telluride detector (ATR objective GMBH, Germany) was

pH Complex Cu (%) Water

of α-D-glucose Fourier transform infrared spectrum deconvolution: comparison

with experimental and theoretical data. Spectrochim. Acta A, 55(1), 229-238.

Precision Quantitative Proteomics with Fourier-

2. Protein turnover analysis with FTMS

2.2 High-resolution mass spectrometry for protein turnover analysis

Intensity (x10 counts)

120 Shocked (pH5) 30 Control (high-iron)

Fig. 2. Calculation of isotopologue intensities for a representative tryptic peptide

Isotopomer Number (M)

electrophoresis for separation and fluorography and autoradiography for quantitation of

2.3 Implication of proteome turnover studies in mycobacteria

3. Label-free quantitation of proteins

3.1 Purpose of the study

3.2.1 Source of variability

(II) Protein extraction

(III) LC/MS samples

(IV) LC/MS injections SP,1 SP,2 RP,1 RP,2

(VI) Protein XIC AS A

For determining positives

Unlabeled protein samples from [15N]-labeled protein samples

3.2.2 Extended selection of differentially regulated proteins