Terahertz Time-Domain Spectroscopy

Kjell Martin Mølster

September 2019

1 Terahertz Time-Domain Spectroscopy

This is a small, informal, introduction to Terahertz Time-Domain Spectroscopy
(TTDS). It will describe the general idea of how to measure signals and how
to extract optical parameters from these signals. Simulated signals are used
to show the steps, from [9], with which optical parameters are extracted. The
explanation are based on Kjell M. Mølster’s MsC Thesis.

Figure 1: A THz pulse is generated by a generation pulse (pump), propagated

trough a sample, detected at the THz detector with a sampling pulse (probe).
Reference measurements and sample measurements are explained in the next
paragraphs. Figure from [7]

2 Measurement principle
Terahertz Time-domain spectroscopy is a pump-probe spectroscopy technique,
where broadband THz pulses are measured at different points in time, using e.g.
Free-space electro optic sampling (FSEOS). A typical time-domain signal and
its Fourier transform is presented in Figure 2. The probing makes it possible

i
to fully describe the temporal profile of the wave, i.e. both its amplitude and
phase. A typical measurement is a two-step process. Firstly a measurement
of the setup is performed, i.e. without any sample, this is called the reference
measurement. The second measurement is done with a sample in the path of the
THz radiation, known as the sample measurement. By comparing the spectral
profile of the two measurements, one can extract the complex refractive index.

Figure 2: A typical TTDS time trace for sample and reference in transmission,
with their corresponding Fourier-Transform.

Measurements can be done with T-rays reflecting off or transmitting through

a sample. Transmission measurements introduce a relative time delay between
the reference measurement pulse and the sample measurement pulse. Reflec-

ii
tion measurements, with the exception of some special techniques (e.g. ATR),
requires a physical replacement of a mirror with a sample between the two mea-
surements. This replacement often adds to the complexity of the setup and it
introduces an uncertainty in the phase shift due the positional shift between
sample and mirror [2], creating a preference for transmission measurements.
Electromagnetic radiation is subject to attenuation, in an extreme case a
pulse is indistinguishable from the background noise. In the case of TTDS,
pulse attenuation due to absorptive materials can be lowered to an acceptable
level by reducing the thickness of the sample, however, thin samples are prone to
Fabry-Perot effects. For the most absorptive materials, reflection measurements
are necessary.

3 Calculations of optical parameters

Any electromagnetic signal can be described by its Fourier spectrum, given by
its Fourier transform Z ∞
E(ω) = E(t)e−iωt dt. (1)
−∞

In TTDS both measured signals, reference and sample, are Fourier transformed
and compared through division,

Es (ω)
H(ω) = = T (ω)eiφ(ω) . (2)
Er (ω)

Here Es (ω) and Er (ω) are the Fourier transformed sample and reference pulse,
respectively. Where T (ω) is the amplitude and φ(ω) is the phase. Both the
amplitude and unwrapped phase of the Transfer function H(ω) for the signal in
Figure 2 is included in Figure 3. A similar discussion on extraction of optical
parameters from TTDS is described in [8].

3.1 Optically thick samples with small absorption

Considering a sample of thickness d, such that the optical thickness is large
compared to the pulse width of an incoming beam at normal incidence, we can
write the transfer as
ωd dω
T (ω)eiφ(ω) = t12 e−κ(ω) c ei(n(ω)−nair ) c t21 . (3)

Fresnel’s coefficients t12 and t21 are approximately real valued for low absorptive
samples. We can solve analytically for κ(ω) and n(ω),

φ(ω)c
n(ω) = + nair (4)
d
c (n(ω) + nair )2
κ(ω) = − ln [ T (ω)]. (5)
ωd 4n(ω)

iii
Figure 3: The top shows the amplitude of the transfer function, while the
bottom show the unwrapped phase. Unwrapping is used as a means of removing
unwanted discontinuities when the phase jumps one cycle, i.e. when the phase
reach ±2π.

These are the simplest expressions for n(ω) and α(ω) [6]. The resulting optical
parameters for the simulated signals in Figure 2 is shown in Figure 4.
The thickness is the biggest source of error in calculations of n and κ, which
will be addressed in the laboratory lesson.

iv
Figure 4: The Absorption coefficient and refractive index of the simulated sam-
ple in Figure 2. Calculated using equations (4) and (5). From the calculated
refractive index, we can see that the simulated sample is a Lorentz medium,
with the Lorentzian fixed around 2 THz.

3.2 Samples with Absorption

Absorptive materials induce a non-negligible complex part in the transmis-
sion coefficient. As a result the formulas for extracting n and κ become non-

v
analytical. Numerical methods , such as Fixed-point iteration, become a neces-
sity to solve for the complex refractive index. We can represent the equations
by their measurable parts and their theoretical parts as follows,
4 p ω
T (ω) = 2 2
n(ω)2 + κ(ω)2 e− c κ(ω)d (6)
(n(ω) + 1) + κ(ω)
−κ(ω) ω κ(ω)
φ(ω) = arctan( ) + (n(ω) − 1)d + arctan( ).
1 + n(ω) c κ(ω)2 + n(ω) + n(ω)2
(7)

A rough outline of the calculations is included in the thesis.

3.3 Optically Thin samples (Fabry Perot)

In the previous sections we have extracted the complex refractive index n̂ with-
out considering reflection from boundaries inside the sample. This can be done
when the sample’s absorption and thickness will reduce the reflections or when
the thickness is large enough so that for all practical purposes the Fabry-Perot
effect is not observed. The astute reader may consider limiting the time-frame
of the experiment to remove any Fabry-Perot reflections. This is by far the
simplest method to remove the reflections, from both the sample and the elec-
trooptic detection crystal. This kind of time windowing may (a) remove parts
of the pulse (b) reduce the resolution of the scan. The latter will be subject of
a homework task.
Thin samples are defined as samples which have optical thickness comparable
to the pulse width of the beam, and for this section we assume that these samples
show low enough absorption that we see Fabry-Perot reflections. A typical FP
time trace and its spectrum is shown in Figure 5.
Apart from constraining the time window of the measurements, numerical
methods have been developed to handle Fabry-Perot reflections, and were first
mentioned by Duvialleret et al. [3, 4]. A more detailed discussion can be found
in [9].
We can introduce the reflections in the transfer function as follows,

H(ω) = T (ω)eiφ(ω) · F P (ω), (8)

where T (ω) and φ(ω) can be the expressions shown in equations (6-7) or equation
(3), i.e. assuming absorptive or low-absorptive material. We can write the
Fabry-Perot terms as,
i
X n(ω)ωd
F P (ω) = (r10 e−i c )2k . (9)
k=0

Here r10 is Fresnel’s reflection coefficient, and k is the k’th Fabry-Perot reflec-
tion term, whilst i is the maximum number of reflections, the case in Figure 5
gives i = 2. One possible approach to remove FP effects numerically is proposed
in [8], this method is based around making a guess for n and κ, which normally

vi
Figure 5: (TOP) Fabry perot time trace from 300µm GaP (The EO-detection
crystal) (BOTTOM) The absolute value of the Discrete Fourier Transform of
the pulse train.

vii
is done with equations (4-5). Afterwards for each frequency a numerical method
is applied to minimize the error between the measured transfer function and the
analytic, see [9, 8]. Any such numerical solution is not necessary for complet-
ing the laboratory lesson, but it is important that the student understand the
limitations imposed by limiting the time window.

4 Detection of THz by Free-space electro optic

sampling (FSEOS)
One can achieve detection of broadband THz pulses with FSEOS. The scheme
is based around the Pockel’s effect (see section ??). A weak Near-IR pulse and
the pump THz pulse propagates collinearly through an EO crystal. If the probe
pulse is smaller or in proportion to half a cycle of the THz pulse, the probe
sees a ”constant” birefringence induced by the THz pulse [7]. The resulting
instantaneous phase retardance from this birefringence can then be detected.
By adjusting the time delay between the pump and probe one can then sample
the THz pulse ET Hz (t). FSEOS measures both amplitude and phase of an
electric field with a precision of at least 10−2 rad [5]. A typical scheme for
FSEOS is included in Figure 6.

Figure 6: Scheme for FSEOS, an optic sampling pulse and a pump THz pulse
propagates collineraly towards an electro-optic medium, the outgoing pulse from
the EO medium propagates through a compensator, typically a λ4 waveplate.
The wollaston prism (polarizing beam spltter) creates a decomposition of the
pulse in y and x directions. These pulses are fed to two photo-diodes which
are cross connected to an anode and cathode, resulting in balanced detection
I = Iy − Ix . Figure is from [1]

We arrive at (proving this is a homework problem),

Iy − Ix = I ∝ ET Hz . (10)

viii
An important aspect to FSEOS is the alignment procedure, here one blocks
the THz-Pulse shown in Figure 6, and align the λ/4-waveplate such that the
outgoing probe pulse is circularly polarized, making equation (10) equal zero.
After this alignment, one can unblock the THz pulse, and the detected signal is
solely due to the birefringence induced in the EO-crystal by the THz pulse.
The tensor properties of the electrooptic crystal, implies that the orienta-
tion and crystal axes of the EO-crystal is vital to the nonlinear process in the
detection scheme. Two of the normal electro-optic crystal used in FSEOS are
ZnTe and GaP, both of zincblende structure (4̄3m). For the case of a h1 1 0i-
oriented zincblende crystal the THz field and the optical pulse should be aligned
in [1 1 0], and the optical pulse should be parallel to the THz field, to maximize
the birefringence in the zincblende electro-optic crystal [5].

5 Photoconductive antenna
The most widespread generation scheme of THz radiation is the photoconductive
antenna (PCA). The simplest version is shown in Figure 7. Two electrodes
are deposited onto a semiconductor substrate in which free carriers exhibit an
ultrashort lifetime. In the absence of any photo-excitation of charge carriers,
the semiconductor is highly resistive (intrinsic). Since no current flows between
the electrodes, it acts as a dipole, whose dipolar moment p is defined by the
charge difference and distance between the electrodes. In other words, the bias
voltage (DC-Bias) over the gap of distance W ,

Figure 7: A DC-biased PC antenna. The semiconductor substrate is shown

in yellow, metallic electrodes are connected to the DC bias, an optical pulse
(red) is incident on the area between these electrodes and a THz pulse (blue) is
generated. The length between the electrodes are given as W.

ix
 Now, a short optical pulse illuminates the region between the electrodes,
generating free carriers by photo-absorption. The photo induced carrier pop-
ulation first changes the permittivity of the semiconductor and subsequently
accelerates the carriers, whereof some reaches the electrodes. All these phenom-
ena creates a variation in the dipole moment over the electrode gap. It is this
variation which emitte the electromagnetic waves.
One can approximate the radiation from the PC antenna as a Hertzian dipole
where the antenna size is much smaller than the wavelength of the radiated
radiation. This is a valid approximation as the source size is approximately
the same size as the incoming optical pulse spot size (around 10µm), whilst
the wavelength of the THz wave is around 300µm for 1 THz [5]. By only
investigating the far field, i.e. where the propagation distance (r) is much bigger
than the wavelength of THz , we arrive at the equation for far field THz dipole
radiation in free space
2
~ T Hz = µ0 sin θ d [p(tr )]θ̂.
E (11)
4π r dt2r
Here θ is the angle between semiconductor front surface or i.e. the direction of
the THz radiation (θ = π2 for radiation as shown in Figure 7), p(tr ) is the dipole
moment of the source at the retarded time tr = t − r/c. Relating carrier density
ρ(r, t) to the dipole moment , and using the continuity relation

~ · J~ + dρ = 0.
∇ (12)
dt
We arrive at
dp(t)
= W IP C (t). (13)
dt
Where W is the spot size, which is approximated to be the same size as the
distance between the electrodes in Figure 7, and IP C is the photo-current. This
relates the emitted THz radiation to the time derivative of the current flowing
from one electrode to the other by

~ T Hz = µ0 W sin θ d [IP C (tr )]θ̂. ∝ dIP C (t)

E (14)
4π r dtr dt
With small pump power and DC bias the field amplitude increases linearly
with pump power, but is saturated quickly for increasing optical pulse power.
This saturation is due to the small area between the two electrodes, and is
strong relative to other THz emitter. Thus the THz conversion yield is great at
small pump powers.
The free carriers in the semiconductor substrate exhibit ultrashort lifetimes
(the average time for the charge carriers to recombine). This means that when
the carriers are excited in the semiconductor, they even recombine before reach-
ing the electrodes. This means that the pulse of current, i.e IP C (t) is very sharp
and short. This means that its time-derivative is maximized, which from the
above equation leads to a stronger THz electric field.

x
 ~ T Hz , we arrive at the relation
By investigating the Fourier-transform of E
ET Hz (ω) = ωµVDC−Bias Plaser . (15)
Here, we recognize that the THz electric field is proportional to the charge
mobility of the semiconductor µ, the applied electrode voltage VDC−Bias and
the incident laser power Plaser . This means, that the maximum radiation power
is limited by the breakdown voltage and the carrier mobility of the substrate
material. In addition to this, a high dark resistivity (short lifetime), is beneficial
such that there is no current before and after the laser pulse. Such that the
current pulse is as short as possible, and a broader spectrum of THz electric
can be generated. For more details, see [5].

6 Homework Problems
6.1 Free space electrooptic detection
We have already described the typical FSEOS setup (see Fig.6. Let’s assume
the incident detected intensity on each of the photodetectors (in balance) is I20 .
When the THz is incident on the ZnTe crystal, (polarization paralell to [110]),
each of the polarization is retarded, as a result of Pockel’s effect, by sin(∆φ),
such that
I0
Iy = (1 + sin(∆φ)), (16)
2
I0
Ix = (1 − sin(∆φ)). (17)
2
Show that the difference Iy − Ix is proportional to the THz electric field
ET Hz . HINT: You can assume the retardation to be very small. HINT2: You
need to consider the electrooptic tensor.

6.2 Bandwidth and resolution

Explain how you can vary the frequency resolution and bandwidth of a THz-
TDS signal. What will typically set restrictions for your measurements ?. These
are both connected to the translation stage (delay line) for the probe.

6.3 The Photoconductive Antenna

How does the photoconductive antenna work (qualitatively). Starting from the
equation,
ET Hz (ω) = ωµVDC−Bias Plaser . (18)
Explain how the different terms will alter the THz field strength. What limits
the device, and how can you make it better?
For the sake of curiosity, assume that we let the probe and THz pulse be
incident on the PCA whilst exchanging the DC-Bias with an Ampere-meter.
What will happen?

xi
6.4 Extraction of the optical parameters
Starting from

Es (ω)
H(ω) = = T (ω)eiφ(ω) . (19)
Er (ω)
Calculate extraction expressions for n and κ, for optically thick samples, low
absorptive samples. Explain what the different symbols in the equation mean,
and how they are related to the things we measure. What would happen if we
measured the absolute value |ET Hz | instead of ET Hz ?. HINT: You want to
arrive at the formulas in Section 3.1.

7 Laboratory Task
7.1 SAFETY
You will be working with a CLASS 4 Laser, meaning that it is in the class of
most dangerous Lasers. From Wikipedia: Lasers in this class have output
powers of more than 500 mW in the beam and may cause severe,
permanent damage to eye or skin without being focussed by optics
of eye or instrumentation. Diffuse reflections of the laser beam can
be hazardous to skin or eye within the nominal hazard zone. Many
industrial, scientific, military and medical lasers are in this category..
Generally, there will be a supervisor present during the lab, but there are
some general rules to be followed: (1) USE PROTECTIVE EYEWEAR
(2) DONT WEAR RINGS, WATCHES OR OTHER STUFF THAT
MAY REFLECT THE LASER LIGHT (3) DO NOT LOOK INTO
THE LASER BEAM, BEND DOWN TO LOOK AT STUFF IN THE
LASER BEAM HEIGHT OR PICK UP STUFF YOU LOST ON
THE FLOOR. Be careful.

xii
Figure 8: The Class 4 Laser, Tsunami, from spectra physics. Delivering over
500mW avarage power at around 80 MHz repition rate. The pulses are around
60 femtosecond in duration!

xiii
 Figure 9: PROTECTIVE EYEWEAR, should always be used.

7.2 The experimental setup

The THz-TDS setup used in this lab lesson is based on a PCA for generation
and ZnTe for electro-optic detection. The DC-bias is modulated, by a signal
generator and connected to a Lock-in amplifier (SR830). The modulation is
used as a trigger to lower the noise floor of the measured signal.

xiv
Figure 10: Experimental setup, THz-TDS. The first paraboloidal mirror is sit-
uated in such a way that they collect the radiation from the PCA such that it
is collimated. There is a focus in the region where the sample is situated. The
sample can move in and out of the THz path with a electrical motor. The THz
is then collimated again, before being focused and impinging on the Electrooptic
crystal. The probe beam, is propagating on the right hand side of the picture,
through the delay line and then through one of the last paraboloidal mirrors,
before meeting up with the THz radiation at the electrooptic crystal.

Two optical paths has to be considered in the optical setup, firstly, the THz
beam radiates from the PCA , and is collimated, focused, collimated and finally
focused again, by 4 off-axis paraboloidal mirrors. The first region of focus is
where the sample will be placed. The sample is mounted on a electric motor.
Secondly, the probe pulse, that is used to sample the THz by FSEOS is delayed
by a electrical motor, such that the THz electric field is detected as a function
of distance. This distance can easily be converted to time by considering the
velocity of light in air. All the pertinent programs are available on the computer
located next to the LIA/function generator. The programs are programmed
directly in MATLAB.

xv
7.3 Setting up the PCA

Figure 11: The Photoconductive Antenna (PCA) , mounted on a translation

stage

We will now proceed to do preliminary measurements of the PCA (Fig.11). The

most important is to check that the PCA is working as described my the man-
ufacturer. The PCA, as discussed earlier, should have a high dark resistance.
The specified dark resistance of this antenna is 1.5M Ω. The way this is mea-
sured is with a conventional multi-meter connected to a BNC adapter as seen
in Fig.13.

xvi
 Figure 12: Multimeter

Connect the multi-meter to the PCA BNC and measure the resistance with
no incident laser light on the Photoconductive Antenna.(Q1) Is the value rea-
sonable? (Q2) Why is it important with a large dark resistance? The equipment
is very expensive, be careful when connecting the BNC to the multimeter.

xvii
 Figure 13: BNC from PCA

We’re now interested in how the conductivity (i.e. resistance) changes when
a laser pulse is impinges on the surface of the antenna. You will now be handling
the laser beam itself, and it is important that you talk to your lab supervisor,
which will explain to you what to do. Measure the incident laser power before
the lens, this should be above 0.5W . Now, measure the resistance of the antenna.
(Q3) Why do you think the resistance change?
You will now put a bias over the electrodes of the antenna. The bias will be
modulated as a square wave by a signal generator, this modulation frequency
is fed into a Lock-in Amplifier to increase Signal-to-Noise ratio of the measure-
ments. The signal generator is situated on top of the Lock-in Amplifier as shown
in Figure.14

xviii
 Figure 14: Signal generator on top of the Lock-in amplifier

The Lock-in amplifier should be turned on, if not ask the lab supervisor to
help you. The power button of the signal generator is located on the front,
white button on the left of the grey panel. The modulation voltage should not
exceed 15V, as this is the limit of the PCA. (Q4) What sets the limit of the Bias
Voltage? (Q5) How would you expect the THz power to change with increasing
DC-voltage? Set the voltage to a reasonable amount, so that you may increase
it later, but not beyond 15 Volt. You need to turn on the output, but for now
you can keep it off.

7.4 Balance of photodetectors

We will now check that equal amounts of probe light is hitting both the photo-
detectors. (Q6) Why is this important?. In order to detect anything, you first
need to turn on both photodetectors. The photodetectors are shown in Fig.15.

xix
 Figure 15: The photodetectors used for balanced detection

See if you can see the beams that are hitting the two detectors. (Q7)
Explain the difference in polarization between them. (Q8) When comparing to
the homework, what does these beams represent?
We will now lock-in to another reference, namely a chopper reference. Con-
nect the chopper output trigger to the lock-in input trigger. Turn on the chopper
and verify that the trigger works. (Q9) Do you see any signal on the lock-in?
(Q10) What happens if you block one of the photo-detectors? Reason. (Q11)
What would happen if you rotate wave-plate in the polarization sensitive detec-
tion setup?

7.5 Measuring the first signal

Change the lock-in trigger to the signal generator, and turn on the output.
The detected signal is traversed using a delay line, in this setup for the probe
beam, implying that a scan of the line is necessary to position the best temporal
overlap between THz and IR-probe. By doing so, one can optimize the setup
alignment while letting the delay line rest on the position of highest measured
birefringence, known as resting on the peak. Use the program SimpleScan.mat
to find the peak and place the delay line on the position of largest signal.

xx
7.6 Rotating the ZnTe and ramping DC-bias
As you are resting the delay line on the peak, rotate the ZnTe crystal. (Q12)
Make a graph and explain what you observe. (Q13) Make a graph of THz
power as a function of DC-bias DO NOT EXCEED 15V.

7.7 Optimizing and measuring the reference and sample

While the setup is optimized, measure, with reasonable time constant (100-
300ms), the reference and sample measurements, by running the program
THZTDS.mat. (Q14) Use the data to extract the optical parameters (n and
κ) (first time the program will calculate it automatically). (Q15) Explain the
modulation of the spectra, how does it change the extraction. Looking at the
extraction calculations, the thickness of the sample is essential, (Q16) measure
the thickness using a caliper and microscope and calculate the refractive index
using both of these thicknesses.

7.8 Analysis
There are multiple ways of getting rid of the spectral ringing. One way is to limit
the scan region. There are two trivial ways of doing that. (a) Physically, but
shortening the delay line scan or (b) post-processing by limiting the time vector
in the program. Either way is fine. (Q17) Use one of these approaches and
explain how it affects the resolution and bandwidth of the experiment. (Q17)
What kind of limits does this set on the experiment? You can use the program
Extraction.mat to extract n and κ. (Q18) Compare your values to literature
and explain possible discrepancies.

7.9 Nitrogen Purge

A plexiglass box surrounds the THz-TDS setup. It is used to purge the air
inside the box such that there is less water in the air. (Q19) Why do you think
water absorbs THz while Nitrogen does not. (Q20) Make a new measurement
using a purged chamber. Calculate n and κ and explain the discrepancies.

References
[1] http://photonicswiki.org/images/e/eb/FSEOS of THz.png. Checked
09.09.2019.
[2] H. Němec F. Kadlec A. Pashkin, M. Kempa and P. Kužel. Phase-sensitive
time-domain terahertz reflection spectroscopy. REVIEW OF SCIENTIFIC
INSTRUMENTS, 74(11), 2003.
[3] L. Duvillaret, F. Garet, and J. L. Coutaz. A reliable method for extrac-
tion of material parameters in terahertz time-domain spectroscopy. IEEE
Journal of Selected Topics in Quantum Electronics, 2(3):739–746, Sep 1996.

xxi
 [4] Lionel Duvillaret, Frédéric Garet, and Jean-Louis Coutaz. Highly precise
determination of optical constants and sample thickness in terahertz time-
domain spectroscopy. Appl. Opt., 38(2):409–415, Jan 1999.
[5] Yun-Shik Lee. Principles of Terahertz Science and Technology. Springer,
2009.
[6] David G. Cooke Peter Uhd Jepsen and Martin Koch. Terahertz spec-
troscopy and imaging – modern techniques and applications. Laser Pho-
tonics Rev, 5:124–166, 2011.
[7] Rohit P. Prasankumar and Antoinette J . Taylor. Optical Techniques for
Solid-State Materials Characterization. CRC Press, 1st edition, 2011.
[8] Maik Scheller. Data extraction from terahertz time domain spectroscopy
measurements. Journal of Infrared, Millimeter, and Terahertz Waves,
35(8):638–648, 2014.
[9] Trygve Sørgård. ”terahertz time-domain spectroscopy of thin material sam-
ple, master’s thesis. Norwegian University of Science and Technology, 2015.
[10] Withawat Withayachumnankul and Mira Naftaly. Fundamentals of mea-
surement in terahertz time-domain spectroscopy. Journal of Infrared, Mil-
limeter, and Terahertz Waves, 35(8):610–637, Aug 2014.

8 Appendix
8.1 Unwrapping and Extrapolation of phase and the im-
pact on n and κ
This section is all about the phase φ(ω) term in the following equation

Es (ω)
H(ω) = = T (ω)eiφ(ω) . (20)
Er (ω)

φ is describing the ratio between the complex and real part of the trans-
fer function trough an angle in complex space. Using the phase term directly
in calculations of n and κ gives discontinuities phase jumps due to the phase
jumping from −π to π or vice versa. To prevent this, one use unwrapping which
makes the phase term continuous. Unwrapping is done by adding/subtracting
2π as the phase is 2π shifted each jump.
However, this may impose a 2πn bias in the phase throughout the whole
spectrum. Specially in the lower frequencies as there are more uncertainty in
the measurement due to lower spectral amplitude. A discussion on this can also
be found in [10].
Some authors argue that using a linear extrapolation from the middle fre-
quencies, where the phase is set to zero at ω = 0 is a good solution. This solution
is used by some authors and is discussed in detail in [3]. In this lab we do not

xxii
use any extrapolation. However, for the readers appreciation we have included
two figures that show how this impacts n and κ for a particular measurement
in this setup.
We extrapolated the measurement by using the phase between 0, 09 to
4.5THz. Then we use the extrapolated values for frequencies below 0.09THz.

xxiii
Figure 16: (TOP) Extrapolated phase and initial phase (MID) n for both cases
(BOT) Absorption coefficient for both cases.

xxiv