INSTITUTE OF PHYSICS PUBLISHING SEMICONDUCTOR SCIENCE AND TECHNOLOGY
Semicond. Sci. Technol. 20 (2005) S246–S253 doi:10.1088/0268-1242/20/7/015
Chemical recognition in terahertz
time-domain spectroscopy and imaging
B Fischer
1
, M Hoffmann
1
,HHelm
1
, G Modjesch
2
andPUhdJepsen
1
1
FMF and Department of Molecular and Optical Physics, Albert-Ludwigs-Universit
¨
at,
D-79104 Freiburg, Germany
2
Pharmaceutical Institute, Albert-Ludwigs-Universit
¨
at, D-79104 Freiburg, Germany
Received 21 January 2005
Published 8 June 2005
Online at stacks.iop.org/SST/20/S246
Abstract
In this paper, we present an overview of chemical recognition with
ultrashort THz pulses. We describe the experimental technique and
demonstrate how signals for chemical recognition of substances in sealed
containers can be obtained, based on the broadband absorption spectra of the
substances. We then discuss chemical recognition in combination with THz
imaging and show that certain groups of biological substances may give rise
to characteristic recognition signals. Finally, we explore the power of
numerical prediction of absorption spectra of molecular crystals and
illuminate some of the challenges facing state-of-the-art computational
chemistry software.
1. Introduction
The recently developed technology of generating and detecting
ultrashort and coherent electromagnetic pulses in the very-
far-infrared region [1] opens a practical avenue to chemical
imaging with very-far-infrared radiation (FIR). The short
duration of the pulses results in a bandwidth covering the
region from 0.1 to 5 THz of the electromagnetic spectrum. The
broadband spectral nature of these pulses permits recording
of the dielectric function (absorption coefficient and index
of refraction) from the modification of the shape of an
electromagnetic pulse transmitted through the sample [2]. This
concept uses the coherent nature of the source to directly trace
the electric field of the electromagnetic pulse (rather than its
intensity). It also exploits the subpicosecond pulse length to
implement time-gated detection, thereby eliminating thermal
blackbody noise at these frequencies. The latter feature
enables us to operate source, sample and detector at ambient
temperature, a great advantage in practical applications.
In this paper, we will give an overview of chemical
recognition with ultrashort THz pulses. Section 2 describes
the experimental technique and demonstrates how signals for
chemical recognition of substances in sealed containers can
be obtained. Section 3 discusses a practical implementation
of spatial imaging combined with chemical recognition.
Section 4 discusses the possibility of obtaining characteristic
recognition signals from biological substances. In section 5,
we explore the power of numerical prediction of absorption
spectra of molecular crystals and illuminates some of the
challenges facing state-of-the-art computational chemistry
software.
In order to observe distinct spectral features in the far-
infrared absorption spectrum of a substance some sort of long-
range order is required. Substances in the condensed phase
are held together by either ionic, covalent or electrostatic
forces, and therefore the lowest frequency modes will be
associated with intermolecular motion. A medium with
long-range ordering of its molecular constituents can support
phonon-like intermolecular modes at discrete frequency bands,
whereas an amorphous medium will display a continuum
of strongly damped intermolecular vibrational modes. At
higher frequencies, where intramolecular modes are active,
this picture is no longer valid. Here, the vibrational modes
of the isolated molecules are found, albeit influenced by the
surroundings.
Although THz spectroscopy has been proposed to be able
to identify and investigate a wide range of substances, we will,
for the reason given above, focus our attention on compounds
with crystalline structure. Since the mass of the unit cell must
be high and the intermolecular forces must be small in order
to move vibrational frequencies into the terahertz frequency
regime, we will further restrict our discussion to hydrogen-
bonded organic crystals.
0268-1242/05/070246+08$30.00 © 2005 IOP Publishing Ltd Printed in the UK S246
Chemical recognition in terahertz time-domain spectroscopy and imaging
THz emitter
THz detector
Sample
Optical
delay line
femtosecond
laser pulses
Beamsplitter
Figure 1. Experimental set-up for chemical recognition of a sample
using far-infrared radiation in the transmission mode.
2. Chemical sensitivity
To demonstrate the concept of chemical recognition in the
THz frequency domain [3–5], we investigated the pulse
modification caused by selected sealed samples using the set-
up shown in figure 1. This terahertz time-domain spectrometer
uses photoconductive switches for generation and detection
of the far-infrared light [6]. The switches are operated with
optical (800 nm) laser pulses of 15 fs duration. This instrument
allows us to record the absorption coefficient and index of
refraction of samples with a spectral resolution of 0.5 cm
−1
.
This is achieved by tracing the temporal shape of the electric
field with sample E
s
(t) and without sample E
r
(t),wheret is
the optical delay time of the gating laser pulse and the far-
infrared pulse at the detector. The ratio of these fields in the
frequency domain ν
E
s
(ν)
E
r
(ν)
=
4n
(n +1)
2
exp
[
−αd/2+i2πν(n − 1)d/c
]
(1)
contains the absorption coefficient α(ν) and index of refraction
n(ν) of the sample. The sample thickness is d and c is the speed
of light in vacuum.
Two typical traces for E
r
(t) and E
s
(t), transmitted through
an empty envelope and an envelope containing a small amount
of lactose powder, are shown in figure 2. The modification
of the pulse shape due to the presence of an infrared active
compound in the envelope is apparent from the difference
in electric field traces. The pulse is delayed and attenuated;
both are measures of the dispersion and the absorption by
the sample. The spectral key is contained in the pronounced
ringing of the sample trace at later times. This ringing is due
to the free induction decay of the coherently excited sample
[7] which appears as constructive interference in the direction
of propagation of the FIR beam.
Figure 3 gives typical spectra of the product αd obtained
for several polycrystalline substances sealed in bags made
of polyethylene. The absorption spectrum of the empty bag
is given as the bottom trace. It is smooth and practically
featureless in the low-frequency range. The spectra of lactose,
cocaine and morphine show distinct and different features in
0 5 10 15 20
-1
0
1
2
3
4
5
6
7
8
9
Pulse through PE bag with lactose
Reference pulse
THz signal [nA]
Time delay [ps]
Figure 2. The temporal profile of the electric field strength, here
represented as the induced photocurrent in the detector, of the FIR
pulse transmitted through an envelope containing lactose and
through an empty envelope. The sample trace is offset vertically by
5 nA for better visibility of the traces.
this spectral range. Practically identical spectral signatures are
obtained when the samples are enclosed in paper envelopes. A
molecular dynamics assignment of the individual resonances
in this very-low-frequency range will remain a challenging
task, but is irrelevant for the suitability of these spectral
fingerprints as a key in chemical recognition.
The identity and purity of the drugs used, cocaine–HCl
(Merck) and morphine–HCl (Synopharm), are according to
the prescriptions of The European Pharmacopoeia [15]. The
other samples were α-lactose monohydrate (glucose free,
4% β-anomer content), sucrose (purity > 99.5%), aspirin
(acetylsalicylic acid, purity > 99.5%) and D-tartaric acid
(purity > 99%).
The course grained samples (10–100 mg) were placed
in polyethylene (LDPE) bags with single wall thickness of
55 µm. The filled bags (thickness ≈ 0.5 mm) were placed at
the position labelled in figure 1, shaded by a copper plate with
an aperture of 8 mm diameter. The aperture is filled by the
lateral profile of the far-infrared light pulse. The samples were
at ambient temperature (≈298 K) when recording the spectra.
No specific precautions were taken to prepare the samples
in an optically flat manner when recording the traces shown
in figures 2 and 3. To our knowledge no spectral studies have
previously been reported for α-lactose, cocaine and morphine
in this frequency range to which our data could be compared.
It is interesting to note that the recently published spectrum of
β-lactose [16] differs significantly from the spectrum of the
α-anomer used in this work. Some reasons for this difference
are that the two anomers differ in the spatial arrangement
of atoms and that the β-anomer in contrast to the α-anomer
crystallizes without water molecules in the unit cell structure.
Several technical aspects and limitations of our
technology need to be mentioned. The typical energy of a
single far-infrared pulse in our spectrometer is at the level of
10
−17
J. Since the spectrometer is operated at a repetition rate
of 10
8
s
−1
, the sample is exposed to a radiation power of a few
nW, a level well below the radiation level experienced by the
S247
B Fischer et al
0.5 1.0 1.5 2.0 2.5 3.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0 2.5 3.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0 2.5 3.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0 2.5 3.0
0
2
4
6
8
10
cocaine
Absorption αd
morphine
lactose
Absorption αd
Frequency [THz]
PE plastic bag
Frequency [THz]
Figure 3. Frequency spectra of the absorption recorded for the empty envelope and identical envelopes containing small amounts of
α-lactose, cocaine and morphine. The error bars represent one standard deviation from the mean of typically ten measurements.
01234
1E-13
1E-12
1E-11
1E-10
1E-9
1E-8
1E-7
1E-6
1E-5
1E-4
1E-3
THz beam, peak power
P
peak
=27µW
Blackbody background
P(
ν
< 4 THz) = 0.38 mW @ 300 K
THz beam, average power
P
av
= 2.67 nW @ 100 MHz
Spectral power density [W/THz]
Frequency [THz]
Figure 4. Comparison of blackbody emission from an ideal black
surface with 1 cm
2
area and the average power of our THz beam and
peak power of the ultrashort THz pulses.
sample from the thermal background. Even the peak power
(taking as pulse width a value of 0.5 ps) is only 27 µW. These
estimates are quantitatively illustrated in figure 4, where the
spectral power density of the thermal background radiation
emitted from a 1 cm
2
area at 300 K is compared to the average
and peak spectral power densities experienced by a sample
with an area of 1 cm
2
in our spectrometer. The radiation
load experienced by a sample from the thermal background
is several orders of magnitude larger than the average power
incident on the sample from the spectrometer. Besides being
a non-invasive, contact-free technique, THz-TDS is therefore
also quite harmless.
When implemented in the transmission mode, a natural
limitation of our approach occurs in optically thick samples.
Our current dynamic range for values of αd is from 0.05
to 20, the lower limit providing ample room for technical
improvement. The upper limit for αd is more fundamental,
and is linked tightly to the frequency-dependent signal-to-
noise ratio of the spectrometer. The limitation in penetrating
optically thick samples with a signal sufficient for pulse
characterization can be circumvented by operating in reflection
geometry [8].
3. Chemical recognition in terahertz imaging
It is tempting to explore further possible applications of this
technology. The potential of technical imaging in the very-
far-infrared (FIR) range has been considered as early as 1975.
Owing to the lack of easy-to-use sources at that time this
topic had not been pursued past an initial proof of principle
[9]. Recent advancements in ultrashort electromagnetic pulse
generation has revived this subject [10, 11]. In particular,
electro-optical techniques are now capable of upconverting an
FIR radiation field pattern into the visible domain [12, 13],
permitting live viewing of the spatial and temporal
distributions of the electric field strength. Very recently,
distinction between different types of biological material in
pulsed THz imaging [17], as well as detection of specific
chemicals in scanning continuous-wave THz imaging [18, 19],
was demonstrated.
In the following, we will demonstrate that THz imaging
combined with a straightforward THz-TDS analysis of the
transmitted THz pulses allows contact-free and reliable
recognition of drugs and other chemicals hidden in containers.
The applied recognition algorithm is simple and very fast, and
scales linearly with the number of chemical substances in the
reference database.
In figure 5(a) a photograph of a sample, consisting of
four pellets glued to a piece of paper, is shown. Each pellet
contains 60 mg of either lactose, aspirin, sucrose or tartaric
S248
Chemical recognition in terahertz time-domain spectroscopy and imaging
0 1020304050
0
10
20
30
40
50
X [mm]
Y[mm]
0
0.5000
1.000
(
a
)
(
b
)
Figure 5. (a) Visible image of sample with four pellets containing
different chemicals and (b) an image of the THz transmission
amplitude.
acid, mixed and pressed with 80 mg of polyethylene powder
for mechanical stability.
We record the full THz pulse shape (truncated 5 ps before
the main pulse and with a length of 70 ps) transmitted through
the sample. We use a 25 × 25 pixel grid, with a grid spacing of
2 mm. THz images can then be formed by different methods,
as will be discussed below.
In figure 5(b) the THz transmission image of the sample is
shown, obtained by recording the peak amplitude of the THz
pulse after transmission through the sample, which is mounted
on a 1 cm thick styrofoam plate and covered with a second
sheet of paper during the measurements. The four pellets are
clearly visible in the THz image, but no distinction between
the chemicals is possible by this simple imaging technique.
In order to identify and distinguish the chemicals in the
sample, we define a recognition coefficient R proportional to
the height of a spectral feature with respect to its baseline,
R = α(ν
2
) −
1
2
(α(ν
1
) + α(ν
3
)), (2)
where α is the absorption coefficient defined in equation (1). If
more than one clear spectral feature is present in the absorption
spectrum then R may be taken as the sum or product of several
peak heights, thereby increasing the specificity. For samples
of unknown thickness or irregular shape only the product
αd is measured. In this situation, the product of absorption
and sample thickness can be used in equation (2), as shown
below.
The absorption spectra of lactose, aspirin, sucrose and
tartaric acid shown in figure 6 are extracted from the THz
pulses transmitted through the relevant regions of the sample.
The solid lines show the absorption averaged over the
pellet areas (20–30 pixels), with vertical bars indicating the
corresponding standard deviation. This standard deviation
is significantly larger than that in figure 3 due to spatial
inhomogeneity of the sample. The lower absorption curve
in each panel shows the background absorption of the packing
material. The standard deviation is indicated for an average
over 30 pixels. The vertical lines indicate the frequencies
chosen for chemical recognition.
In figure 7 maps of recognition coefficient R for the four
chemicals in the sample are shown, using the frequencies
indicated in figure 6. For aspirin and tartaric acid we use
one peak for recognition, and for lactose and sucrose the sum
oftwopeaksisused.
In spite of its simplicity, the recognition strategy presented
here is clearly capable of identifying the four different
chemicals contained in the sample. Due to the prominent
spectral features in the absorption spectra of lactose and tartaric
acid, the signal-to-noise ratio of the recognition signals for
those substances is strong and clear. In spite of the weak
spectral features and relatively large background absorption of
aspirin and sucrose the recognition strategy is still successful,
although with slightly lower signal-to-noise ratio.
Within the limits of the dynamical range of the
spectrometer, the recognition coefficient is proportional to the
concentration of the chemical. With the proper calibration,
this strategy can therefore also be used to determine the
concentration of the chemical. This capability has been
demonstrated in experiments with quasi-continuous-wave THz
radiation by Kawase and co-workers [18, 19]. We note that
owing to the relatively broad line shapes in the THz absorption
spectra, chemical recognition with THz radiation will probably
find its best applications in situations where a known and
limited range of possible substances may or may not be present
in a sample.
4. Spectral signatures of biomolecules
Having seen that polycrystalline samples of organic molecules
offer specific responses to THz radiation, it is of interest
to investigate if specific responses can be obtained from
biological materials. However, since biological systems
typically lack a crystalline structure it can be expected that
only a limited group of biological materials will possibly show
distinct resonance frequencies in the THz range.
Biopolymers are a group of biologically interesting
materials with a natural long-range order. Among the
most important biopolymers are carbohydrate energy storage
molecules such as cellulose which is one of the most abundant
organic compounds in the biosphere, and chitin which is
responsible for the structural strength of exoskeletons of
insects and crustatea. DNA is another biopolymer of immense
importance for all life.
In order to investigate the possibility of finding spectral
signatures of biomolecules in the THz range, we have
measured the absorption spectra of cellulose, chitin and a
small oligonucleotide. The molecular structures of cellulose
and chitin are shown in figure 8, along with the chemical
structure of a small artificial single-stranded oligonucleotide
S249
B Fischer et al
0.5 1.0 1.5 2.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0
0
2
4
6
8
10
0.5 1.0 1.5 2.0
0
2
4
6
8
10
ν
1
ν
2
ν
3
ν
4
ν
5
ν
6
ν
1
ν
2
ν
3
ν
1
ν
2
ν
3
,
ν
4
ν
5
ν
6
ν
1
ν
2
ν
3
Lactose pixels
Absorption
α
d
Aspirin pixels
Sucrose pixels
Absorption
α
d
Frequency [THz]
Tartaric acid pixels
Frequency [THz]
Figure 6. Solid lines show the average absorption of lactose, aspirin, sucrose and tartaric acid in the sample. The lower curve in each panel
shows the absorption of the packing material. The error bars represent one standard deviation from the mean of typically 20–30
measurements. The indicated frequencies are used for chemical recognition.
0 5 10 15 20 25
0
5
10
15
20
25
Lactose
Y[mm]
0 1020304050
0
10
20
30
40
50
Aspirin
0
0.5000
1.000
0 1020304050
0
10
20
30
40
50
Sucrose
X[mm]
Y[mm]
0 1020304050
0
10
20
30
40
50
Tartaric acid
X[mm]
Figure 7. Maps of the recognition coefficient R for the four different chemicals in the sample.
with the base sequence A–T–A–T–A. The absorption spectra
of the three substances are shown in figure 9. The upper panel
shows the absorption spectra at room temperature and the
lower panel shows the corresponding spectra recorded at 10 K.
The samples consisted of 115 mg cellulose, 25 mg chitin
and less than 1 mg of the oligonucleotide. The chitin and
the oligonucleotide material were mixed with polyethylene
powder and all samples were pressed to pellets to ensure plane
interfaces of the samples. At room temperature, the absorption
spectra of all samples are dominated by a monotonously
increasing absorption. This absorption slope may be the result
of scattering since the particle size in the samples becomes
comparable to the wavelength at the highest frequencies.
Another contribution to this background absorption may come
from amorphous phases present in the samples. The spectral
shape resembles closely that of rapidly cooled, and therefore
amorphous, sugar melts [5].
When the samples are cooled to 10 K (lower panel of
figure 9), distinct absorption features emerge on top of the
broad background absorption. In cellulose, we observe an
absorption band centred at 2.15 THz with a FWHM width
of 0.35 THz. In chitin, we observe a weaker band centred
at 1.7 THz with a FWHM width of 0.45 THz. The
oligonucleotide displays an absorption feature at 2.4 THz with
S250