FACHBEREICH PHYSIK
UNIVERSITA¨T DORTMUND
Design and Realisation of a new
AMANDA Data Acquisition System
with Transient Waveform Recorders
Dissertation zur Erlangung des Doktorgrades
vorgelegt von
Wolfgang Wagner
Dortmund/Wuppertal, Oktober 2004
Contents
1 Introduction 1
2 Particle Physics and Cosmic Rays 5
2.1 The birth of Particle Physics . . . . . . . . . . . . . . . . . . . . 5
2.2 The spectrum of Cosmic Rays . . . . . . . . . . . . . . . . . . . 6
2.3 Particle propagation and detection . . . . . . . . . . . . . . . . 8
2.3.1 Neutrino production . . . . . . . . . . . . . . . . . . . . 12
2.4 Possible sources of high energy neutrinos . . . . . . . . . . . . . 12
2.4.1 Objects with black holes . . . . . . . . . . . . . . . . . . 13
2.4.2 Supernovae . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.3 Weakly Interacting Massive Particles . . . . . . . . . . . 18
3 The AMANDA experiment 21
3.1 Neutrino detection technique . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Ice properties . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 The MuonDaq . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.1 Time calibration . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 pADC calibration . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4.1 Track reconstruction . . . . . . . . . . . . . . . . . . . . 33
3.4.2 Energy reconstruction . . . . . . . . . . . . . . . . . . . 35
4 The new AMANDA Data Acquisition system TWRDaq 37
4.1 Motivation for a new DAQ system . . . . . . . . . . . . . . . . . 37
ii CONTENTS
4.2 The TWRDaq system . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 The Transient Waveform Recorder TWR . . . . . . . . . 41
4.2.2 Trigger logic - Veto logic . . . . . . . . . . . . . . . . . . 45
4.2.3 The TWRDaq readout system . . . . . . . . . . . . . . . 46
4.2.4 Verification of the synchronisation . . . . . . . . . . . . . 56
4.2.5 Filtering and Merging . . . . . . . . . . . . . . . . . . . 57
4.3 System stability tests . . . . . . . . . . . . . . . . . . . . . . . . 59
5 Reconstruction with Waveforms 63
5.1 Data processing and analysis . . . . . . . . . . . . . . . . . . . . 63
5.1.1 Hit extraction . . . . . . . . . . . . . . . . . . . . . . . . 63
5.1.2 Time synchronisation and calibration . . . . . . . . . . . 67
5.1.3 Photon counting . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Comparison of the performance of MuonDaq and TWRDaq . . . 72
5.2.1 Hit efficiency . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.2 Amplitude and dynamic ranges . . . . . . . . . . . . . . 76
5.2.3 Muon track reconstruction . . . . . . . . . . . . . . . . . 82
5.3 Estimation of the reconstruction quality . . . . . . . . . . . . . 85
5.3.1 Systematic error . . . . . . . . . . . . . . . . . . . . . . 87
5.4 Investigating signal characteristics using the TWRDaq . . . . . 88
5.4.1 Cross talk . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6 Proposal for a TWR based trigger system 91
6.1 Motivation and general idea . . . . . . . . . . . . . . . . . . . . 91
7 Re´sume´ and outlook 97
8 Outlook – Further work 101
A Data format 103
A.1 Description of the config file TWR.cnf . . . . . . . . . . . . . . 106
A.1.1 Name convention and content of the raw data files . . . . 108
A.2 Additional diagrams . . . . . . . . . . . . . . . . . . . . . . . . 110
A.2.1 Hit efficiency . . . . . . . . . . . . . . . . . . . . . . . . 110
CONTENTS iii
List of Figures 115
List of Tables 119
Bibliography 120
B Acknowledgement 125
iv CONTENTS
1Introduction
Some of the most exciting problems in astrophysics are the origin of the cosmic
rays and the understanding of the accelerating processes. Probably neutrinos
can play a role in answering these questions.
The charged part of the cosmic rays, e.g. nuclei, protons and electrons, are
deflected by magnetic fields and lose all directional information, while stable
neutral particles, i.e. neutrinos and photons, traverse the magnetic fields un-
deflected. In addition, charged particles and photons interact with matter and
various radiation fields and are absorbed or deflected. In contrast, neutrinos
have a very small cross section and can traverse the Universe mostly undis-
turbed.
Since many astronomical objects are obscured by matter and photon shields,
neutrinos are the only particle type to escape from the inner core of the ob-
jects, delivering information about the relevant processes.
Due to the small cross section of neutrino interactions, neutrino detectors have
to be very large in order to observe a sufficient number of events. High en-
ergetic neutrinos interacting with matter will produce secondary leptons, i.e.
muons, electrons and taus, which can be observed by these detectors. The
high energetic leptons can be detected in transparent media, i.e. water or ice,
by the emitted Cˇerenkov light.
Neutrino detectors suffer from a large muon background produced by cosmic
ray showers in the atmosphere. Therefore, the detectors are located under
ground to shield them from the atmospheric muon background. In order to
identify muons produced by neutrinos, the Earth is used as a filter. Muons
propagating upwards through the detector are considered as secondary parti-
cles from neutrino interactions, while downwards moving muons are related to
the atmospheric background.
The demands on neutrino detectors are
• a sufficient angular resolution < 5◦ in order to identify point sources
2 1. Introduction
accurately. This can only be achieved by a good relative time resolution.
• a sufficient energy resolution to measure the diffuse spectrum of neutrinos
over a wide energy range. Thus, the dynamic range of the detector has
to be large.
• a high sensitivity at low energies to enable detection of low energetic
neutrinos from supernovae and exotic particles.
• a dead-time free operation of the detector.
One of these underground detectors is the Antarctic Muon and Neutrino
Detector Array AMANDA, a Cˇerenkov telescope, located in the deep ice at the
geographical South Pole. The detector consists of 19 strings of photomultiplier
tubes PMTs. The signals from the PMTs are measured by Time to Digital
Converters TDCs and peak sensing Analog to Digital Converters pADCs.
AMANDA is a working neutrino telescope, but many opportunities remain to
improve the detector. The present Data AcQuisition DAQ system – MuonDaq
– limits the detector capabilities. The major critical points can be summarised:
• The detector saturates for high energy events above 1016 eV limiting the
sensitivity for high energy events and the energy resolution. Since the
search for the sources of the extraterrestrial neutrinos is the main goal of
neutrino astronomy, it is important to make detectors sensitive for high
energy neutrinos [LM00]. It is expected that at an energy above 1016 eV
the muon flux from neutrino interactions exceeds the atmospheric muon
flux allowing for a better background rejection.
• The dead-time of the MuonDaq amounts to 15% [Sch02].
• The fact that the pADCs measure only the maximum amplitude causes
ambiguities for complex signals.
The purpose of this thesis is the design and realisation of a new DAQ for the
AMANDA detector in order to avoid the limitations of the present system
and to extend the capabilities of the whole detector. The new system uses
Flash Analog to Digital Converters FADC to digitise the full signal of the
light sensors in each event.
A short introduction to high energy cosmic ray physics is given in chapter 2.
Possible sources for extraterrestrial neutrinos are discussed and different
models of neutrino fluxes from these sources are presented.
In chapter 3, a description of the AMANDA detector and the present DAQ
system will follow. The principles of calibration and reconstruction are also
explained in this chapter.
3The layout of the improved new AMANDA DAQ system – TWRDaq – is
decribed in chapter 4. The first steps towards an analysis of the data from
TWRDaq are presented in chapter 5. The first results of a comparison
between MuonDaq and TWRDaq are also presented in chapter 5.
Finally, the first steps to build a new trigger system based on an online analysis
of the captured data are described. This system will be capable of search-
ing for coincident hits within the detector volume and is presented in chapter 6.
4 1. Introduction
2Particle Physics and Cosmic
Rays
2.1 The birth of Particle Physics
With the discovery of natural radioactivity, in 1896 by Henri Becquerel, and
the discovery of the first elementary particle, the electron by Sir Joseph J.
Thomson in 1897, the era of particle physics began. These discoveries gave
the first hints that the atom is not the smallest constituent of matter.
In 1900 it was already known that the discharging rate of an electroscope
depends on the amount of ionised particles in the air. Furthermore, it
was found that an electroscope discharges in the presence of radioactivity.
Initially, the origin of ionisation in air was assumed to be due to natural
radioactivity in the air or from the Earth. A first measurement on the Eiffel
tower showed a slight decrease of the discharging rate with height, but there
was still a high level of ionisation. The absorption of radiation emitted by
natural radioactivity sources in the air was measured on Earth. It was soon
clear that the origin of the ionisation in high altitudes could not be the Earth.
Initially, the radioactivity of air itself was considered as the likely source
of the ionisation. However, it was demonstrated that the level of radiation
(ionisation rate) was too large to be accounted for by radioactivity in the
air [YS85]. In 1911 and 1912, Victor Francis Hess studied the ionisation of
the air during several balloon flights. He discovered, that in contrast to his
expectations, the discharging rate of an electroscope decreases only for the
first several hundred meters in altitude; for higher altitudes it increases again.
He showed that at a height of 1800 m the ionisation reaches the same level
as at the Earth’s surface. At 5000 m, he observed an ionisation level which
was much higher than at sea level. At this time it was clear that Hess had
discovered a new type of radiation coming from space, now termed cosmic
6 2. Particle Physics and Cosmic Rays
rays today [YS85,Hes11,Hes12].
By experimenting with cosmic rays, many new elementary particles, from the
positron to the strange quark, were discovered during the following years.
With the development of new techniques to artificially produce and accelerate
particles, the era of accelerator physics began. Charged particles like electrons
and protons are accelerated to very high energies up to several TeV producing
new particles in interactions.
In the intervening years, many more particles have been discovered, ever
larger particle accelerators and detectors were built to examine their proper-
ties. With the development of the Standard Model in the early 1970s, the
understanding of elementary particles has reached a high level. For some time
cosmic ray physics stood in the shadow of accelerator and particle physics.
However, many open questions remain, which can be answered only by
experiments measuring high energy particles. Nature provides us with a beam
of high energy particles beyond the reach of man-made accelerators – the
cosmic rays.
2.2 The spectrum of Cosmic Rays
The ionisation measured by Victor Hess originated from secondary particles,
which are produced in interactions of high energy particles with air molecules
in the upper atmosphere. These primary cosmic rays are produced in distant
sources such as supernovae or stars. The interaction of a primary cosmic ray
particle produces a cascade of secondary particles, which develops in the atmo-
sphere. Figure 2.1 summarises the current understanding of the development
of air-showers. First hadronic interactions dominate the shower producing
pi0, pi±, K0, K± and other hadrons. The decay of these hadrons produces a
large number of leptons and photons.
Since the discovery of the cosmic rays, a variety of detectors have mea-
sured their spectrum with high precision. Cosmic rays arrive isotropically
from all directions at the Earth and their flux can be described by a power
law: dN/dE ∝ E−γ . Above 109 eV, the primary cosmic rays consist of roughly
98% of nuclei accompanied by 2% of leptons. The flux of nuclei consists of
about 87 % of protons, 12 % α particles and 1 % of heavier nuclei [Lon92].
Neutrinos, which also arrive in large numbers at the Earth, interact rarely with
matter and are not contained in this calculation. Apart from neutrinos from
the sun and from one supernova (SN 1987 A), no extraterrestrial neutrinos
have yet been detected [Sup98].
The primary cosmic ray energy spectrum, beginning at a few hundred GeV,
is displayed in figure 2.2. While the low energy evolution is governed by the
Earth’s magnetic field and solar modulation, blocking particles with energies
2.2. The spectrum of Cosmic Rays 7
+
Κ
−
Κ
0
Κ
−pi +pi
−
+
pi
0
Κ−
+
Κ
−
+µ
pi0
−
e
−
e e
+
e
+
e
+−
e
−
ee
+−
e e
+−µ νµ +µ
−νµ −
+
Κ −
+
pi
−
+µ
primary particle
γγ
γ γγ γn p
hadronic cascade
Cherenkov
light
hadronic interaction
with nuclei
electromagnetic cascade electromagnetic cascade
Figure 2.1: A Cosmic Ray shower in the atmosphere.
below several GeV [Gai90], the high energy spectrum reaches energies up to
1020 eV.
The behaviour of the spectral index γ is interesting, being stable over several
orders of magnitudes but showing two major breakpoints, where γ changes sig-
nificantly. At 1015 eV the spectral index changes from γ = 2.63±0.02(stat.)±
0.09(syst.) to γ = 3.10 ± 0.10(stat.) ± 0.10(syst.) and decreases again above
1019 eV [Wie98]. The break region around 1015 eV is called the knee, while
the region around 1019 eV is called the ankle [Gai90].
The power law behaviour of the cosmic ray spectrum can be explained by
the diffuse shock acceleration model. Particles are accelerated in successive
processes gaining in each step a small amount of energy, proportional to the
actual energy of the particle, in every step. With a constant probability to
escape, this process results a power law energy spectrum. However there is
a certain maximum energy for a specific accelerating system that depends on
the average magnetic field and the size of the object [Gai90].
It is expected that a supernova shock wave can produce the cosmic ray
8 2. Particle Physics and Cosmic Rays
spectrum up to 1019 eV. The breakpoint at the knee can be explained by
the leaky box model: Particles produced in local sources are confined by
the average galactic magnetic field to the local galaxy, with a certain small
probability they can escape from the galaxy. The mean time of residence in
the leaky box, until the particle escapes, can be estimated as τesc = 2 · 107
years [Gai90].
At a certain energy the gyro-radius of the particle exceeds the radius of
the galaxy and the particles escape from the confined galaxy. For energies
above 1015 eV, the particles are no longer confined to the galaxy, lowering
significantly the flux at higher energies.
Acceleration to higher energies above the ankle up to 1020 eV, can possibly be
explained by processes in Active Galactic Nuclei AGN.
Several different mechanisms can accelerate particles in the most luminous
sources. The understanding of these mechanisms is an important point in
particle- and astrophysics.
Greisen, Zatsepin and Kuzmin predicted in 1966 an upper limit to the cosmic
ray spectrum at ∼ 1020 eV, which is also referred to as the Greisen Zatsepin
Kuzmin GZK cutoff. Protons above this cutoff energy are attenuated due to
interactions with the CMB, resulting in a loss of energy and producing a Delta
resonance,
p + γ −→ ∆+ −→ n + pi .
The mean free path length for protons with energies above 1020 eV is predicted
to be about several Mpc only.
In the past years, the knowledge of cosmic rays has increased significantly.
However there are still many open questions concerning the cosmic rays of the
highest energies around the ankle. Since the origin of cosmic rays is still un-
known, an important goal of astronomy is to identify the cosmic ray sources.
Furthermore, the acceleration mechanisms that produce particles with the
highest energies are still unknown and have to be studied. Due to the low
fluxes and limited experimental sensitivities, the energy spectrum of cosmic
rays around 1020 eV is still uncertain. To answer these important questions,
different types of radiation can be used. The next section gives an overview of
the particles and their propagation.
2.3 Particle propagation and detection
Figure 2.3 presents an overview of astroparticle physics. The paths of differ-
ent particles are shown from the source to the detectors on Earth. Detection
2.3. Particle propagation and detection 9
(1 particle per m2-second)
Knee?
(1 particle per m2-year)
? Ankle?
(1 particle per km2-year)
Energy (eV)
F
lu
x 
(m
2  
sr
 s
 G
eV
)-1
109              1011            1013           1015           1017             1019           1021
104??
102??
10-1
10-4
10-7
10-10
10-13
10-16
10-19
10-22
10-25
10-28
Figure 2.2: The Cosmic Ray Energy spectrum. The major breakpoints, the
knee at ∼ 1015 eV and the ankle at ∼ 1019 eV, are clearly visible [Kam01].
of each type of radiation has particular advantages and disadvantages as de-
scribed below.
• Photons give mostly information regarding the surface of opaque ob-
jects. Furthermore, compact sources are often surrounded by matter or
dust clouds, obstructing the path of photons. Thus, it is not possible to
10 2. Particle Physics and Cosmic Rays




















   
   
   
   
   















  
  
  
  
  
B
Source region, e.g.
surrounding dustclouds,
galaxy
fluorescence
detector
dust clouds
interstellar
detector
airshower
intergalactic
magnetic fields
µ
µ
γ
ν
ν
AGN, GRB, TD
Source, e.g.
supernova,
underground 
detector
detector
satellite
carged particles
protons and other
atmosphere
airshower
γ
γ
ν
Earth
airshower
Figure 2.3: Survey of astroparticle physics. Only neutrinos can traverse
through dust clouds and are not affected by magnetic fields.
look into the core of the source.
Photons with an energy above 2 · 1014 eV interact with low energy
photons from the Cosmic Microwave Background CMB producing an
electron – positron pair. Figure 2.4 illustrates the energy dependence
of the horizon for interactions with background radiation fields in the
Universe including photons produced by the source itself.
γ + γbg −→ e
+e−
The flux of high energy photons from distant sources is limited by these
reactions. The horizon for a 1015 eV photon is only about 7 kpc.
• High energy charged particles are deflected by interstellar and inter-
galactic magnetic fields. Therefore, charged particles arrive isotropically
at the Earth without any memory of their initial direction. Only parti-
cles above 1019 eV are fast enough to retain directional information.
The GZK cutoff limits the flux of high energy protons above 1020 eV, as
2.3. Particle propagation and detection 11
Figure 2.4: The energy dependence of the horizon for gamma astronomy cor-
responding to optical depth τγγ = 1 assuming Ω = 1, q0 = 0.5 and h = 0.75.
The dashed line shows a second calculation using a different diffuse background
model. The shaded area is not visible for photon astronomy [Man96].
described above. Therefore, only very nearby sources are expected to be
seen from Earth [Gre66].
• Free neutrons quickly decay to protons with a half-life period of
885.7 ± 0.8 s [Gro04]. From the galactic centre very high energy neu-
trons above 1019 eV are capable of reaching the Earth before decaying.
Thus, only a small part of the Universe is visible with neutrons.
• Neutrinos interact with matter via weak interactions with extremely
small cross sections. In comparison to other kinds of radiation, neutrinos
are not affected by dust clouds or magnetic fields and propagate from
the source in a straight line. Neutrinos, which are produced in a source
can escape from the core of that source.
Neutrinos have been an exciting subject of physics research since their
discovery [R+56]. The experimental discovery of neutrino oscillations
with solar and atmospheric neutrinos provides an insight into physics
beyond the standard model [SNO02,Sup98].
12 2. Particle Physics and Cosmic Rays
Neutrinos offer the potential to discover sources which are not visible with
other methods and may allow the observation of the active core of a source
surrounded by dust shields. However, the small cross sections of neutrino
interactions are also a disadvantage, neutrino detectors have to be very large
in order to measure a sufficient number of events. Although the detection of
neutrinos is difficult, it offers exciting prospects for exploring the Universe.
2.3.1 Neutrino production
Neutrinos have no electrical charge and cannot be accelerated directly, but are
produced in the decay of primary accelerated particles. The most important
source for high energy neutrinos are weak decays of charged pions (and kaons)
pi+ −→ µ+ νµ −→ e+ νeνµ νµ
pi− −→ µ− νµ −→ e− νe νµ νµ
p −→ n e+ νe .
These pions are produced in hadronic interactions of protons, neutrons and
photons. In one scenario, high energy hadrons are accelerated and interact
with each other or with photons in the source or the surrounding region:
p + p −→ X + pi±,
p + γ −→ X + pi±,
and analogous processes involving neutrons. Thus, in order to produce the
neutrinos hadronic interactions are necessary. However, neutrinos can be
also produced in decays of heavy particles, e.g. heavy relics from the Early
Universe, see section 2.4.3. On the following pages a selection of possible
neutrino sources are discussed.
2.4 Possible sources of high energy neutrinos
In this section the most important models for the production of high energy
neutrinos are presented. In the subsection 2.4.1 models based on black holes are
discussed, while further subsections present supernovae and exotic particles.
2.4. Possible sources of high energy neutrinos 13
2.4.1 Objects with black holes
Some of the most fascinating objects in the Universe are black holes. First,
the English astronomer G. Michell considered in 1783 the possibility that a
large massive star can attract and even trap light. In 1916, K. Schwarzschild
developed a metrics describing the vicinity of a so called black hole. The radius
of this object is described by the Schwarzschild - radius Rs
Rs =
2GM
c2 ,
where G is the gravitational constant, M is the mass of the black hole and c
is the speed of light in a vacuum. In 1963, R.P. Kerr found a new metrics
including the angular momentum of black holes. It turns out that several
high luminosity sources in the sky can be explained by assuming a rotating
black hole as a central engine [Mir04].
Active Galactic Nuclei
Active Galactic Nuclei AGN are one of the most powerful sources in the Uni-
verse. Today’s standard theory is that they can be described by assuming a
galaxy with a super-massive black hole in the centre (M ≈ 106 − 108M)1.
The black hole attracts matter from the surrounding host galaxy forming an
accretion disk around the black hole [CMU95, Man97]. It is suspected that
most galaxies have a massive black hole in their centre. However to produce
the observed luminosity, the accretion rate has to be sufficiently high. In the
middle of figure 2.5 a sketch of an AGN is shown. Perpendicular to the accre-
tion disk, the AGN develops two relativistic jets which can be observed due to
their radio emission. The accretion disk emits at UV and optical wavelengths,
while the torus, which surrounds the centre, emits infrared light.
Due to their anisotropic structure, AGN show different appearances depend-
ing on the orientation with respect to the observer. An AGN with a jet pointing
directly towards the Earth is observed as a blazar; these objects vary rapidly
in intensity. An AGN observed perpendicular to the jets will be seen as a
radio galaxy. In this case the end points of the jets are visible due to radio
emission, while the core emits UV and optical light. There are further classi-
fications according to the galaxy type, features of the emission spectrum and
the intensity in radio emission, which are not discussed here. The production
of neutrinos in AGN is of particular interest to neutrino astrophysics. The
neutrino fluxes are tested assuming either a diffuse flux arriving isotropically
from all directions or as a flux from certain point sources. Since the emission
1M is the mass of our sun.
14 2. Particle Physics and Cosmic Rays
Figure 2.5: There are three different sources in the sky, which can be explained
by the same mechanism, a black hole accreting matter. Inter alia, they differ
in size and lifetime [Mir04]. The supermassive black hole in the center of the
middle object is called an Active Galactic Nuclei AGN. An AGN pointing with
its jet towards the observer is termed Blazar, while an AGN seen from the side
is considered as a Quasar.
processes inside the AGN are not fully understood, it is important to know
whether the acceleration of hadrons or leptons is the dominant source of the
observed radiation. In hadronic processes it is expected that a significant neu-
trino flux will be produced due to pion decay. Therefore, the detection of an
AGN as extraterrestrial neutrino source would help to understand the charac-
ter of the sources.
In figure 2.6, various extragalactic neutrino flux predictions and experimental
limits are shown. The models presented are listed below:
1. In the model of Becker, Biermann and Rhode (BBR) [JKB], flat and steep
spectrum sources are considered using the jet-disk symbiosis model by
Falcke et al. [FB95].
2.4. Possible sources of high energy neutrinos 15
log(Eν /GeV)
 
Φ
 
E ν
2  
/(G
eV
 (s
r s
 cm
2 )-
1 )
¶
·
¸
¹
º
»
¶
·
¸
¹
º
»
BBR
SS
MPR
M
RB
WB
limit (2000)
Atmospheric
charm
AMANDA II, unfoldedl
10
-10
10
-9
10
-8
10
-7
10
-6
10
-5
10
-4
2 3 4 5 6 7 8 9 10 11 12
Figure 2.6: Muon neutrino fluxes - (1): p p interactions; Becker, Bier-
mann and Rhode (BBR) [JKB]; (2): Stecker and Salamon, AGN neu-
trino flux (p γ) [SS96]; (3): p γ interactions, Mannheim, Protheroe, Rachen
(MPR) [MPR01]; (4): Mannheim (M) [Man95], , p γ in blazar and p p in its
host galaxy; (5): Rachen & Biermann (RB) [RB93], p γ; (6): Waxman and
Bahcall (WB) [WB99]; Prompt neutrinos from GRBs, assuming that these are
the sources producing the Cosmic Ray spectrum at E > 1019 eV. The dashed
lines give the atmospheric flux prediction from light baryon (i.e. no charm con-
tribution) decay (horizontal and vertical) [VZ80,H+95]. Dotted lines indicate
the atmospheric spectrum from heavy baryons, i.e. baryons having also charm
quarks as constituents. Amanda limit and data for muon neutrinos [M+].
2. The quasar model by Stecker and Salamon (SS) [SS96] is normalised to
the X-ray flux of AGN sources. It can be excluded by current AMANDA
limits [M+].
3. Mannheim, Protheroe and Rachen (MPR) [MPR01] predict a maximum
flux from blazar sources for p γ interactions.
4. The model by Mannheim (M) [Man95] includes p γ interactions in blazars
as well as protons from blazars interacting with protons in the host galaxy
16 2. Particle Physics and Cosmic Rays
of the active nuclei.
5. Rachen and Biermann (RB) [RB93] calculate the number of neutrinos
from AGN proton interactions with the CMB.
6. Waxman and Bahcall (WB) [WB99] give an estimation of the GRB neu-
trino flux, assuming that the Cosmic Ray spectrum at E > 1019 eV
results from these sources.
The dashed lines in the figure represent the conventional atmospheric neu-
trinos, as calculated by Volkova [VZ80]. The upper line gives the horizontal
neutrino flux while the lower line is the calculation for a vertical flux compo-
nent. The neutrino flux from baryons with charm contribution in the atmo-
sphere is indicated as dotted lines. Heavy baryons decay faster than light ones
(τ ∼ 10−12 s compared to τ(conventional)∼ 10−8 s). The spectrum is there-
fore flatter because the particles including charm quarks do not interact with
nucleons in the atmosphere. The model is given by [MRS03], the two lines in-
dicate the uncertainties in the model. The atmospheric neutrino spectrum as
observed by AMANDA is shown in figure 2.6. The data points are calculated
unfolding the detector spectrum [M+,Gee03]. The limit is calculated from the
data.
Gamma Ray Bursts
Gamma Ray Bursts GRBs were first detected by a satellite in 1967 as a high
intensity pulse of γ–rays. Due to the short time scales, of about ms to 100 s,
it is expected that the source is a very small and compact object. GRBs are
distributed homogeneously over the sky without any clustering in the galactic
plane indicating that they originate from extragalactic sources. Today, a rate
of about 670 GRB per year is detected [P+99].
Two classes of GRBs have been detected, long enduring GRBs with a time
scale ≥ 2 s and short GRBs with a shorter timescale. An afterglow detected
at optical and radio wavelengths has been measured from long enduring GRBs.
This made it possible to measure the distance to these objects using the red-
shift.
Since their luminosity is very high reaching up to 1052 erg, it is likely that
GRBs are objects, which emit most of the radiation in jets. The currently
favoured model is the Fireball model FM, which explains the development of
the jets, but does not specify the engine of the GRB. However, an object in-
cluding a black hole is likely [Lon92].
A possible scenario for a GRB is shown in figure 2.5. Due to the lack of hy-
drogen and helium emission lines in the spectrum, it is assumed that GRB are
related to SN type Ic. The progenitors of these supernovae are massive rotat-
ing binaries with little hydrogen and helium. A GRB on March 29th 2003 was
2.4. Possible sources of high energy neutrinos 17
identified as SN 2003 dh, by an unambiguous SN light bump in the spectra of
the GRB [S+03].
It is expected that due to the strong shockwave protons are accelerated, cre-
ating pions, which decay to large numbers of neutrinos. Thus, GRBs are also
an interesting prospective source for neutrino astronomy. Figure 2.6 includes
the flux of prompt neutrinos as predicted by the model of Waxman and Bah-
call [WB99].
So far, no significant neutrino flux from GRBs has been detected.
Microquasars
Microquasars are galactic objects and have a similar structure as AGN.
A stellar mass black hole or a neutron star is at the centre of the Micro-
quasar [Mir04]. Since the centre is not visible, this question can only be
inferred indirectly. The order of magnitude of the jet length is of the order of
light years and the diameter of the accretion disk is about 1000 km. Figure 2.5
(left side) shows a Microquasar.
Due to their small size the fluxes of microblazars show a high variabil-
ity [Mir04].
2.4.2 Supernovae
A star is formed by a hydrogen gas cloud, which contracts until fusion reactions
start, due to the high temperature and pressure in the core of the cloud. The
interactions of the hydrogen nuclei produce helium. During this period, the
star is in equilibrium since the pressure caused by the fusion reaction balances
the gravitational force.
If the temperature and mass of the star is high enough, it contracts again
until the next fusion reaction is started. Depending on the mass further fusion
reactions start until iron is produced. Since iron has the highest binding energy
per nucleon, a further fusion reaction will not gain more energy and the star will
finally collapse. During the collapse, which is called a supernova SN explosion,
the pressure inside the star will become very high initiating a process called
deleptonisation:
e− + p −→ n + νe ,
resulting in a (proto-) neutron star and a large neutrino burst. Although
neutrinos have a small cross section, they are trapped in the dense core of the
supernova and diffuse out over a time interval of several seconds [A+02c,B+91].
In addition, the material will heat up during contraction, photons produce
18 2. Particle Physics and Cosmic Rays
e+e−-pairs which are mostly reabsorbed due to the high density. However via
the following process they can produce neutrinos:
e+ + e− −→ να + ν¯α ,
which can escape from the centre of the supernova. In this process all neutrino
flavors (α = e, µ, τ) are produced in equal numbers [A+02c].
It is expected that from a distant supernova, the neutrino signal will be visi-
ble minutes to hours before the optical signal. Therefore a neutrino detector
observing this signal can be used as an early warning system for a supernova
in the future. The neutrinos emitted during the hot phase of the supernova
have an energy between 11 MeV and 25 MeV [A+02c].
In addition, high energy cosmic rays can be produced by shock acceleration in
supernovae shells. The accelerated particles can decay producing further high
energy neutrinos.
2.4.3 Weakly Interacting Massive Particles
An interesting problem in astrophysics is the quest for dark matter. Due to
observations of the velocity distribution of stars and galaxies, it has been found
that the mass of the visible matter is not sufficient to explain the dynamics of
galaxies and galaxy clusters. Apart from baryons and neutrinos, there must
be a large amount of invisible dark matter [Lon92].
So far, the nature of dark matter has not been identified. It is predicted that a
large portion of this matter is non-baryonic. Thus, the search for dark matter
concentrates on the so called Weakly Interacting Massive Particles WIMPs.
One possible scenario is that WIMPs are neutralinos, which are particles pre-
dicted by supersymmetric extensions to the standard model of particle physics.
Their mass is predicted to be in the TeV range.
WIMPs interact with surrounding matter only weakly or gravitationally.
When traversing through matter, WIMPs can lose energy by elastic scattering
off nucleons. Due to their large mass, WIMPs can be gravitationally trapped
in large massive objects like the stars or planets and accumulate in the centre,
where interactions between WIMPS may become more likely. They can anni-
hilate pairwise producing neutrinos [A+02a].
One possibility of detecting WIMPs is the detection of an excess neutrino
signal from the centre of the Earth or the sun [Eks04].
Taking into account the variety of possible sources presented on the preceeding
pages, a possible neutrino detector has to satisfy certain requirements:
• A sufficient angular resolution < 5◦ in order to identify point sources
accurately. This can only be achieved with good time resolution.
2.4. Possible sources of high energy neutrinos 19
• A high energy resolution is necessary to measure the diffuse spectrum of
neutrinos. Thus, the dynamic range of the detector has to be large.
• The sensitivity should be high at low energies to enable detection of
supernovae and WIMPs.
20 2. Particle Physics and Cosmic Rays
3The AMANDA experiment
The advantage of neutrino astronomy, the fact that neutrinos travel un-
hindered through matter and magnetic fields, also results in the technical
challenge of their detection. The detector volume has to be very large in
order to obtain a sufficient number of events.
The aim of the Antarctic Muon and Neutrino Detector Array AMANDA is to
detect high energy neutrinos. AMANDA is a Cˇerenkov detector, which uses
the deep ice at the geographical South Pole as active volume. The detector
consists of 19 chains – strings – of light sensors located between 1150 and 2350
m below the ice surface. Fig. 3.1 shows the detector in its present configura-
tion. The sensors were deployed between November 1995 and February 2000.
The detector has been chosen to be located at ∼ 1500 m below the surface,
because of the clear and bubble free ice at this depth. In addition, the ice
overburden shields the detector from high energy muons produced by cosmic
ray interactions above the detector.
This section contains a short survey of the detector and its operation.
3.1 Neutrino detection technique
Neutrinos interact in matter via charged current reactions, producing a sec-
ondary charged lepton ` accompanied by a hadronic cascade, where the lepton
takes a large fraction of the energy:
ν` + nucleon −→ ` + hadronic cascade .
The lepton loses its energy via continuous ionisation and stochastic processes,
e.g. bremsstrahlung, pair production, nuclear interactions and Compton scat-
tering.
22 3. The AMANDA experiment
Figure 3.1: The AMANDA detector since 2000.
If the velocity of a charged particle exceeds the velocity of light in the sur-
rounding matter, a cone of Cˇerenkov light is emitted. The angle θc of the
Cˇerenkov light emission with respect to the particle direction is determined
3.1. Neutrino detection technique 23
by the velocity of the particle v and the refractive index of the surrounding
matter n and is described by
cos θc =
1
βn
with β = v/c. The number of photons N per unit length x and wavelength
interval dλ due to Cˇerenkov radiation is given by [Leo94]
d2N
dxdλ =
2piαz2
λ2
(
1 − 1β2n2
)
, (3.1)
where, α is the fine structure constant and z is the particle charge. The
detection of charged particles e.g. electrons, muons and taus can be achieved
using water or ice as the detector medium, which are transparent for Cˇerenkov
photons. The emitted light can be detected using photomultipliers with high
gain.
Figure 3.2 shows the detection of particles with light sensors.
Muons lose their energy continuously via ionisation. Beyond several hundred
GeV stochastic processes, e.g. bremsstrahlung, nuclear intractions and pair
production, become more important. The muon energy loss can be described
by
−dEdx = a(E) + b(E) · E , (3.2)
where a(E) represents the muon energy loss by ionisation and b(E) describes
the energy loss by stochastic processes. The parameters a and b vary only
slowly with energy [Gro04]. In addition, a cone of Cˇerenkov will be observed
in transparent matter, which is displayed in figure 3.2(a). While the energy
loss due to Cˇerenkov is negligible, the Cˇerenkov cone allows the particles to
be detected and their trajectory to be measured. For particle which travel far
enough in the detector, it is also possible to detect the particle direction.
The mean deviation angle Ψ between the direction of the incident νµ and the
produced muon is expected to be [LM00]
Ψ = 0.7◦ · EνTeV
−0.7
.
In contrast, electrons lose energy very quickly in a small volume and produce
a short cascade of secondary particles, which is displayed in figure 3.2(b).
The tau produced by a ντ have short decay lengths causing a similar cascade.
Taus with an energy above 1015 eV will produce a unique signature in the
detector – two bright cascades connected with a short track, a so called “dou-
ble bang”. These high energy tau-induced cascades can be distinguished from
24 3. The AMANDA experiment
µ
p
ν
n ν
n
µ
e
(a) (b)
e
Figure 3.2: Detection principle of charged particles with light sensors using the
Cˇerenkov technique. A muon-neutrino produces a Cˇerenkov cone (a), while
electron-neutrinos produce a short cascade (b).
electron-induced cascades [W+02,JL95].
Reconstruction of particle trajectories and energies is performed taking into
account the arrival times and the number of photons in each photomultiplier.
The background for underground Cˇerenkov detectors mainly consists of atmo-
spheric muons, which are produced by cosmic ray showers in the atmosphere.
While hadrons, photons and electrons, produced by cosmic rays, are absorbed
in the atmosphere or at the surface, muons can travel larger distances in matter
until they either decay directly or come to rest and decay. Only neutrinos can
traverse the entire Earth. For neutrinos of energy below 1015 eV, the Earth is
nearly transparent. Therefore, muons with directions originating from below
the horizon are considered as secondary particles produced by the interaction
of a neutrino. Thus, the Earth acts as a filter. The ratio between background
atmospheric muons and neutrino induced muons near the detector is about
106 [A+02b].
For neutrinos with an energy above 1016 eV, which are referred to as Ultra
High Energy neutrinos UHE, the Earth becomes progressively more opaque.
3.1. Neutrino detection technique 25
The relation between the transmission coefficient for neutrinos traversing the
Earth and the zenith angle is shown in figure 3.3. Therefore, for identification
of UHE neutrinos, different methods must be employed. Muons from these
neutrinos mostly reach the detector from close to the horizon [H+01].
10
-17
10
-15
10
-13
10
-11
10
-9
10
-7
10
-5
10
-3
10
-1
10
90 100 110 120 130 140 150 160 170 180
Figure 3.3: Transmission coefficient for neutrinos when traversing the
Earth [Bec04].
3.1.1 Ice properties
Table 3.1 shows a comparison of the optical properties of ice and water. The
absorption length in ice is much larger than in water. In contrast, the effective
scattering length1 in water is much larger than in ice. Due to the large back-
1The effective scattering length Λeff was originally defined by optical oceanographers to
account for the mean scattering angle θscatt: Λeff = Λscatt1 − 〈cos(θscatt)〉
26 3. The AMANDA experiment
ground radiation produced by bioluminescence and K40 decay in salt-water,
ice as detector material proves to be a good choice.
Property Water Ice
absorption length 50 - 60 m 90 - 100 m
effective scattering length 2000 m 25 - 30 m
Table 3.1: Optical properties in water and ice. [LM00]
Figure 3.4: The depth dependence of the scattering length. The center of the
detector corresponds to a depth of about 1700 m.
The Antarctic ice formed in several thousand years influenced by weather
conditions and volcanic erruptions. Fluctuations of the dust density in the
atmosphere caused dust layers in the ice. The depth dependence of the
3.2. The MuonDaq 27
scattering length for Antarctic ice at the South Pole is displayed in figure 3.4.
The scattering increases with dust concentration. Above 1400 m scattering
is dominated by air bubbles in the ice, while below the air bubbles disappear
and scattering is mostly caused by dust layers. These dust layers are also
observed in AMANDA data by an increased hit probabilty of OMs at the
depth of these dust layers [W+99].
It was shown that in depths between 1400 m and 2300 m the ice properties
are suitable for a neutrino observatory [W+99].
3.2 The MuonDaq
In total AMANDA is composed of 677 Optical Modules OMs, which are
displayed in figure 3.1 (right). The OMs contain a photomultiplier tube
PMT and the high voltage electronics consisting of a voltage divider. The
AMANDA detector is equipped with 8 inch PMTs (5912-2) from Hamamatsu
with 14 dynodes. The PMTs are connected to the pressure glass sphere with
a silicon gel and can be operated at a high gain of about 1 · 109. In addition
most of the OMs are equipped with a nylon ball serving as diffuser, which
is connected via an optical fibre to the surface and is used for calibration
purposes.
This section presents an overview of the current AMANDA Data Acquisition
DAQ system, which is referred to as the MuonDaq. The reconstruction uses
the arrival time and the number of photons represented by every PMT pulse
from each OM.
Strings transmission technology number of modules deployment
1 - 4 coaxial cables 80 1995/96
5 - 10 twisted pair 222 1996/97
11 - 13 optical fibre 123 1997/98
14 - 19 optical fibre 252 1999/2000
Table 3.2: Comparison of the AMANDA strings [W+01].
Table 3.2 lists the different building stages of AMANDA. Construction can
only take place during the Antarctic summer season from November to Febru-
ary.
The analogue PMT signals are transmitted to the ice surface. Taking into
account new experience gained and new developments in electronics, various
techniques have been used for the analogue transmission over the years.
28 3. The AMANDA experiment
The first four strings deployed in the Antarctic season 1995/96 used coaxial
cables for signal transmission. The detector was enlarged in the following sea-
son 1996/97 by 6 strings equipped with twisted pair cables. This detector is
called AMANDA B10. It turned out that the twisted pair cables are more
vulnerable to electromagnetic interference between cables - called cross talk -
and electromagnetic radiation from outside the detector.
PMT pulses have a typical length of about 20 ns for a single photon electron
1.p.e. pulse. Due to dispersion, a PMT pulse sent along a 2 km electrical
cable will widen significantly from 20 ns to about 200 ns [A+00]. Thus, the
resolution in pulse separation is of the order of ∼200 ns.
Finally, the recent enlargements of the detector were perfomed using optical
fibres for signal transmission. The electrical signal current of the PMT drives
a light emitting diode LED or a laser diode, which is located in the OM. The
light signal is coupled into an optical fibre, which transmits the signal to the
surface, where a Positive intrinsic Negative diode PiN converts it again into
an electrical signal. For these OMs the pulse resolution is much better. Two
pulses arriving with a time delay of more than 20 ns can be resolved. The
detector in its present configuration is called AMANDA II [W+01].
A schematic overview of the current MuonDaq, is shown in figure 3.5. The
electronics at the surface are located in the MAPO2 laboratory, located at a
distance of 800 m from the geographical South Pole.
The analogue signal from the OM (1) is amplified at the surface. Electrical sig-
nals are decoupled from the cable and amplified by the so called Swedish Ampli-
fiers SWAMP (2), while the optical signals are converted back into an electrical
signals and amplified by the Optical Receiver Board ORB (3). While the ORBs
are equipped with two prompt and one delayed output, the SWAMPs provide
three prompt and one delayed output. The amplification for the prompt out-
put and the delayed output is different. In most cases, the amplification for
the prompt output is larger. The delay for the delayed output is ∼ 2µs.
One of the prompt outputs is connected to a discriminator (4), which is inte-
grated in the AMANDA trigger system. A further output of the discriminator
to a to a Time to Digital Converters TDC 3 (5) measuring the arrival time
and pulse length of the PMT pulses with a time resolution of ∼ 1ns. More
precisely, the TDCs measure the positive and negative crossing point of the
signal at a certain threshold, as determined by the discriminator. The TDCs
are capable of storing up to 16 edges. Assuming one leading and trailing edge
for each pulse, the TDC will record 8 complete pulses. However, since in some
cases one of the edge can be lost, the place in the memory can be filled with
an additional leading edge of a further pulse. Additional pulses are lost. This
limits the dynamic range and the sensitivity of AMANDA especially for Ultra
High Energy UHE events, since high energy events may contain more than
2named after Martin A. POmerantz, the first astronomer working in Antarctica
3The currently used TDC is a V693 from CAEN with 128 channels (VME module).
3.2. The MuonDaq 29






 
 
 






 
 
 
Trigger logic
(NIM)
Time to Digital 
Converter TDC
Analog to Digital
peak sensing
Converter pADC
save timestamp
GPS Latch
Transient Waveform
Recorder TWR
4
save timestamp
GPS Latch D
at
a 
re
ad
ou
t
TW
R
D
aq
Prompt Delayed
SWAMP
Swedish
Amplifiers
Prompt Delayed
ORB
Receiver Board 
Optical 
1
Discriminator
6
SPASE
external triggers
calib.
D
at
a 
re
ad
ou
t
M
uo
nD
aq
TDC STOP
2km
ADC GATE
GPS Latch
TWR trigger
New
el
ec
tri
ca
l
ca
bl
e
o
pt
ic
al
fib
re
Veto
Supernova
System SN
String Trigger
Multiplicity
Multiplicity
Adder
as electrical channels
Discriminator and Multiplicity ADer DMAD−2000
2 3
4
GPS
5
7
8
9
10
11
Figure 3.5: The current AMANDA DAQ system - MuonDaq
30 3. The AMANDA experiment
eight individual pulses in many PMTs.
The delayed output of the SWAMPs and the ORBs is connected to a peak
sensing Analog to Digital Converter pADC 4 (6), which samples the maxi-
mum pulse amplitude in a time window of about 6 µs around the trigger time.
Thus, for OMs with more than one pulse per event the association between
TDC times of a pulse and pADC pulse amplitude is ambiguous.
The global trigger time is recorded by a GPS Latch (7), which is described
in detail in section 4.2.1. The GPS Latch is synchronised with the central
AMANDA GPS clock (8).
The Discriminator and Multiplicity ADder DMAD-2000 is composed of dis-
criminators (4) and the Multiplicity Adder unit (9), providing several trigger
signals with different multiplicity thresholds. AMANDA uses a multiplicity
trigger with a sliding time window of 2.5 µs. The DMAD requests M ≥ hit
OMs within the time window. Currently, the threshold is set to M = 24. The
DMAD also provides an advanced local coincidence trigger called the string
trigger, which is sensitive to coincidences within a single string.
An additional trigger logic (10) consisting of various NIM modules is used to
produce the final trigger signals, which start the TDC, pADC and GPS Latch
and data readout. This trigger logic uses the trigger signals from the DMAD
and additional external triggers, e.g. an external trigger from the South Pole
Air Shower Experiment SPASE and calibration triggers.
As described in section 2.4.2, a supernova will produce many neutrinos with
an energy of about 107 eV, which produce short tracks of lengths of about 10
cm. Since the signals from these short tracks will not produce coincidences,
the multiplicity trigger is not able to detect a supernova. Nevertheless, the
number of arriving neutrinos is very large and will affect the signal rate of
each PMT in the detector. A system of scalers monitors the noise rate of each
PMT and searches for an enhancement in the full detector in certain time
slices. This so called Supernova System (11) is directly connected to the dis-
criminators [A+02c].
The dead time of the MuonDaq is currently dominated by the dead time of
the readout system. The deadtime amounts to 2.2 ms [Sch02] representing
about 15 % of the events. The multiplicity trigger threshold (M=24) has been
chosen to compromise between high energy and low energy efficiency. If the
multiplicity threshold is too low, the efficiency for low energy events with a
small number of participating OMs is high, but the probability for losing a
high energy event is high. Vice versa, a too high multiplicity threshold results
in a high efficiency for high energy events and a low efficiency for low energy
events.
4The 7164 Peak-Sense Analog-to-Digital Converter from Philips with a 12 bit resolution
and a voltage range of 4.096 V is used.
3.3. Calibration 31
3.3 Calibration
3.3.1 Time calibration
 delta t
threshold
Figure 3.6: Time slewing due to fixed threshold.
About 2 km of cable separate the OMs from the readout electronics at the
ice surface. In order to achieve sufficient timing resolution for reconstruction,
the arrival times of the pulses must be corrected for the time offset t0. These
offsets are measured by sending light pulses from a YAG laser5 to a nylon ball
located below every OM. This ball serves as a light diffuser [A+00].
With an Optical Time Domain Reflectometer the propagation time of the light
pulse down to the ball can be determined accurately. The light pulses propa-
gate a short distance through the ice and enter the PMT. By measuring the
arrival time of the arriving PMT pulses at the surface, the time delay t0 can
be measured [A+00].
An additional time-slewing appears due to the fixed discriminator threshold
and the dispersion of the pulses in the cable. The time-slewing is illustrated in
figure 3.6. A small pulse exceeds the discriminator threshold later than a large
pulse causing a systematic time shift. It was found that the time-slewing cor-
rection can be described as α/
√
pADC, where α is a constant for each channel.
This procedure is termed α-calibration.
The parameter α can be determined by the arrival time distribution displayed
in figure 3.7. A linear regression yields the value of the constant. The relative
time-shift between a laser pulse and the crossing of the discriminator threshold
versus 1/√pADC can be approximated by a linear function. The crossing of
the Y-axis gives the time offset t0.
The complete time correction applied can be written as
ttrue = tmeasured − t0 − α/
√
pADC . (3.3)
5Yttrium, Aluminium and Garnet laser. This term usually refers to an laser with a
wavelength of 532 nm. In fact it should be Nd (neodymium) YAG but Nd is usually omitted.
The name describes components of the synthetic gemstone used in the laser.
32 3. The AMANDA experiment
1/sqrt(ADC)
20
60
40
80
120
100
0.04 0.06 0.08 0.10
le
ad
in
g 
ed
ge
 (0
.5 
ns
)
Figure 3.7: Relative time shift between a laser calibration pulse from the
surface and the crossing time of the signal over the discriminator threshold
versus 1/√pADC [A+00].
with the true arrival time ttrue, the arrival time measured at the surface
tmeasured, the time delay due to cable transit times t0 and the measured pADC
value representing the pulse amplitude. Applying this technique a time reso-
lution between 4 ns and 7 ns can be achieved [A+00].
3.3.2 pADC calibration
The number of detected photons is an important piece of information. It is
expected that N photons arriving at almost the same time result in a PMT
pulse with an amplitude N times larger than a single photon pulse amplitude.
Since most of the PMT pulses are single photon pulses, the size of a single
photon amplitude can be extracted from the pADC spectrum.
Figure 3.8 shows the distribution of pADC values for one channel. The single
photon bump is visible. The mean pulse amplitude for a single photon is
determined by a fit. The number of photons contained in a pulse is determined
by dividing the measured pADC value by the fitted value for a single photon.
For larger pulses containing several photons, the number of photons can be
underestimated if some of the photons arrive with a time delay. The maxima
of the single photon pulses will not match exactly in time, resulting in a wider
pulse with a smaller amplitude. In these cases, the integrated pulse rather
than the pulse amplitude would be a better measure for the number of photons.
3.4. Reconstruction 33
pADC value
0 200 400 600 800 1000 1200 1400
co
u
n
ts
0
50
100
150
200
250
300
Figure 3.8: The distribution of pADC values for OM 15.
3.4 Reconstruction
The analysis of the data from AMANDA mainly focuses on track reconstruc-
tion and extraction of the particle energy. A good survey of the reconstruction
algorithms is found in [A+04a].
3.4.1 Track reconstruction
Particle tracks are reconstructed using a maximum likelihood method, based
on probability density functions p.d.f. of the arrival time of individual photons.
Assuming that a particle with energy E0 traverses the detector on a track
~r(t) = ~r0 + cvac(t − t0) · pˆ , (3.4)
in the direction defined by the unity vector pˆ, with nearly the vacuum velocity
of light cvac, the arrival time tgeo of an unscattered photon is given by,
tgeo = t0 +
pˆ(~ri −~r0) + d tan θc
cvac
,
34 3. The AMANDA experiment
cθ
cθ
OM
µ
d
x
Cherenkov light
t  ,    , E0     0      0
PM
T-axis
η p
r
ri
Figure 3.9: Cherenkov cone: definition of track parameters.
where ~ri is pointing to the OMi and ~r0 defines the point, where the Cˇerenkov
photon is emitted. The delay between the expected arrival time as calculated
from a chosen track and the measured arrival time in a single OM is labelled
time residual: tres = thit − tgeo. A schematic overview of the track geometry is
given in figure 3.9.
The Likelihood function is defined as
L(x|~r0, t0, pˆ, E0) =
∏
i
p(xi|~r0, t0, pˆ, E0) (3.5)
where xi describes all information delivered from AMANDA, i.e. arrival time
ti, pADC value pADCi and the time over threshold TOTi of the OMi. The
p.d.f.
p(xi|~r0, t0, pˆ, E0) (3.6)
represents the probability density of observing a photon at time ti in OMi for
a track with parameters (~r0, t0, pˆ, E0) . The p.d.f. is a function depending not
only on the track parameters, but also on the PMT orientation and the ice
properties. Currently, a simple parameterised p.d.f. based on a homogeneous
ice model is used. This function, named Pandel function, uses a small set of
empirical parameters and the absorption length λa. The Pandel function used
by the reconstruction program has been obtained from simulated data [K+97].
The likelihood function used in the analysis presented in this thesis is a Single
Photo Electron SPE function, which considers every arriving photon to be
3.4. Reconstruction 35
independent of the other photons and describes only single photon pulses.
In order to achieve accurate track reconstruction, a hit-cleaning procedure is
applied to the data. During the hit-cleaning OMs with a high noise rate are
excluded from the reconstruction. In addition, noise hits and after-pulses are
also excluded from the analysis. In principle, hits that are isolated in time and
space are considered as noise hits [A+04a]. Afterpulses are late pulses inherent
to PMTs and do not contain any extra information6.
Since the reconstruction only accounts for the arrival times and the number
of photons in the OM, these values have to be calibrated from the raw data.
During calibration the arrival times are corrected for the offset due to cable
length and for time slewing due to different signal amplitude as described in
section 3.3. The number of photons is determined by dividing the measured
pADC amplitude by the expected size of a single photon pulse.
The reconstruction algorithm has to be initialised by a first guess track as a
seed for the maximum likelihood fit. In this thesis, the direct walk algorithm
was used. Data selection of, e.g. upgoing tracks, can be accomplished based
on the first guess algorithm [A+04a].
The final track parameters are found by maximising the likelihood function. So
far, track reconstruction is often performed using the first pulse. In addition,
several new techniques have been developed to use the information from pulses
following the first pulse. A detailed review of muon track reconstruction in
AMANDA is given in [A+04a]. The reconstruction of cascades follows a similar
procedure and is described in [A+03] and will not be discussed here in detail.
One algorithm which allows the separation of cascade and track like events,
is the “tensor of inertia fit”: All hit OMs are weighted with their measured
pADC value and considered as virtual mass points in a volume. The tensor
of inertia calculated from this virtual mass distribution has three main axes
corresponding to the eigenvalues. While for a cascade the three axes have
about equal size, one value will exceed the other values for a track.
However, the accuracy for track reconstruction is low for a single cascade,
because the Cˇerenkov light appears as a bright, spherical distribution in the
detector, which is illustrated in figure 3.2.
3.4.2 Energy reconstruction
While the track reconstruction is based mainly on the accurate arrival times
of the PMT pulses, the energy reconstruction also relies on the number of
6One possible cause for after-pulses are residual gas atoms in the PMT. A fast electron
moving from the cathode to the first dynode can ionise an atom. The ionised atom will be
accelerated in the electric field between dynode and cathode. When hitting the cathode a
number of electrons are ejected from the cathode forming a secondary cascade of pulses which
are called after-pulses. The time between the first direct pulses from the photon and the
after-pulses and the number of after-pulses are a characteristic for each PMT type [Leo94].
36 3. The AMANDA experiment
photons. Currently, three different approaches for the energy reconstruction
are used [A+04a]:
• The density of emitted photons along the muon track depends on its ac-
tual energy. An OM nearby the track without a hit is less probable. The
energy can be estimated from the likelihood function taking in account
the number of hit OMs and the number of OMs without a hit along the
track. The basic idea is that a track of a high energy particle is visible
in a larger volume than that of a low energy particle.
• An estimate of the number of photons is given by the pADC values.
Employing these values in the likelihood function results in an improved
energy resolution. This method is referred to as Full Ereco.
• Another method to measure the energy is to use a neural network
[Gee03]. This has been performed using several energy dependent input
variables. One of these variables is, for example, the number of pulses in
all OMs. The determination of the energy spectrum of νµ is performed
using regularised unfolding based on the work described in [Blo84].
The energy spectrum of νµ was measured from 1012 eV up to 1014 eV
with a high precision, shown in figure 2.6. However, it turned out that
the dynamic range of the energy reconstruction is limited for energies
above 1016 eV by saturation of input variables.
4The new AMANDA Data
Acquisition system TWRDaq
4.1 Motivation for a new DAQ system
It has been demonstrated in [A+01] and [A+00] that AMANDA is a fully
functioning neutrino telescope. Nevertheless, some opportunities remain to
improve AMANDA’s capabilities. The primary objective of AMANDA is to
identify astronomical sources of neutrinos. Figure 2.6 shows the measured en-
ergy spectrum of atmospheric neutrinos and several predictions of extragalactic
neutrino fluxes. It is expected that the extragalactic neutrino flux exceeds the
atmospheric neutrino flux above several PeV. Thus, the identification of ex-
tragalactic neutrinos is more promising at higher energies. In addition, it has
been shown that the effective volume of the AMANDA detector increases with
energy [B+01]. The muon energy is correlated to the integrated number of
photons in the detector.
The AMANDA detector uses ice as an active medium with non-negligible
scattering. The time spread of the PMT pulses due to scattering is larger than
the resolution of the pulse separation. Thus, the signal waveform produced by
a PMT, can be complex. A typical complex signal waveform of an electrical
channel is displayed in figure 4.1. The present AMANDA DAQ system – the
MuonDaq – has limited capabilities for these complex waveforms. The main
problems and ideas for improvement are outlined in the following points:
• The pADC delivers only one value per channel per event. Thus, the α-
calibration is ambiguous for channels with more than one hit and photon
counting is not possible. The goal is to utilise all of the information
available from the sensors in the ice without losing important information
due to such ambiguities.
38 4. The new AMANDA Data Acquisition system TWRDaq
event time /ns
20000 22000 24000 26000 28000
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2400
2600
2800
3000
3200
3400
Figure 4.1: A typical complex signal waveform from an electrical channel (OM
283).
• The current procedure to extract the number of photons from the pADC
works only if the photons arrive almost at the same time. If not, the
pulse amplitudes will be smaller and the pulses wider. This leads to
a systematic underestimation of the number of photons for pulses that
contain more than one photon.
• The TDC is only capable of measuring up to 16 edges, limiting the
dynamic range to 8 pulses. High energy events saturate the TDCs
leading to a non optimal resolution for these most interesting events.
For the energy and the track reconstruction of UHE particles, it is
necessary to have a large dynamic range for the number of photons
detected by each PMT. It is required to extend the dynamic range for
the number of detected pulses by a factor of ∼ 10. Due to more accurate
information with respect to the pulse amplitude, the dynamic range for
the integrated number of photons, increases even more by a factor of ∼
100. This helps to improve the sensitivity for high energy muons and
cascade events. Since ντ and νe are detected by observing the induced
cascades, with a large amount of light in a small volume, the increased
dynamic range is crucial.
The saturation of energy depending variables above a certain energy
was demonstrated using the neural network method for energy recon-
struction (see section 3.4.2). Figure 4.2 displays the relation between the
total number of PMT pulses expected in the detector – Nhit – and the
simulated muon energy Eµ. The saturation above 1016 eV can be clearly
seen. Due to the saturation of the input parameter Nhit, the dynamic
range of the energy reconstruction is likewise limited. Figure 4.3 shows
the muon energy Ereco, as reconstructed by the neural net, versus the
true muon energy Eµ from the simulated data. For this analysis the
4.1. Motivation for a new DAQ system 39
Figure 4.2: Relation between total number of hits and simulated muon energy
[Gee03].
neural net is only trained up to 5 · 1015 eV, where the saturation of the
input parameters begins.
• The dead time of the current system is approximately 15 %, resulting in
significant loss of data. The dead time is mainly caused by the readout
process of the DAQ. It is desired to reduce electronics-induced dead time
to a negligible level of less than 1 % using the TWRDaq.
These problems indicate the necessity to install a substantially improved
DAQ system. In actual fact, these goals can be achieved using the Transient
Waveform Recorder TWR [Sys03] – a Flash ADC, which is capable of record-
ing the complete waveform of a PMT. During the Austral summer 2001/2002
a test setup with 48 channels was installed. This test version of the TWRDaq
ran stably throughout the winter. A test on the time resolution compared the
arrival times of pulses from both DAQ systems. The test showed a good and
40 4. The new AMANDA Data Acquisition system TWRDaq
Figure 4.3: Relation between the muon energy Ereco as reconstructed by the
neural net versus the true muon energy Eµ (simulated data). The dashed line
is a fit through the MC data [Gee03].
stable correlation with an accuracy of about 30 ns. Figure 4.4 presents two
typical waveforms containing a single photon pulse from an electrical channel
(above) and an optical channel (below).
This system was extended in the following summer season 2002/2003 to read
almost the entire AMANDA detector. The TWRDaq was running in parallel
with the MuonDaq.
Finally, during the following summer season 2003/2004, this system was
upgraded again, increasing the readout frequency from 90 Hz to 150 Hz, which
allowed the multiplicity threshold to be lowered. The number of connected
channels increased to 597 out of 677. Since most of the reliable channels
were connected to the systems in 2003 and 2004 a full track reconstruction is
possible and can be compared to the MuonDaq.
Both the TWRDaq systems, running in 2003 and 2004, are described in this
document. Section 4.2 describes the physical layout of the system, while the
4.2. The TWRDaq system 41
event time /ns
22500 22600 22700 22800 22900 23000
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2600
2700
2800
2900
3000
event time /ns
22800 22900 23000 23100 23200 23300 23400 23500 23600
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
3030
3040
3050
3060
3070
3080
3090
3100
3110
Figure 4.4: Two typical waveforms containing a single photon pulse from an
electrical channel (OM 155 – top) and an optical channel (OM 622 – bottom).
following chapters discuss the extraction of information from the waveforms
and first experiences regarding the reconstruction using waveforms.
4.2 The TWRDaq system
4.2.1 The Transient Waveform Recorder TWR
The Transient Waveform Recorder TWR built by SIS1 is a 12 bit Flash ADC2
sampling the incoming signal continuously at a rate of 100 MHz, in a ring
buffer of 1024 samples, covering a total time window of 10.24 µs. The TWR
is a VME3 module, which is configured, controled and read out via the VME
1Struck Innovative Systems in Hamburg/Germany
2A parallel encoding Analog to Digital Converter ADC, sampling the incoming signal by
comparing it to a cascade of reference voltages with a certain frequency.
3Versa Module Europa
42 4. The new AMANDA Data Acquisition system TWRDaq
Parameter value
Number of channels 8
Sampling frequency 100 MHz
Time window 10.24 µs
Samples per waveform 1024
Resolution 12 bit
Voltage Range 5V (+1 V to -4 V)
Memory banks 2
Events per memory bank 128
Table 4.1: Characteristics of the TWR.
7 µsec3 µsec
T
Common Stop Trigger from MuonDaq
Time window TWR 10.24 µsec
without delay in TWR
of about 7 µsec
internal delay in TWR
with programmable
Figure 4.5: Adjustment of the event time interval with respective to the trig-
ger time. An appropriate time window around the trigger time is chosen by
delaying the incoming trigger signal.
bus connector. The characteristics of the TWR are summarised in table 4.1.
TWR is triggered externally. After each trigger pulse (event-trigger), which
serves as a common stop trigger, the TWR switches to the next ring buffer
and starts sampling the next event. Every TWR contains 8 channels. The
TWR has a total voltage range of 5 V reaching from 1 V to -4 V.
In order to measure a certain time window before and after the trigger pulse,
the incoming trigger signal is delayed by a programmable number of clock
cycles - trigger delay - inside the TWR. This is illustrated in figure 4.5. Since
the OMs are located at different depths, the incoming signals arrive with
different delays at the surface. In order to measure a similar time window
for every OM4 the OMs are distributed to the TWRs according to their
signal delays. The individual trigger delay for each TWR is programmed to
4Intended are 3 µs before the trigger and 7 µs after the trigger.
4.2. The TWRDaq system 43
accomplish this condition.
For the optical channels of AMANDA the fast output of the Optical Readout
Board ORBs are used as the input signal, while for the electrical channels the
fast output of the SWAMPs is used.
Every TWR has two memory banks each containing 128 events. The TWRs
are running in Autobankswitchmode, which reduces the dead time of the
system by separating the readout and the data-taking process. After starting
the sampling, the first memory bank is filled. During read out of the first
memory bank, the second memory bank is filled. As long as the readout time
for one memory bank is not longer than the time for receiving 128 trigger
signals, the data-taking and the readout process are independent and the
dead time is negligible. This case is very rare.
The TWR is equipped with a timestamp counter, which allows the measure-
ment of the time between events. In normal mode, it runs at 100 MHz. The
frequency can be down-scaled in order to measure larger time differences. The
sample clock frequency is not affected by down-scaling the timestamp clock
speed. This feature of the TWR enables the TWRDaq to find synchronisation
errors between the TWRs by comparing the time differences between adjacent
events in different TWRs.
In the current TWRDaq system, the absolute timing information is given by
the GPS time (see below), while the timestamp counter of the TWR is only
used to test the synchronisation within the present TWRDaq.
Every channel has an independent programmable detection threshold thdet,
which is set during initialisation. While the baseline of the PMT signals is
about 0 V, the signal amplitude is negative. The thresholds are adjusted
according to the individual noise behaviour of each OM. For a normal event
only 6 % - 8 % of the OMs contain a PMT signal below the threshold.
Therefore, it is essential to mark waveforms containing a pulse below thdet
and store this information in a dedicated register. A waveform is valid, if at
least one of the 1024 values is below thdet. During the readout process only
these valid waveforms are read out.
The GPS latch GPS2VME
The GPS2VME [Lei03] latch is basically a clock which is synchronised ex-
ternally by a Global Positioning System GPS clock. The synchronisation is
done via three electrical signals delivered by the GPS clock, located in the
AMANDA counting room. The different signals are listed in table 4.2. The
achieved accuracy of the time signal is 150 ns.
The current version of the GPS2VME is a VME module, which is configured
and read out using the VME bus. The GPS2VME is triggered by an external
trigger signal. For each trigger the GPS2VME latches the actual time infor-
44 4. The new AMANDA Data Acquisition system TWRDaq
Signal Description Frequency
1PPS Pulse Per Second 1 Hz
synchronising begin of second
10MHz 10 MHz clock pulse 10 MHz
fine synchronisation pulse
Timestamp serial timestamp 1 Hz
synchronise day, hour, minute
Table 4.2: Signals delivered from the GPS clock to synchronise the GPS2VME
module.
mation. This timestamp consists of 4 long-words, the number of seconds of
the actual day, the 10 MHz counter of the actual second, an event counter
and single bits indicating the stability of the incoming synchronisation signals.
The timestamps are stored in a FIFO, which is read out via the VME bus.
For more information about the GPS2VME see [Lei03].
Another function of the GPS2VME latch is to produce a trigger veto during
emission of strong radio signals5, which interfere with the data taking process.
The SIS3100 interface and the Digital Signal Processor DSP
To interconnect and control the various remote VME crates of the system,
holding the TWRs and the GPS latch, a VME to VME interface is used. A
complete VME/VME interface consists of two VME interface modules SIS3100
and an optical cable with two fibres linking them. In order to connect the PC
to the system, a similar PCI/VME interface is used, consisting of a SIS3100
and a PCI interface card SIS1100 with an optical cable in between. The VME
interfaces SIS3100 are used to build a network of VME crates. Some of the
SIS3100 contain a Digital Signal Processor DSP, which is optimised for pro-
cessing large amounts of data using fast and simple algorithms.
The DSPs programs are written in assembler and are loaded and started dur-
ing initialisation. For communication between the TWRDaq software and the
DSP, several registers located in the DRAM of the SIS3100 are used.
In addition, the SIS3100 modules in the remote crates provide several pro-
grammable NIM6 input and output registers, which are used to disable the
5The Very Low Frequency VLF experiment studies the lower ionospheric and mesospheric
signatures of charged particle using a radio beacon signal transmitted from South Pole. It
turned out that this signal results in a non negligible amount of cross talk in the twisted-pair
signal cables of strings 5 to 10 averting normal data taking. The AMANDA trigger has to be
blocked during the VLF emission, which take place four times per hour for about 1 minute.
6Nuclear Instruments and Methods signal level – logic ’1’ = -0.8 V, logic ’0’ = 0 V at
50 Ω termination [Leo94].
4.2. The TWRDaq system 45
trigger signal during initialisation [Sys02].
4.2.2 Trigger logic - Veto logic
LRS 621 AL
Discriminator
in
out
M−Mult 18
~150 Hz
M−Mult 80
~ 2 Hz
1
µ − DAQ
µs
Coincidence Unit
LRS 365 AL
out
in
coinc level = 3
veto
M−Mult 24
Stringtrigger
downscaled
trigger
TW
R
TW
R
TW
R
TW
R
G
PS
TW
R
Crate 0
Co
nv
er
te
r
LRS 429
in
Logic Fan out
2m
Trigger with
out
crates 1−5
Common Stop
Trigger enable
SIS3100 lemo out 2
12.0m
VLF trigger veto from GPSAMD
VETO−Logic for TWR system
Muon Trigger
located in NIM crates above DMADD
~90 Hz
~90 Hz
M−Mult 24
DMAD
TWR Multiplicity = 18 in 2004
TWR Multiplicity = 24 in 2003
2 3
4
5
SPASE
TWR Trigger
2ns
OM signal in
2004
2003
Gate =  25
NIM Logic
Dual Gate
Lecroy
222
out*start
out
30ns
located in the NIM Crate below TWRdaq
VETO DISABLE 1ns
Figure 4.6: Trigger setup.
The current version of the TWRDaq is triggered externally by the trigger logic
of the MuonDaq – the DMAD–2000. In 2003 the same multiplicity trigger
level as in the MuonDaq was used (i.e. M = 24 hit OMs in a time window of
about 2.5 µs with a trigger frequency of ∼90 Hz). Since February 2004 a lower
multiplicity level M = 18 has been used, resulting in a trigger frequency of
about 150 Hz, while the MuonDaq still uses the M = 24 trigger. The trigger
signal is distributed to all TWRs and the single GPS latch simultaneously.
A description of the trigger system including all important time-delays can be
seen in figure 4.6. The system is described in the text. The original trigger
signal from the DMAD goes through a veto-logic to stabilise the system. The
46 4. The new AMANDA Data Acquisition system TWRDaq
DMAD trigger logic issues a trigger signal when approaching the multiplicity
threshold. For a single event this can happen several times producing a burst
of trigger signals. An artifical dead time of about 20 µs is used to prevent
these multiple triggers. Trigger signals starting the TWRDaq in a short time
distance lead to half filled buffers and thus result in corrupted data.
The external trigger pulse from the DMAD (event-trigger) goes to a discrim-
inator (1) which serves as a pulse shaper. This signal is used as input for
a level three coincidence logic (2). Only, if the DMAD trigger signal, the
TRIGGER ENABLE from the TWRDaq and the VETO DISABLE shows a
logic 1, does the trigger pass starting the TWR trigger.
The TWR trigger signal launches a gate generator (3), which forms the VETO
DISABLE. The inverse output of the gate generator is connected to the logic
coincidence and disables the output of the coincidence module for the chosen
length of the gate signal. The delay between the TWR trigger signal and the
start of the VETO DISABLE is 30 ns, determining the length of the TWR
trigger signal.
The passing trigger is distributed by a logic fan out (4) to the different crates
holding the TWRs and GPS latches. In order to reduce the input cabling of
the crates and to make the system more reliable, the trigger is distributed via
the P2 bus7 on the VME backplane. The trigger is received by a converter
board (5) in the crates changing the signal level to ECL8 and transmitting it
via the backplane.
The incoming trigger signal is delayed inside the TWR by a programmable
number of clock cycles (internal clock of the TWR at 100 MHz) and serves
as a COMMON STOP TRIGGER to the TWR, which switches to the next
ring buffer and starts sampling for the next event. This external trigger
signal is delivered simultaneously to all TWRs and to the GPS latch, which is
synchronised by a connected GPS clock and stores the actual timestamp for
each trigger in a FIFO.
4.2.3 The TWRDaq readout system
The TWRDaq is a network of VME crates including one master crate con-
nected to several remote crates. The crates are connected by SIS3100 inter-
faces. A survey of the complete system of 2003 is shown in figure 4.7. The
local SIS3100 modules in the remote crates contain the DSP and act as a bus
master. The remote crate no. 3 also contains the GPS latch. The trigger is
distributed to all TWRs and the single GPS latch simultaneously. Thus, a
TWR event contains a GPS timestamp and a number of waveforms. As ad-
7Second free connector of the VME bus system.
8Emitter Coupled Logic
4.2. The TWRDaq system 47
SIS3300
TWR
DSP
SIS3100
SIS3300
TWR
SIS3300
TWR
DSP
SIS3100
SIS3300
TWR
SIS3100
FIFO
SIS3100
FIFO
  Polchomper
  WaveMerger
VME−VME
Bridge
SIS3100
FIFO
VME−VME
Bridge
Ethernet
Optical Fibre
.
.
.
VME VME
.
.
.
.
.
.
VME VME
Latch
GPS2VME
excternal trigger from MuonDaq
.
.
.
.
.
.
.
.
.
RIO 3
Power PC
Ethernet
PCI board
extracting of PMT pulses from waveform 
Feature Extraction:
Remote Crate 4
Remote Crate 3
Remote Crate 0
Master Crate
Figure 4.7: Survey of the TWRDaq system in 2003.
ditional timing information, the timestamps of the TWRs are stored. These
timestamps are currently used for testing only.
After 128 event triggers, the buffer of the TWR is full and the data is read out
by the DSP. The DSP program sorts the data according to events and sends it
via two optical fibres to the other side of the SIS3100 interface, located in the
master crate, which receives the data and stores it into a FIFO. In order to save
time and disk space, the DSP only reads waveforms from those TWRs, that
contain a pulse below the detection threshold. The information about which
waveforms are “valid” is delivered by the TWR. This procedure is called Zero
suppression.
In 2003, the TWRDaq readout was accomplished by two programs, the reader
and the eventbuilder. The system was controlled by the reader running on a
rio3 Power PC located in the master crate. The reader was a software package
embedded in the Multi Branch System MBS 9 [B+00]. The data were read
from the SIS3100 modules located in the master crate.
Finally, the data had to be collected from the rio3 and written into the final
data format. The rio3 was connected to a PC running Linux via Ethernet,
and the eventbuilder program, reading the data directly from a shared mem-
ory segment in the memory of the rio3, working as a client of the MBS system.
The current TWRDaq system has been running since February 2004. It is
displayed in figure 4.8 and is a modified version of the DAQ running in 2003.
9DAQ software package developed by the Gesellschaft fu¨r SchwerIonenforschung GSI in
Darmstadt, Germany.
48 4. The new AMANDA Data Acquisition system TWRDaq
SIS3300
TWR
DSP
SIS3100
SIS3300
TWR
SIS3300
TWR
DSP
SIS3100
SIS3300
TWR
SIS3100
FIFO
SIS3100
FIFO
SIS3100
DSP
Ethernet
Optical Fibre
SIS1100
PCI board
  Polchomper
  WaveMerger
VME−VME
Bridge
SIS3100
FIFO
VME−VME
Bridge
.
.
.
VME VME
.
.
.
.
.
.
VME VME
VME−PCI
Bridge
Feature Extraction:
extracting of PMT pulses from waveform 
Latch
GPS2VME
excternal trigger from MuonDaq
.
.
.
.
.
.
.
.
.
Remote Crate 5
Remote Crate 0
Remote Crate 3
Master Crate
Figure 4.8: Overview of the TWRDaq system in 2004.
Limitations in performance of the rio3 Power PC and the MBS system led
to the decision to upgrade the system. The changes made the system more
reliable and made it possible to increase the trigger rate up to 200 Hz.
A major change is the replacement of the rio3 Power PC by an additional
VME/PCI interface, consisting of a SIS1100 PCI card and a SIS3100 VME
board.
The system in 2004 is now started and controlled fully by the eventbuilder
software on the Linux PC. It is connected to the master crate by an additional
bridge, consisting of a SIS3100 module in the master crate and a SIS1100
(PCI card) in the PC, linked by an optical fibre.
The state diagram of the TWRDaq is displayed in figure 4.9. When
starting the TWRDaq, the TWR configuration file TWR.cnf is read and the
baselines are measured as described in section 4.2.3. Afterwards the TWRs
are initialised and the sampling starts by enabling the trigger signal. This
state is referred to as initialisation. The TWRDaq is running and waits for
the DSP to signal a full buffer of data.
After reading the data, the synchronisation is verified between TWR and GPS
time. In case of synchronisation errors the system is restarted (all modules
are reset and the system is started again), otherwise the system will wait for
the next data buffer. This procedure is explained later in this section. If an
unrecoverable hardware failure is detected, the system is aborted. In addition
the system can be stopped by an external interrupt.
Table 4.3 summarises the main features and differences between the TWRdaq
4.2. The TWRDaq system 49
synchronisation
Verify
Initialisation
Read TWR.cnf
Get baseline
Start TWRs
Start DSP program
Enable Trigger
TWRDaq running
Wait until DSP ready
OK
Start
successful
Controled shutdown VME or hardware problems
not 
successful
AbortStop
Reinitialise the system
Enable trigger
Disable trigger
Restart
next read not successful
Sync error
Data confirmed/write
Read data
Figure 4.9: State diagram of the TWRDaq.
TWRDaq in 2003 TWRDaq in 2004
number of remote crates 5 6
number of remote TWR s 72 75
number of connected channels 576 597
artificial dead time 20 µs 20 µs
DMAD multiplicity level 24 18
average trigger frequency ∼ 90 Hz ∼ 150 Hz
internal trigger delay 7µs individual for each TWR
feature extraction done by TWRDaq software DSP program
on rio3 in remote SIS3100
Table 4.3: Main features and differences between TWRDaq in 2003 and 2004
season.
in the 2003 and 2004 seasons. The first TWRDaq in 2003 used a fixed internal
trigger delay of about 7 µs, which led to a time window of about 3 µs before
and 7 µs after the trigger time for the signals at the surface. Due to different
50 4. The new AMANDA Data Acquisition system TWRDaq
cable delays for the OMs, an individual trigger delay for each TWR has been
used since 2004.
The further data flow is well described in the manual pages of the data-
handler [Ack]. The raw TWR data is written to tape and it is also sent to
other related processes. The final data format is displayed in figure A.2 in
the appendix. The filename convention for the raw data files is described in
section A.1.1 in the appendix.
The feature extraction
After zero suppression, a further data compression is necessary and is explained
in this paragraph. While the hit information of the MuonDaq only consists of
a few numbers, the data volume of a waveform is substantially larger. In the
present configuration, a full uncompressed waveform of one channel consists of
2048 bytes. Since the pulses are rather short, the waveform mostly contains
baseline noise.
In order to reduce the data amount significantly, an online data compression
algorithm is applied to the data, which is referred to as Feature Extraction FE
throughout this document. The basic idea is to extract the signal pulses from
the waveform including a number of preceeding and succeeding values. In 2003
the compression was implemented in the software running on the rio3, while
in 2004 it was implemented in the DSP program. Since the PMT pulses of an
electrical channel have a width of about 300 ns and the pulses of an optical
channel only about 30 ns, this is an effective way to reduce the data amount
at an early stage.
The complete feature extraction algorithm is shown in figure 4.21. Figure 4.10
n
succ
n
succ
precn
thend
*
*
*
*
*
* *
* *
* *
* *
*
*
*
*
*
**** * * * * * *
* * * * *
*
*
*
*
*
*
* *
* * *
* * * * * * * * * * *
detection threshold
end threshold
thdet
one fragment
baseline brun
Figure 4.10: Different thresholds are applied to the signal during the extraction
of pulses.
4.2. The TWRDaq system 51
visualises the different thresholds applied to the signal. The algorithm searches
for a value below the detection threshold thdet. If a value below this threshold
is found, a waveform fragment is created. First, a programmable number of
preceeding nprec values are added and all following values are added to the
fragment until the end of the pulse is detected. Since the slope of the leading
edge of the PMT pulse has a smaller slope and may overshoot the baseline
brun, an absolute threshold thend is defined for the end of the pulse. It is close
to the baseline and is calculated as
thend = brun − 4 digits .
In order to obtain the full pulse including the tail of the pulse, the end of
the pulse is detected by applying thend. Finally, nsucc samples are added to
the fragment after the end of a pulse. The thresholds are set in digits, where
1 digit corresponds to 1.22 mV.
If an adjacent pulse is detected within the nsucc samples, i.e. a sample surpasses
thdet again, the waveform fragment is continued with this pulse. Otherwise the
fragment is closed. Any further pulse is recorded in an additional fragment.
The waveform can consist of more than one fragment, which itself can contain
more than one PMT pulse.
The algorithm was not changed between the seasons 2003 and 2004, except
that in 2003 the complete waveform was written out, if the number of waveform
fragments per waveform was larger than 4. The extracted pulses are stored in
a binary data format described in figure A.3 in the appendix.
The parameters used for the feature extraction in 2003 and 2004 are listed in
channel type Parameter season 2003 season 2004
electrical number of preceeding samples 20 12
number of following samples 4 8
optical number of preceeding samples 20 3
number of following samples 4 4
Table 4.4: Parameters for the Feature Extraction used in 2003 and 2004.
table 4.4. The algorithm depends on the applied thresholds, which have to be
optimised to detect single photon pulses.
Determining the detection threshold thdet
The detection threshold thdet has to be determined for every individual channel
in the TWR. In order to make the system independent of baseline changes,
thdet is calculated from a measured baseline brun and a fixed threshold thamp as
thdet = brun − thamp, .
52 4. The new AMANDA Data Acquisition system TWRDaq
The threshold thamp is a fraction of the single photo-electron pulse and is
integrated charge / a.u.
0 1 2 3 4 5 6
Co
un
ts
0
10
20
30
40
50
60
70
integrated charge / a.u.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Co
un
ts
0
50
100
150
200
250
Figure 4.11: Distribution of integrated charges of an electrical (OM 1 – top)
and an optical channel (OM 625 – bottom).
obtained from the charge distribution histogram, see figure 4.11. The charge
of a pulse is defined as the area between baseline and the actual pulse. The
threshold is optimised in such a way that the single photon pulse is visible
while the PMT noise is minimised. Typical distributions of integrated charges
are displayed in figure 4.11, for an optical and an electrical channel.
The values of thdet have a fixed value for a whole season. They are listed in
the TWR.cnf file, which is explained in detail in the appendix. The value
of brun is determined once per run, during initialisation, by calculating the
average value of 128 waveforms. Since most of the waveforms are empty and
the pulses are small compared to the total waveform window, this procedure
delivers a good estimate for the baseline. Thus, brun has a fixed value for each
run10, which is contained in the header of each raw data file.
10The length of a run is 24 hours.
4.2. The TWRDaq system 53
Since the baseline can vary with time, an actual estimate for the baseline is
wfm[i_value]
< Threshold
sum := 0;
i_value < 32
wfm[i_value];
sum := sum +
i_value := 0;
sum := 0;
i_value := 992;
sum := sum +
wfm[i_value];
i_value < 1024
yes
no no
no
yes
yes
baseline :=
sum / 32;
Baseline based on last 32 values
Baseline based on last 32 values
Figure 4.12: Procedure for estimating the baseline. The full waveform con-
taining 1024 samples is stored in the wfm field. If no pulse is detected below
the threshold within the first 32 samples of the waveform, the baseline is cal-
culated as average of the first 32 samples. Otherwise the average of the last
32 samples is considered as baseline estimate.
necessary for data analysis. The value of brun does not give any information
about short term baseline shifts and the noise behaviour of the individual
channel. Thus, for every single waveform an estimated baseline is determined,
which is called dynamic baseline bdyn throughout this document. In the 2003
season the value of bdyn was determined by calculating the average value of
the first samples prior to the first pulse in a waveform. If a pulse appears
54 4. The new AMANDA Data Acquisition system TWRDaq
OM number
3660 3680 3700 3720 3740 3760 3780
dy
na
m
ic
 b
as
el
in
e 
/ m
V
1
10
210
OM number
3720 3740 3760 3780 3800 3820
dy
na
m
ic
 b
as
el
in
e 
/ m
V
1
10
210
310
Figure 4.13: Distribution of bdyn of an electrical (top) and an optical channel
(bottom). The dashed line shows the brun.
within the first 20 values of the waveform, the part of the waveform after the
last detected pulse was used. The calculation was done on the rio3 Power PC.
In the 2004 season the procedure was changed slightly in order to implement
the algorithm in the DSP. The procedure of estimating the baseline is sketched
in figure 4.12. At first, the DSP starts to loop over the first 32 values of the
waveform. If no value below the threshold is detected, the average value of
these 32 values is considered as an estimate for the baseline. If at least one
value below the threshold is detected, the baseline is determined by the last
32 values of the waveform. The probability to detect a pulse within the first
32 samples and within the last 32 samples of the waveform is small.
Figure 4.13 shows typical distributions of bdyn, for an electrical and an optical
channel. The deviation of the average value of bdyn from the brun is displayed
in figure 4.14 for each channel. A systematic shift appears, which is caused by
the procedure determining the brun. As explained before, brun is determined
4.2. The TWRDaq system 55
OM number
100 200 300 400 500 600
ba
se
lin
e 
sh
ift
 / 
m
V
-5
-4
-3
-2
-1
0
1
2
3
4
5
Figure 4.14: Difference between brun and bdyn versus channel number.
by integrating over 128 waveforms with a possible background of pulses in the
waveforms. Pulses that appear in the waveforms shift brun to a lower value
than the true baseline. Since this shift is small and stable, it is included in
the detection threshold.
Since the memory space for a program in the DSP is limited, the algorithm
has to be simple. The current algorithm, as displayed in figure 4.21, may
produce overlapping fragments. In order to reduce the amount of data and to
simplify the further analysis, such overlapping fragments are merged to one
compact fragment by the eventbuilder.
The efficiency of the compression mainly depends on the pulse widths and
the resulting fragment lengths. Figure 4.15 shows the distribution of the
fraction of remaining data as a percent of the initial data volume for an
electrical and an optical channel. As the electrical channels produce wider
pulses, the waveform fragments are longer and the compression efficiency is
lower for these channels. The average fraction of remaining data is displayed
in figure 4.16 for all channels. It can be seen that the compression efficiency
differs for different types of analogue transmission. While the first four strings
(OM 1 – OM 80), which use coaxial cables produce longer pulses and result
in a compression of about 8 % of the initial data, the waveforms from strings
5 to 10, using twisted pair cables for analogue transmission, have shorter
pulses resulting in a lower amount of data after reduction of about 5 %. The
waveforms of the channels using optical transmission (most of the OMs above
OM 302) are compressed to about 3 %. Some channels above OM 302 have
broken optical fibres and are read out by twisted pair cables. These OMs can
be identified in figure 4.16 by the larger amount of remaining data.
56 4. The new AMANDA Data Acquisition system TWRDaq
Fraction of remaining data %
0 10 20 30 40 50 60 70 80 90 100
Co
un
ts
1
10
210
310
Fraction of remaining data %
0 10 20 30 40 50 60 70 80 90 100
Co
un
ts
1
10
210
310
Figure 4.15: Remaining data amount in percent of the initial data illustrat-
ing the compression efficiency of an electrical (top) and an optical channel
(bottom).
4.2.4 Verification of the synchronisation
The data are read out by the DAQ in bunches of 128 triggered events. The
trigger signal arrives synchronously at the TWR modules and the GPS latch.
However, there is a small chance of a lost or an additional trigger, e.g. due
to electronics noise. A synchronisation error which is not noticed will lead
to a misalignment in the data. With an additional trigger in one TWR
after the Nth event, all following waveforms from this specific TWR will be
associated with a wrong event leading to corrupted data for all following
events. Thus, a misalignment must be avoided. However, the TWRs are
equipped with an intrinsic measure of time which can be used for relative
time measurements, the timestamp counter register. A synchronisation error
4.2. The TWRDaq system 57
OM number
100 200 300 400 500 600
Fr
ac
tio
n 
of
 re
m
. d
at
a 
%
0
5
10
15
20
25
30
Figure 4.16: Average remaining data amount of all channels.
is detected by inspecting the time difference between event 127 and event
128 using the timestamp counter of each TWR and the GPS timestamps.
First, the timestamp differences of all TWRs are compared to a reference
timestamp difference calculated from TWR 1 in the first crate. Then, the
reference timestamp difference is compared to the time difference of the GPS
timestamps.
In case of synchronisation errors, the system is stopped, the trigger is disabled
and the system will be reinitialised and restarted. The data is not written
out.
4.2.5 Filtering and Merging
In order to verify the data quality, events belonging to the same trigger signal
sampled by both MuonDaq and TWRDaq, are merged to one event. The
events are identified by the GPS time. This is performed by the wavemerger
program using the GPS times of both DAQ. A comprehensive survey of the
wavemerger and the further datahandling can be found in [A+04b].
The further data handling is shown in figure 4.17. The data from the
eventbuilder is written to disk and is collected by the Polechomper program.
This is a program running at the South Pole network, which is responsible
for data transfer and storage. All raw data from MuonDaq and TWRDaq is
stored on tape. The tapes are sent to the Northern hemisphere in the following
Antarctic summer. The whole year, a stream of data is sent continuously
via satellite to the outside world. Due to the limited bandwidth for the
data transmission from the Pole to the outside world, it is essential to filter
58 4. The new AMANDA Data Acquisition system TWRDaq
Disk
TAPES
SATELLITE
TRANSFER
TWRDaqMuonDaq
Disk
BACK OF SCIENCE
MAPO
F2kF2k
Monitoring files
Disk
raw data binary format
N_Wfm > 120
WaveMerger
READER
Cut in 2004:
+every 25th event
Cut in 2003:
N_OM > 140
+every 100th event
ROOT format
Figure 4.17: Survey of the online data-flow at the Pole.
the data and to transmit only the most interesting events, i.e. high energy
events. The wavereader reads the raw data from TWRDaq and sends it to
the wavemerger.
In 2003 all data from the TWRDaq was sent to the wavemerger, while
only events with more than 140 hit OMs in the MuonDaq were sent to the
wavemerger.
However, due to the higher data rate, this was not practicle in 2004. In 2004,
all data from the MuonDaq was sent to the wavemerger, while only events
with more than 120 hit OMs in the TWRDaq were sent to the wavemerger.
Events passing this cut are sent to the merging process, which searches the
data from the MuonDaq for events matching the TWR events.
In addition to these high energy events, a down-scaled stream of normal
events are also merged. Currently every 25th event is selected, which helps
to understand these low energy events in the Northern hemisphere. However,
many of these low energy events are not written to the standard AMANDA
format (f2000), because there are no matching events from the MuonDaq.
The data produced by the merging process is send to the satellite link and is
transmitted to the Northern hemisphere.
In order to monitor the system, the reader produces histograms which are
written into a root format and later merged with other root monitoring files
from the MuonDaq to produce a monitoring file for each run. These files are
sent via satellite to the AMANDA monitoring system [Ahr01].
4.3. System stability tests 59
4.3 System stability tests
In the Antarctic season 2003/2004 the multiplicity threshold M for the
TWRDaq was lowered from M = 24 to M = 18. Furthermore, some test
runs were made with a trigger multiplicity of M = 16. Figure 4.18 shows the
distribution of trigger rates for different multiplicities. For M = 16 the trigger
rate can rise up to 220 Hz without causing instabilities in the readout process.
The dead time of a data taking system can be determined by analysing the
average trigger rate / Hz
0 50 100 150 200 250 300
co
u
n
ts
1
10
M=24
M=16
M=18
Figure 4.18: Trigger rates using different levels of multiplicity from the trigger
of the MuonDaq. The trigger rate is determined in a time window of one
second. The measured trigger frequencies below 70 Hz occur just before or
after the VLF veto window (for details see section 4.2.1).
time differences between adjacent events [Leo94], which is shown in figure 4.19
(top). As already explained in section 4.2.2, an artificial dead time of about
20 µs was chosen to avoid overlapping waveforms. In figure 4.19 (lower), a
magnification of the first bin of the upper histogram is shown. A sharp cutoff
is visible, which is caused by the artificial dead time. Assuming that this dead
time is exactly 20 µs, the real dead time in percent of the total number of
triggers can be estimated to be 0.015±0.01 % for M = 24 and 0.018±0.013 %
for M = 18. Thus, the TWRDaq in the present configuration operates almost
without dead time.
The increase in events of smaller size, i.e. fewer waveforms, for low trigger
multiplicities can be seen in figure 4.20. It can be seen that, for a lower trigger
60 4. The new AMANDA Data Acquisition system TWRDaq
time distance / sec
0.02 0.04 0.06 0.08 0.1
co
u
n
ts
1
10
210
310
410
M=24
M=16
M=18
time distance / sec
0.001 0.002 0.003 0.004 0.005
-210×
co
u
n
ts
0
5
10
15
20
25
30
35
40
45
M=24
M=16
M=18
Figure 4.19: Time difference between adjacent events for various multiplicity
triggers (top). A magnification of the first bin is displayed below. The artificial
dead time of about 20 µs can be seen as a cutoff in the figure below.
multiplicity these events with a small number of hit OMs dominate. Further-
more, it was shown that using a multiplicy trigger with M = 16, the events are
still not dominated by noise events [Mes03]. Thus, they contain meaningful
data.
4.3. System stability tests 61
number of OMs with waveform
0 50 100 150 200 250 300
co
u
n
ts
1
10
210
310
M=24
M=16
M=18
Figure 4.20: The distribution of number of hit OMs per event in the TWRDaq.
62 4. The new AMANDA Data Acquisition system TWRDaq
i_value < 1024
yes
−−> output waveform in mem − array complete waveform in wfm − array
wfm[i_value]
< Threshold
i_value < 1024
yes
no
yes
i_value := 0;
i_mem := 0;
i_value <
N_PREC
yesno
i_frag := i_frag := 0;
i_value−N_PREC;
mem[i_mem] :=
i_frag < i_value
yes
i_mem++;
i_value++;
wfm[i_value]
< Baseline−4
no
no
wfm[i_frag];
i_frag++;
wfm[i_value]
< Threshold
wfm[i_value];
i_value++;
i_mem++;
i_after++;
i_after
< N_FOLL
yes
no
no
no
i_value < 1024
yes
yes
yes
no
i_after := 0;
Fragment Header
mem[i_mem] :=
0x4000 + i_frag;
i_mem++;
no
mem[i_mem] :=
0x2000 + i_value;
i_mem++;
mem[i_mem] :=
0x8000;
i_mem++;
Waveform End
Fragment End
wfm[i_value];
i_mem++;
i_value++;
mem[i_mem] :=
mem[i_mem] :=
Figure 4.21: The complete Feature Extraction Algorithm.
5Reconstruction with Waveforms
This section surveys on the reconstruction results as obtained with data from
TWRDaq. First, the hit extraction algorithm, which analyses the waveform
fragments and extracts hits, is described in section 5.1.1.
The quality of the analysis relies on the resolution of the measured pulse
start time. The time calibration of the TWRDaq is described in detail in
section 5.1.2.
In order to verify the correctness of the data from the TWRDaq, it is
compared with the data from the MuonDaq in section 5.2.
Finally, a first approach to the use of TWR data for track reconstruction
is described in section 5.2.3. The goal is to demonstrate that a track
reconstruction with data from the TWRDaq gives identical results to the
MuonDaq. The reconstruction uses the existing AMANDA software.
5.1 Data processing and analysis
5.1.1 Hit extraction
For the reconstruction of particle tracks and their energy in Cˇerenkov detectors,
the arrival time of the pulses and the number of photons in each pulse has to
be extracted from each OM. Thus, a hit in an OM consists of a timestamp and
a number of photoelectrons extracted from a PMT pulse.
First, the pulses in the waveform have to be identified by an algorithm. A
waveform pulse may be the superposition of a number of single photoelectron
pulses, which arrive at different times. Therefore, the resulting shape of the
waveform can be rather complex. In order to obtain the maximum information,
a good pulse separation is desirable for reconstruction, i.e. to extract arrival
times of pulses, even if they have only a small time separation.
64 5. Reconstruction with Waveforms
m−th_1
th_2
lm−th_2
th_1
lm
m
th_e
m−th_e
m     baseline
th_1  threshold 1st pulse
th_2  threshold following pulse
th_e  threshold pulse end
lm    local minimum between pulses
Figure 5.1: Relative thresholds used for hit extraction.
While the narrow pulses of the optical channels result in good resolution for
pulse separation, the electrical channels show a different behaviour. Due to
dispersion in the electrical cables, the received signals at the surface are wider
and the resolution for the pulse separation is worse.
The hit extraction uses different thresholds, which are illustrated in figure 5.1.
Similar to the feature extraction algorithm, the first pulse is found with a fixed
threshold, th 1, relative to the baseline m. Typical values for the threshold are
180 mV for the electrical channels and 18 mV for optical channels. Thus, the
applied absolute threshold is m - th 1 for a first pulse.
In order to identify an adjacent pulse in the falling edge of the previous pulse,
the algorithm searches for a local minimum, lm, by applying a different thresh-
old, th 2, relative to any local minimum. Since th 1 takes into account the
amplitude of a single photo-electron pulse from each individual channel, th 2
was chosen as a fraction of th 1. The pulse end is detected if the signal ex-
ceeds another threshold level, the e. After detecting the end of a pulse, the
algorithm searches for a next pulse, again using th 1 as the threshold level.
For the analysis in this thesis, the optimised thresholds (see section 4.2.3) for
the feature extraction are used for detection of the first pulses. For the pulse
end, a reduced relative threshold of th e = th 1 / 10 is used.
The pulse separation benefits from the extra information contained in the
waveforms as compared to the MuonDaq. For overlapping pulses, the local
minimum between the pulses serves as the baseline and the linear fit method
(see below) can be applied to all pulses. The advantage of the pulse separation
using the dynamic threshold is demonstrated in figure 5.2. While the TDC
detects only two pulses (lower dashed lines), the waveform analysis can resolve
four pulses (middle dash-doted lines).
The number of photo-electrons in each pulse is obtained by integrating the
5.1. Data processing and analysis 65
event time /ns
23600 23800 24000 24200 24400
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2200
2400
2600
2800
3000
Figure 5.2: A waveform measured with OM 180. The lower (dashed) vertical
lines show the leading and trailing edge as measured by the TDC, while the
middle (dash dotted) lines indicate the pulses found by the hit extraction.
The pulse separation benefits from the increased information contained in the
waveforms. The solid lines show the result of the linear fit method.
event time /ns
21200 21300 21400 21500 21600 21700
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2750
2800
2850
2900
2950
3000
3050
event time /ns
23200 23250 23300 23350 23400 23450
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2960
2980
3000
3020
3040
3060
3080
3100
Figure 5.3: Linear fit to determine the pulse onset time. An electrical pulse of
OM 115 (top) and an optical pulse of OM 521 (bottom).
area of the waveform between the baseline and the pulse itself and dividing
66 5. Reconstruction with Waveforms
it by the mean charge for a single photo electron 1 p.e. pulse. The charge
of each pulse is proportional to the area between the baseline and the pulse
itself. In order to get a meaningful unit for the charge, the area is transformed
into nC. The analogue signals are transmitted from the ORBs and SWAMPs
to the TWRs via coaxial cables, which are terminated by a 50 Ω resistor. The
charge C in nC is calculated as
C = 10ns50Ω ·
5V
4096 ·
n
∑
i=0
(ai − bdyn) ,
where ai are the sampled values of the TWR, n is the number of sampled
values in the pulse and bdyn is the dynamic baseline. The pulse onset time as
determined with a fixed threshold is not sufficient for the analysis. Sampling
the complete shape of the PMT pulses provides the opportunity to improve
the time resolution itself.
Linear fit method
The determination of the pulse onset time is performed using a linear fit to
the leading edge. The algorithm searches for the maximum rise between two
values, xi and xi+1, within the leading edge. In the case of an electrical
channel, the first sample before the maximum rise, xi−1, and the first sample
after the maximum rise, xi+2, are used to determine the tangent. For the
optical channels, the tangent is calculated with just the two values xi
and xi+1 of maximum rise. The difference in the procedures, for the optical
and electrical channels is due to the maximum number of values within the
leading edge. In the case of an optical channel, the pulse width is shorter
and therefore, the maximum number of values is smaller than in the electrical
case.
The crossing point of the tangent from the linear fit and the baseline is
considered as a good estimate of the pulse onset and is shown in figure 5.3 for
an optical and an electrical channel. This procedure would lead to incorrect
results for an adjacent pulse located on the falling edge of the predecessor
pulse. The pulse onset time of these is determined by the crossing point of
the linear fit and the local miminum, lm. Due to dispersion in the cables, the
electrical channels have a slower leading edge rise-time. Therefore, the time
difference between the beginning of the pulse found with the fixed threshold
and the pulse onset as determined with the linear fit method, is larger for
electrical channels. In contrast, the difference is very small for the optical
channels, since pulses from these channels have a small width and a steep
leading edge. In most cases, the leading edge of these pulses contains only 2
or 3 samples. The α-calibration used by the MuonDaq (see section 3.3.1), can
be replaced by the linear fit method.
5.1. Data processing and analysis 67
t_delay
T_waveform T_trigger
T_fragment
T
10.24 µsec
pos*10 nsec
TWR time window
Figure 5.4: Time correction for the waveforms.
An important piece of information contained in the TWR raw data is
an estimate for the current baseline, called the dynamic baseline bdyn, as
introduced in section 4.2.3. This baseline estimate, bdyn, may be used in the
hit extraction analysis to give a value for the baseline, m. The merged data
of 2004 include bdyn, while the merged data of 2003 do not. However, in 2003
about 20 samples before the first pulse were included in every fragment. The
average of the first five values in the fragment forms a reliable estimate for bdyn.
5.1.2 Time synchronisation and calibration
The global time information for an event which is included in the TWR
raw data is the GPS timestamp of the trigger time Ttrigger. The start time
of every waveform fragment has to be determined. Figure 5.4 illustrates
the correction procedure. As described in section 4.2.1, the incoming trig-
ger signal is delayed by a time interval, tdelay, in order to sample a time
interval around the trigger time. The parameter, pos, is stored in the TWR
raw data and gives the position of the waveform fragment relative to the
total waveform window. pos is counted in sample points. Thus, the start
time of every waveform fragment has to be corrected using the folowing formula
Tfragment = Ttrigger − (10.24µs − tdelay) + (pos × 10ns) . (5.1)
As already mentioned in section 3.3.1, the transmitted analogue signals expe-
rience a delay, t0, due to different cable delays in the ice. In addition, various
delays are caused by the electronics. For reconstruction, these delays have to
be determined in order to obtain the exact arrival times of the photons at the
location of the OM.
68 5. Reconstruction with Waveforms
TDC time window
Specific time offset for 
every single OM
TWR time window
Figure 5.5: A first approach to calibrate the timing of the system using the t0
calibration of the MuonDaq.
The t0-calibration has been done for the data from the MuonDaq using a cal-
ibration laser signal [A+00]. For the analysis of the MuonDaq data, the pulse
onset time is represented by the leading edge as measured by the TDC. Since
the TDC time is determined using a discriminator with a fixed threshold, the
arrival time has to be corrected by the α-calibration, depending on the pulse
amplitude (pADC value). This correction is defined by
tcorrected = traw − t0 −
α√pADC . (5.2)
However, the association between the pulse and the pADC value is ambiguous
for more than one pulse per waveform leading to incorrect time corrections for
multiple pulses.
A first approach to the calibration of the TWRDaq is to calibrate the time
offsets with respect to the MuonDaq and to use the calibration constants for
the MuonDaq.
In order to find these offsets, δt0, the merged data stream is used since it
combines data from MuonDaq and TWRDaq. The main idea is to calculate
the time difference between the first appearing pulse in the waveform and
the leading edge of the first pulse in the TDC data for single OM, which is
connected to both TWRDaq and MuonDaq, .
Although both DAQs use different thresholds for detecting pulses and since
the TDC time window (about 34 µs) is much longer than the time window of
the TWR (10.24 µs), it is assumed that most of the pulses are assigned to the
same physical pulse. The occurence of a peak justifies that assumption.
In order to enhance the time accuracy, the α-calibration is applied to the data
from the MuonDaq. The exact pulse start time of the waveform pulses is
determined using the linear fit method as described above.
The distribution of δt0 for an electrical and an optical pulse are shown in
figure 5.6. The value of δt0 can be determined by a Gaussian fit, which is also
displayed in the diagram.
5.1. Data processing and analysis 69
delay in nanoseconds
320 340 360 380 400 420
hi
ts
1
10
delay in nanoseconds
2340 2360 2380 2400 2420 2440
hi
ts
1
10
Figure 5.6: The time difference between the first pulse in a waveform and the
first leading edge in the TDC window. Above an electrical channel (OM 5),
below an optical channel (OM 423) are shown.
The distribution of δt0 for the whole detector for the year 2003 is shown in
figure 5.7.
The accuracy of this calibration can be estimated using the width of the δt0
distribution, which is displayed in figure 5.8. While the optical pulses show
a typical width between 3.5 ns and 4.5 ns, the distributions of the electrical
70 5. Reconstruction with Waveforms
OM number
100 200 300 400 500 600
de
la
y 
/ n
se
c
0
1000
2000
3000
4000
5000
6000
Figure 5.7: The distribution of the δt0 versus OM number (results from 2003
data).
OM number
100 200 300 400 500 600
w
id
th
 o
f d
is
tri
bu
tio
n 
/ n
se
c
0
5
10
15
20
Figure 5.8: The distribution of the σ of the time calibration fit versus OM
number (results from 2003 data).
channels are wider, between 5 ns and 8 ns. As described in section 5.3.1,
the error of the TWR synchronisation dominates the optical channels, while
5.1. Data processing and analysis 71
the width of the electrical channels can be related to a lower accuracy of the
applied α-calibration or the linear fit method.
However, it was found that the calibration can be improved by using only very
large pulses having sharper leading edges, especially for strings 1 to 4.
5.1.3 Photon counting
Charge calibration
For N single photons arriving within a small time window, T, the signals
will superimpose to one pulse in the PMT and cannot be separated. If the
photons arrive almost synchronously, the amplitude will reach a value of N
times the amplitude of a single photon pulse. However, as the arrival times
differ more, the superposition results in a flatter and wider pulse. Thus, the
amplitude is not a good measure of the number of photons.
For larger time delays between the arriving photons, the pulses can be
separated. While the pulses of OMs with optical transmission are measured
with a width of a few 10 ns, the pulses of OMs with electrical transmission
are much wider. In this case, the separation is less efficient.
While the MuonDaq uses the pulse amplitude for determining the number of
photons contained in the pulse, the reconstruction of waveforms may utilise
the integrated charge of the pulse, excluding the overshoot. A clear advantage
is the fact that, for overlapping single photon pulses arriving with a time
delay, td, and an average pulse width, twidth, with td < twidth, the integrated
charge rather than the amplitude, gives a more reliable measure for photon
counting. Figure 5.9 shows a typical distribution of the integrated charge of
one OM.
The size of the single photo electron pulse is determined by a gaussian fit,
which is displayed in figure 5.9. The distribution of the single photo electron
pulse can be seen in figure 5.10 for the electrical and the optical channels.
Since the electrical pulses are much wider and the amplification gain is larger
for the SWAMPs than for the ORBs, the charge is much larger for the
electrical channels.
The reconstruction in this thesis still uses the amplitude to determine the
number of photons. Since the presently applied reconstruction does not take
into account the number of photons, the results will not change. However,
a more advanced reconstruction would benefit from the expected improved
resolution.
72 5. Reconstruction with Waveforms
integrated charge / nC
0 10 20 30 40 50 60 70
co
u
n
ts
0
100
200
300
400
500
Figure 5.9: The charge distribution of OM 56 (electrical channel). The size of
the single photo electron pulse is determined by a gaussian fit. The line shown
at the upper end of the distribution represents the largest charge observed in
this run.
5.2 Comparison of the performance of Muon-
Daq and TWRDaq
The first test on the TWR data consists of a comparison of the data taken with
MuonDaq and TWRDaq. Again the combined data set, called merged data,
is used. The goal is to confirm that the merging procedure is working well and
to investigate the dynamic ranges and the stability of the measurements.
Using the algorithm as described above, the pulses are detected and the hit
information is extracted from the waveforms.
5.2.1 Hit efficiency
While the TDC collects the pulses in a time window of 34 µs, the TWR in its
present configuration measures the incoming signals in a total time window
of 10.24 µs around the trigger. Figure 5.11 displays the distribution of the
pulse onset times obtained from TDC and TWR after time calibration. The
limited TWR time window is clearly visible. The TWRDaq loses mainly
after-pulses, which arrive several microseconds after the event. In order to
obtain an estimate for the hit efficiency and pulse resolution of the TWRDaq,
5.2. Comparison of the performance of MuonDaq and TWRDaq 73
OM number
100 200 300 400 500 600
1 
p.
e.
 c
ha
rg
e 
/ n
C
0
2
4
6
8
10
OM number
300 350 400 450 500 550 600 650
1 
p.
e.
 c
ha
rg
e 
/ n
C
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Figure 5.10: The mean charge of a single photo electron pulse of the electrical
(top) and optical channels (bottom).
the number of pulses found by both DAQs in the TWR time window has to
be compared. Figure 5.11 shows the leading edge distribution of the TDC hits
and the TWR pulses for one OM. The TWR time window is clearly visible
and is determined and reduced in order to avoid side effects on both sides by
about 200 ns. The relative pulse rate ratio, η, of the sum of all hits in the
reduced time window of both DAQs gives an estimate of the hit efficiency. η
74 5. Reconstruction with Waveforms
OM number
0 5000 10000 15000 20000 25000 30000
N
um
be
r o
f h
its
1
10
210
310
TWRDaq
MuonDaq
Figure 5.11: The leading edge distribution of OM 516 using data from the
MuonDaq and the TWRDaq. The dashed histogram contains the hits mea-
sured by the TWRDaq.
is defined as
η = detected pulses in TWRDaqdetected pulses in MuonDaq .
Since the AMANDA detector uses several types of analogue signal transmis-
sion techniques with different resolutions for the pulse separation, the results
are split into several plots for each type of signal transmission. The hit effi-
ciency for the first four strings of the detector, using coaxial cables for signal
transmission, is displayed in figure 5.12 (top). The diagrams for all other types
of signal transmission can be found in the appendix A.2.1. In order to get an
estimate for hit efficiency of every group of channels, the distribution of pulse
rate ratios is displayed in figure 5.12 (bottom). The mean and the RMS of
these distributions are summarised in table 5.1.
In principle, the relative pulse rate ratio is larger for optical signal transmis-
sion, which can be explained by the lower detection threshold for the TWRDaq
applied to these channels. While the TWRDaq uses thresholds between 15 mV
and 25 mV for the optical channels, the MuonDaq threshold for these chan-
nels varies between 50 mV and 60 mV. In contrast, the TWRDaq detection
threshold for the electrical channels is higher in order to reduce the amount of
data. While the threshold is about 180 mV for the TWRDaq, the threshold
5.2. Comparison of the performance of MuonDaq and TWRDaq 75
OM number
10 20 30 40 50 60 70 80
et
a
0
0.2
0.4
0.6
0.8
1
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
2
4
6
8
10
12
14
16
18
20
22
Figure 5.12: Hit efficiency for the first 4 strings. The ratio of pulses, η, for
each OM is shown above, while the distribution of η is shown below.
String Type of analogue Average η RMS of η thdet thdet
transmission TWRDaq MuonDaq
mV mV
1 - 4 electrical 0.97 0.08 180 100
5 - 10 electrical 1.08 0.05 180 140
11 - 13 optical 1.14 0.04 15 - 25 60
11 - 13 electrical 0.99 0.08 180 100 - 150
14 - 19 optical 1.18 0.09 15 - 25 50
14 - 19 electrical 0.98 0.11 180 200
Table 5.1: The distribution of relative pulse rate ratios for different types of
analogue signal transmission.
76 5. Reconstruction with Waveforms
TWR pulse amplitude / mV.
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
pA
DC
 v
al
ue
0
500
1000
1500
2000
2500
3000
3500
4000
OM 127
Figure 5.13: The relation between the pulse amplitude of the largest pulse in
the waveform and the pADC value from MuonDaq for an electrical channel
(OM 127). The pADC value corresponds to mV.
for the MuonDaq is about 100 mV for the first four strings and about 140 mV
for the strings 5 to 10. This effect is partly balanced by the higher resolution
in pulse separation that increases the number of pulses in the TWRDaq.
One result of this investigation is the loss of pulses due to higher TWR thresh-
olds, thdet, for the electrical channels. The value of thdet will be re-optimised
in the coming season.
5.2.2 Amplitude and dynamic ranges
While the pADC uses the delayed output, the TWR is connected to the prompt
output of the SWAMPs or ORBs. Both outputs have different amplification
characteristics. The gain of the prompt output of the SWAMP is larger than
the gain of the delayed output by a factor of ∼3.
A simple test of the stability of both DAQs is to compare the observed
pulse amplitudes. Figure 5.13 shows the relation between the amplitude of
the largest pulse found in the waveform and the pADC value for an electrical
channel. The correct association of the pulses results in a linear relation. The
histogram clearly shows the limitation of the voltage range of the TWR. The
fact that the TWR saturates earlier is caused by the higher amplifier gain of
the prompt output. The total voltage range of the TWR is 5 V, reaching
from approximately +1.2 V to -3.8 V. The amplitude of the waveform pulses
saturates at around 3.7 V.
5.2. Comparison of the performance of MuonDaq and TWRDaq 77
TWR pulse amplitude / mV.
0 200 400 600 800 1000 1200 1400 1600 1800
pA
DC
 v
al
ue
0
500
1000
1500
2000
2500
3000
OM 516
TWR pulse amplitude / mV
0 200 400 600 800 1000 1200 1400 1600 1800
pA
DC
 v
al
ue
0
500
1000
1500
2000
2500
3000
OM 516
Figure 5.14: The relation between the pulse amplitude of the largest pulse in
the TWR and the pADC value from MuonDaq for an optical channel (OM 516).
The diagram above shows the relation without application of the amplitude
fit procedure and the lower diagram shows it using using the procedure. The
pADC value corresponds to mV.
The saturation of the pADC 1 is also visible around pADC = 4096 mV.
Figure 5.14 (top) displays the corresponding relation for an optical channel.
While the signals of the electrical channels are comparatively large and often
exceed the dynamic range of the TWR and the pADC, the signals of the opti-
cal channels are smaller and are limited by the maximum output voltage of the
ORB. The linear correlation is not as good as for the electrical channels. Due
to the narrow pulses, the TWR, with its limited sampling resolution of 10 ns,
may miss the pulse maximum, which leads to an underestimation of the pulse
1The pADC has a digital range of 12 bit (0..4095 digits).
78 5. Reconstruction with Waveforms
event time /ns
21100 21200 21300 21400 21500 21600 21700
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2500
2600
2700
2800
2900
3000
3100
Figure 5.15: A waveform of OM 516 demonstrating the degradation of the
amplifier gain for larger amplitudes.
amplitude. In case of the ORB the relation is more complicated. The pADC
is capable of measuring the maximum amplitude of the PMT pulse only if the
pulse width is larger than 50 ns. The PMT pulses of the optical channels have
a typical width of 10 ns to 30 ns. A pulse shaper, contained in the ORBs,
solves this problem by spreading the pulses. In case of multiple pulses, the
widened pulses overlap and the correct amplitude information is lost.
In addition, the large spread of the pADC values at a given waveform ampli-
tude can be understood by the error caused by missing the pulse maximum
amplitude in the TWR. The amplitude resolution of the TWR can be im-
proved by applying a linear fit of the leading and the trailing edge of the pulse.
The crossing point of the tangents is considered as a better estimate for the
pulse amplitude. The result is displayed in figure 5.14 (lower diagram). The
dynamic range of the TWR amplitude has increased. However, the spread of
the ratio of the pADC and TWR amplitude is still large.
In addition, the amplifier gain of the ORBs changes for large amplitudes.
A very large waveform pulse of OM 516 is shown in figure 5.15. Above an
amplitude of 400 mV the pulse shape degrades and starts to saturate. Due
to the higher amplification gain, the prompt output starts to saturate much
earlier as can be seen in the figure.
Finally, the maximum number of recorded pulses has to be considered. The
total number of recorded edges in the TDC is limited to 16 edges correspond-
ing to 8 pulses with a leading and a trailing edge each. The time window of
the TDC is about 34 µs, while the TWR samples the signal in a window of
about 10 µs.
Figure 5.16 shows the distribution of the number of pulses found in the
waveforms and the number of leading edges in the TDC. As can be seen, the
number of pulses in the TDC shows a peak around eight and decreases rapidly
5.2. Comparison of the performance of MuonDaq and TWRDaq 79
number of pulses in event
0 10 20 30 40 50 60 70 80
co
u
n
ts
1
10
210
310
410
510
610
710
TWRDaq
MuonDaq
Figure 5.16: The distribution of the number of pulses per channel.
for larger numbers, while the number of pulses in the waveform extends to up
to 30.
As described in section 3.2, the TDC may miss trailing edges making it possi-
ble to store additional leading edges in the free memory space. A total number
of up to 16 pulses can appear. OMs with more than eight leading edges do
not contain the full information of each pulse, since the corresponding trailing
edges are lost.
The integrated charge forms can be used as a measure for the number of
photons contained in a pulse. Figure 5.17 displays the relation between the
integrated charge of the largest pulse in the waveform and the pADC value
from the MuonDaq. Again, the limitation of dynamic range for the pADC is
clearly visible for the electrical channel. Due to the higher amplification gain
of the prompt output, the relation between charge and pADC value slightly
degrades above pADC = 2000 when reaching the voltage limit of the TWR.
An important property of the DAQ systems is the dynamic range for the
photon counting. Using the pulse amplitude as a measure for the number of
photons in a pulse, the dynamic range is determined by the ratio of the size
of the largest obtained pulse and the average size of a single photo electron
pulse. The distributions of the dynamic ranges for the pADC and the TWR
are displayed in figure 5.18. It is clearly visible that the dynamic range for
the TWR is apparently smaller by a factor of about 2 to 3. While the total
voltage ranges of both devices are of the same order (pADC: 4.096 V; TWR:
∼3.8 V ), the amplification gain of the prompt output used by the TWR and
the delayed output used by the pADC differ by a factor of ∼3. Thus, in order
to obtain a larger dynamic range for the TWR, the amplification gain of the
prompt output or the HV of the PMT has to be reduced.
80 5. Reconstruction with Waveforms
pADC value / mV
0 1000 2000 3000 4000 5000
pu
ls
e 
ch
ar
ge
 T
W
R 
/ n
C
0
20
40
60
80
100
120
OM 097
pADC value / mV
0 500 1000 1500 2000 2500 3000 3500
pu
ls
e 
ch
ar
ge
 T
W
R 
/ n
C
0
0.5
1
1.5
2
OM 441
Figure 5.17: The relation between the integrated charge of the largest pulse in
the waveform and the the corresponding pADC value from the MuonDaq. The
diagram above shows an electrical channel (OM 97), below an optical channel
(OM 441).
5.2. Comparison of the performance of MuonDaq and TWRDaq 81
OM number
0 100 200 300 400 500 600 700
dy
na
m
ic
 ra
ng
e 
in
 p
ho
to
el
ec
tro
ns
0
10
20
30
40
50
60
70
80
OM number
0 100 200 300 400 500 600 700
dy
na
m
ic
 ra
ng
e 
in
 p
ho
to
el
ec
tro
ns
0
5
10
15
20
25
Figure 5.18: The dynamic range for photon counting using the pulse amplitude
as a measure for the number of photons with the MuonDaq. Above shows the
distribution of the dynamic ranges of all channels using the pADC, below the
distribution using the TWRs.
82 5. Reconstruction with Waveforms
number of OMs in MuonDaq
0 50 100 150 200 250 300 350 400
n
u
m
be
r o
f e
ve
nt
s
1
10
210
310
Figure 5.19: The distribution of the number of hit OMs per event in the
MuonDaq for “merged data” in 2003.
5.2.3 Muon track reconstruction
The aim of this first track reconstruction using data from TWRDaq is to
demonstrate that the reconstruction with data from TWRDaq gives the same
result as with data from the MuonDaq. For this purpose “Merged data” con-
taining both data from MuonDaq and from TWRDaq is used. The filtering and
merging process is described in section 4.2.5. The comparison is an essential
test for all other following analyses with TWR data. It has to be mentioned
that the merged data set is not a pure sample of unbiased events, but is en-
riched with high multiplicity events. For the analysis presented in this thesis
data from 2003 was used. The selected high multiplicity events have more
than 140 hit OMs in the MuonDaq. Figure 5.19 shows the distribution of hit
OMs in the MuonDaq per event.
For the reconstruction the original AMANDA reconstruction software is used,
which uses the information that is available in the MuonDaq. Hits from the
MuonDaq contain the leading edge time and the time over threshold, TOT.
The pADC value is usually associated with all hits within a time window
around the trigger time. From the TWR data a similar dataset for every hit
has been extracted using the algorithms, as described in the previous sections.
Thus, the data from TWRDaq differs from the MuonDaq dataset by
• individual amplitude values, which are extracted from each pulse shape,
• a different procedure to determine the leading edge time,
• and a better separation of overlapping pulses.
5.2. Comparison of the performance of MuonDaq and TWRDaq 83
 0
 50
 100
 150
 200
 250
 300
 350
 400
 0  20  40  60  80  100  120  140  160  180
co
u
n
ts
zenith angle [deg]
Zenith distribution of 13174 Events (2003)
TWR
MDAQ
Figure 5.20: The distribution of the zenith angle obtained from reconstruction
with data from TWRDaq and MuonDaq.
For the reconstruction algorithm, which is applied in this thesis, only the lead-
ing edges are utilised. Channels with a high noise rate or instabilities are
excluded from the reconstruction. This procedure is termed hit cleaning. Only
the TOT and the pADC value are used for the hit cleaning.
Figure 5.20 shows the distribution of the zenith angle obtained from re-
construction with TWR data and MuonDaq. The results for both DAQs are
similar and do not show any systematic features. Thus, the hit extraction
followed by the standard reconstruction with data from TWRDaq gives the
same result as with MuonDaq.
The distribution of the angle between the MuonDaq and TWRDaq recon-
struction results is displayed in figure 5.21, while the corresponding cumulative
distribution is shown in figure 5.22. The results show that the median of the
angular difference is about 1◦, which is less than the angular resolution of the
AMANDA detector, which is assumed to be ∼ 3◦ on tracks2. Thus, the recon-
struction with data from TWRDaq gives similar results to the reconstruction
with data from the MuonDaq. This analysis has been done without applying
quality checks on the reconstructed tracks.
2This assumption is based on Monte Carlo data. Muon tracks have been generated using
the software package dCorsika (2001) and the response of the AMANDA has been simulated.
The median of the angular difference between simulated track and reconstructed track is
assumed as measure for the angular resolution.
84 5. Reconstruction with Waveforms
 0
 100
 200
 300
 400
 500
 600
 700
 800
 900
 0  2  4  6  8  10
Co
un
ts
Angle between TWR and MDaq reco [deg]
Figure 5.21: The distribution of the angle between the MuonDaq and
TWRDaq reconstruction.
 0
 10
 20
 30
 40
 50
 60
 70
 80
 90
 100
 0  2  4  6  8  10  12  14
Pe
rc
en
ta
ge
 w
ith
 s
m
al
le
r d
ev
ia
tio
n
Angle between TWR and MDaq reco [deg]
Cumulative Distribution
Figure 5.22: The integrated distribution of the angle between the MuonDaq
and TWRDaq reconstruction representing the percentage of events with a
smaller angular difference.
5.3. Estimation of the reconstruction quality 85
5.3 Estimation of the reconstruction quality
In experimental data, the true muon track direction is not known and it
is difficult to determine the angular resolution of a reconstruction. In fact,
this is possible for simulated data with a known track. As the Monte Carlo
implementation into the detector simulation is still in a testing phase, the
accuracy is probably not sufficient for a full comparison of the resolution of
MuonDaq and TWRDaq. However, it is possible to investigate improvements
and degradation of the reconstruction with respect to the MuonDaq using
the following method. The general idea is to split all hits of an event into
two subsamples of equal size. Since the hits originate from the same event
being distributed randomly into two subsamples, a reconstruction with the
subsamples of hits should yield a similar result. Of course, this procedure
only gives good results for events with a larger number of hits, since each
subsample will contain only half of the hits and the resolution will degrade
significantly for a smaller number of hits [Rib].
All hit channels are ordered by the time of their first hit in a numbered
list. In order to produce the subsamples, all hits of even numbered channels
are combined into one subsample, even, whereas the other hits are collected
in a subsample called odd. In order to produce subsamples containing the
complete time window of the physical event, the hits are ordered by time.
OMs which are not connected to both DAQs are excluded from the analysis.
The resulting distribution of the angular difference between odd and even
reconstruction is displayed in figure 5.23. Smaller differences correspond to a
better intrinsic angular resolution. The lower diagram shows the integrated
distribution of both DAQs. The median of this distribution can serve as an
indicator for a comparison. As can be seen, the result is slightly worse for
the TWRDaq. While for the MuonDaq the median is around ∼ 4.1◦, the
TWRDaq is ∼ 4.4◦ including all events.
The dependence of the median of the angular difference distribution and
the number of channels in the event is displayed in figure 5.24 (top). The
reconstruction results are stable above a number of participating OMs of about
40, corresponding to an average number of channels of 20 in each subsample.
Furthermore, with increasing number of channels, the reconstruction becomes
more stable, as the median angle between odd and even decreases.
Despite the fact that the reconstruction does not take into account the
additional information delivered by the TWRDaq, the results show that the
reconstruction is comparable to the MuonDaq results. In addition, figure 5.24
shows an important result. It can be seen that for events with more than 30
hit OMs, the median angle is still below 10 degrees. This means that for the
corresponding subsamples with more than 15 hit OMs, the reconstruction still
gives reliable results. This is an important piece of information and suggests
that a further decrease of the multiplicity threshold is promising.
86 5. Reconstruction with Waveforms
 0
 50
 100
 150
 200
 250
 0  5  10  15  20  25  30
Co
un
ts
Angle between odd and even reco [deg]
TWR
MDAQ
 0
 20
 40
 60
 80
 100
 0  5  10  15  20  25  30
Pe
rc
en
ta
ge
 w
ith
 s
m
al
le
r d
ev
ia
tio
n
Angle between odd and even reco [deg]
Cumulative Distribution
TWR
MDAQ
Figure 5.23: The distribution of the angular difference between odd and even
sample (top). Below the cumulative distribution. The ordinate corresponds to
the percentage of events of smaller angular difference.
5.3. Estimation of the reconstruction quality 87
 0
 5
 10
 15
 20
 25
 30
 35
 40
 45
 50
 0  50  100  150  200  250
M
EA
D
IA
N
 A
N
G
LE
 B
ET
W
EE
N 
O
DD
 A
ND
 E
VE
N 
RE
CO
NCH after hit cleaning
Merged Data 2003
MDAQ
TWR
 0
 200
 400
 600
 800
 1000
 1200
 0  50  100  150  200  250
co
u
n
ts
NCH after hit cleaning
Merged Data 2003
MDAQ
TWR
Figure 5.24: The median of the angular difference distribution versus the num-
ber of participating channels in the event (top). The corresponding distribution
of number of channels in the event (bottom).
5.3.1 Systematic error
The TWR samples the PMT signals with a frequency of 100 MHz which
results in a binning of 10 ns and a limited time accuracy. An additional
88 5. Reconstruction with Waveforms
TWR2
TWR1
dt_fragment_1
dt_fragment_2
Trigger pulse time
Figure 5.25: Systematic error due to the binning of the TWR.
problem arises from the lack of synchronisation between different TWRs.
Every TWR is equipped with a 100 MHz oscillator, which serves as an internal
clock. Figure 5.25 illustrates the problem. An incoming trigger is latched
by the TWR at the following rising edge of the internal clock. Due to the
independent clocks, the time difference between the actual trigger signal and
the latching time in a single TWR, a jitter of about 5 ns, limits the time
resolution [Nam].
This is the dominant error in time resolution for the optical channels. A
solution for this problem will be a global synchronisation of all TWRs using
an external oscillator.
5.4 Investigating signal characteristics using
the TWRDaq
After-pulses
After-pulses may be caused by residual gas atoms in the PMT. The time
between the first direct pulses caused by the photon(s) and the after-pulses
are characteristic for each PMT [Leo94]. The distance between the the first
pulse in a waveform and every following pulse is displayed in figure 5.26.
While the hits with a time distance of less than 3 µs are presumably related
to the event, the later pulses can be mostly considered as after-pulses. For
5.4. Investigating signal characteristics using the TWRDaq 89
distance from 1st pulse
0 2000 4000 6000 8000 10000
co
u
n
ts
1
10
210
310
OM 140
distance from 1st pulse
0 2000 4000 6000 8000 10000
co
u
n
ts
1
10
210
310
OM 640
Figure 5.26: The distance between the first pulse in a waveform and every
following pulse. Above shows an electrical channel (OM 140), below an optical
channel (OM 640).
most of the OMs, a second maximum in the hitrate at about 6 µs after the
first occuring pulse is visible. It has to be mentioned that the drop in the
hitrate is caused by the time window of the TWR of 10 µs. Since most of the
first hits in the TWR window are occurring around 2 µs to 3 µs in the the
window, the maximum distance to the window end is about 8 µs.
The TWRDaq can be used to study the shape and the occurrence of after-
pulses in order to develop an efficient filter for these pulses.
90 5. Reconstruction with Waveforms
event time /ns
24650 24700 24750 24800 24850 24900
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2980
3000
3020
3040
3060
event time /ns
19350 19400 19450 19500 19550 19600 19650 19700 19750
PM
T 
si
gn
al
 a
m
pl
itu
de
 /d
ig
its
2850
2900
2950
3000
3050
3100
3150
3200
Figure 5.27: Two waveforms with a crosstalk pulse. The crosstolk pulse can
be distinguished from a normal PMT pulse by a positive overshot.
5.4.1 Cross talk
Electro-magnetic interference, called cross talk, between signal cables or in the
dAQ electronics may cause additional pulses. These pulses differ from standard
PMT pulses by their appearance, which is clearly visible in the TWR signal
shape. Typically, a positive pulse directly followed by a negative pulse of
about equal size is visible. A simple filter algorithm may identify cross talk
This fact can probably be used to identify these pulses to clean the data and
make the reconstruction more precise. A cross talk pulse from an optical and
an electrical channel are displayed in figure 5.27.
6Proposal for a TWR based
trigger system
6.1 Motivation and general idea
Presently, the AMANDA detector is triggered by the multiplicity trigger or
the string-trigger, which is based on local coincidences within one string, see
section 3.2. Particles which do not exceed the multiplicity trigger threshold or
the string-trigger threshold will not be recorded by the AMANDA detector.
The ability to trigger on local coincidences across strings would be a useful
trigger for many analyses. Unfortunately, this is not available in the existing
system. It is in principle possible via an online analysis of the recorded data.
It is currently being investigated, how such a trigger system can be realised
and it has been proposed to utilise the TWRDaq to build a software trigger
system. Each TWR samples the incoming signal and, as soon as a pulse be-
low the threshold is detected, writes out a hit containing a timestamp and a
waveform. The data is collected on a PC, where the trigger software searches
the data stream for possible trigger conditions. This chapter describes the first
steps towards a software trigger system. The system is still under construc-
tion [Mes03].
The layout of the readout system is basically the same as of the current
TWRDaq system. Figure 6.1 shows an overview of the proposed system. One
major change is the firmware of the TWRs, which has to be changed to enable
the TWRs to sample each channel independently. As before an individual
fixed threshold is applied to detect signals in each channel. Inside the TWR,
the 8 channels are grouped in 4 double-channels, whose signals are sampled
in parallel. During readout, the channels have to be separated, since in most
cases only one channel contains a signal below the threshold. This task will
be done in the DSP.
92 6. Proposal for a TWR based trigger system
SIS3300
TWR
FPGA
SIS3300
TWR
FPGA
SIS3300
TWR
FPGA
SIS3300
TWR
FPGA
DSP
SIS3100
DSP
SIS3100
.
.
.
SIS3100
FIFO
SIS3100
FIFO
SIS3100
FIFO
DSP
SIS3100
Optical Fibre
LEMO / NIM
.
.
.
.
.
.
.
.
.
clock
external 100 MHz
VME
VME
Crate 1
Crate 6
Collecting data / Separation of double channels
Latch
GPS
Fan out
VME
VME
TWRs sampling continuously / Only pulses are stored
PCI board
SIS1100
Master Crate
Synchronuous
bank switch
Figure 6.1: The proposed TWR trigger system.
A TWR hit consists of a header, a timestamp and the waveform. The header
contains the OM number. The internal timestamp of the TWR is used as a
measure of the start time of the waveform. In order to guarantee the synchro-
nisation within the whole system, all TWRs are synchronised by an external
100 MHz clock.
Similar to the feature extraction, the waveform contains a programmable num-
ber of preceeding and following samples around the pulse. This stream of TWR
hits is collected by the DAQ software on the connected PC.
In contrast to the MuonDaq, the timestamps are corrected for cable delays be-
fore the trigger search process in order to allow scanning for local coincidences.
The TWR hits in each channel are filled in a time-ordered FIFO pipeline. For
the trigger search only the start time of the waveform is used, because it is not
possible to extract the exact pulse start time from each waveform due to the
large amount of data.
Most of the waveforms contain only noise hits. In order to identify time in-
tervals, which possibly contain an event, a pre-trigger search is applied. This
process is illustrated in figure 6.2. The timestamps are filled in a histogram
with a bin size of several µs. Time intervals of interest are identified by apply-
ing a threshold. If the contents of one or several bins exceed the threshold, a
6.1. Motivation and general idea 93
Waveform
Timestamp
Om number
OM 680
OM 679
OM 1
FIFO containing time ordered raw data 
for each OM
ra
w
 d
at
a 
is 
fil
le
d 
in
to
 F
IF
O
s
tim
es
ta
m
ps
 a
re
 c
or
re
ct
ed
 fo
r T
0 
TWR Daq system
Time interval of interest
H
its
Time
Threshold
Figure 6.2: Pre-trigger search with TWR data.
time interval of interest is formed including the adjacent bins before and after
the bins over threshold. The TWR hits stored in these bins are filled in a
histogram and sorted by time.
This list of time sorted events forms the basis for the following search for
events. Figure 6.3 illustrates the process of building the multiplicity trigger.
For each hit appearing a trigger window of 2.5 µs is added to the multiplicity
function. Several hits occurring in one OM, during this trigger window, extend
the trigger window further. A trigger is launched if the multiplicity function
exceeds a preset threshold.
In the new TWR trigger system two thresholds are used. If the multiplicity
function exceeds the higher threshold thmult, a multiplicity event is built from
all hits around the trigger time. Events that exceed only the lower thlocal are
tested for local coincidences using more advanced algorithms.
94 6. Proposal for a TWR based trigger system
Time
Time
Hits in time interval of interest 
M
ul
tip
lic
ity
2.5 usec
thmult
localth
Figure 6.3: Multiplicity trigger search with TWR hits. Each hit OM adds a
2.5 µs “rectangle” starting with the waveform start time to the multiplicity
function. A second hit in the same OM extends the rectangle of the OM but
does not add a new rectangle. A trigger is found if the multiplicity function
reaches a certain threshold.
In principle, a variety of different trigger conditions are possible. Among them,
a multiplicity trigger, considering n channels with at least one hit in a time
window dT serves as a basic condition. Since accidental coincidences will dom-
inate the recorded events at low multiplicity thresholds, an additional trigger
condition is useful to detect local coincidences and distinguish them from ac-
cidental coincidences. The additional trigger condition can:
• be a local coincidence within one string. N hit OMs within a row of M
adjacent OMs in one string – string trigger.
• require in a time window dT that more than N hit OMs in a certain
(spherical) volume V with radius R and centre ~X – cluster trigger I:
M = {~xk| |~xk − ~X| < R } with ‖M‖ > N ,
where ‖M‖ is the number of elements in M.
6.1. Motivation and general idea 95
• require in a time window dT that more than N hit OMs which fulfil a
compactness condition in space – cluster trigger II:
n
∑
k=0
|~xk − ~X|
n < CII
where ~X is the centre of gravity of the hit OMs and CII is the compactness
threshold.
• require in a time window dT that more than N hit OMs which fulfil a
compactness condition in space and time – cluster trigger III:
n
∑
k=0
|~xk − ~X| · |tk − Tm|
n < CIII
where Tm is the average time of the hits in the time interval dT , ~X is
the centre of gravity of the n hits and CIII is the compactness threshold.
• require in a time window dT more than N OMs, with at least one hit OM
in a volume with radius R around the OM – neighbourhood trigger:
M = {~xk| ∃i with i < k ∧ |~xk − ~xi| < R}with ‖M‖ > N ,
where ‖M‖ is the number of elements in M .
• be an external triggers e.g. laser calibration or random trigger.
• be a supernova trigger based on the noise rate of individual channels.
• be a special triggers for slow moving particles.
Since the trigger is formed within an online software analysis the algorithm
has to be rather simple. While most of the trigger conditions require a large
amount of computing time, the multiplicity trigger and the neighbourhood
trigger are based on simple algorithms. Figure 6.4 illustrates the neighbour-
hood trigger. Hit modules with at least one hit module within a spherical
volume with radius R are counted (solid circles) while isolated ones are not
counted (dashed circles). This trigger is sensitive to cascade and track-like
events.
96 6. Proposal for a TWR based trigger system
Figure 6.4: The neighbourhood trigger. Only hit OMs with at least one hit
OM within a spherical volume with radius R are counted.
7Re´sume´ and outlook
This thesis describes the design and realisation of a new data taking system
for the AMANDA detector. The original DAQ system – the MuonDaq – uses
TDCs and peak sensing ADCs to sample the arrival time and amplitude of
PMT pulses.
The MuonDaq limits the detector capabilities for various reasons. The in-
formation recorded by the MuonDaq is incomplete and ambiguous for high
energy events. In addition, the system has a limited dynamic range for photon
counting of complex signals.
The main goal of this thesis was to design and install a new AMANDA DAQ
system, without the limitations of the MuonDaq. The new system – the
TWRDaq – uses Flash ADCs to record the complete waveform of the PMTs.
The work started with the installation of a first test system in the Antarctic
summer 2001/2002 season with 48 channels. Based on this work, a new system
was developed and finally installed in the 2003/2004 season and connected to
576 channels. After one year of successful operation, this system was improved
again to achieve higher trigger rates. The number of connected channels in-
creased to 597. This system has now been operating since February 2004.
Most of the goals have been fulfilled.
It was shown that the dead time of the new TWRDaq is about 20 µsec and
the real dead time in percent of the total number of triggers amounts to only
0.015 ± 0.01 % and 0.018 ± 0.013 % for trigger thresholds of M = 24 and
M = 18 respectively. Thus, the TWRDaq in the present configuration oper-
ates almost without dead time. In contrast, the dead time of the MuonDaq
amounts to 2.2 msec and results in a data loss of about 15 %.
The current system is externally triggered by the multiplicity trigger of the
MuonDaq. In 2003 the multiplicity threshold was set to 24 with a trigger
frequency of about 90 Hz. Due to the negligible dead time of the TWRDaq,
the multiplicity threshold was lowered from 24 to 18 in the Antarctic summer
2003/2004 resulting in a trigger frequency of 150 Hz. Thus, the sensitivity of
98 7. Re´sume´ and outlook
AMANDA for low energy events increased. A hit extraction algorithm and fur-
ther tools to determine the exact arrival times of the photons and the number
of photons in each pulse have been developed. The algorithm uses a dynamic
threshold to detect adjacent pulses. Pulses that overlap can be separated by
this algorithm. The muon track analysis described in this thesis is based on
this algorithm.
While the TDC of the MuonDaq is only capable of measuring up to 16 edges,
limiting the dynamic range to 8 pulses, the number of pulses in a waveform
from the TWRDaq can exceed 60. High energy events will benefit from this
extended dynamic range. The hit extraction algorithm delivers a value for the
amplitude of each individual pulse in a waveform. This makes the estimation
of the number of photons contained in each pulse more precise, which is im-
portant for the energy resolution.
Furthermore, the consistency of the data from the TWRDaq was tested by
comparing hits measured by both DAQs. Events from both DAQs are associ-
ated by their GPS time and written in a combined data stream, termed merged
data. One requirement for track reconstruction is an accurate determination
of the cable delays between the PMTs. This time calibration was performed
before for the MuonDaq. The basic idea of the time calibration presented in
this thesis is to determine the time offset between MuonDaq and TWRDaq
for each channel, using the merged data. Then the cable delays determined
with the MuonDaq are incorporated to determine the final cable delay for each
channel. The results show that a timing accuracy between 3.5 ns and 4.5 ns
for the optical channels and about 5 ns to 8 ns for the electrical channels can
be achieved. This is accurate enough for track reconstruction.
It was shown that the angular difference between reconstructed tracks using
TWR data and data from MuonDaq is smaller than the expected angular res-
olution of AMANDA.
In addition, a different approach was used to estimate the intrinsic track re-
construction quality in both DAQs. Hits from a single track were split into
two sub-samples. Track reconstruction was performed with both sub-samples.
The angular difference between both reconstructions gives a measure of the
track quality. The analysis presented showed that, despite the fact that the
reconstruction does not take into account the additional information delivered
by the TWRDaq, the reconstruction is comparable to the MuonDaq results.
For events with more than 30 hit OMs, the analysis of the corresponding
subsamples with more than 15 hit OMs show a good correlation. This is an
important piece of information and suggests that a further decrease of the
multiplicity threshold is promising.
The improvements of the detector capabilities and further plans to extend the
maximum trigger rate to up to 200 Hz results in a significant increase in data
rate, which can be solved without loss of data only by further improvements of
the feature extraction, a self-developed compression algorithm and by increas-
99
ing the readout speed of the system.
This thesis describes the successful design, installation and first tests of the
new TWRDaq. A complete investigation of all features of the new system is
a subject of future work within the AMANDA collaboration.
100 7. Re´sume´ and outlook
8Outlook – Further work
The present system is still triggered by the trigger system of the MuonDaq,
using global multiplicity conditions within the whole detector or local coin-
cidence conditions within a single string. An advantage of using devices like
TWRs, containing programmable logic like FPGAs, is the potential to build a
fast software trigger system. A new firmware for the TWR FPGAs has already
been developed [Mes]. First successful tests have already been performed in the
2003/2004 Antarctic summer season. It is proposed to upgrade the TWRDaq
in the following Antarctic season again, in order to generate a trigger signal.
The present reconstruction is mainly based upon the leading edge time of the
first pulses for each channel. Advanced reconstruction using information from
the following pulses is necessary for a further improvement of the reconstruc-
tion.
The bandwidth for data transmission from the South Pole to the Northern
hemisphere is limited. Therefore, it is necessary to develop an improved com-
pression algorithm for the waveform data [Ref].
The improvement in energy resolution, due to the improved amplitude mea-
surement and the larger dynamic range, will be investigated using the neural
net approach [Mun]. A first point source analysis will follow soon after [Gro].
Finally, it has to be mentioned that the experience gained from waveform re-
construction will be an important opportunity for the IceCube collaboration.
102 8. Outlook – Further work
Appendix A
Data format
Event Configuration Register
Connected OM for channel 1
Connected OM for channel 8
Threshold channel 1
Baseline channel 1
Threshold channel 8
Baseline channel 8
Baseaddress of bridge module
Number of TWR modules in crate
TWR 0
TWR 1
Last TWR
VER 1.0 6.10.02Header for the binary datafiles of the new TWRdaq    
Clockdivider
Number of crates
Crate 1
Crate 2
Last crate
Length of header**
Begin mark* Baseaddress of TWR module
Statusregister
Module ID
Acquisition Control Register
Extern Start Delay Register
Extern Stop Delay Register
Id of TWR***
*     0xaaaaaaaa
**   in bytes
*** number of crate *16 + number of TWR in crate 
Baseaddress of GPS module
Baseaddress of Clock module
OM_is_optical 8
OM_is_optical 1
Figure A.1: Header of binaryfile
104 Appendix A. Data format
VER 1.3.2 8. 1. 2003
eventcounter
GPS Event 1 − lword 1
GPS Event 1 − lword 2
GPS Event 1 − lword 3
GPS Event 1 − lword 4
TWR 0
TWR 1
TWR 2
last TWR
TWR 5
TWR 4
TWR 3
value1value2
value3value4
value n value n−1
1st waveform
2nd waveform
last waveform
Final dataformat from the eventbuilder
TWR number
TWR−begin markEventbegin mark
length of event block*
eventtimestamp
Reserved
FEATURE EXTRACTION !!
waveform compressed, that
means only the interesting
parts (pulses) are saved
channel number
number of waveforms
length of first waveform*
length of second waveform
TWR−beginmark 0xffffffff
Eventbegin mark 0xbbbbbbbb
* in bytes
length of last waveform
Event−Header
Figure A.2: Dataformat of binaryfile
105
Waveform length
Begin of new waveform
f    0    a    3(1) WAVEFORM BEGIN MARK
4    1    a    0
Begin of waveform fragment
End of waveform fragment
Position of waveform fragment endinside waveform 0..1023
Position of waveform fragment inside waveform 0..1023
Values of peak 1 
4    b    5    3
0    c    e    4
0    c    e    3
2    b    7    0
2    1    b    5
End of waveform
Values of peak 2
0    c    e    4
0    c    e    3
8    0    0    0 
(3)
WAVEFORM END MARK
(4)
(5)
(6)
(7)
WAVEFORM FRAGMENT END MARK
WAVEFORM FRAGMENT BEGIN MARK
WAVEFORM FRAGMENT BEGIN MARK
WAVEFORM FRAGMENT END MARK
only the interesting peaks are saved 4. 7. 02
VER 1.1
0    c    e    2
Mean value of waveform
Each waveform begins with WAVEFORM BEGIN MARK accompanyed by the length of the whole waveform in bytes 
(2)
waveform and terminated with a WAVEFORM FRAGMENT END MARK (4) and (6). The waveform is closed by 
with WAVEFORM FRAGMENT BEGIN MARK (3) and (5) accompanyed by the position of the fragment inside the
followed by the mean value of the waveform (2). Two waveform fragments are stored in this example beginning  
a WAVEFORM END MARK (7). To align the waveform to full 32 bit words, a filling word 0xaaaa is inserted if ne− 
cessary.
Waveform data format with feature extraction
Figure A.3: Dataformat of compressed waveform
106 Appendix A. Data format
A.1 Description of the config file TWR.cnf
The configuration file TWR.cnf contains important information on the
system like the number of the connected OMs, the TWRs and the applied
threshold. The clock predivider sets the speed of the internal timestamp
counter relative to the TWR frequency of 100 MHz. A value of 100 means
that the timestamp-counter is running with a frequency of 100 MHz / 100 =
1 MHz and has an accuracy of 1µsec.
A typical TWR.cnf can be seen here:
GENERAL
/* PCI driver name */
PCI DRIVER NAME /usr/local/sis1100/nod/sis1100
Directory in which the SIS1100 driver is located.
/* Data output directory */
DATA OUT DIRECTORY /export/seal1/datadir/
Output directory for the data.
/* Number of 128-event blocks per file */
EVENTBLOCKS PER FILE 40
This number multiplied by 128 gives the number of events per
rawdata file.
/* VLF trigger stuff */
VLF TRIGGER VETO 0
DURATION OF VETO 60
OVERLAP 3
TIME 0 0
TIME 1 15
TIME 2 30
TIME 3 45
The GPS latch will produce 4 trigger vetos with a length of
DURATION seconds beginning at TIME 0, TIME 1, TIME 2, TIME 3
minutes of each hour. This veto is extended in both directions
by OVERLAP seconds.
/* DSP parameters for Feature Extraction */
PRECEEDING VALUES OPTICAL 0x3
FOLLOWING VALUES OPTICAL 0x4
PRECEEDING VALUES ELECTRICAL 0x10
A.1. Description of the config file TWR.cnf 107
FOLLOWING VALUES ELECTRICAL 0x12
These parameters define how the Pulse Extraction in the DSP
works. See 4.2.3
/* clock predivider for internal TWR clock */
CLOCK PREDIVIDER 1
The clock predivider sets the speed of the internal timestamp
counter relative to the TWR frequency of 100 MHz. A value of 100
means that the timestamp counter is running with a frequency of
100 MHz / 100 = 1 MHz and has an accuracy of 1µsec.
STATUS REG 0x102
ACQ CONTROL REG 0x2b4
EVT CONFIG REG 0x2c
TFCC 1022
These values define various parameters for the TWR, for example
the waveform window length.
N CRATES 6
The number of crates holding TWRs.
CRATE 0
/* local VME base of bridge module in master crate */
BASE BRIDGE 0x04000000
The base address of the local SIS3100 bridge module in the
master crate connecting to this remote crate.
/* Sharc program */
/* has to be located in direcory current daq */
DSP PROGRAM s3300v61.ldr
The DSP file for the remote SIS3100 bridge module in this crate
BASE GPS 0x10000000
The remote VME base address of the GPS2VME in this remote crate.
If it is set to 0xffffffff there is no GPS2VME in this crate
N TWR 0x10
The number of TWRs in this crate
BASE TWR 0x00000000
The VME base address of the first TWR
EXTERN STRT DEL 0
108 Appendix A. Data format
The start delay for this TWR. It has no meaning for the current
DAQ.
EXTERN STOP DEL 700
The stop delay for this TWR. This number multiplied by 10nsec
defines the delay time for an incoming trigger. A value of 700
means, that the waveform
covers about 3000nsec before the trigger and 7000nsec after the
trigger.
TWR OM 640 641 643 644 645 646 647 648
The number of the conected OM beginning with 1
OPTICAL 1 1 0 0 0 0 0 0
If the OM is read out by an optical fibre, the value is set to 1
otherwise 0
TWR BASELINE 4000 4000 4000 4000 4000 4000 4000 4000
The baseline for the OM. Currently this value has no meaning,
since the baseline is determined automatically at the begin of
each run.
TWR THRESHOLD 20 20 20 20 20 20 20 20
The threshold for the channel is determined by:
<determined baseline> - TWR THRESHOLD and is programmed into
the TWR. The thresholds are manually set and optimized for every
single OM.
...
BASE TWR 0xf0000000
EXTERN STRT DEL 0x0
EXTERN STOP DEL 700
TWR OM 123 124 125 126 127 128 129 130
OPTICAL 1 1 0 0 0 0 0 0
TWR BASELINE 4000 4000 4000 4000 4000 4000 4000 4000
TWR THRESHOLD 20 20 20 20 20 20 20 20
CRATE 2
BASE 100MHz 0x0
BASE GPS 0x10000000
N TWR 0x10
...
A.1.1 Name convention and content of the raw data files
The convention for the binary filenames is
A.1. Description of the config file TWR.cnf 109
twr yyyy ddd rrrr ffff.dat in 2003
twr yyyy ddd rrrr ffff bbbbb eeeee.dat in 2004
with
yyyy year
dddd day in year
rrrr run number as received from muon daq
ffff file number incremented with every new file of actual run
bbbbb time of first event in file in seconds of actual day
eeeee time of last event in file in seconds of actual day
The header of the raw data file contains the whole structure of the system
and all important numbers, like the threshold applied for the pulse detec-
tion used for Feature Extraction. Figure A.1 shows the structure of the header.
110 Appendix A. Data format
A.2 Additional diagrams
A.2.1 Hit efficiency
OM number
100 150 200 250 300
et
a
0
0.2
0.4
0.6
0.8
1
1.2
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
10
20
30
40
50
Figure A.4: Hit efficiency for strings 5 to 10. The the ratio of pulses η for each
OM is shown above, while the distribution of η is shown below.
A.2. Additional diagrams 111
OM number
320 340 360 380 400 420
et
a
0
0.2
0.4
0.6
0.8
1
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
2
4
6
8
10
12
Figure A.5: Hit efficiency for the electrical OMs in strings 11 to 13. The the
ratio of pulses η for each OM is shown above, while the distribution of η is
shown below.
112 Appendix A. Data format
OM number
320 340 360 380 400 420
et
a
0
0.2
0.4
0.6
0.8
1
1.2
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
5
10
15
20
25
30
Figure A.6: Hit efficiency for the optical OMs in strings 11 to 13. The the
ratio of pulses η for each OM is shown above, while the distribution of η is
shown below.
A.2. Additional diagrams 113
OM number
450 500 550 600 650
et
a
0
0.2
0.4
0.6
0.8
1
1.2
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
1
2
3
4
5
6
7
8
Figure A.7: Hit efficiency for the electrical OMs in strings 14 to 19. The the
ratio of pulses η for each OM is shown above, while the distribution of η is
shown below.
114 Appendix A. Data format
OM number
450 500 550 600 650
et
a
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
eta
0.4 0.6 0.8 1 1.2 1.4 1.6
co
u
n
ts
0
10
20
30
40
50
Figure A.8: Hit efficiency for the optical OMs in strings 14 to 19. The the
ratio of pulses η for each OM is shown above, while the distribution of η is
shown below.
List of Figures
2.1 A Cosmic Ray shower in the atmosphere. . . . . . . . . . . . . . 7
2.2 The Cosmic Ray Energy spectrum. . . . . . . . . . . . . . . . . 9
2.3 Survey of astroparticle physics. . . . . . . . . . . . . . . . . . . 10
2.4 The energy dependence of the horizon for gamma astronomy. . . 11
2.5 Three different sources in the sky, which can be explained by a
black hole. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Diffuse muon neutrino fluxes. . . . . . . . . . . . . . . . . . . . 15
3.1 The AMANDA detector since 2000. . . . . . . . . . . . . . . . . 22
3.2 Detection principle of charged particles with light sensors using
the Cˇerenkov technique. . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Transmission coefficient for νµ when traversing the Earth. . . . . 25
3.4 The depth dependence of the scattering length. The center of
the detector corresponds to a depth of about 1700 m. . . . . . . 26
3.5 The current AMANDA DAQ system - MuonDaq . . . . . . . . . 29
3.6 Time slewing due to fixed threshold. . . . . . . . . . . . . . . . 31
3.7 Relative time shift between a laser calibration pulse from the
surface and the crossing time of the signal over the discriminator
threshold versus 1/√pADC. . . . . . . . . . . . . . . . . . . . . 32
3.8 The distribution of pADC values for OM 15. . . . . . . . . . . . 33
3.9 Cherenkov cone: definition of track parameters. . . . . . . . . . 34
4.1 A typical complex signal waveform from an electrical channel . . 38
4.2 Relation between total number of hits and simulated muon energy. 39
4.3 Relation between the muon energy Ereco as reconstructed by
the neural net versus the true muon energy Eµ. . . . . . . . . . 40
116 LIST OF FIGURES
4.4 Two typical waveforms containing a single photo-electron pulse. 41
4.5 Adjustment of the event time interval with respective to the
trigger time. An appropriate time window around the trigger
time is chosen by delaying the incoming trigger signal. . . . . . 42
4.6 Trigger setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.7 Survey of the TWRDaq system in 2003. . . . . . . . . . . . . . 47
4.8 Overview of the TWRDaq system in 2004. . . . . . . . . . . . . 48
4.9 State diagram of the TWRDaq. . . . . . . . . . . . . . . . . . . 49
4.10 Different thresholds are applied to the signal during the extrac-
tion of pulses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.11 Distribution of integrated charges. . . . . . . . . . . . . . . . . . 52
4.12 Procedure for estimating the baseline. . . . . . . . . . . . . . . . 53
4.13 Distribution of bdyn. . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.14 Difference between brun and bdyn versus channel number. . . . . 55
4.15 Remaining data amount in percent of the initial data. . . . . . . 56
4.16 Average remaining data amount of all channels. . . . . . . . . . 57
4.17 Survey of the online data-flow at the Pole. . . . . . . . . . . . . 58
4.18 Trigger rates using different levels of multiplicity from the trig-
ger of the MuonDaq. . . . . . . . . . . . . . . . . . . . . . . . . 59
4.19 Time difference between adjacent events for various multiplicity
triggers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.20 The distribution of number of hit OMs per event in the TWRDaq. 61
4.21 The complete Feature Extraction Algorithm. . . . . . . . . . . . 62
5.1 Relative thresholds used for hit extraction. . . . . . . . . . . . . 64
5.2 A waveform measured with OM 180. . . . . . . . . . . . . . . . 65
5.3 Linear fit to determine the pulse onset time. . . . . . . . . . . . 65
5.4 Time correction for the waveforms. . . . . . . . . . . . . . . . . 67
5.5 A first approach to calibrate the timing of the system using the
t0 calibration of the MuonDaq. . . . . . . . . . . . . . . . . . . 68
5.6 The time difference between the first pulse in a waveform and
the first leading edge in the TDC window. . . . . . . . . . . . . 69
5.7 The distribution of the δt0 versus OM number (results from
2003 data). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
LIST OF FIGURES 117
5.8 The distribution of the σ of the time calibration fit versus OM
number (results from 2003 data). . . . . . . . . . . . . . . . . . 70
5.9 The charge distribution of OM 56 (electrical channel). . . . . . . 72
5.10 The mean charge of a single photo electron pulse of the electrical
(top) and optical channels (bottom). . . . . . . . . . . . . . . . 73
5.11 The leading edge distribution of OM 516 using data from the
MuonDaq and the TWRDaq. The dashed histogram contains
the hits measured by the TWRDaq. . . . . . . . . . . . . . . . . 74
5.12 Hit efficiency for the first 4 strings. . . . . . . . . . . . . . . . . 75
5.13 The relation between the pulse amplitude of the largest pulse
in the waveform and the pADC value from MuonDaq for an
electrical channel. . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.14 The relation between the pulse amplitude of the largest pulse
in the waveform and the pADC value from MuonDaq for an
optical channel. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.15 A waveform of OM 516 demonstrating the degradation of the
amplifier gain for larger amplitudes. . . . . . . . . . . . . . . . . 78
5.16 The distribution of the number of pulses per channel. . . . . . . 79
5.17 The relation between the integrated charge of the largest pulse
in the waveform and the the corresponding pADC value from
the MuonDaq. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.18 The dynamic range for photon counting using the pulse ampli-
tude as a measure for the number of photons with the MuonDaq. 81
5.19 The distribution of the number of hit OMs per event in the
MuonDaq for “merged data” in 2003. . . . . . . . . . . . . . . . 82
5.20 The distribution of the zenith angle obtained from reconstruc-
tion with data from TWRDaq and MuonDaq. . . . . . . . . . . 83
5.21 The distribution of the angle between the MuonDaq and
TWRDaq reconstruction. . . . . . . . . . . . . . . . . . . . . . . 84
5.22 The integrated distribution of the angle between the MuonDaq
and TWRDaq reconstruction. . . . . . . . . . . . . . . . . . . . 84
5.23 The distribution of the angular difference between odd and even
sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.24 The median of the angular difference distribution versus the
number of participating channels. . . . . . . . . . . . . . . . . . 87
5.25 Systematic error due to the binning of the TWR. . . . . . . . . 88
118 LIST OF FIGURES
5.26 The distance between the first pulse in a waveform and every
following pulse. . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.27 Two waveforms with a crosstalk pulse. The crosstolk pulse can
be distinguished from a normal PMT pulse by a positive overshot. 90
6.1 The proposed TWR trigger system. . . . . . . . . . . . . . . . 92
6.2 Pre-trigger search with TWR data. . . . . . . . . . . . . . . . . 93
6.3 Multiplicity trigger search with TWR hits. . . . . . . . . . . . . 94
6.4 The NEXT neighbourhood trigger. . . . . . . . . . . . . . . . . 96
A.1 Header of binaryfile . . . . . . . . . . . . . . . . . . . . . . . . . 103
A.2 Dataformat of binaryfile . . . . . . . . . . . . . . . . . . . . . . 104
A.3 Dataformat of compressed waveform . . . . . . . . . . . . . . . 105
A.4 Hit efficiency for strings 5 to 10. The the ratio of pulses η for
each OM is shown above, while the distribution of η is shown
below. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A.5 Hit efficiency for the electrical OMs in strings 11 to 13. The
the ratio of pulses η for each OM is shown above, while the
distribution of η is shown below. . . . . . . . . . . . . . . . . . . 111
A.6 Hit efficiency for the optical OMs in strings 11 to 13. The
the ratio of pulses η for each OM is shown above, while the
distribution of η is shown below. . . . . . . . . . . . . . . . . . . 112
A.7 Hit efficiency for the electrical OMs in strings 14 to 19. The
the ratio of pulses η for each OM is shown above, while the
distribution of η is shown below. . . . . . . . . . . . . . . . . . . 113
A.8 Hit efficiency for the optical OMs in strings 14 to 19. The
the ratio of pulses η for each OM is shown above, while the
distribution of η is shown below. . . . . . . . . . . . . . . . . . . 114
List of Tables
3.1 Optical properties in water and ice. [LM00] . . . . . . . . . . . . 26
3.2 Comparison of the AMANDA strings [W+01]. . . . . . . . . . . 27
4.1 Characteristics of the TWR. . . . . . . . . . . . . . . . . . . . . 42
4.2 Signals delivered from the GPS clock to synchronise the
GPS2VME module. . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Main features and differences between TWRDaq in 2003 and
2004 season. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Parameters for the Feature Extraction used in 2003 and 2004. . 51
5.1 The distribution of relative pulse rate ratios for different types
of analogue signal transmission. . . . . . . . . . . . . . . . . . . 75
Bibliography
[A+00] E. Andres et al. The amanda neutrino telescope: Principle operation
and first results. Astroparticle Physics, 13:1–20, 2000.
[A+01] E. Andres et al. Observation of high-energy neutrinos using
cherenkov detectors embedded deep in antarctic ice. Nature, 410:441
– 443, 2001.
[A+02a] J. Ahrens et al. Limits to the muon flux from wimp annihilation
in the center of the earth with the amanda detector. Phys.Rev.,
D66(032006), 2002.
[A+02b] J. Ahrens et al. Observation of high energy atmospheric neutrinos
with the antarctic muon and neutrino detector array. Physical Review
D, 66, 2002.
[A+02c] J. Ahrens et al. Search for supernova neutrino-bursts with the
amanda detector. Astroparticle Physics, 16:345, 2002.
[A+03] J. Ahrens et al. Search for neutrino-induced cascades with the
amanda detector. prd, 67(012003), 2003.
[A+04a] J. Ahrens et al. Muon track reconstruction and data selection tech-
niques in amanda. Nuclear Instruments & Methods, 524:169–194,
2004.
[A+04b] J. Ahrens et al. Twr software manual. 2004.
[Ack] M. Ackermann. Datahandler manual pages. DESY-Zeuthen,
http://www.ifh.de/nuastro/protected/datahandler.html.
[Ahr01] J. Ahrens. Ein webbasiertes u¨berwachungsystem fr das amanda
neutrinoteleskop. Master’s thesis, Johannes Gutenberg-Universitt
Mainz, 2001.
[B+91] A. Burrows et al. The future of supernova detection. Physical Review
D, 45(10), 1991.
BIBLIOGRAPHY 121
[B+00] R. Barth et al. GS Multi-Branch System User Manual. Gesellschaft
fu¨r Schwerionenforschung mbH GSI, Planckstr. 1 D-64291 Darmstadt
Germany, 2.2 edition, 2000.
[B+01] S. W. Barwick et al. Potential of amanda ii in he neutrino astro-
physics. In Proceedings of the 27th International Cosmic Ray Con-
ference, Hamburg, Germany, pages 1–4, 2001.
[Bec04] J. K. Becker. Calculation of the agn neutrino flux and of event rates
for large volume neutrino telescopes. Master’s thesis, University of
Wuppertal, 2004. WU D 04-06.
[Blo84] V. Blobel. Unfolding methods in high–energy physics experiments.
Proceedings of the 1984 CERN School of Computing, CERN, 1984.
[CMU95] P. Padovini C. Megan Urry. Unified schemes for radio-loud active
galactic nuclei. Publ.Astron.Soc.Pac., 107(803), 1995.
[Eks04] P. Ekstroem. A Dark Matter Search with AMANDA. PhD thesis,
Stockholm University, 2004.
[FB95] H. Falcke and P. L. Biermann. The jet-disk symbiosis. I. Radio to X-
ray emission models for quasars. Astronomy & Astrophysics, 293:665,
1995.
[Gai90] Thomas K. Gaisser. Cosmic Rays and Particles. Cambridge Univer-
sity Press, 1990.
[Gee03] H. Geenen. Atmospheric neutrino and muon spectra measured
with the AMANDA-II detector. In Proceedings of the 28th ICRC,
Tsubuka, 2003.
[Gre66] K. Greisen. End to the Cosmic-Ray Spectrum? Physical Review
Letters, 16, 1966.
[Gro] A. Groß. Phd thesis in preparation.
[Gro04] Particle Data Group. Particle Physics Booklet. Springer, 2004.
[H+95] M. Honda et al. Calculation of the flux of atmospheric neutrinos.
Physical Review D, 52:4985, 1995.
[H+01] S. Hundertmark et al. A method to detect uhe neutrinos with
amanda. In Proceedings of the 27th International Cosmic Ray Con-
ference, Hamburg, Germany, 2001.
[Hes11] V. F. Hess. u¨ber die absorption der g–strahlung in der atmospha¨re.
Physikalische Zeitschrift, 12(22,23):998–1001, 1911.
122 BIBLIOGRAPHY
[Hes12] Viktor Franz Hess. u¨ber beobachtungen der durchdringenden
strahlung bei sieben freiballonfahrten. Physikalische Zeitschrift,
13:1084–1091, 1912.
[JKB] P. L. Biermann and W. Rhode J. K. Becker. In Prep.
[JL95] S. Pakvasa J.G. Learned. Detecting ντ oscillations at pev energies.
Astroparticle Physics, 3(3):267–274, 1995.
[K+97] A. Karle et al. Analog optical transmission of fast photomultiplier
pulses over distances of 2 km. Nuclear Instruments & Methods,
387:274–277, 1997.
[Kam01] K.-H. Kampert. Cosmic rays and particle physics. Heavy Ion Phys.,
14:203–215, 2001. invited talk presented at the ”Symposium on Fun-
damental Issues in Elementary Matter”.
[Lei03] H. Leich. GPS2VME User’s Guide. DESY Zeuthen, January 2003.
[Leo94] W.D. Leo. Techniques for Nuclear and Particle Physics Experiments.
Springer Verlag, 1994.
[LM00] J. G. Learned and K. Mannheim. High-energy neutrino astrophysics.
Annu. Rev. Nucl. Part. Sci., 50:679, 2000.
[Lon92] M.S. Longair. High Energy Astrophysics, volume 1. Cambridge Uni-
versity Press, 1992.
[M+] K. Mu¨nich et al. In Prep.
[Man95] K. Mannheim. High-energy neutrinos from extragalactic jets. As-
troparticle Physics, 3:295, 1995.
[Man96] K. Mannheim. Beacons at the gamma ray horizon. ArXiv Astro-
physics e-prints, 1996.
[Man97] K. Mannheim. Agn models: High-energy emission. ArXiv Astro-
physics e-prints, 1997.
[Mes] T. Messarius. Phd thesis in preparation.
[Mes03] T. Messarius. Analyse von zufa¨lig getriggerten daten des amanda-
detektors. Master’s thesis, University of Wuppertal, 2003. WU D
03-02.
[Mir04] I. F. Mirabel. Microquasar – agn – grb connections. app, 2004.
BIBLIOGRAPHY 123
[MPR01] K. Mannheim, R. J. Protheroe, and J. P. Rachen. Cosmic ray bound
for models of extragalactic neutrino production. Physical Review D,
63:23003, 2001.
[MRS03] A. D. Martin, M.G. Ryskin, and A.M. Stasto. Prompt neutrinos
from cc and bb and the gluon at very small x. ArXiv Astrophysics
e-prints, 2003.
[Mun] K. Munich. Phd thesis in preparation.
[Nam] J. Nam. Private communication. UCI Irvine,
http://www.ps.uci.edu/ jwnam.
[P+99] W. S. Paciesas et al. The fourth batse gamma-ray burst catalog
(revised). ArXiv Astrophysics e-prints, 1999. accepted for publication
in Ap. J. Suppl.
[R+56] F. Reines et al. Detection of the free neutrino: A confirmation.
Science, (124):103, 1956.
[RB93] J. P. Rachen and P. L. Biermann. Extragalactic Ultra-High Energy
Cosmic-Rays - Part One - Contribution from Hot Spots in Fr-II Radio
Galaxies. Astronomy & Astrophysics, 272:161, 1993.
[Ref] F. Refflinghaus. Diploma thesis in preparation.
[Rib] M. Ribordy. A new method to estimate the angular resolution. web-
page.
[S+03] K.Z. Stanek et al. Spectroscopic discovery of the supernova 2003dh
associated with grb 030329. Astrophys.J., 591:L17–L20, 2003.
[Sch02] Th. Schmidt. Aufbau und Funktionsnachweis eines Optischen Moduls
mit optisch-analoger Pulsu¨bertragung fu¨r den AMANDA–II– und
ICECUBE– Detektor. PhD thesis, Humboldt Universita¨t Berlin, Ger-
many, 2002.
[SNO02] SNO Collaboration. Direct Evidence for Neutrino Flavor Transfor-
mation from Neutral-Current Interactions in the Sudbury Neutrino
Observatory. Physical Review Letters, 89:11301, 2002.
[SS96] F. W. Stecker and M. H. Salamon. High Energy Neutrinos from
Quasars. Space Science Reviews, 75:341, 1996.
[Sup98] SuperKamiokande Collaboration. Evidence for Oscillation of Atmo-
spheric Neutrinos. Physical Review Letters, 81:1562, 1998.
124 BIBLIOGRAPHY
[Sys02] Struck Innovative Systems. SIS1100/SIS3100 Standard Design
Firmware Version 5 User Manual. http://www.struck.de, version:
1.30 edition, August 2002.
[Sys03] Struck Innovative Systems. SIS3300/SIS3301 65/80/100 MHz VME
FADCs Users Manual, Dedicated AMANDA Firmware Version:
1.00. http://www.struck.de, version: 1.00 edition, January 2003.
[VZ80] L. V. Volkova and G. T. Zatsepin. Promt lepton generation-
atmospheric muon and neutrino spectra at high-energies. Soviet
Journal of Nuclear Physics, 37:212, 1980.
[W+99] K. Woschnagg et al. Optical properties of south pole ice at depths
from 140 to 2300 meters. In Proceedings of the 26th International
Cosmic Rays Conference, volume 2, page 200. ICRC XXVI, 1999.
[W+01] R. Wischnewski et al. Performance of the amanda ii detector. In Pro-
ceedings of the 27th International Cosmic Ray Conference, Hamburg,
Germany, pages 1105 – 1109, 2001.
[W+02] C. Wiebusch et al. Results from amanda. Modern Physics Letters
A, 17(31):2019–2037, 2002.
[WB99] E. Waxman and J. Bahcall. High energy neutrinos from astrophysical
sources: An upper bound. Physical Review D, 59:23002, 1999.
[Wie98] B. Wiebel Sooth. Measurement of the allparticle energy spectrum and
chemical composition of cosmic rays with the HEGRA detector. PhD
thesis, University of Wuppertal, 1998.
[YS85] H. Elliot Y. Sekido, editor. Early History of Cosmic Rays. D. Reidel,
1985.
Appendix B
Acknowledgement
This work has been pursued in the framework of the AMANDA collaboration
and I would like to thank all former and present members. After three years,
it is difficult to remember all persons who have contributed to this work. I
am sure that I will forget someone.
If there are two persons, I have to thank most, these are Prof. Wolfgang
Rhode and Karl-Heinz Becker. I would like to thank them for their continuous
and unrelenting support during the last three years.
When I first arrived at the South Pole, I felt a little lost. I was a member
of the collaboration for about five month, but my knowledge about the
detector and the conditions in Antarctica were still very limited. Without the
invaluable help from the Mainz collaborators, Tom Feser, Marc Hellwig and
Jens Ahrens, the TWR project would not have been a success. Together we
managed to solve a lot of problems. I also have to thank Henrike Wissing and
Jiwoo Nam for their inestimable help.
I have to thank Steven W. Barwick for his support for the TWR project.
The two weeks which I spent in Irvine helped a lot to increase my knowledge
about AMANDA detector. He was of invaluable help in finding weak points
in the TWR system and giving impulses to solve them. I am grateful to Jiwoo
Nam for helping me to find critical bugs in the DAQ system and for many
interesting discussions.
I also feel grateful to all my colleagues from the Astroparticle group in
Wuppertal. Especially, I have to thank my room mates Kirsten Mu¨nich,
Julia Becker, Heidrun Bojahr and Matthias Bartelt for the nice and friendly
atmosphere.
Furthermore, I am indebted to Andreas Gross for his great work on track
reconstruction with TWR data and to Timo Messarius for his extensive
work on the TWR system and the new trigger system. I also have to thank
Christopher Wiebusch for his encouraging support for our project.
Special thanks go to Tino Ha¨upke and Matthias Kirsch for their helpful
126 Appendix B. Acknowledgement
support in critical situations at the Pole and numerous discussions about
further possibilities of the TWR.
I also would like to thank Dr. Nikolaus Kurz from the Gesellschaft fu¨r
Schwerionenforschung for his support with the MBS system.
This work has profited a lot from discussions with Markus Ackermann,
Stephan Hundertmark, Tonio Hauschildt, Mathieu Ribordy, Thierry Cast-
ermans, Christine Burgess and others at the Pole and in the Northern
hemisphere.
Thanks to Oana Tas.ca¯u, I did not forget that there still exists a world outside
the University.
To my family, thank you very much. Although my Grandmother was worried,
when I first told her about going to the Pole, she always supported me.