Feldmann 2011 Dissertation

Advances in GPS based
Time and Frequency Comparisons

for Metrological Use
Von der Fakultät für Mathematik und Physik der

Gottfried Wilhelm Leibniz Universität Hannover
zur Erlangung des Grades
Doktor der Naturwissenschaften

Dr. rer. nat.
genehmigte Dissertation
von
Dipl.-Phys. Thorsten Feldmann
geboren am 11. Juli 1977 in Duisburg
2011
Referent: PD Dr. Ekkehard Peik
Korreferent: Prof. Dr. -Ing. Steffen Schön
Tag der Promotion: 29.06.2011
Abstract
The global satellite navigation system GPS is the main tool for remote time and
frequency comparisons of atomic timescales and oscillators. The comparisons are
realized by referring the local timescales at both sites to the atomic clocks in the
satellites or to a common reference time by measuring a binary code transmitted
by the satellites and exchanging the data afterwards. Besides measuring the code
modern GPS receivers are also capable to track the phase of the carrier frequency,
but with an unknown initial number of cycles of this frequency between receiver
and satellites. By combining these carrier-phase measurements with the code mea-
surements and correction data calculated from a worldwide network of reference
stations, the positions and the time offsets of the receivers can be estimated with
high accuracy within a process called Precise Point Positioning (PPP).
The main issues of this work are to provide a highly precise frequency at any
location, which can be used for other scientific applications, and the precise cal-
ibration of operational time links. In this context, the mathematical models and
the GPS receiver hardware are examined in detail from the viewpoint of time and
frequency comparisons.
Based on a passive hydrogen maser and a state-of-the-art time and frequency
transfer receiver a highly precise mobile frequency reference was realized at PTB.
It was carried to the Institute of Quantum Optics (IQO) at the Leibniz Universität
Hannover (LUH) and is used as the reference source in the framework of developing
optical clocks. The evaluation of the performance of the passive maser at PTB and
at LUH is shown. Due to PPP the maser located at LUH can be referenced to
PTB’s primary standards at an averaging time of 104 s with the help of PPP at a
relative frequency instability of better than 10−14 .
GPS time and frequency comparison is a one-way technique, since signals are only
transmitted by the satellites and received on the ground. Thus, physical influences
occuring olong the signal path and relativistic effects on the satellite clocks have
to be corrected by mathematical models. Different PPP software packages are
analyzed in view of all parameters which have to be estimated. The influence of
the models is studied in detail by using only the carrier-phase measurement on
baselines with different lengths. This is possible, if the ambiguous carrier-phase
measurement does not deviate too much from the code measurement.
Subsequently, different receivers are analyzed in view of long and short term
stability. The requirements on the equipment for time and frequency transfer
are outlined and new concepts for improvements are proposed. The first GPS
common-clock measurement between the remote sites PTB and IQO is shown.
The local clocks at PTB and IQO are canceled out by comparing the GPS data to
the measurements using an optical fiber which is operated for frequency transfer
between these institutes.
Finally, calibrations of operational GPS links used by the International Bureau
for Weights and Measures (BIPM) for the realization of the international atomic
time scale (TAI) by means of a traveling GPS receiver are demonstrated. An
uncertainty at the level of 1 ns was achieved thanks to employing a state-of-the-art
receiver. Based on such a receiver and a time interval counter a new setup for link
calibration was realized at PTB, which further reduces the uncertainty.
Keywords: Time Comparisons, Frequency Comparisons, GPS

Zusammenfassung
Das globale Satellitennavigationssystem GPS ist das wichtigste Instrument zum
Vergleich entfernter atomarer Zeitskalen und Oszillatoren. Der Vergleich wird rea-
lisiert durch die Referenzierung der lokalen Zeitskalen auf beiden Seiten zu den
Atomuhren in den Satelliten oder zu einer gemeinsamen Referenzzeit, indem ein
von den Satelliten ausgesendeter binärer Code gemessen wird und die Messdaten
anschließend ausgetauscht werden. Neben der Codemessung können moderne GPS-
Empfänger auch die Trägerphase nachverfolgen, allerdings ohne die anfängliche
Zahl der Zyklen dieser Phase zwischen Satelliten und Empfänger zu kennen. In-
dem diese Trägerphasenmessung mit der Codemessung und Korrekturdaten eines
weltweiten Netzwerkes von Referenzstationen kombiniert wird, können die Posi-
tionen und die Zeitbasis der Empfänger mit hoher Genauigkeit bestimmt werden.
Dieser Prozess wird ”Precise Point Positioning” (PPP) genannt.
Die Hauptfragestellungen dieser Arbeit sind die Bereitstellung einer hochpräzisen
Frequenz an jedem beliebigen Ort, die für Forschungszwecke genutzt werden kann,
sowie die präzise Kalibrierung von Zeitverbindungen. In diesem Zusammenhang
werden auch die mathematischen Modelle und die GPS Empfänger vom Stand-
punkt der Zeit- und Frequenzvergleiche her analysiert.
Aufbauend auf einem passiven Wasserstoffmaser und einem modernen Zeit- und
Frequenzempfänger wurde an der PTB eine hochpräzise transportable Frequenzre-
ferenz realisiert. Diese wurde zum Institut für Quantenoptik (IQO) an der Leibniz
Universität Hannover (LUH) gebracht und dient dort als Referenz bei der Ent-
wicklung optischer Uhren. Die Ermittlung der Betriebseigenschaften des passiven
Masers an der PTB und an der LUH wird gezeigt. Mit Hilfe des PPP-Verfahrens
kann der Maser an der LUH mit einer Mittelungszeit von 104 s an die primären
Frequenznormale der PTB angebunden werden.
GPS Zeit- und Frequenzvergleiche sind eine Einwegtechnik, da Signale nur von
den Satelliten ausgesendet werden und am Boden empfangen werden. Daher müssen
physikalische Einflüsse auf den Signalweg und relativistische Effekte auf die Sa-
tellitenuhren durch mathematische Modelle korrigiert werden. Verschiedene PPP-
Softwarepakete werden im Hinblick auf alle zu bestimmenden Parameter unter-
sucht. Der Einfluss der Modelle wird detailliert untersucht, indem nur die Träger-
phasenmessung auf Basislinien verschiedener Länge benutzt wird. Dies ist möglich,
wenn die mehrdeutige Trägerphasenmessung nicht zu sehr von der Codemessung
abweicht.
Anschließend werden verschiedene Empfänger im Hinblick auf Kurz- und Lang-
zeitstabilität untersucht. Die Anforderungen an die Geräte im Rahmen von Zeit-
und Frequenzvergleichen werden deutlich gemacht und neue Konzepte zur Verbes-
serung werden vorgeschlagen. Das erste GPS-Experiment mit gemeinsamer Refe-
renz zwischen den entfernten Orten PTB und LUH wird vorgestellt. Die Einflüsse
der lokalen Uhren an der PTB und an der LUH werden durch den Vergleich der
GPS-Daten mit den Messungen einer optischen Faser, die zur Frequenzübertragung
zwischen diesen Instituten dient, beseitigt.
Schlussendlich wird die Kalibrierung von GPS Verbindungen, die vom interna-
tionalen Büro für Maß und Gewicht zur Realisierung der internationalen Atomzeit
TAI ausgewertet werden, mithilfe eines reisenden GPS Empfängers gezeigt. Durch
die Verwendung eines modernen Empfängers liegt die Unsicherheit im Bereich von
1 ns. Basierend auf einem solchen Empfänger und einem Zeitintervallzähler wurde
an der PTB ein neuartiger Aufbau entwickelt, mit dem die Unsicherheit weiter
reduziert werden kann.
Stichworte: Zeitvergleiche, Frequenzvergleiche, GPS

Contents
List of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1. Introduction 1
2. State-of-the-Art in remote Time and Frequency Comparisons 5

2.1. Characterization of Frequency Instabilities . . . . . . . . . . . . . . 5
2.2. Today’s remote Time and Frequency Comparison Methods . . . . . 7
2.3. GPS and Time and Frequency Transfer . . . . . . . . . . . . . . . . 8
2.3.1. Brief GPS System Overview . . . . . . . . . . . . . . . . . . 9
2.3.2. Coordinate Systems and Satellite Motion . . . . . . . . . . . 13
2.3.3. Observation Model . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.4. Time and Frequency Transfer Receivers . . . . . . . . . . . . 15
2.4. Signal Delays and Correction Models . . . . . . . . . . . . . . . . . 20
2.4.1. Atmospherical Corrections . . . . . . . . . . . . . . . . . . . 20
2.4.2. Relativistic Corrections . . . . . . . . . . . . . . . . . . . . . 22
2.4.3. Multipath . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.4. Site Displacement . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.5. Other Delay Sources . . . . . . . . . . . . . . . . . . . . . . 25
2.5. The International GNSS Service (IGS) . . . . . . . . . . . . . . . . 25
2.6. Precise Point Positioning . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7. GPS and UTC Generation . . . . . . . . . . . . . . . . . . . . . . . 31
3. A Highly Precise Mobile Frequency Reference 34

3.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2. Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3. Performance of the passive Hydrogen Maser . . . . . . . . . . . . . 36
3.4. Performance of the GTR50 Time and Frequency Transfer Receiver . 39
3.5. Installation of the Frequency Reference at the IQO . . . . . . . . . 40
4. Software Assessment 44
4.1. Common-Clock quasi zero Baseline Experiments . . . . . . . . . . . 44
4.2. Disconcontinuous measurement data and drifts . . . . . . . . . . . . 47
4.3. Clock comparisons and PPP parameters . . . . . . . . . . . . . . . 49
4.3.1. Software and Experimental Setup . . . . . . . . . . . . . . . 49
4.3.2. Time Scale Comparisons . . . . . . . . . . . . . . . . . . . . 52
v
4.3.3. Position Estimates . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.4. Troposphere Zenith Path Delay Estimates . . . . . . . . . . 56
4.4. Code free Carrier-Phase Frequency Comparisons . . . . . . . . . . . 58
4.4.1. Description of the Method . . . . . . . . . . . . . . . . . . . 59
4.4.2. Application to different Baselines . . . . . . . . . . . . . . . 62
5. Hardware Experiments 67
5.1. Performance of Receivers . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1.1. Instability of Time and Frequency Receivers . . . . . . . . . 68
5.1.2. QUEST Experimental Receivers . . . . . . . . . . . . . . . . 70
5.1.3. Temperature Sensitivity . . . . . . . . . . . . . . . . . . . . 72
5.2. Suggestions for Improvements in Frequency Transfer . . . . . . . . . 73
5.2.1. Phase Comparator Measurements . . . . . . . . . . . . . . . 73
5.2.2. Stabilization of Cables . . . . . . . . . . . . . . . . . . . . . 74
5.3. The first Long Baseline Common-Clock Experiment . . . . . . . . . 76
6. Relative Calibration of TAI links 80

6.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2. Description of the Calibration Procedure . . . . . . . . . . . . . . . 81
6.2.1. CV versus AV . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.2.2. Code Based Time Transfer and Position Errors . . . . . . . 86
6.2.3. PPP and Antenna Phase Center Corrections . . . . . . . . . 88
6.2.4. Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.3. Uncertainty Estimation . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.4. PTB’s Calibration Set-up . . . . . . . . . . . . . . . . . . . . . . . 95
6.5. METAS-PTB Calibration . . . . . . . . . . . . . . . . . . . . . . . 98
6.6. USNO-PTB Calibration . . . . . . . . . . . . . . . . . . . . . . . . 107
6.7. Suggestions for further Reductions of the Uncertainty . . . . . . . . 115
7. Summary and Outlook 117
Appendix 119
A. Mathematical Description of the Coordinate Systems . . . . . . . . 119
B. Mathematical Description of the Satellite Motion . . . . . . . . . . 121
C. CGGTTS Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
D. RINEX Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
E. P3 linear combination . . . . . . . . . . . . . . . . . . . . . . . . . 125
F. Brief Derivation of Relativistic Corrections . . . . . . . . . . . . . . 127
G. Estimation of the Ephemerides of Sun and Moon . . . . . . . . . . 129
H. BIPM’s Circular T . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Bibliography 134
vi
List of Acronyms
1 PPS one pulse per second signal
ADEV Allan Deviation
AV all-in-view
BIPM Bureau international des poids et mesures
C/A-code coarse/aquisition code
CCD common-clock difference
CCTF Consultative Committee for Time and Frequency
CGGTTS CCTF Working Group on GPS and GLONASS Time Transfer Stan-
dards
CV common-view
dCCD difference between two CCDs
DMTD dual mixer time difference technique
FR fixed receiver
GLONASS Global Navigation Satellite System, Russian GNSS system
GNSS Global Navigation Satellite System
GPS Global Positioning System, US GNSS system
IERS International Earth Rotation and Reference Systems Service
IfE Institut für Erdmessung, geodetic institute at the LUH
IGS International GNSS Service
ITRF International Terrestrial Reference Frame
IQO Institute of Quantum Optics at the LUH
LUH Leibniz Universität Hannover
METAS Swiss Federal Office of Metrology, operating the timescale UTC(CH)
vii
MDEV Modified Allan Deviation
NICT Japanese National Institute of Information and Communications Tech-
nology, operating the timescale UTC(NICT)
MJD Modified Julian Day
NRCan Natural Resources Canada
ODEV Overlapping Allen Deviation
ORB Observatoire Royal de Belgique, operating the timescale UTC(ORB)
P-code precision code
PPP Precise Point Positioning
PRN pseudorandom noise
PTB Physikalisch-Technische Bundesanstalt, operating the timescale UTC(PTB)
PZ-90 Russian Earth parameter system 1990
QUEST Centre for Quantum Engeneering and Space-Time Research
RINEX Receiver Independent Exchange Format
ROA Real Instituto y Observatorio de la Armada, Spanish institute oper-
ating the timescale UTC(ROA)
SD standard deviation
TAI temps atomique international
TDEV Time Deviation
TR travelling receiver
TIC time interval counter
TWSTFT Two-Way Satellite Time and Frequency Transfer
UFE Czech Institute of Photonics and Electronics, operating the timescale
UTC(TP)
USNO United States Naval Observatory, operating the timescale UTC(USNO)
UTC Coordinated Universal Time
WGS84 World Geodetic System 1984
viii
1. Introduction
Precise time and frequency are of paramount importance in technology and sci-
ence [1, 2]. Typical technical applications for accurate and/or stable frequencies
are for instance the steering of electrical power grids and modern digital telecom-
munication networks [3]. Accurate timing is for instance the basis of navigation
using Global Navigation Satellite Systems (GNSS), like the US Global Positioning
System (GPS) [4, 5]. Also industrial processes have to be steered with high preci-
sion and accuracy. In science the applications range from test of general relativity
and Earth science to atomic spectroscopy and elementary particle physics. Fur-
thermore, for many activities in the modern society an authoritative time reference
is mandatory.
The unit of time, the second, is a fundamental unit of the international system
of units (SI) [2, 6]. Since the unit of length, the meter, is defined via the velocity
of light, it is also derived from a precise frequency.
In order to establish reliable standards, many countries in the world maintain a
national time and frequency laboratory where atomic clocks serve as the national
standards. Such a laboratory is PTB. The dissemination of legal time and of the
unit of time and frequency is one of the prominent tasks of PTB. The transmission
of standard frequency (77.5 kHz) and coded time informations is the most popular
service of PTB [7].
An atomic clock uses an atomic transition to lock an electronic oscillator to it.
The output frequency is mostly 10 MHz or 5 MHz and is shared in the laboratory by
a distribution system. Special electronic generators undertake the task of counting
the frequency zero crossings and generate a pulse signal with fast rising time every
second (1 PPS signal).
Different types of clocks with different levels of accuracy are available commer-
cially and primary frequency standards based on caesium are developed by the
time laboratories [1, 2, 8].
The atomic clocks are compared locally with each other. These comparisons can
either by done with frequency comparators using the dual mixer time difference
(DMTD) technique [9] or with time interval counters (TICs) which measure the
difference between two 1 PPS signals by counting the accumulated phase of a
reference frequency. The latter technique is less precise due to a higher noise
level induced by the electronics, but each pulse can be unambiguously related to a
certain event.
1
2 1. Introduction
Additionally, the timescales of the national time and frequency laboratories are
compared with each other. The combination of local and remote comparisons
enables the intercomparison of all atomic clocks and the generation of a world-wide
reference timescale, the Coordinated Universal Time (UTC). The UTC calculation
is done by the Bureau international des poids et mesures (BIPM) in France. The
UTC is published on a five day basis as UTC-UTC(k) in the Circular T [2, 10].
Here ”k” stands for the different time laboratories. Details are given in Section 2.7.
In the early days of atomic time and frequency comparisons the standards of the
time and frequency laboratories were compared by using portable caesium clocks
[11] since no adequate transmission technology was available. With the advances
in the technology two methods for remote comparisons were available. One of
these methods made use of the signals of the Long Range Navigation (LORAN)
system, an Earth based system originally developed for navigation at sea consisting
of long wave radio station chains along coastlines transmitting pulse series steered
to atomic clocks [12]. The second method used the synchronization signals of
the color subcarrier of a television signal [13]. Today, remote time and frequency
comparisons are mainly done by using satellites [14].
Two complementary methods for satellite based remote time and frequency com-
parisons exist. The so-called Two-Way Satellite Time and Frequency Transfer
(TWSTFT) utilizing geostationary telecommunication satellites, and the compar-
isons using the signals of the satellites of a GNSS system. TWSTFT is a point
to point technique, since both sites are receiving as well as transmitting a signal.
In contrast, the GNSS technique is a one-way method. In principle an unlimited
number of sites everywhere on Earth can receive the satellites’ signals and compare
their local times and frequency to it. The comparison between the remote labs is
established only by the exchange of the measurement data. However, corrections
to the measurement data are necessary since propagation effects do not cancel out
in this one-way technique. A short overview is given in Section 2.2.
The main issues of this work are to provide a highly precise frequency at any
location, which can be used for other scientific applications, and the precise cali-
bration of operational time links. In this context, the mathematical models needed
in the framework of GPS and the GPS receiver hardware are examined in detail
from the viewpoint of time and frequency comparisons.
The usage of the signals of a GNSS system, namely of GPS, is the main tool for
time and frequency comparisons. The functionality and the current status of this
technique is outlined in Section 2.3.
After the overview of the current status of the remote time and frequency com-
parisons with GNSS systems this work comprises of four main parts. The first is
the description and evaluation of a new mobile frequency reference, which allows
to provide a highly stable frequency at any desired location by making use of the
GPS technology. Then data analysis methods for high precision GPS time and
frequency comparisons and mathematical correction models are studied in detail.
3
After this the hardware for GPS based time and frequency comparisons is exam-
ined. Finally the calibration of operational time links by means of GPS, including
a new hardware concept for the reduction of the uncertainty, is demonstrated on
two examples. In the following some more details on these chapters is given.
As mentioned above, GPS allows comparing frequency standards at every loca-
tion on the Earth. In this context, a mobile frequency reference consisting of a
passive hydrogen maser (an atomic clock producing a stable frequency in the short
term) and a state-of-the-art GPS receiver was developed. Thus, a frequency which
is referenced to PTB’s primary standards can be provided at any location and
be used for scientific applications requiring highest accuracy and stability. This
mobile frequency reference is currently operated at the Institute of Quantum Op-
tics (IQO) at the Leibniz University Hanover (LUH) as part of the involvement
in the Center for Quantum Engineering and Space-Time Research (QUEST). The
experimental setup, local test of the equipment, and results of the measurements
between Hanover and PTB are subject of Chapter 3.
The usual GPS observable is the measurement of the signal transmission time
between satellite and receiver based on a binary code emitted by the satellites. The
accuracy of these measurements is limited by the chip rate of the code. However,
modern receivers can also measure the microwave carrier frequency of this trans-
mission with high accuracy. A combination of the code and the carrier frequency
measurement is used within an analysis strategy called Precise Point Positioning
(PPP) [15]. PPP was used to evaluate the data of the mobile frequency reference
in order to obtain frequency differences between the remote clocks at PTB and
LUH with superior accuracy.
The PPP method was originally developed for geodetic science in order to de-
termine a position with sub-centimeter accuracy. It is a post-processing approach,
where correction data calculated from networks of globally distributed GPS re-
ceivers are combined with the measurement data. Different software concepts exist
which use different solution algorithms and different correction models. Several of
these software packages were evaluated with regard to all relevant parameters and
their capabilities for time and frequency comparison were tested. The results are
outlined in Chapter 4. In Section 4.4 a simple algorithm for frequency comparisons,
which uses only the carrier-phase observable, is demonstrated and compared to the
PPP method.
For highly accurate and stable frequency comparisons not only the data analysis
strategy is important, but also the involved hardware. In Chapter 5 different GPS
receivers are used to assess their capability for remote frequency comparisons. Some
of these receivers were provided by and operated in cooperation with the IQO and
Institut für Erdmessung (IfE) at the LUH. To characterize time and frequency
transfer GPS receivers, two or more devices are usually connected to the same
time and frequency source (common-clock setup) and the antennas are installed
4 1. Introduction
nearby each other, such that error influences originating from the GPS signal path
are minimized (short baseline).
By using an optical fiber, which was operated between PTB and the IQO in the
framework of QUEST, it was possible to obtain frequency differences between the
two remote clocks with an accuracy superior to any GNSS measurement [16] and
the influence of the remote clocks could be eliminated. The results of this first
common-clock long baseline experiment are shown in Section 5.3.
In contrast to pure frequency comparisons, in the framework of time and timescale
comparisons it is necessary to relate distinct events with each other, in other words,
to compare 1 PPS signals. Hence it is important to have a knowledge about the in-
ternal delays of the time transfer equipment involved in such comparisons. For the
evaluation of the links used for the UTC generation by the BIPM, this knowledge
is mandatory. This requires the calibration of the equipment or of the complete
link. Recently, with the use of new GPS receivers, an uncertainty of less than
2 ns for a relative link calibration between the Royal Spanish Naval Observatory
(Real Instituto y Observatorio de la Armada, ROA) and PTB performed in 2008
has been reported [17]. The further reduction of the uncertainty due to a revised
concept including hardware improvements developed in the context of this work is
demonstrated in Chapter 6.
2. State-of-the-Art in remote
Time and Frequency
Comparisons
Before the techniques for remote time and frequency comparisons and the GPS
system are explained, a few words on the analysis of the instabilities of clocks and
time and frequency links are necessary. Subsequently the different techniques for
remote time and frequency comparisons are briefly introduced. Then the function-
ality of the GPS system as the primary tool for time and frequency comparisons
is described. Recent techniques to improve the performance are outlined. Finally,
the importance of GPS time comparisons in UTC generation by the BIPM and the
need for accurate calibration of the time links involved is explained.
2.1. Characterization of Frequency Instabilities

Variances over samples of frequency data are used to characterize the instability.
Very often it is not appropriate to use the common standard deviation (SD) for
this task, because it is non-convergent for some noise types found in frequency
measurements [18, 19]. A better approach is to use the so-called Allan Deviation
(ADEV) or two-sample variance [20]. Is is defined as
X M
1
σy 2 (τ ) = (ȳi+1 (τ ) − ȳi(τ ))2 , (2.1)
2 (M − 1) i=1
where ȳi is the ith of M relative frequency averages, averaged over the sampling
interval τ from frequency data y(t) [19]. The smallest averaging value τ0 can never
be smaller than the spacing of the data. The frequency data can be calculated
from phase data x(t) according to
x(t) − x(t − ∆t)

y(t) = , (2.2)
∆t
where ∆t is the spacing of the data.
The Allan variance is usually calculated for a few sampling intervals and then
depicted in a double logarithmic plot. With the help of the slope of the data points
5
6 2. State-of-the-Art in remote Time and Frequency Comparisons
it is possible to distinguish between the noise types affecting the data. By applying
theoretical statistical considerations, confidence intervals can be calculated, which
appear as error bars in the plot [19].
In order to improve the confidence and the ability to distinguish between differ-
ent noise types, more averaging processes have been introduced which led to the
Overlapping Allan Deviation (ODEV) and the Modified Allan Deviation (MDEV).
Details can be found in the references [18, 19]. Furthermore, √ to characterize the
time error, the Time Deviation (TDEV), which is defined as (τ / 3)MDEV(τ ), was
invented. Table 2.1 shows the slopes of a double logarithmic plot corresponding
to different noise types for ADEV, MDEV, and TDEV [18]. The TDEV is defined
Table 2.1.: The slope of the different deviations depending on the noise type.
White frequency noise can also be denoted as random walk phase
noise, random walk frequency is also known as white frequency aging.
Noise ADEV MDEV TDEV

white phase -1 -3/2 -1/2
flicker phase -1 -1 0
white frequency -1/2 -1/2 1/2
flicker frequency 0 0 1
random walk frequency 1/2 1/2 3/2
in a way that it is equal to the SD in the case of white phase noise. Only in
this case the slope of the double logarithmic plot is negative. The minimum of
the double logarithmic plot indicates the averaging interval at which this type of
Gaussian distributed noise can be removed from the data without loosing other
relevant information.
Although the deviations are defined for evenly spaced data, they can also be used
in the case of missing measurements if the data gap is not too big and τ0 is replaced
by an effective average value. Details are explained in the references [18, 19].
In the framework of time and frequency metrology it is typical that the timestamps
of measurement data are related to the Modified Julian Day (MJD). It is a con-
tinuous numbering of the days starting at November 17, 1858. The January 1st,
2011 was MJD 55562.
2.2. Today’s remote Time and Frequency Comparison Methods 7
2.2. Today’s remote Time and Frequency

Comparison Methods
The TWSTFT method is an active technique where respective ground stations are
transmitting and receiving signals that are referenced to the local clock signals.
The difference between the two clock signals can be calculated from the difference
of the signal propagation delays (Figure 2.1) measured on both sites. This requires
expensive and labor intensive ground installations.
Figure 2.1.: TWSTFT (left) and GNSS time and frequency transfer (right).
In contrast, the Earth installations in the framework of GNSS time and frequency
transfer are only consisting of receivers. Each GNSS satellite is transmitting sig-
nals which contain information about its position and the offset of its internal clock
with respect to a common system time. Thus, each receiver can perform a signal
propagation time measurement with respect to the satellite’s clock or to the com-
mon system time, referenced to the local clock connected to the receiver. The link
between the local clock signals at two sites is calculated afterwards by exchanging
the measurement data. Thereby the satellite clocks cancel out.
A GNSS link can in principle either be evaluated in common-view (CV) or in
all-in-view (AV) mode. CV means that first the difference between two receivers
is calculated for each satellite seen by both receivers independently at each epoch
and that the mean value is calculated afterwards. In the AV mode a solution
is at first independently calculated for each receiver with respect to the GNSS
system time including all tracked satellites, and differences are made based on the
averages [21,22]. For a CV comparison in principle the signal of one single satellite
is sufficient.
In Table 2.2 the typical operational distances of the different time and frequency
link techniques are listed. Since the CV method requires that the satellites are
commonly visible at both sites the operational distance is limited, similar as for
TWSTFT. Due to referencing each station to the common time reference in AV
mode, clocks located at the opposite sites of the Earth can be compared, but noise
of the GNSS system time is partially transferred to the time comparison results.
Table 2.2.: Typical operational distances of the different time and frequency com-
parison techniques.
Technique Distance
GPS AV 20000 km
GPS CV 10000 km
TWSTFT 10000 km
Optical fiber 300 km
The advantage of the TWSTFT method is that signals are propagating simul-
taneously in both directions and errors due to the atmosphere of the Earth cancel
out to the first order. However, GNSS time transfer is less cost and labor intensive,
because only receiving equipment is needed. A GNSS receiver can in principle be
operated everywhere on the Earth. There is no signal emission, just reception.
New methods of comparing two remote clocks by an optical signal that is sent
through an optical fiber have recently been developed. Mainly experiments have
been done to compare remote experimental optical frequency standards [16]. First
experiments related to time transfer via optical fibers using a two-way technique
have also been performed [23, 24]. However, time and frequency transfer using
an optical fiber is a point-to-point technique, currently just applicable for short
distance links, and a very expensive method compared to GPS.
2.3. GPS and Time and Frequency Transfer

Currently the US Global Positioning System (GPS) is the main tool used for time
and frequency transfer between remote sites. A second system is operated by
Russia and named Global Navigation Satellite System (GLONASS). Unfortunately
the constellation was incomplete during many years and the usage was restricted
to military users. Thus, this system did not become very popular and currently
few receivers are on the market. Additional systems are planned, like the European
Galileo system or the Chinese COMPASS, but these systems are currently all in
the development phase.
In this work the focus is on the GPS system. Most statements made in this
Section are in principle also valid for the GLONASS system, except some technical
details.
2.3. GPS and Time and Frequency Transfer 9
2.3.1. Brief GPS System Overview

The nominal GPS system consists of a constellation of at least 24 satellites orbiting
the Earth in 6 orbital planes. The orbits are approximately circular and equally
spaced around the equator at a 60◦ separation with an inclination relative to the
equator of 55◦ . The mean distance of a satellite to the center of mass of the Earth
is about 26600 km [4]. Figure 2.2 gives an artist impression of the GPS satellite
constellation, which is also called the space segment of the GPS system [4]. Due
to this constellation, far in the north and in the south regions exist where the
satellites are never visible in the zenith for an observer on the Earth. This is the
so-called ”north hole” or ”south hole”, respectively.
Figure 2.2.: The GPS satellite constellation. This image is by courtesy of the US
government.
On board of each satellite atomic clocks are operated. The type of atomic
clocks (caesium, rubidium) depends on the satellites within a specific production
series. The nominal output frequency of the clocks is f0 = 10.23 MHz. From
this fundamental frequency the two microwave frequencies f1 = 1575.42 MHz and
f2 = 1227.60 MHz are derived by multiplying f0 by 154 and 120, respectively.
These two frequencies are the carrier frequencies on the L1 and the L2 band. Since
they are below 2 GHz no directional antenna is needed to receive the signal.
Before being transmitted by an antenna array the two carriers are phase modu-
lated with pseudorandom noise codes, the so-called PRN-codes. These are binary
codes with a chip rate of 1.032 MHz on the f1 frequency, named coarse/aquisition
(C/A) code, and a binary code with 10.23 MHz chip rate on both frequencies,
called precision (P) code. These codes are unique for each satellite, continuously
repeated, and used for the identification and the tracking of the satellites by a
correlation process performed by a receiver. All satellites transmit their signals

on the same frequencies. This technique is called Code Division Multiple Access
(CDMA). To suppress cross-correlation between the PRN-codes, so-called Gold
codes are used. Mathematical details and details on the signal generation are out-
lined in reference [4]. The term chip is used instead of bit, because these codes
contain no additional data information. The modulation is a Bi-Phase Shift Key
(BPSK) modulation [4].
An additional 50 bits per second binary code is superimposed to the PRN-codes
by a binary addition modulo 2 (exclusive-or process). This code is called the navi-
gation message and contains information about the status of the individual satellite
as well as of the constellation as a whole, the satellites’ positions (ephemeris data),
the offset of the satellites’ clocks with respect to a common reference timescale (the
GPS system time), and information about the ionosphere [25].
The P-code and the C/A code signal are modulated to the carrier frequency in
a way that there is a phase shift between the two codes of 90◦ . Due to the satellite
antenna array the signals are right-handed circular polarized to suppress Faraday
rotation in the ionosphere.
The satellites are controlled and monitored by the so-called ground segment,
consisting of several monitor stations, a few uplink stations, and a master control
station which is located at Schriever Air Force Base near Colorado Springs, USA.
Here the satellite ephemerides are forecasted, the offset of the individual clocks with
respect to the system time is calculated, and clock steering commands are sent to
the satellites, if necessary. The system time is an ensemble timescale of all clocks
in the satellites and the ground stations. It is kept in close agreement with the
time scale of the United States Naval Observatory (USNO), UTC(USNO). Since
there is more than one atomic clock in each satellite, the master control station
also decides which clock is used for the signal generation depending on the clock
performance.
All systems which are capable of receiving the GPS signal form the user segment.
Since the transmitting power of the GPS satellites is less than 40 W, the signal is
below the thermal noise on the Earth. As mentioned before, the signal reception
and demodulation is done with the help of a correlation process. The receiver
generates a local copy of the PRN-code derived from its internal oscillator. This
local copy is electronically shifted in time and multiplied with the incoming an-
tenna signal. If the received satellite PRN-code, which is hidden in the noise, and
the replica signal coincide, the result of the multiplication, called the correlation
function, is at a stable maximum and the receiver’s tracking loops can lock to the
satellite. It is obvious that this process also requires a local replica of the received
carrier-frequency. The carrier replica has to be shifted in frequency, because the
received satellite carrier frequency is affected by the Doppler effect.
Figure 2.3 shows the correlation principle exemplarily for one chip of the code.
The black rectangular function is the incoming code chip. If the local copy (blue
rectangle) is in phase with the incoming code chip, the correlation function is at
its maximum (blue arrow), as depicted in the right picture. In the left picture the
replica code is 1/4 code chip too early.
In fact, a GPS receiver uses two additional replica codes to detect the maximum
of the correlation function. One of them is earlier (green rectangle) than that
which should be in phase with the incoming code and one is later (red rectangular)
with a fixed offset (0.5 code chips in Figure 2.3). Consequentially, the additional
copies of the code are called early and late replica, respectively, while the copy
that should be in phase with the incoming signal is called prompt replica. If late
and early correlation function (green and red arrow, respectively) are in agreement,
the maximum correlation of the prompt replica is accomplished. Thus, the receiver
searches for the zero crossing of the difference of early and late correlation function.
This process has to be done in several receiver channels simultaneously in order to
allow the receiver to track the signals of multiple satellites at the same time. The
length of the code chips determines the width of the correlation function and thus
the achievable precision.
The correlation process allows the receiver to measure the phase of the satellite
code with respect to its local clock. The length of one PRN C/A-code is 1023
code chips and the repetition rate is 1 ms. Thus, these measurements have to be
aligned to the GPS system time by the help of the navigation message in order
to get an unambiguous measurement. This is possible with the help of a 20 ms
data sequence in the navigation message, the so-called handover word, which is
periodically repeated and allows for aligning one of the 20 C/A-code sequences
within this message to the GPS system time, as transmitted by the satellite. With
this information and the correlation measurement the receiver is able to calculate
the apparent signal propagation time between satellite and receiver. Multiplying
this propagation time with the speed of light yields the range between receiver and
satellite. However, there are propagation errors on the signal path and the clocks
are not perfectly synchronized. For this reason the measured quantity is called
”pseudorange”.
The correlation and the demodulation of the signals is usually done by a micro-
processor. The antenna signal is converted to an intermediate frequency (IF) by
mixing it with a frequency derived from the receiver’s local oscillator, as done in
an ordinary radio, but then the IF is directly sampled and converted to a digital
stream. The receiver channels and the correlators are only virtually existing. A
good explanation of the details of the correlation measurement, the demodulation
of the navigation message, and technical details of GPS receivers are given in ref-
erence [4]. The structure and the contents of the navigation message are described
in the documentation [25].
It has to be outlined here that the P-code has a repetition rate of one week
and that its PRN sequence is encrypted. Thus a direct tracking of the P-code is
restricted to authorized users with a special key. However, modern receivers can
1.5 1.5
1.0 1.0
Correlator output
Correlator output
0.5 0.5
0.0 0.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Code chips Code chips
Figure 2.3.: The correlation principle for one code chip. The code chips are sym-
bolized by the rectangular functions. The black function depicts the
incoming code chip, the green, blue, and red functions are the early,
prompt, and late replicas, respectively. The triangular function is
the correlation function. The arrows show the correlator output re-
lated to the distinct replica. The left picture shows a situation in
which the prompt replica is in phase with the incoming signal, while
in the right picture the replicas are 1/4 code chip too early. This
figure is similar to the figure given on page 148 of reference [4].
track the P-code on both frequencies after locking to the C/A-code by performing
a cross-correlation between the signals on the two frequencies. As explained later,
receiving a signal on both frequencies has the advantage of providing a correction
of the signal path delays induced by the ionosphere. Receivers which are tracking
the P-code with the help of the C/A-code show the same measurement noise as
ordinary C/A-code receivers. They cannot benefit from the narrower correlation
peak due to the higher chipping rate of the P-code providing a higher precision,
because the C/A-code is always needed to keep the replica carrier frequency and
the cross-correlators in the correct phase relationship.
Since the receiver needs to lock on the carrier frequencies of the satellites, it can
provide a measurement of the phase of the received signal with an ambiguity of
multiples of cycles of the carrier frequency. Modern receivers can do this with a
precision of better than 1% of the wavelength [5], but the initial phase cannot be
related to the GPS system time and is always ambiguous.
In principle the operation of GLONASS is similar to that of GPS, but GLONASS

uses the same PRN code on all satellites. Instead, the satellites use different carrier
frequencies. This technique is called Frequency Division Multiple Access (FDMA).
The advantage is that if there is an interfering signal on one of the frequencies
not the complete system will fail to operate. The disadvantage is that different
frequencies have different delays within the analog part of the receiver which makes
it much more difficult to calibrate a GLONASS receiver. The GLONASS space
segment consists only of three orbital planes. and the satellites are controlled by a
master control station nearby Moscow in Russia. A description of GLONASS and
its frequencies is summarized in the reference [4].
All GNSS systems planned for the future will use the CDMA technique. Hence
they can provide some of their signals on the same frequencies as GPS and a modern
receiver in principle just needs an update of the internal processing software to be
able to track the new satellites.
In addition to the GNSS system local augmentation systems based on geosta-
tionary satellites exist. These systems mostly transmit signals on the f1 frequency.
However, these systems are not used for time and frequency comparisons and are
not subject of this work.
2.3.2. Coordinate Systems and Satellite Motion

In this subsection the nomenclature and the satellite motion used throughout the
remainder of the work (e.g. in Chapter 4) are explained. Basic relations for the
coordinate systems and the transformations between them are given in Appendix A.
A short mathematical description of the satellite motion is outlined in Appendix B.
It is needed later in Chapter 4
The position of the receiver’s antenna and also the position of the satellites can be
represented in a right-handed Cartesian coordinate system (x, y, z) with the origin
located at the center of mass of the Earth. The x-axis is pointing to the Greenwich
meridian in the meridian plane, while the z-axis coincides with the rotational axis
of the Earth. Thus the (x, y)-plane rotates along with the Earth. The position of
a receiver located at a fixed point on the Earth surface does not change with time
in these Earth Centered Earth Fixed (ECEF) coordinates, neglecting very small
effects at the cm/year level due to continental drifts and tidal effects on the solid
Earth surface induced by the gravitational force of the Moon.
The ECEF system can be expressed with the geodetic coordinates ϕ, θ and h,
representing the geodetic latitude, longitude and height, respectively. The flatten-
ing of the Earth has to be taken into account.
Different realizations of the ECEF system with different parameters for the flat-
tening of the Earth are in use. The ECEF reference frame used by the GPS control
segment to predict the ephemeris data broadcasted with the navigation message is
the World Geodetic System 1984 (WGS 84), while the frame used by GLONASS is
the PZ-90 system [4]. To improve the precision of position estimates and/or time
and frequency comparison results ephemeris data provided by a different source
are often employed in post-processing applications. The reference frames of this
ephemeris data can differ from the reference frames broadcasted by the satellites,
e.g. the frame used by the International GNSS Service (IGS) is the International
Terrestrial Reference Frame (ITRF) which is in agreement with the conventions
of the International Earth Rotation Service (IERS). The IGS and the usage of its
data are described in Section 2.5.
The motion of a satellite around the Earth can be described in a right handed
coordinate system using the Earth’s center as the origin. A convenient choice is
the so-called Earth-Centered Inertial Coordinate System (ECI) with the x-axis
pointing to the vernal equinox and the z-axis defined as the Earth’s rotational axis
on January 1st 2000, 12:00 UTC [5]. The satellite motion is very well approximated
by Newtonian mechanics. The ECI system is e.g. needed to derive relativistic
effects (see Appendix F).
The azimuth and the elevation of the satellite are defined in a local left-handed
coordinate system with the receiver antenna’s being located at the origin.
2.3.3. Observation Model

The observation model, connecting the measurements of a receiver to the geometry
and delays induced by physical influences are essential to understand the function-
ality of GNSS system. Thus, they are introduced briefly in the following.
The geometric range between one satellite s at the time of signal emission ts and
the receiver’s antenna at the time of signal reception tr is given by
p
ρs (ts , tr ) = (xs (ts ) − xr (tr ))2 + (ys (ts ) − yr (tr ))2 + (zs (ts ) − zr (tr ))2 , (2.3)
and the observation model for the code measured pseudorange Cs (in units of
meters) is given by [5]
Cs (ts , tr ) = ρs (ts , tr ) + (δtr − δts )c + ∆C + ε . (2.4)
Here Cs can stand for the measurement of the C/A-code as well as for the P1 and
P2 code measurement, respectively, or for linear combinations of the data which are
explained later. δtr and δts denote the offset of the receiver and the satellite clock
with respect to a common reference time (e.g. the GPS system time), respectively,
∆C summarizes all physical influences on the signal path and the satellite clock,
ε denotes the noise of the measurement process, and c is the speed of light in
vacuum. This equation is valid for all satellite navigation systems, including the
Russian GLONASS.
The delay due to physical influences on the code measurement can be modeled
as
∆C = δion + δtro + δtide + δmul + δrel , (2.5)
with δion and δtro being the signal delays induced by the refractivity of the ionosphere
and the troposphere, δtide denoting variations due to deformation of the Earth’s
surface by the gravitational potential of the Moon and the Sun and ocean tides,
and δmul denoting multipath delays due to signal reflections at buildings, trees, and
hard grounds nearby the receiver’s antenna and diffraction effects. δrel summarizes
the relativistic effects.
The model for the carrier-phase measurement is given by
λΦs (ts , tr ) = ρs (ts , tr ) + (δtr − δts )c + λNs + ∆Φ + ε . (2.6)
Φs is measured in cycles of the carrier frequency. As in equation (2.4), it can

be the phase measurement on the f1 or f2 frequency or a linear combination of
these measurements in cycles of the carrier frequency. λ is the corresponding
wavelength. Since the initial number of carrier frequency cycles between satellite
and receiver antenna are unknown, λΦ does not directly represent the pseudorange.
This ambiguity is expressed by Ns .
For the carrier-phase measurement the respective delay due to physical influences
reads
∆Φ = −δion + δtro + δtide + δmul + δrel + δAnt + δpwu . (2.7)
In contrast to the code measurement the ionospheric error has the opposite sign,
because the group velocity of the satellite signal is different from the phase velocity
in the ionosphere, which is explained below. δAnt is related to changes of the
antenna phase center for different elevation angles of the incoming signal and δpwu
denotes phase wind up effects.
In terms of positioning, the equations (2.4) and (2.6) are usually solved by a
linearization. After an expansion to a Taylor series either a least-square estimation
algorithm [5], or, particularly in case of a moving receiver, a Kalman filter [26] is
used. The measurements of at least four satellites are necessary in order to solve
for the three coordinates and the receiver clock. In case of time and frequency com-
parisons, in principle the signals of one satellite are sufficient, since the receiver’s
antenna position is fixed and well known.
Depending on the desired precision more or less of the delays in (2.5) and (2.7)
have to be removed by mathematical models, in terms of positioning as well as in
terms of time and frequency comparisons.
2.3.4. Time and Frequency Transfer Receivers

In contrast to positioning, where the pseudorange observations (2.4) are used to
determine the position of the receiver, the task in the framework of time and
frequency comparisons is to solve for the receiver clock offset δtr . The position
of the antenna is usually static within the reference frame, neglecting the small
variations mentioned above. For this account, GNSS receivers used for time and
frequency comparisons must allow for referencing the measurements to an external

timescale physically provided by an 1 PPS signal.1
A typical timing receiver system used in a timing laboratory consists of the
receiver itself, which is usually located in a temperature stabilized environment, and
an antenna which is located outside, preferably at a site with minimum multipath
effects. In case of some receivers additional computers to collect and process the
measurement data are needed.
In the majority of cases the antenna and the antenna cable are not temperature
stabilized and have to meet strict conditions to the stability. As outlined in refer-
ence [27] the delay of the antenna cable should vary less than 0.05 ps per degree
per meter over the range -30◦C to +40◦ C and the antenna should tolerate diurnal
temperature changes of 40◦ C.
Two different basic design concepts for a time and frequency transfer receiver are
possible. Figure 2.4 shows the more traditional set-up. The GPS measurements are
at first referenced to the time of an internal oscillator. The internal oscillator can
either be synchronized to an external frequency or to the frequency derived from the
satellite’s signals. The dashed lines indicate that in principle both synchronization
methods can be utilized. The internal timescale is synchronized to the GNSS time.
An external 1 PPS signal is compared to the 1 PPS signal generated from the
internal oscillator with a TIC. Since the GNSS measurements are also referenced
to the internal oscillator, it cancels out after the TIC and GNSS data are combined.
This design can be realized differently. For example, it is not necessary that
the TIC and the computer that combines the GNSS and TIC data are integrated
in the same housing together with the GNSS receiving and analyzing parts of the
system. A system can be assembled from multiple devices.
Integrated time and frequency receivers of this type are available on the market
since decades. Some older devices were so-called single-channel receivers. These
receivers were capable to track only the C/A-code signals of one satellite at the same
time, which is in principle sufficient for time transfer. For CV time comparisons
between remote sites with these receivers, a schedule was needed to ensure that
both sites track the same satellite at the same time. Such schedules are still
provided by the BIPM for international timescale comparisons, but the usage of
that kind of receivers diminishes.
Newer multi-channel receivers are capable to track several satellites at the same
time [28]. Modern state-of-the-art time and frequency receivers which are con-
structed according to the set-up depicted in Figure 2.4 are usually based on geodetic
receiver circuit boards. These boards are originally intended for precise position-
ing applications together with a post-processing of the output data, but output a
1 PPS signal derived from the internal clock. They are capable to track the C/A-
1
If only a frequency comparison is desired, it is sufficient to reference the receiver’s measurement
to an external reference frequency with ambiguous phase.
Figure 2.4.: Simplified schematic of a traditional time and frequency transfer re-
ceiver. All parts of the receiver which are analyzing and processing
the GPS signal are summarized as ”correlator”.
code as well as the P-code on both frequencies and to measure the carrier-phase.
The boards are integrated together with the TIC and a computer which combines
the data.
While older receivers which are only tracking the C/A-code are pure GPS re-
ceivers, some of the modern geodetic receivers are capable of tracking the signals
of the other GNSS systems. If the internal delays of receivers designed as shown
in Figure 2.4 are well known, the 1 PPS input connector can be defined as the
receiver’s reference point to which all measurements are related.
Another design concept for time transfer based on geodetic receivers is depicted
in Figure 2.5. Some of these receivers allow for synchronizing the internal timescale
to an external 1 PPS. The internal timescale is thereby related to the first zero
crossing which follows the external 1 PPS signal. Hence the 1 PPS signal and
the reference frequency must be coherent, which means that both signals have to
originate from the same source.
The advantage of a receiver designed according to Figure 2.5 is that no additional
noise of a TIC affects the measurements. However, the definition of a reference
point is more difficult, because in a timing laboratory the relation between 1 PPS
signal and reference frequency is a priori not fixed. It depends on the cable lengths
and the delays of the signal distribution equipment. Thus an additional measure-
ment is required which must be repeated every time the setup is changed.
The delay between the 1 PPS and the zero crossing of the frequency can directly
be measured using an oscilloscope or a fast TIC. If the result is applied to all
Figure 2.5.: Simplified schematic of a typical time and frequency transfer receiver
based on geodetic technologies.
measurements, the 1 PPS input connector of the receiver can such be considered
as the reference point. Alternatively, the delay between the input 1 PPS and the
output 1 PPS can be measured with a TIC if the receiver provides an 1 PPS output.
In this case the GPS measurements are related to the output 1 PPS.
In both design concepts the internal delays of the system should be well defined
and provided by the manufacturer, as well as the delay of the antenna cable and
the antenna. This includes the signal propagation time difference between the two
frequencies inside a modern receiver. The value of the trigger level of the 1 PPS
input detection should be known or adjustable [27].
According to the recommendations of the Consultative Committee for Time and
Frequency (CCTF) a time receiver provides its code based measurements in the
so called CGGTTS standard [29]. The data are given in 16 min intervals for each
satellite and are computed using pseudorange measurements recorded in 1 s inter-
vals. A detailed description of the CGGTTS file can be found in the reference [29].
The relevant parameters are the time differences between the satellite clock and
the receiver timescale, which is related to the external timescale, and between the
GNSS system time and the receiver timescale, respectively. A cutout of a typical
CGGTTS file is shown in Appendix C.
The internal processing software of the receiver reduces the data by applying
quadratic fits to 15 successive measurements. These 15 s averages are corrected
for the atmospherical delays and the clock corrections taken from the navigation
message in order to get access to the GNSS system time. The known internal delays
of the receiver, the antenna delay, and the delay of the reference signal are applied.
The geometric range between the antenna and the satellite (2.3) is calculated by
using the broadcast ephemeris data and a fixed position for the antenna which is
manually entered into the processing software.
The final results are obtained by a linear fit to 52 successive 15 s averages. Thus,
the total averaging time is 780 s. The gap of 3 minutes between the 16 minute
data spacing and the averaging time is inherited from the use of old single-channel
receivers which needed this time to detect and lock to a signal of a new satellite.
Besides the CGGTTS data a second data standard is used within the frame-
work of time comparisons. Originally defined for geodetic applications, modern
receivers record their measurement data in the Receiver Independent Exchange
Format (RINEX) [30]. For each satellite in view the raw measurement data of the
C/A, P1, P2 pseudoranges in meters and the cycles of the carrier-phase on both
frequencies are recorded in the RINEX observation file. Some receivers provide
additional information, like the Doppler effect on the signal, the signal-to-noise
ratio, or the pseudorange measurements of a regional satellite based augmentation
system.
The spacing of the data is not prescribed in the RINEX standard [30]. Usu-
ally 30 s intervals are used. The timetags can be related either to the GPS or
the GLONASS timescales, but mostly the GPS time is utilized as receiver time
scale. The timetags are given as year, month, day, hour, minute, and second. The
broadcast satellite ephemeris and clock data are recorded in a separate file, the
RINEX navigation file. The observation data of a multi-system receiver contain
the measurement data of all GNSS systems, while the navigation files are generated
separately for GPS and GLONASS. A cutout of a typical 30 s GPS RINEX file is
shown in Appendix D.
The original idea of the RINEX definition was to enable post-processing of the
measurement data for high precision positioning applications and network solu-
tions, independent from the hardware and correction models favored by the hard-
ware manufacturer.
A lot of geodetic receivers designed as depicted in Figure 2.5 do not provide
CGGTTS data, because the CGGTTS standard is used in the timing community
only, which is small compared to the geodetic community. Thus, a software called
RINEX2CGGTTS was developed at the Royal Observatory of Belgium (Observa-
toire Royal de Belgique, ORB) which allows for generating standard CGGTTS files
from daily 30 s RINEX observation data by combining it with two days of data of
the navigation files [31]. The fixed antenna position must be entered to the this
software. The internal delays, the antenna delay, and the reference delay have also
to be entered to the software, because they are usually not applied to the RINEX
data since they are of no interest in positioning applications.
2.4. Signal Delays and Correction Models

The GNSS technique, in terms of positioning as well as in the framework of time
and frequency comparisons, is a one-way technique. Thus, all signal delaying effects
and influences on the clocks in the transmitting satellites are affecting the results,
if they are not removed by models or technical tricks.
In the following the different effects are briefly introduced and explained. All
correction models which are studied in detail within a simple frequency transfer
algorithm developed at PTB (Section 4.4) are explicitly outlined as mathematical
expressions. Further informations are given in the appendix.
2.4.1. Atmospherical Corrections

The propagation of the satellite signals is affected by the ionosphere, which is the
region of the atmosphere between about 70 km and 1000 km above the Earth sur-
face [4], and the troposphere, which is in terms of satellite navigation the complete
region of the atmosphere below the ionosphere [5].
The change of the apparent length of the path of a signal in a medium with
refractivity n compared to a signal propagating in the vacuum reads
Z
∆r = (n − 1) ds . (2.8)
Path
Since the troposphere affects the satellite signals by the presence of neutral atoms
and molecules [5], the troposphere is a non-dispersive medium for frequencies below
15 GHz [4], which means that the effect is the same on all GNSS frequencies. The
group and phase velocity of the signals are equal.
The refractivity of the troposphere is usually rescaled by N = 106(n − 1) and
sometimes separated into wet and dry parts N = Nw + Nd . Thus equation (2.8)
for the troposphere becomes
Z Z Z
−6
δtro = Tro = 10 Nds = Nw ds + Nd ds = Trow + Trod . (2.9)
Path Path Path
The wet part of the troposphere is the region below 10 km and contributing with
about 10 % to (2.9), while the remaining 90 % are due to the dry part of the
troposphere.
The integration is usually done along the zenith path and aligned to the signal
path by a mapping function afterwards. The mapping function is depending on
the satellite elevation.
The most commonly used solutions for (2.9) are the Saastamoinen [32] and the
Hopfield [33] model. Sometimes a simplified model is sufficient [34], depending on
2.4. Signal Delays and Correction Models 21
the desired precision. Together with the related mapping functions it can be given
by
Troz,w = 0.1 m , Troz,d = 2.31e−h/7000 m m , Fw = Fd = 1/ sin(el) , (2.10)
where the index z denotes the tropospheric delay in the zenith direction. Fw and
Fd are the mapping functions. h is the geodetic altitude of the receiver’s antenna,
and el is the satellite’s elevation.
In contrast to the troposphere, the ionosphere affects the satellite signals by free
electrons caused by ultraviolet radiation from the sun which ionizes gas molecules.
Usually a GPS receiver uses forecasted parameters broadcasted in the GPS nav-
igation message to correct for this delay. Details of this model can be found in
the references [5, 25]. The GLONASS navigation message does not include such
parameters [35]. Modern receivers which are capable to receive both systems si-
multaneously can use the parameters in the GPS navigation message also for the
correction of the GLONASS signals.
The ionosphere is a dispersive medium, group velocity and phase velocity are
influenced with opposite sign. This property can be used, if the receiver is capable
to receive signals on both frequencies f1 and f2 , to remove the ionospherical delay
from the observation equations in first order by a linear combination. If P1 is
the P-code pseudorange on f1 and P2 the P-code pseudorange on f2 , the linear
combination is given by
f1 2 P1 − f2 2 P2
P3 = . (2.11)
f1 2 − f2 2
A brief derivation is given in Appendix E.
Equation (2.11) is called the ionosphere free linear combination since the higher
order effects are marginal under normal conditions. Small significant effects occur
only in the case of high ionosphere activity due to solar storm events [36].
Inserting the GPS frequencies the combination (2.11) can be written as
1542f0 2 P1 − 1202 f0 2 P2 1542 P1 − 1202 P2
P3 = = . (2.12)
f0 2 (1542 − 1202 ) 9316
For the carrier-phase measurement one can define L1 = λ1 Φ1 and L2 = λ2 Φ2,
with Φ1 being the measurement at the frequency f1 with respect to one satellite
and Φ2 being the measurement at the frequency f2 with respect to the same satel-
lite. λ1 and λ2 are the corresponding wavelengths. Then a ionosphere free linear
combination can also be constructed for the carrier-phase measurements:
f1 2 L1 − f2 2 L2
L3 = . (2.13)
f1 2 − f2 2
With the assumption that the pseudorange measurements P1 and P2 are affected
by independent additive noise with variances [37]
Var [P1] = Var [P2] = ε2 , (2.14)
the variance of the P3 combination is

2
f1 P1 − f2 2 P2 f1 4 + f2 4 2
Var [P3] = Var = ε , (2.15)
f1 2 − f2 2 (f1 2 − f2 2 )2
according to the calculation rules of variances. After inserting the values of the
GPS frequencies the P3 variance becomes
Var [P3] = 8.87ε2 , (2.16)
with the standard deviation
εP3 = 2.98ε . (2.17)
Equation (2.15) is also valid for the L3 combination (2.13). The benefit of canceling
out the ionosphere error in first order due to the usage of (2.11) and (2.13) has to
be paid for with increased noise.
It has to be noted that higher order contributions to the ionospherical refractiv-
ity, which are neglected in (E.19) and (E.20), can only be removed by using more
than two frequencies or model values for the Earth magnetic field. In terms of a
future modernization of the GPS system the transmission of a signal on a third
frequency f3 = 115f0 is planned [5] and all newly developed GNSS systems will
broadcast signals on more than two frequencies.
2.4.2. Relativistic Corrections

Relativistic effects arise due the gravitational potential of the Earth, the motion
of the satellite and the rotation of the Earth.
From the point of view of a user on the Earth the satellite clocks are shifted by
general relativistic effects due to the mean gravitational potential of the Earth and
the mean velocity of the satellite on its orbit. This correction has not to be taken
into account by the user, because it is already applied in the design of the GPS
satellites. The clocks in the satellites are adjusted to provide an output frequency
of about 10.22999999543 MHz. An observer on the Earth geoid thus observes the
nominal frequency of 10.23 MHz [4, 5, 38].
Since the satellite orbits are not perfect circles but ellipses, the gravitational
potential and the velocity undergo variations. Thus, also the frequency observed
by a user on the Earth varies. This effect can be modeled approximately as
2
∆te = ~rs · ~vs . (2.18)
c2
It is called the eccentricity correction [4, 38].
A second effect that has to be corrected by the user is the so-called Sagnac effect.
It is related to the rotation of the Earth. A possible approximation is
1
∆tS = ~ωE · [~rR × (~rR − ~rT )] , (2.19)
c2
2.4. Signal Delays and Correction Models 23
with the angular velocity vector of the Earth ~ωE and ~rR and ~rT the receiver’s
and satellite’s position at reception and transmission time, respectively. For many
applications it is sufficient to estimate ~rR − ~rT from the pseudorange measure-
ment (2.4).
Appendix F outlines a brief derivation of the three corrections introduced above.
2.4.3. Multipath
The multipath delay is one of the most difficult to predict delay sources [4, 5].
The satellite signal could be reflected at objects nearby the antenna. These could
be buildings, trees, hard grounds, or mainly metal objects, respectively. The re-
flected signal and the direct signal interfere with each other at the antenna phase
center and the correlation measurement, and thus the pseudorange measurement
is affected. These delays are geometry, site, and time dependent. It is obvious
that this delay cannot be corrected with a mathematical model. But it can be
minimized with some technical tricks.
A first reduction is achieved due to the right-handed circular polarization of the
satellite signals. The receiving antenna attenuates the reflected signal, because it
is designed for right-handed polarization and the polarization has likely changed
after the reflection.
Further reduction is accomplished by using choke ring antennas or so-called ”pin-
wheel” antennas [39]. The former are antennas equipped with vertical metal rings
aligned concentrically around the receiving part. These rings are connected to
electrical ground and signals with negative elevation angles are attenuated. The
latter are based on a commercial patent-registered design consisting of a phase ar-
ray of aperture-coupled spiral slots optimized for right-handed circularly polarized
signals.
Modern receivers can minimize the impact of multipath on the correlation mea-
surement by using the concept of ”narrow correlators” [40]. The reflected signals
have at most 1/3 of the signal strength compared to the direct signals [5]. A su-
perposition of the direct and the reflected signal deforms the resulting correlation
function [4], as depicted in Figure 2.6.
The receiver’s tracking loops are thus unable to find the correct maximum, if the
spacing between early and late correlator is too big. For smaller spacing this effect
could partially compensated. This requires a higher bandwidth of the receiver,
as demonstrated in reference [40]. Since the correlation process is usually done
completely digital, this means a higher sampling rate and the capability of the
receiver processor to analyze a high amount of data, which has become possible
only through the progress in modern microelectronics.
Figure 2.6.: The deformation of the correlation function due to multipath. The
green line shows the correlation function for the direct received sig-
nal, the yellow line is the multipath signal correlation function, the
black line is the superposition. The blue arrows depicts the corre-
lator output for 1.0 chip early and late correlator spacing. The red
arrows depict the correlator output for narrow early and late corre-
lator spacing.
2.4.4. Site Displacement

For high precision applications effects related to dynamical deformations of the
Earth and forces on the satellites due to massive bodies in the solar system (mainly
the Moon and the Sun) have to be considered. The knowledge of the latter is
especially important within network applications, where the ephemerides of the
satellites are to be determined with very high precision.
The deformation of the Earth surface is caused by the gravitational potential of
the Moon and the Sun acting on the Earth’s elastic body. The displacement of the
station coordinates due to the changing gravitational force can be modeled and
calculated from the positions of the Moon and the Sun. An approximation of the
vector displacement ∆~r of the station is given by [5]
X2 2 1 h i
Mn aE 4 3
∆~r = h2 r̂ R̂n · r̂ − + 3l3 R̂n · r̂ R̂n − (Rn · r̂) r̂ ,
n=1
mE Rn 2 2
(2.20)
with aE the equatorial radius of the Earth (semi-major axis of the Earth ellipsoid,
cf. Appendix A), n = 1, 2 indices for the Moon and the Sun, respectively, Mn the
masses of the Moon and the Sun, mE the mass of the Earth, and Rn the distances
of Moon and Sun with respect to the center of mass of the Earth. The constants
h2 = 0.6078 and l2 = 0.0847 are the so-called degree two Love and Shida numbers,
respectively. R̂n are the unit vectors of the distance vectors of the astronomical
2.5. The International GNSS Service (IGS) 25
objects (Moon, Sun) with origin at the center of mass of the Earth, and r̂ is the
unit vector of the antenna’s position vector.
Since the antenna position is known in ECEF coordinates, also the positions of
the Sun and the Moon have to be known in this coordinate system. In Appendix G
it is explained how the positions of these astronomical bodies are estimated in
the ECEF system. More details concerning the model and the calculation of the
momentary position of Moon and Sun are given in reference [5]. In time transfer
on long intercontinental baselines the Earth displacement has an impact of up to
3 ns if it is not modelled, as demonstrated in Section 4.4.
Another deformation of the solid Earth surface is caused by the falling and
rising tides of the ocean, caused by the gravitational potential of the Moon. The
periodical redistribution of the water masses causes periodic loading of the ocean
bottom. Also the solid Earth deforms under this load. These ocean loading effect
can be modeled, but are different at each location. Details of general models and
the calculation of the coefficients are explained in the references [5, 41, 42]. The
coefficients can be calculated for each station via an on-line program provided by
the Onsala Space Observatory in Sweden [43].
2.4.5. Other Delay Sources

For high precision orbit determination the perturbation of the gravitational po-
tential seen by the satellites due to the forces of the Moon, the Sun, and other
planets in the solar system can be corrected by a model. Furthermore, a model
for the solar radiation pressure caused by the sunlight should be applied. GPS
preprocessing analysis software often used algorithms to detect and remove phase
wind-up effects. Details are extensively explained in reference [5].
Usually the ephemeris of a satellite are referenced to the center of mass of the
satellite, but the pseudorange measurement is referenced to the phase center of
the satellite’s antenna array. Depending on the production series of the satellites
the center of mass and the phase center do not coincide. This can be corrected
by rotations, taking into account that the satellites’ antennas are always pointing
in the direction of the center of the Earth [5]. The variations of the receiver
antenna are usually corrected by applying correctin data, which is explained in the
subsequent Section.
2.5. The International GNSS Service (IGS)

In kinematic GPS positioning the accuracy of the solution can be improved by using
correction data from a fixed source with known position [4]. However, the accu-
racy of code based GPS applications is limited by the accuracy of the correlation
measurement, which is determined by the chip rate of the PRN codes [4]. In order
to benefit from the carrier-phase measurements, which are up to 100 times more
accurate than the code measurements [5], precise information about the satellite’s
ephemeris and the satellite’s clocks are necessary. Otherwise the accuracy would
be limited by the predicted broadcasted parameters. This can be achieved by oper-
ating the receiver as part of a network and solving the equations (2.4) and (2.6) for
the complete set of measurements together with all relevant modeled corrections
using an adjustment calculus [44]. Furthermore, correction data extracted from
such a network can be used together with the measurements of a single receiver
in a post-processing approach. Such a network is operated by the International
GNSS Service (IGS).
The IGS is a voluntary federation of more than 200 institutes and agencies world-
wide, collecting RINEX observation and navigation data of a worldwide network
of reference stations [45]. Figure 2.7 shows the distribution of the stations. These
stations are equipped with geodetic GNSS receivers and the data are uploaded to
local data centers. Some of the receivers are combined GPS and GLONASS re-
ceivers. Furthermore, some stations are equipped with additional sensors providing
meteorological data.
Figure 2.7.: The IGS network [46] in December 2010. Each dot indicates an IGS
reference station.
Regional data centers collect the data from the local data centers, pass them to
global data centers, and act as an archive for the data. Finally the GPS data are
used by several analysis center for a precise determination of the satellite ephemeris,
the satellite clocks, and the Earth orientation parameters. Each analysis center uses
a different software package, applying the correction models introduced before and
solving the equations (2.4) and (2.6) after linearization with an adjustment calculus
for the satellite clocks, the ephemeris, and station positions and clocks, using the
raw GNSS measurement data.
2.5. The International GNSS Service (IGS) 27
A set of stable sites, which have to comply with strict requirements, is used for
the reference frame determination. The geodetic reference frame of the IGS is a
realization of the ITRF [47] (see Subsection 2.3.2).
Finally a weighted average of the solutions of the analysis centers is calculated,
with the weights depending on the quality of the solutions. In the context of the
IGS these combined solutions are called ”products”, while the raw GNSS measure-
ments are called ”data”. The products and data can be downloaded from the global
analysis centers free of charge and can be used for high precision post-processing
GNSS applications. In addition to the IGS products, some analysis centers also
provide estimated ionosphere and troposphere parameters. Furthermore, the navi-
gation data of the receivers are combined to global and regional RINEX navigation
files, containing the broadcasted ephemeris of all satellites or of all satellites visible
in a particular region of the Earth, respectively. A list of the data centers, the
analysis centers, and the institutes participating in the IGS can be found in the
reference [48], where the regional data centers are called ”operational data centers”.
A description of the data formats used by the IGS is given in the reference [49].
The IGS provides the products in three different levels of precision. The Final
Products (IGS) have the highest precision and internal consistency. The Final
Products are the basis for the IGS reference frame. They are made available
on a weekly basis with a latency of 12 days for the last day of the week. The
Rapid Products (IGR) are available with a latency of 17 hours on a daily basis
and their quality is nearly comparable to that of the IGS Final Products. For
most positioning applications it is sufficient to use the IGR products. The Ultra
Rapid Products (IGU) are updated four times a day at at 03:00, 09:00, 15:00, and
21:00 UTC. They are intended for near real time applications with lower precision
and are computed half with observed data and half on the basis of prediction. The
latency of the IGS Final Products is mainly due to the latency of the observation
data of several stations [15]. Thus the IGR Rapid Products are computed from a
lower number of stations of the global network (Figure 2.7). For GLONASS only
Final Products are available.
The IGS as well as the IGR products are computed on a daily basis, which means
that the solutions for each day are calculated from the RINEX files provided by
the IGS stations for this particular day. The orbit files contain the ephemeris data
of each satellite in Cartesion ECEF coordinates within the IGS reference frame
and the satellite clock with respect to the IGS timescale with low precision in 15
minute intervals timetaged to the GPS time. The satellite’s position represents
the center of mass of the satellite. The clock files contain a daily realization of
the IGS reference frame given as the positions of all contributing IGS stations and
the high precision satellite and station clocks and their rates. In case of the IGS
products, the clock file is available with 5 minute or 30 second intervals, while the
IGR clock file is provided with 5 minute intervals only. IGU clock files are not
available. GLONASS clock products are not yet provided by the IGS.
The IGS timescale is a weighted mean of the satellite clocks and the stations
equipped with external frequency standards [50]. Stations connected to clocks
with a good short term stability gain the highest weight, because the computation
is done on a daily basis. The timescales of the IGS and IGR products can differ by
several ten nanoseconds [50]. In the long term, the timescales are aligned to the
broadcasted GPS system time.
Several of those stations that are connected to external frequency standards
which are collocated at timing laboratories together with other GPS equipment.
For example, at PTB the station ”PTBB” is operated, consisting of an Ashtech
Z12-T receiver, a meteorological sensor providing outside temperature, humidity,
and pressure measurements, and the antenna installation. Since the PTBB receiver
is connected to UTC(PTB), its data are provided to the BIPM in parallel, and
CGGTTS P3 files are generated with the RINEX2CGGTTS software. In case of
the P3 data the receiver is called ”PT02”. In principal the clock products can be
used for a timescale comparison between the timing institutes in which an IGS
station is located, but due to the daily analysis strategy the solution is affected by
a phase discontinuity each day at midnight [50].
The satellite velocity is not included in the orbit product, but needed for the
calculation of the relativistic eccentricity correction 2.18. Since the velocity ~vs at
the epoch t is the derivative of the satellite position ~rs
d~rs (t) ~rs (t + ∆t) − ~rs (t)
~vs (t) = ≈ , (2.21)
dt ∆t
it can be calculated from the ephemeris data. The position of the satellite can be
interpolated from the 15 minute ephemeris data, for instance using the Lagrange
interpolation [5]. The Lagrange polynomial is given by
m
X m
Y t − tj
xn (t) = x̂n (tj ) , k 6= j , n = 1, 2, 3 , (2.22)
j=0
tj − tk
k=0
with xn (t) being on of the three component of the satellite position vector ~rs at the
time t and x̂n (tj ) on of the three components of the satellite ephemeris data taken
from the orbit file at the epoch tj . m is the order of the polynomial and usually
selected as 7 or 9 [5]. Also other interpolation techniques can be used, like Newton
or Chebyshev polynomials.
Precise products with the same format that uses the IGS are also provided in-
dependently from the IGS by other institutes and agencies. For example, the
US National Geospatial-Intelligence Agency (NGA) and the Russian Information-
Analytical Center (IANC) provide such products. The NGA ephemeris data are
referenced to the antenna phase center of the satellite, but no clock files are gen-
erated. The IANC data are with less precision, because only a few of the stations
of the IGS network are included in the computation. In contrast to the IGS, the
2.6. Precise Point Positioning 29
NGA products include the satellite velocity and the IANC offers clock products
for the GLONASS system. The products of other agencies are related to different
reference frames and transformations according to equation (A.5) are necessary to
enable comparability.
The advantage of using IGS products is that these products are the weighted
average of several analysis centers, while the products of other agencies are usually
done with one single processing strategy. Thereby errors are minimized and the
reliability is increased. However, on the IGS data centers also the products of some
of the participating analysis centers are available.
A further product of the IGS are files containing information about the Earth
rotation and the polar motion, calculated from the network. This Earth rotation
parameters can be used for the transformation between the ECI and the ECEF
coordinate system (see Appendix A).
In addition, the IGS offers correction data for the antenna phase center of the
satellites and for a variety of receiver antennas. The antenna phase center of the
receiving antenna depends on the angle of the incoming signals and thus on the
satellites’ elevations.
2.6. Precise Point Positioning

Precise Point Positioning (PPP) is a post-processing method which was originally
developed within the geodetic community to enable the determination of a position
with sub-centimeter precision [15]. It makes use of the P3 code as well as of the
carrier-phase measurements of a GPS receiver, taken from the RINEX data. The
equations (2.4) and (2.6) are solved simultaneously together with all implemented
correction models together with IGS like data after a linearization. For this pur-
pose, either the least-square estimation method [5] or a Kalman filter [26] is used.
The ambiguities of the phase of the carrier are estimated with the help of the P3
code. The code itself is weighted down in such a way, that the stability of the result
is mainly that of the carrier-phase measurement. Different software concepts for
the PPP method exist and are explained later in Chapter 4.
The PPP method can obviously be used for time and frequency comparisons, if
the processing software outputs also the receiver clock estimations. In contrast to
positioning of stationay antennas, where the solution can be improved by averaging
over longer measurement data, for time and frequency transfer a solution for each
epoch is needed. Since the ambiguities have to be estimated each time a PPP
process is started, the clock estimates of continuous measurement data show a phase
jump between the distinct independent PPP solutions, because the ambiguity can
not be perfectly solved due to the noise of the code.
Figure 2.8 illustrates this discontinuity. Here the data of two receivers at PTB
have been processed with the NRCan-PPP software [26], which was developed
Figure 2.8.: Comparison between single day and multi day batch PPP process-
ing. The daily solution shows daily phase discontinuities. The data
are calculated from a comparison of two identical receivers which
were connected to the same clock in a short baseline setup (distance
between the antennas was just a few meters).
at the geodetic institute at Natural Resources Canada (NRCan). This software

package is analyzed in detail in Chapter 4. Mostly the GPS receivers produce daily
RINEX data with 30 s data spacing. If the PPP process is done independently
for each daily file, the solutions can diverge from each other by a few hundred of
picoseconds, although the receiver was continuously measuring. This is called the
day-boundary phase jump.
The NRCan software allows for processing of longer periods than one day. One
can assemble multiple RINEX files to a multi-day batch. The solution of the multi
day batch shows a jump within MJD 54581, which is caused by a measurement
discontinuity. The phase jump shows a different sign for the daily and the multi day
processing. This shows that the phase jumps are statistically distributed. Despite
of some weakness, PPP is currently the most stable method for time comparisons.
Geodetic GPS receiver models designed according to Figure 2.5, and only few
time and frequency transfer receiver models apply the internal delays to the RINEX
data, because the RINEX standard [30] was originally developed for geodetic ap-
plications where absolute phase values are not needed. The internal delays can be
applied manually after the PPP processing. The total correction D, which has to
be added to the PPP results, is given by
1542 · DP1 − 1202 · DP2
D= + DCab − DRef (2.23)
9316
2.7. GPS and UTC Generation 31
with the P1 and P2 delays DP1 and DP2 , respectively, the estimated antenna cable
and antenna delay DCab , and the delay of the receiver’s reference point with respect
to the local UTC reference DRef . The frequency dependence of the antenna cable
is absorbed in DP1 and DP2 .
Some receivers do not provide the P1-code. Then the differential code delay
between the C/A-code and the P2-code has to be known.
Since PPP estimates the receiver clock independently for each receiver at any
epoch, time and frequency comparisons using this method are always AV processes.
Besides using the PPP solution directly for time and frequency comparisons,
PPP software could also serve to monitor the antenna positions used for the code
based time transfer and for their periodical updating.
2.7. GPS and UTC Generation

The Coordinated Universal Time UTC is the world-wide reference timescale. Its
generation and dissemination is one of the main tasks of the BIPM Time De-
partment [10]. A physical realization of UTC is not existing. Instead, individual
laboratories ”k” realize an approximation to UTC, named UTC(k). The BIPM
computes and publishes UTC-UTC(k) and its uncertainties [51] in five day inter-
vals referenced to MJDs at 00:00 UTC in the monthly published Circular T (e.g.
see Appendix H).
Figure (2.10) shows the international network of time links on April 2006. The
timescales of the participating laboratories are compared to regional and global
pivot laboratories using the TWSTFT and the GPS method. The a priori choice
of the pivot laboratories (e.g. PTB as the global pivot laboratory) is arbitrary and
could by the BIPM if necessary. Few laboratories are equipped with TWSTFT
stations. The main tool for international time comparisons is GPS. The use of
GLONASS is in an experimental state, used only for the link between UTC(PTB)
and the Russian timescale UTC(SU), and planned for the future [52].
By combining the results of the international time comparisons with the local
clock intercomparison in the laboratories each clock can be compared with each
other. The participating laboratories provide their GPS, TWSTFT, and clock
comparison data to a central BIPM data storage. Usually more than one GPS
receiver is operated in a time laboratory and different kinds of data (CGGTTS
P3, CGGTTS C/A, RINEX) are available. For a link computation the BIPM uses
the best available method. If data of an operational TWSTFT link are available,
this method is preferred. The combination of GPS and TWSTFT is subject of
studies [53].
To calculate the UTC timescale an algorithm called ALGOS is used by the
BIPM [54]. It is a step-by-step algorithm. The link and the clock comparison
data are filtered and cleaned from outliers. Single-frequency C/A-code data are
Figure 2.9.: Deviation from UTC of some selected timescales within 100 days
in 2010, taken from the Circular T. UTC(PTB), UTC(USNO),
UTC(CH), and UTC(ROA) are the local UTC realizations of the
PTB, the USNO, the Swiss Federal Office of Metrology (METAS),
and the Royal Spanish Naval Observatory (ROA), respectively. The
uncertainty is represented by ”error bars” and reflects the quality of
time link calibrations (see Chapter 6).
corrected using ionospheric maps produced by the Center for Orbit Determina-
tion in Europe (CODE), which is one of the IGS analysis centers, located at the
Astronomical Institute of the University Bern in Switzerland. Then the clocks
are weighted depending on their instabilities and a free atomic timescale (Échelle
atomique libre, EAL) is calculated as the weighted average of the more than 350
contributing clocks. Thereby the weighting procedure and the clock frequency pre-
diction are chosen such that EAL is optimized for long-term stability. An upper
limit for the weight of each clock is implemented [55]. By a comparison with the
primary caesium clocks on a monthly basis a frequency steering is applied to EAL
in order to ensure that the scale unit is close to the SI second on the geoid. The
corrected timescale is called international atomic time (Temps atomique interna-
tional, TAI). Finally, UTC is obtained by adding leap seconds to TAI as directed
by the International Earth Rotation and Reference Systems Service (IERS). UTC
is thus kept in agreement with the astronomical time UT1 within ±0.9 s.
Besides publishing the difference UTC-UTC(k) for each participating laboratory
together with the related uncertainty in the Circular T, the BIPM also provides
the weights of the participating clocks and their rates with respect to TAI.
2.7. GPS and UTC Generation 33
Figure 2.10.: The BIPM international network of time links in April 2006. This
image is by courtesy of the BIPM.
The calculation of UTC-UTC(k) and its uncertainties requires the knowledge of

the internal delays of the time transfer equipment involved or the uncertainty of the
calibration of the time links. Up to now, operational links have been calibrated
either by cost and labor intensive two-way satellite time and frequency transfer
(TWSTFT) calibration campaigns, reaching the uncertainty of 1 ns [56], or by
more easy to handle circulations of traveling GPS receivers [57]. The uncertainty
of the latter has been estimated to be 5 ns, based on BIPM’s long-term experience
during the past decades.
The concept of BIPM GPS calibrations is explained in detail in Chapter 6.
Advances based on experiments carried out with the ROA are shown. Here a
further reduction of the uncertainty due to a revised concept including hardware
improvements developed at PTB is demonstrated later in Chapter 6.
3. A Highly Precise Mobile
Frequency Reference
3.1. Motivation
The PTB disseminates a standard frequency (77.5 kHz) and coded time informa-
tion via the long wave transmitter DCF77 located at Mainflingen nearby Frank-
furt/Main [7]. However, this service is not capable to provide a sufficiently stable
and accurate reference for many scientific applications.
The absolute measurement of optical frequencies requires a highly precise refer-
ence frequency with relative uncertainty of measurement ideally far below 10−14
at one day averaging time. Such experiments are e.g. carried out at the LUH in
the framework of the Center for Quantum Engineering and Space-Time Research
(QUEST). A sufficiently stable transportable reference frequency standard and
the possibility of comparisons to PTB’s primary frequency standards are helpful
in these activities.
On that account a mobile frequency reference based on a passive hydrogen maser
was constructed in PTB. The connection to PTB is established by using GPS PPP
as described in Section 2.6.
3.2. Equipment
The passive hydrogen maser VCH-1006 manufactured by the Russian company
”Vremya-Ch” was chosen. A Maser (Microwave amplification by stimulated emis-
sion of radiation) uses dissociated (atomic) hydrogen gas in a microwave cavity and
the 1420,405 MHz ground state hyperfine transition. They are available as passive
and active devices. In active devices the cavity system oscillates by itself and an
electronic circuit is steered to the cavity output signal. A passive maser’s cavity
system does not oscillate by itself. Instead of that, a 1420,405 MHz interrogation
signal is needed to lock a quartz oscillator frequency to the atomic line. Thereby
the frequency accuracy is degraded but passive masers could be made smaller and
lighter than active devices which are typically not mobile.
In Figure 3.1 PTB’s maser room is shown. The four active masers are all manu-
factured by ”Vremya-Ch”. The masers H4 and H5 are of the type series VCH-1005
34
3.2. Equipment 35
and H6 and H8 are of VCH-1003A. It is clearly visible that these bulky devices can-
not be used for mobile applications. Although the commercial caesium clocks are
small and portable, they do not fulfill the requirements on the short term stability
required for optical frequency measurements.
Figure 3.1.: PTB’s maser room with four active masers and three commercial
caesium clocks.
On the left photo in Figure 3.2 the VCH-1006’s front view is visible. The device
is just a little bit larger than a commercial caesium clock. It weights about 30 kg.
With the help of a small keyboard the clock time of the display can be adjusted and
the output frequency generated by a digital synthesizer can be changed in relative
steps of 10−15 within 10−10 .
The right photo shows the electrical connections on the back side. Power can be
provided both as 220 V AC and 27 V DC, so that a continuous operation can be
guaranteed. Via RS232 interface a computer can monitor important operational
parameters like molecular hydrogen pressure, ion pump current, and signal level
voltages in a fixable interval. In case of any malfunction these parameters are useful
to localize the source of the problem. We decided for a normal notebook PC, as
it is simply transportable and offers its own uninterrupted power supply due to its
accumulator. If it is connected to the internet, other computers can monitor the
maser all over the place.
As GPS receiver the Dicom GTR50 time and frequency transfer receiver was
chosen (left Photo of Figure 3.3). The receiver module is included in the case
of an industrial personal computer that is operating on common Linux system
software. It can be controlled by a web interface or directly by connecting monitor,
keyboard, and mouse. As depicted in the photo, all connections (1 PPS, antenna)
36 3. A Highly Precise Mobile Frequency Reference
Figure 3.2.: Maser VCH-1006: The left photo shows the front view, the right
photo points out the multiplicity of 1 PPS and frequency outputs.
The signals can be distributed to the GPS receiver and a measuring
system without additional amplifiers.
Figure 3.3.: GPS-receiver GTR50 and notebook: The left photo shows the
GTR50. The notebook is placed on top receiver on the right photo.
are accessible from the front panel. CGGTTS, L3P, and RINEX files can be created
at the same time in adjustable intervals. The main advantage is that the GPS
receiver module is heated to a stable operating temperature at 45◦ C. As antenna
the Novatel GPS-702 GG together with a 50 m FSJ1 cable is used.
3.3. Performance of the passive Hydrogen Maser

In a first step the passive maser was characterized at PTB by comparing it to
different frequency standards. In Figure 3.4 in the left plot the frequency instability
in terms of the Allan Deviation of the comparison of the passive maser, denoted
3.3. Performance of the passive Hydrogen Maser 37
internally as H7, with respect to PTB’s active maser H6 and the primary frequency
standard CSF1 is depicted. Precision frequency comparators were used for the
comparisons. The caesium fountain CSF1 is currently the frequency source with
the best long term stability available at PTB. The frequency instability of H7 is
better than 3 · 10−15 at one day. Thus, its performance exceeds the manufacturer’s
specifications of 3 · 10−15 in one day [58].
Figure 3.4.: Relative frequency instability in terms of the Allan Deviation of the
passive maser H7 with respect to an active maser and the primary
frequency standard CSF1 (left plot) and comparison of the passive
maser with respect to UTC(PTB) (right plot). After an averaging
time of 105 s the instability is limited by the passive maser.
The plot includes the specifications of a commercial caesium clock of type 5071
high performance option, manufactured initially by Hewlett Packard and now by
Symmetricom, to demonstrate that the frequency instability of the passive maser
is more than one order of magnitude better for short averaging times of less than
a few days. A passive maser represents the most stable transportable frequency
standard available today. Also in the satellites of the European GNSS system
Galileo passive masers will be employed as the local clocks.
The right plot shows a long term comparison of the H7 maser’s 1 PPS output
with respect to UTC(PTB) in April and May 2008 for 30 days. The measurement
was performed utilizing a time interval counter. UTC(PTB) was derived from the
primary standard CS2 that was steered to UTC by an high resolution frequency
offset generator. A relative frequency drift of 4.87 · 10−15 /day was observed. Such
drifts are typical for masers.
In order to use the maser as a mobile reference, which means that the maser
has to be transported in a vehicle, a special mount was constructed that protects
the maser against heavy mechanical shocks. The maser is clamped between metal
chains which are connected to an outer metal frame with four springs in such a way
that the maser can oscillate mechanically inside the outer structure. Handholds
are mounted to the structure to ensure that it can easily be carried by at least two
persons. Figure 3.5 shows the construction while it is located in the luggage space
of a car. The metal springs are clearly visible.
Figure 3.5.: The transportable maser inside the luggage space of a car.
The operation of the maser should never be stopped, otherwise it could take
several weeks until the performance of the continuous operation (Figure 3.4) is
restored. In Figure 3.5 the rechargeable batteries which provide the maser with
27 V DC are indicated. If the maser is transported with a car it can also be
connected to 230 V AC generated from the car’s electrical system by a DC to AC
converter.
In order to monitor whether the operational parameters of the maser are affected
by the transportation or not, the notebook is also operated continuously.
The first test on the transportability of the maser was conducted in 2008. It
was taken to the LUH, which is about 70 km away from the PTB. The device was
shown there on an exhibition for several hours and then transported back to PTB.
Shortly before the transport the frequency was compared to the active maser H5
by a high resolution phase comparator. Measurement values were recorded every
second. Directly after the maser returned to PTB this comparison was repeated.
The operational parameters were recorded during the complete experiment.
Figure 3.6 shows the relative frequency difference between H7 and H5 before and
after the transport. The absolute frequency of the passive maser has changed by
8.6·10−14, but the noise after the transport experiment is at the same level as before.
Furthermore no abnormality was detected in the operational parameters. Since the
maser performance reaches its limit at 104 s (cf. Figure 3.4) it is sufficient to show
only the standard deviation. This shows that the maser can survive a transport
without damages, but the monitoring with GPS is essential in order to get the
3.4. Performance of the GTR50 Time and Frequency Transfer Receiver 39
Figure 3.6.: Relative frequency differences between H7 and H5 before and after
the transport to Hanover.
absolute frequency value at any time, also because of the drift noted in normal
operations (Figure 3.4).
3.4. Performance of the GTR50 Time and

Frequency Transfer Receiver
Since the environmental conditions at a remote location are unknown it is nec-
essary to monitor the maser’s frequency output with respect to PTB’s frequency
standards. For this reason the transportable maser was combined with the Di-
com GTR50 time and frequency transfer receiver [59]. This receiver consists of
a common industrial computer chassis (Figure 3.3, left photo) and a temperature
stabilized box including a Javad GGD 112-T receiver board and a time interval
counter card. The GTR50 is designed according to Figure 2.4, but uses state-of-
the-art electronic circuits.
If the antenna is connected, the internal 20 MHz oscillator of the GPS board is
synchronized to the GPS time and a 1 PPS signal is derived. This 1 PPS signal
is compared to an external 1 PPS signal with the internal time interval counter
card. The measurement data of the time interval counter and the GPS data are
stored on the hard drive of the computer. The data are combined afterwards by
a processing software and can be output in the standard CGGTTS [29] and the
RINEX [30] format.
The time interval counter uses a new patent-registered technique based on surface
acoustic wave (SAW) filters [60], but with less precision than the device described
in the reference. The precision of the time interval counter is specified in the
manual as <50 ps RMS [59].
Figure 3.7 allows a view inside the temperature stabilized box, housing the re-
ceiver board and the time interval counter. The receiver board is clearly identifi-
Figure 3.7.: A view inside the temperature stabilized box.
able. On the right side the outside signal connectors are visible (cf. Figure 3.3), on
the left side two connectors for the RS232 interconnections between the computer
and the box are situated. The time interval counter is not visible, because it is
located below the receiver board. For a uniform temperature distribution the air
inside the box is circulated by a small fan and the heater produces stable conditions
of 45◦ C ± 1◦ C. Thus, the performance of the GTR50 receiver is very stable and
independent from the environmental conditions. It is thus a well suited receiver
for a mobile frequency reference.
The performance of this type of receiver is analyzed in detail in Chapter 4. A
relative frequency instability of better than 10−15 at one day is possible with two
GTR50 receivers in a short baseline common-clock experiment and data evaluation
utilizing the PPP method.
3.5. Installation of the Frequency Reference at

the IQO
The Institute of Quantum Optics (IQO) of the LUH is engaged in developing optical
clocks based on neutral magnesium atoms [61]. In addition, an optical fiber link for
frequency comparisons based on frequency combs can be operated between IQO
and PTB [16]. For both applications, the local frequency comb measurements [62]
and the evaluation of the optical fiber, a precise microwave reference frequency is
needed at the IQO. For this purpose the highly precise frequency reference was
installed at the optical clock laboratory at the IQO.
Figure 3.8 shows the mobile frequency reference at the IQO (left photo) and the
antenna on the roof of the IQO’s building (right picture).
3.5. Installation of the Frequency Reference at the IQO 41
Figure 3.8.: The highly precise mobile frequency reference at the optical clock
laboratory at the IQO of the LUH and the antenna on the roof of
the building.
The maser, the GPS receiver, and the monitoring PC are visible. Additionally,
a battery buffered uninterrupted AC power supply (UPS) and a DC power supply
(DCPS) are available. The DCPS charges the batteries which provides the maser
with 27 V DC (cf. Figure 3.5). These batteries are not visible here.
The antenna is mounted nearby an old chimney together with an other antenna.
The plain antenna on the right side is the antenna which is connected to the GTR50
receiver. The old rod antenna on the left side is connected to a GPS disciplined
rubidium clock with low precision that was used as the IQO’s reference source
before the H7 maser was installed.
In Figure 3.9 the relative frequency instability of the comparison of the passive
maser at IQO with the active H5 maser at PTB in 2008 is depicted and the results
for a local comparison at PTB with a high resolution frequency comparator is
shown.
The GPS data were analyzed using the Japanese Concerto C4 software [63]. De-
tails of the software are explained in Chapter 4. The receiver at PTB was also of
type GTR50. It was connected to UTC(PTB) and the frequency differences be-
tween UTC(PTB) and the H5 maser were additionally measured with a frequency
comparator.
The results shown demonstrate that it is possible to provide a highly precise
reference frequency at a remote location through a line with two GTR50 that can
be referenced to PTB’s masers exceeding an averaging time of 104 s. Up to this
averaging time the results are dominated by white noise of the GPS measurement.
For longer averaging times the instability is completely governed by that of the
passive maser.
It has to be outlined, that such a frequency connection can also be done with
institutes that operate their own equipment. Figure 3.10 shows the frequency
Figure 3.9.: Relative frequency instability of the comparisons of the mobile maser
to an active maser at PTB (red dots) and with GPS PPP between
PTB and LUH (blue dots).
instability of a comparison of an active maser at the Max-Planck Institute for

Quantum Optics (MPQ) in Garching with an active maser at PTB.
Figure 3.10.: Relative frequency instability of the comparison of active masers

located at MPQ and PTB with GPS PPP.
The MPQ operates its own active maser and its own GPS receiver of type Septen-
trio PolaRx. The comparison to PTB is easily established just by exchanging
RINEX data with PTB. At PTB an Ashtech Z12-T receiver was used. The Po-
laRx and the Ashtech receiver are designed according to Figure 2.5 and are not
limited by the performance of a time interval counter in the short term (see Chap-
3.5. Installation of the Frequency Reference at the IQO 43
ter 5). The data were evaluated at PTB using the PPP method (NRCan software
package, see Chapter 4). The results demonstrate that also on a 400 km baseline
GPS PPP frequency comparisons are capable to reference maser frequency stan-
dards to PTB’s standards at averaging times of 104 s. Since two active masers are
compared, the instability is one order of magnitude better for averaging times of
more than 104 s.
4. Software Assessment
Different software packages for the PPP analysis of GPS data are available. In this
Chapter some of these packages, which are used in time and frequency laboratories,
are analyzed in view of their capabilities for time and frequency comparisons.
In the first Section the results of two software packages, namely the NRCan-
PPP software [15] and the Concerto C4 software package [63] are compared in
common-clock zero baseline experiments. While the former package was originally
developed for geodetic applications, including the processing of carrier-phase data
of dynamic (moving) receivers, the latter is especially designed for time and fre-
quency comparisons.
The NRCan-PPP software is of particular importance, because it is used by
the BIPM for the evaluation of the links contributing to UTC with carrier-phase
data [64]. To enable high precision processing of the data of a receiver on a moving
platform it is based on a Kalman filter [26]. In the Section 4.2 it is outlined that
special care has to be taken when data which are affected by corrupted measure-
ment data are evaluated with this software package.
In Section 4.3 five different software packages have been analyzed in view of all
relevant parameters which have to be estimated by a PPP software package. Not
only the clock solution, but also other estimated parameters like positions and
zenith tropospheric zenith path delay estimates are compared in order to find the
software that is most suitable for time transfer. Especially the repeatability of the
position could be a good indicator for quality. Dicom GTR50 time and frequency
transfer receivers connected to local time scales in different laboratories were used.
In Section 4.4 a simple algorithm developed at PTB, which uses only the carrier-
phase measurements is introduced. The results of frequency comparisons on base-
lines with different lengths are compared to the solutions of the NRCan-PPP soft-
ware and the influence of the correction models is analyzed.
The results of the studies presented in this Chapter have already been partially
published in the references [65, 66].
4.1. Common-Clock quasi zero Baseline

Experiments
In this study two different software packages are used, the NRCan-PPP soft-
ware [15] and the Concerto software package developed at the Japanese National
44
4.1. Common-Clock quasi zero Baseline Experiments 45
Institute of Information and Communications Technology (NICT) [63] which main-

tains Japan’s timescale UTC(NICT). The NRCan-PPP software was originally de-
signed for geodetic needs. Therefore it uses a Kalman-Filter, because this is the
way to enable high precision processing of data of a receiver in motion. The NICT
C4 software is especially developed for time and frequency transfer and uses a
simpler least squares estimation algorithm.
If a PPP process is started, the ambiguities of all satellites have to be initially
estimated. Since the RINEX data are usually recorded on a daily basis and the
ambiguities are estimated with the help of the P3-code, a phase discontinuity
between two consecutive days occurs. This problem is well known as the so-called
day boundary problem [67]. Both the NRCan and the C4 software package are of
special relevance for time and frequency comparisons, because they allow to use
processing information of the previous day as input for the actual processed day.
In NRCan-PPP the phase jumps are removed by adopting a file which is called
the ”final normal equation of the Kalman-Filter” of the previous day to the actual
processed day (cf. Figure 2.8). The C4 software simply restores the ambiguity
in a file that is employed by the actual process. This update was implemented
in summer 2008 while the developer, Tadahiro Gotoh from NICT, was staying at
PTB as a guest researcher [68].
As depicted in Figure 4.1 the solution of the NRCan-PPP shows transient oscil-
lations in the beginning. This is typical for a Kalman-Filter [26]. For this reason
there is also the option of using a backward smoothed solution. By concatenating
multiple RINEX files, a multi-day batch solution is possible with this software.
In Figure 4.1 and in the following the time differences as the integral quantity of
the frequency are plotted, since this allows for a better detection of irregularities.
Figure 4.1.: One day comparison of two GTR50 receivers in a common-clock quasi
zero baseline setup with NRCan-PPP. The forward solution shows
initially transient oscillations. Since no calibration of the receivers
has been made, the offset is about 34 ns.
46 4. Software Assessment
As shown in Figure 4.1 the forward solution and the backward smoothing con-
verge not until half a day. Because of that the forward solution is just useful if the
processing time is longer than one day and the first days can be neglected.
In Figure 4.2 the relative frequency instability of the comparison of two receivers
in a common clock quasi zero baseline setup (the antenna positions differ only by
a few meters) is shown.
Figure 4.2.: Relative frequency instability of comparison of two GTR50 receivers

obtained with NICT C4 (blue dots) and NRCan-PPP (red squares)
results.
Because the 30 second measurement data were taken from a longer period ranging
from MJD 54713 to 54732 and the first day is neglected, the NRCan-PPP backward
and forward solution give exactly the same result and can thus not be discriminated
in Figure 4.2. From the Allan deviation comparison between the NRCan-PPP and
the C4 solution it is not possible to decide which software concept is better and
should be preferred. For this reason further examinations are needed. But the
results demonstrate that it is possible to compare frequencies with an uncertainty
level of 10−15 at an averaging time of about one day with both software concepts
and GTR50 receivers.
The solutions provided by the two software concepts sometimes show a small
divergence. In Figure 4.3 a cutout from a longer data batch is plotted. This
divergence is typical, a difference larger than 400 ps has not been observed in
different proper data records from different receivers. Normally this divergence
will increase after three months of continuous processing, because then the results
of the Kalman-Filter are no longer reliable, as specified by the developer [69].
Unfortunately there are not yet such long data records with continuous phase
measurements and without any gap available. If there is one small gap in the
data both software concepts restart the processing, which means resetting of all
ambiguities.
4.2. Disconcontinuous measurement data and drifts 47
Figure 4.3.: Cutout from a longer period with typical divergence between C4 and
NRCan-PPP common clock zero baseline solution. The receivers
have not been calibrated, the offset is about 34 ns.
The maximum difference (≈ 0.3 ns) between the two solutions show that the
numerical processes have an influence on the results. Effects due to different cor-
rection models are equal for both receivers in the quasi zero baseline common-clock
setup and cancel out. In Chapter 6 the calibration of operational TAI links, which
are evaluated by the BIPM with the PPP method, using common-clock differences
on both sites are demonstrated. The software comparison depicted in Figure 4.3
shows that the usage of exactly the same software which is used for the evaluation
of the operational links is mandatory for the calibration, in order to accomplish a
calibration with an uncertainty at the nanosecond level.
4.2. Disconcontinuous measurement data and

drifts
Until now continuous measurement data were used, that means that the 30 s
code and phase measurement taken from both receivers without interruption are
processed over the complete period. Gaps in the measurement data make for dis-
rupting the processing and the ambiguities are reset. It is a well known fact that
this causes phase jumps of up to 1 ns in the results [67]. However, it is an inter-
esting question how the different numerical concepts are responsive to corrupted
measurement data.
To find this out a 25 day data batch of a common clock experiment with two
GTR50s was processed. Due to experiments with the receivers and the receiver
software, the measurement was interrupted in average every 2 hours during the
first 7 days (Figure 4.4).
Figure 4.4.: A 25 day data batch with corrupted measurements in the first 7 days
processed with C4 and the two NRCan-PPP options. The receivers
have not been calibrated.
As expected the NRCan-PPP Kalman-filter has to engage very often and this
makes the solution very noisy. Surprisingly the Kalman-Filter does not engage cor-
rectly in the period with continuous measurement data and produces an oscillation
on a larger scale. In backward smoothed solution the noise is carried forward to the
time interval with continuous measurement. However, if the starting point is lo-
cated in the time interval with good measurement data it equals the C4 solution up
to 300 ps. In contrast, the C4 software is much more resistant against corrupted
measurement data. Corrupted data means that these data are not gathered in
routine laboratory conditions.
RINEX data which are suspected to be affected by measurement interruptions
and are processed with a Kalman filter based software should always be checked
with a second software package, e.g. the Japanese C4.
In Figure 4.5 a Septentrio PolaRx receiver was compared to a GTR50 in a
common clock experiment. The clock signals were provided by different pulse and
frequency distribution amplifiers and the cables were a few 10 meters long. This
apparently caused a phase drift between the two signals to be compared of order
1 ns during the 9 days. Figure 4.5 shows that the slope of the solutions depends on
the starting point. It is thus not determinable which solution represents the real
slope for a given day. Comparing the red and blue plotted data for the days 3 to
7, respectively, the slopes differ from 1.6 · 10−15 in four days for the blue data and
10−15 in 4 days for the red ones. This effect has only be found in the backward
smoothed solution of the NRCan-PPP software.
It has to be noted here that the BIPM uses 30 day batches of RINEX data and the
NRCan backward smoothed solution to evaluate the links for the TAI computation.
To find out whether the effect presented before influences the solutions in case of
timescale comparisons it is necessary to to compare data on long baselines.
4.3. Clock comparisons and PPP parameters 49
Figure 4.5.: The NRCan-PPP backward 6 day smoothed solutions starting one
day displaced from each other.
4.3. Clock comparisons and PPP parameters

Common-clock experiments in a short baseline setup are important for assessing
the general limitations of the measurement process and the data processing strate-
gies. However, figuring out the performance of the PPP software in case of long
baseline time comparisons is very important, because in this case the influence of
the different correction models used in the different software concepts does not
cancel out.
Figure 4.6 shows the timescale comparison UTC(ROA) and UTC(TP) (TP de-
notes Tempus Pragense) of the Royal Spanish Naval Observatory (ROA) and the
Czech Institute of Photonics and Electronics (UFE), respectively. On all sites
GTR50s are connected to the local realizations of UTC derived from commercial
cesium clocks. IGS rapid clock and ephemeris data were used, but no Earth ro-
tation parameters and ocean loading coefficients (cf. Section 2.5). The standard
deviation of the differences between NRCan-PPP and C4 results is 0.275 ns.
The effects induced by the correction models make an impact. This is the so-
called ”software noise” and depicted in the plot on the right handed side of Fig-
ure 4.6. It exceeds the variations shown in Figure 4.3 and the effect demonstrated
in Figure 4.5.
One cannot distinguish which of the two solutions is closer to the reality. Thus,
the NRCan-PPP and the C4 software package are compared to three other software
packages which are used within the timing community in a more detailed analysis.
4.3.1. Software and Experimental Setup

For this analysis data of three Dicom GTR50 time and frequency transfer receivers
located at PTB, at NICT, and at ROA, have been used. The 1 PPS inputs of
Figure 4.6.: Takeout from a longer measurement ranging from MJD 54584 to
54648 between ROA and UFE. The results are on a 5 min. basis.
This data are by courtesy of ROA and UFE.
these receivers were connected to the local UTC realizations (Table 4.1). The
initial position is the position that is known before the analysis and stored in the
receivers for the computation of the CGGTTS data.
Table 4.1.: Initial positions of the stations and timescale input.
Station Initial position 1 PPS input

x = 3844056.75 m UTC(PTB): Steered primary
PTB y = 709664.09 m clock CS2
z = 5023131.73 m
x = −3942087.86 m UTC(NICT): Active
NICT y = 3368252.55 m H-Maser steered by clock
z = 3702001.32 m ensemble
x = 5105511.68 m UTC(ROA): Steered
ROA y = −555187.02 m Cs 5071 A
z = 3769791.71 m
Five different software packages along with IGS final orbit and clock products and
ocean loading coefficients from the GOT002 model [41] were used. The packages re-
quire different IGS input data due to different interpolation processes (Table 4.2).
For instance, the NRCan-PPP software uses Chebychev Polynomials for the in-
terpolation of the satellite orbits [69]. This interpolation method is more stable
around the end points of the data set than others, so that only the IGS orbit file
of the processed day (pd) is needed. Furthermore the data are processed directly
Table 4.2.: Available software packages and required IGS input. pd denotes
processed day, -pd and +pd denote the previous day and the next
day, respectively. -6h and +6h denotes 6 hours data of the previous
and the next day, respectively.
Software Algorithm File input

Atomium [31] Least square pd RINEX,
PPP estimation -1d, pd, +1d orbit, clock and
earth rotation IGS data
NRCan PPP [15] Kalman filter pd RINEX,
(static mode) pd clock and orbit,
no earth rotation
GIPSY 5.0 [70] Kalman filter pd RINEX,
-1d, pd, +1d orbit, clock and
earth rotation
Bernese GPS Least square pd RINEX,
Software 5.0 [71] estimation -1d, pd, +1d orbit, clock and
earth rotation
NICT C4 [63] Least square -6h, pd, +6h RINEX
estimation -1d, pd, +1d orbit and clock,
-2d, -1d, pd, +1d, +2d
earth rotation
in the ECEF system, so that there is no need for Earth rotation parameters (cf.
Section 2.5).
The C4 software uses an overlapping process, so that also RINEX data of 6 hours
of the day before (-6h) and 6 hours of the day after (+6h) the processed day are
needed. This software is currently available in two versions: An older version with
some bugs related to outlier rejection and correlation avoidance between parame-
ters, and a new optimized version, but C4 is still in an experimental status. For
reasons of comparability, the feature of continuous processing without resetting the
ambiguities has not been used for this analysis.
The GIPSY 5.0 and the Bernese GPS software packages are commercial prod-
ucts. The analysis of the data with these packages was done by Héctor Esteban
at ROA and Ulrich Weinbach at IfE at LUH, respectively, as PTB does not own
the respective software licenses. The Bernese software is similar to the software,
which is used by the Center for Orbit Determination in Europe (CODE), an IGS
analysis center, for generating its network contribution to the IGS products.
The Atomium software was developed at the Royal Belgian Observatory (ORB)
especially for time and frequency comparisons. Similar to the C4 software it uses
a least-square estimation algorithm, but it does not have the capability to remove
the daily phase discontinuity due to the beginning of a new file each day.
4.3.2. Time Scale Comparisons

Two time scale comparisons were made, one for the intercontinental baseline PTB-
NICT (8333 km) and the other for the European baseline PTB-ROA (2182 km)
with all available software packages. To avoid inconsistencies due to data gaps
and receiver problems, a period for data computation was selected where all three
receivers were operating normally. A maximum divergence of about 2 ns between
C4 and the other solutions (Figure 4.7) is visible.
Figure 4.7.: Time scale comparison: The GTR50 receiver at PTB has not been
calibrated. C denotes an additional calibration constant that de-
pends on the cable and internal delays. The color assignment to the
different software packages on the left plot is the same as on the right
plot.
In this picture one can hardly distinguish between the packages of the NRCan,
Atomium, Bernese, and GIPSY software. The divergence of the C4 solution gives
just a first indication that it has some shortcoming.
To cancel the clock noise, it was decided to use the NRCan solution as common
reference and compare all other solutions to it (Figure 4.8). A good agreement of
Atomium, Bernese, and GIPSY solution with NRCan of less than ±350 ps is visible.
The ambiguity estimates seem to be similar in this four packages, in consequence no
significant jumps occur in the double difference data. Since all correction data were
applied to the software packages, the double differences are less noisy compared to
the analysis depicted in Figure 4.6. The C4 solution shows mostly an agreement of
about 1.5 ns with NRCan. Clearly, the ambiguities are solved differently, because
significant jumps with respect to the other solutions occur in case of both links.
A drift with respect to the other solutions is also visible in the PTB-ROA link.
However, such drifts occur only if the solutions are calculated on a daily basis,
as done in the framework of this study. If the feature of continuous time transfer
without resetting the ambiguities is used, these drifts are marginal (cf. Figure 4.6).
Besides the evaluation of the agreement of the GPS solutions with each other,
the study of systematic deviations with respect to other techniques or an available
ground truth is important. For this reason the solutions of the PPP software
packages have been compared to the data of the operational TWSTFT links, which
are operated between the institutes in parallel. The TWSTFT data are usually
measured in 2 hour intervals. Since the data of the IGS clock products are in
5 min intervals, the output of the PPP software packages is also spaced in 5 min
intervals. The GPSPPP data were thus compared to values calculated by a linear
interpolation between two adjacent TWSTFT measurements.
Figure 4.8.: Double time differences NRCan − other software solutions. The color
assignment to the different solutions on the left plot is the same as
on the right plot.
However, TWSTFT is not very helpful for this purpose, because of the noise of
the TWSTFT observations one does not get more information than from Figure
4.8, as shown in Figure 4.9 exemplarily for the link between PTB and ROA. Since
the GTR50 receiver at PTB has not been calibrated before this experiment the
mean offset of about -15 ns occurs in the double difference solution.
It has to be noted here that the NRCan-PPP software package includes the
feature of satellite clock interpolation. This increases the noise level but enables
creating output data with the same data spacing as the input RINEX data. This
feature is also included in some other software packages, e.g. Bernese, but not in
Figure 4.9.: Comparison of GPS PPP with TWSTFT, exemplarily for the link
PTB-ROA.
all packages analyzed here. Thus it was not used within this analysis, because all
packages are to be compared at the same level.
4.3.3. Position Estimates

Another way to assess the reliability of the PPP clock solution is the comparison
of the position estimates, because position and clock offset have to be solved si-
multaneously. A good indicator for the quality of the software is the repeatability
of the estimated positions, because the antennas of time and frequency transfer
receivers are fixed and no significant movement happens within 10 days.
To separate effects related to the vertical and the horizontal position, the position
estimates have to be known in the ellipsoidal coordinate system (A.3). Unfortu-
nately some packages output the results only in Cartesian coordinates. To compare
all software at the same level, the output of all packages was set to Cartesian coordi-
nates and the conversion was done afterwards. For visualization and simplification
the differences of the estimated values are calculated with respect to the initial
position (Table 4.1) on a daily basis.
In a first step (Figure 4.10) the distance between the estimated position vector
and the initial position vector is calculated:
δr = |~r(estimated) − ~r(initial)| . (4.1)
If the initial positions would be known to be true, it would be an estimation for the
absolute positioning error of the software. Unfortunately the 3 GTR50s are not
IGS stations, so that the initial positions cannot be taken from the IGS data, and
the positions might have changed over the years after they were estimated with
other geodetic methods. Furthermore, the dates of the estimation of the initial
coordinates are not traceable for the three laboratories and the related realization of
the coordinate system was unknown. Since in reality the true position is unknown,
the expression (4.1) can just serve as a relative error estimation, i.e. it can show if
a software gives repeatable results.
Figure 4.10.: Divergence of the estimated position with respect to the initial po-
sition. The output of all packages is a daily position estimate.
The NRCan, GIPSY and Bernese solution are in close agreement, but show
variations which are different for each site. It is noticeable that the agreement is
best at the location NICT, while it is worst at ROA. This indicates a relationship
between the quality of the position estimates and the instability of the external
frequency. The average offset of about 9 cm is probably due to the fact that the
initial positions are given in an older realization of the ITRF or in the WGS. Since
this was unknown, no transformation (A.5) was used.
The position estimates of the C4 software show a very unstable characteristic for
all the 3 stations (Figure 4.10). This affects the estimated clock offset in a similar
way. The Atomium solution diverges from the other solutions by 2 cm to 4 cm, but
it is very stable. If these divergence would be the same at each location, the time
comparison solution would not be affected since a consistent error cancels out.
As depicted in Figure 4.11, most of the position variations of the solutions show
up in the height estimate, while latitude and longitude differ only by a few centime-
ters from each other. The Atomium position estimates show a consistent deviation
in the height component for all three sites.
Figure 4.11.: Divergence of the estimated position with respect to the initial po-
sition in ellipsoidal coordinates. λ, ϕ, and h are geodetic latitude,
longitude, and height, respectively. Dots represent the results for
the station PTB, squares for NICT, triangles for ROA. 10−7 ◦ equals
approximately 1.1 cm. An arbitrary offset of 20 cm is applied to
the PTB height data for better visibility.
4.3.4. Troposphere Zenith Path Delay Estimates

The signal delay induced by the troposphere is of the order of more than 2 meters,
which equals more than 6 ns. It has to be modeled and estimated [4,5,32] and the
models depend on the coordinates of the station. It is one of the most important
error sources for time scale comparisons over long baselines. A wrong estimation
leads directly to an offset of the clock estimation. As mentioned in Chapter 2, the
common way to model the troposphere is to use one model for the dry part, an other
one for the wet part of the troposphere zenith path delay (zpd) [4]. The estimation
strategies are different in the software concepts (Table 4.3). The Kalman-Filter
based software estimates a new value as often as a clock offset is estimated. In the
Bernese solution a zpd is estimated every 30 minutes and in the Atomium and C4
solutions every 2 h. The estimated corrections are applied to the solution algorithm
by mapping functions.
Table 4.3.: Zenith path delay estimation in the different software concepts.
Software zpd output

Atomium every 2h wet part estimation,
PPP dry part extracted by Saastamoinen
model
NRCan PPP every 5 minutes
(static mode) modified Hopfield model
GIPSY 5.0 Every 5 minutes
Bernese GPS Every 30 minutes
Software 5.0 [71]
NICT C4 [63] every 2h wet part estimation
dry part extracted by Saastamoinen
model from initial position
The zpd estimates are in close agreement at all stations for all software concepts,
except C4, which shows significant differences from the other packages (Figure 4.12)
in the solutions at PTB and NICT. However, a correlation between the obvious
errors in the time and position estimates and the zpd estimates is not identifiable.
In contrast to the position estimates the zpd estimates show variations at the
same level for the stations NICT and ROA. The solution for PTB is very stable.
A relation between the instability of the reference frequency and the resulting zpd
is thus not possible. Probably, the characteristics of the zpd estimates depends on
the geographic location of the station. PTB is the station which is farthest in the
North.
Further research on the correlation between reference frequency, geographic lo-
cation, position estimates, and zpd estimates is necessary.
Since the parameters estimated by the NRCan-PPP software package, clock
offset, position, and zenith path delay are in good agreement with the estimates of
Figure 4.12.: Zenith path delay estimates.
the two renowned commercial software packages, it can be considered as a reliable

tool for PPP based remote time and frequency comparisons.
4.4. Code free Carrier-Phase Frequency

Comparisons
The main parameters that are estimated by a PPP software are highly correlated.
The position and troposphere estimates are closely joint by the equations (2.11) and
(2.13). An error in the position estimates leads to an error in the position estimates
and vice versa, as demonstrated above. Both errors can feed back on each other
and influence the troposphere zenith path delay estimates. Clock, troposphere, and
position estimates can not be separated. In case of positioning the error can be
reduced by a long averaging time, but in terms of time and frequency comparisons a
clock solution at each epoch is desired to enable the characterization of the involved
frequency standards also at short averaging time.
By definition, PPP is an AV process. Potential benefits of CV on short baselines
are excluded. Furthermore, both code and phase measurements are involved and
the noise of the code partially migrates to the combined solution, e.g. if the receiver
4.4. Code free Carrier-Phase Frequency Comparisons 59
looses the tracking of all satellites and the ambiguities are reset (Figure 2.8). A
jump in the phase domain generates an outlier in the frequency domain.
As demonstrated before, the results of the PPP software packages are influenced
by the solution strategy for the linearized observation equations. Thus, it would be
desirable to have a method which allows to compare remote frequency standards
without linearization. The goal of the work presented subsequently was to reach
the same instability level as with PPP, but with a minimum complexity.
4.4.1. Description of the Method

In order to accomplish this and to enable a detailed analysis of the correction
models the author has developed a simple method which allows comparing the
frequencies of signals connected to two receivers by using only the carrier-phase
measurements in AV as well as in CV mode. Fixed positions are used, as in code
based time transfer, and the equation (2.13) is solved for the receiver clock δtr ,
analog to the solution performed by time and frequency transfer receivers’ internal
software or the RINEX2CGGTTS software [31]. However, the satellites’ clocks
and ephemeris data are taken from IGS data and not from the navigation message.
Beyond this only corrections for the troposphere and the site displacement due to
the gravitational potential of the moon and the sun were used.
The approach described above is of course only possible if the phase measure-
ments are close to the code based pseudorange measurement within several tens
of nanoseconds, which means that the receiver provides a kind of pre-solution for
the ambiguities. Otherwise the Sagnac correction (2.19) and the elevation (A.8),
and thus the troposphere correction, cannot be solved with adequate accuracy. In
some time and frequency transfer receivers, e.g. GTR50, a pre-solution of the am-
biguities is performed. This can easily be verified by multiplying the carrier-phase
measurement of one satellite at one epoch on one of the two frequencies taken
from the RINEX data with the corresponding wavelength and checking that this
product is in quite good agreement with one of the three code-based pseudoranges
(C/A, P1, P3).
In a first step the time difference between a GTR50 receiver’s reference clock
and the IGS timescale was calculated for one satellite. The receiver was located at
PTB and the reference signal was UTC(PTB) derived from the primary frequency
standard CS2. The data were taken from the standard daily 30 s RINEX file. The
satellite’s clock and ephemeris data were extracted from IGS final products. The
fixed antenna position was calculated before with the NRCan-PPP software using
exactly the same input data in order to ensure the consistency of the reference
frame.
To get the position of the satellite at the desired epoch from the 15 min IGS
ephemeris file Lagrange polynomials (2.22) of the order 9. To account for the signal
propagation delay the time of signal emission was estimated by ts = tr −L3/c, where
tr is the time of signal reception, related to the timestamps of the RINEX data.
L3 is the quantity of equation (2.13) and c denotes the speed of light in vacuum.
The satellite clock offset with respect to the IGS timescale was not applied in this
step.
The Sagnac effect was taken into account by rotating the interpolated satellite
position ~rs,Sagnac around the z-axis in the ECEF coordinates:
 
cos(α) sin(α) 0
~rs,Sagnac = − sin(α) cos(α) 0 ~rs . (4.2)
0 0 1
The rotation angle is defined by α = −ωE L3/c, where ωE = 7.29215 · 10−5 /s is the
angular velocity of the Earth. If the trigonometric functions are approximated by
sin(α) ≈ α and cos(α) ≈ 1. , the analogy to (F.39) and (2.19) becomes apparent.
The satellite’s velocity for the eccentricity correction (2.18) was obtained via
equation (2.21) by shifting ts by 0.1 ms and interpolating the satellite position
from the IGS ephemeris file.
The elevation was calculated according to the equations (A.6) and (A.8) and the
troposphere was corrected by applying the model (2.10).
The result is depicted in Figure 4.13 with 5 min spacing due to the 5 min spacing
of the IGS clock file. In order to clarify the difference between the phase and the
code measurement the process described above was also done using the P3-code
(2.11) instead of L3.
Figure 4.13.: The difference between UTC(PTB) and the IGS timescale calcu-
lated from troposphere corrected P3-code and L3-phase for the
satellite transmitting PRN 2.
The chosen satellite was visible two times during one day. While the mean value
of the code measurement is the same before and after the gap, the offset of the phase
measurement differs by more than 10 ns. The pre-solution of the phase ambiguity
done by the receiver board inside the GTR50 is apparently imprecise, as seen in
Figure 4.14. Thus, the method described above cannot be used for calibrated time
comparisons using the carrier-phase measurement, but the relatively low noise level
of the phase results compared to that of the code can enable accurate frequency
comparisons.
To perform AV and CV frequency comparisons a frequency difference between
the reference of the receiver and the IGS timescale can be calculated for each
satellite in view by dividing the corrected L3 data by 300 s, according to equation
(2.2). This should only be done for continuous satellite tracks in order to avoid
outliers and inconsistencies. A gap, like that visible in Figure 4.13, has to be
excluded. It can be bridged by the other satellites visible during this period.
From Figure 4.14, where the code and phase results of all satellites in view are
depicted, it is obvious that the receiver can loose the tracking of particular satellites
for a short period, which can lead to a new phase ambiguity. This can also happen
between the processed 5 min epochs. Therefore it is necessary to check whether
measurement data are available for all the 9 epochs between two processed 5 min
epochs in the standard 30 s RINEX file or not. If data are missing, this epoch has
to be treated as a gap and excluded from further processing.
Figure 4.14.: UTC(PTB)-IGS time derived from troposphere corrected L3-phase

and P3-code measurements for all satellites in view at PTB at one
day.
The frequency difference between two remote receivers, and thus the frequency
difference of the frequency standards connected to these receivers, can now be cal-
culated by applying the process explained above to the L3-phase measurement data
of both receivers. For CV a frequency difference between the two local standards
is processed by subtracting the frequency differences available at the same epochs
in the results of both receivers for each satellite independently. The mean value
of all satellites is calculated afterwards. In terms of the AV method the frequency

average is at first independently calculated at any epoch for each receiver and
differences are made based on the averages.
4.4.2. Application to different Baselines

The first test of the code free frequency comparison method was performed on the
60 km baseline between PTB and the IQO of the LUH. The GTR50 at PTB was
directly connected to an active hydrogen maser (H5) and the data at the IQO were
collected by the mobile frequency reference (cf. Chapter 3) between the MJDs
54792 and 54796 at the end of the year 2008. In a first step the CV solution
was computed only with the relativistic eccentricity correction (2.18). Then the
solution was again computed using the troposphere model (2.10). The frequency
differences of the CV solutions were integrated and are depicted together with the
results of the NRCan-PPP software in Figure 4.15 on the left handed picture. For
better visibility the drift of the masers was removed by a linear fit. The PPP
solution was calculated without applying ocean loading coefficients and antenna
phase center corrections. To enable the identification of all noise types the MDEV
was calculated.
Figure 4.15.: Time differences for PPP and CV solutions and the related MDEV.
The blue plots represents the CV solution with relativistic eccen-
tricity correction, the red plots are the CV solution with eccentricity
and troposphere correction.
The CV solution without troposphere correction is slightly less stable than the
solution with troposphere correction, particularly for medium term averaging times
between 103 s and 105 s. The solution with troposphere correction shows tighter
agreement with the PPP solution and the instability is at the same level as that of
PPP. Small phase jumps are visible in both the CV and the PPP solutions. One
of these jumps at the end of MJD 54793 is not visible in the PPP. Probably some
of the jumps that are related to the GPS measurements are smoothed out by the
Kalman filter.
This analysis has demonstrated that also on this relatively short baselines (60 km)
the troposphere has an impact on frequency comparisons and must be corrected
by a model.
The next comparison using the newly developed software was performed on the
2182 km baseline between ROA and PTB within the period MJD 55231 to 55239
in 2009. The frequency standards and the positions are the same as in Table 4.1.
The frequency instability of the results, expressed as the modified Allan deviation
(MDEV), is depicted in Figure 4.16. The instability of PTB’s primary frequency
standard CS2, locally measured with a high resolution phase comparator with
respect to a H-maser at the same period and the instability of the PPP solution,
obtained with the NRCan-PPP, are also depicted. The instability of the Cs 5071
clock is taken from the manufacturers specification. The PPP solution was again
calculated without applying ocean loading coefficients and antenna phase center
corrections. The code-free frequency comparisons are realized in this case as AV
and CV solutions, applying the eccentricity correction (2.18) and the troposphere
model (2.10).
Figure 4.16.: Frequency instability of UTC(ROA)-UTC(PTB) expressed as the

MDEV for PPP, code free AV, and CV comparisons over the 2182
km baseline PTB-ROA. The MDEV of CS2 is calculated from a
local phase comparator measurement at PTB.
The three frequency instabilities of the remote timescale comparison UTC(ROA)-

UTC(PTB) are similar. The remote comparisons are 2.5 times less stable than the
primary frequency standard CS2, indicating that the remote comparison is limited
by the performance of the commercial caesium standard at ROA. If commercial
caesium clocks are involved, the use of the simple models is almost sufficient on
European baselines. In this particular case the performance of the Cs 5071 clock
is better than specified by the manufacturer.
To figure out the limitations of the code free frequency comparison method and
the correction models used within, it was applied to the intercontinental baseline
between NICT and PTB at the same period as used for the ROA-PTB comparison.
Since UTC(NICT) is derived from an active hydrogen maser (Table 4.1), the results
were combined with the data of the local frequency comparator measurement CS2-
active maser at PTB, such a way that CS2 cancels out and the final solution
represents the comparison of two masers.
In fact, the PPP results reflects the instability of a remote maser comparison (cf.
Figure 3.10), but the code free CV and AV solution seem both to be limited by the
simplicity of the correction models. The AV results are slightly better than the CV
results. This is due to the fact that on this intercontinental baseline mostly one or
two satellites are commonly tracked by both receivers at low elevation angles. Thus,
the AV and CV calculation were done a second time by using an improved model
for the troposphere, taken from the GTR50’s internal code processing software:
−1
sin(el) + 0.00143 cos(el)
δtro = T0 · ,
0.0455 cos(el) + sin(el) (4.3)
−9 −12 s 2 s
T0 = 8.22 · 10 · s − 1.083 · 10 · h · + 7.467 · h · 2 .
m m
The correction is given in units of seconds and h is the height of the antenna above
the Earth ellipsoid (A.3). This model directly includes the mapping function.
The instabilities in Figure 4.17 calculated from solutions obtained with the model
(2.10) are labeled with ”T1”, while the instabilities calculated from solutions ob-
tained with the improved model (4.3) are labeled with ”T2”.
The improvements due to the new troposphere model are marginal. The MDEV
plot indicates a daily effect for the CV and AV solutions. This is due to the
fact that, in contrast to PPP, no correction model for the solid Earth tides was
applied to the code-free solutions. These effect has an opposite sign for PTB and
NICT, because the longitude difference between NICT and PTB is about 125◦ .
Thus, the model for the displacement of the solid Earth due the forces of Sun and
Moon according to equation (2.20) was applied by adding the vector displacement
to the Cartesian coordinates of the antenna before solving equation (2.13). The
estimation of the positions of Moon and Sun in the ECEF system is described
in Appendix G. The model (2.20) includes a constant part which depends only
on the position of the antenna. To be consistent with the ITRF and in case of
calibrated time transfer with PPP, this part has to be removed afterwards, but in
the framework of code-free frequency comparisons these small constant offset do
not have an effect.
Figure 4.17.: Frequency instability of a maser comparison expressed as the

MDEV over the intercontinental 8333 km baseline between NICT
and PTB. ”T1” and ”T2” indicate the troposphere models (2.10)
and (4.3), respectively.
In a first step only the effect due to the Moon was corrected. Then, in a second
step both corrections for Sun and Moon were applied. The phase differences,
obtained by integrating the frequency differences, are exemplarily depicted for one
day in Figure 4.18 together with the PPP solution (black plot) and the AV solution
without solid Earth tide correction (blue plot). The effect of the solid Earth tides
is clearly visible. In the solution with only the part caused by the Moon corrected
(red plot), only parts of the difference to the PPP solution are removed, while the
solution with the complete correction for Sun and Moon (green plot) is in very
good agreement with the PPP solution.
The exercise outlined in this section has clearly demonstrated the importance of
accurate error modeling. Future improvements to this methods could be to solve
for the time of signal emission ts in an iterative process and to apply a correction
for the difference between antenna phase center and center of mass of the satellites.
Furthermore, the code free methods could be made applicable to receivers which
are not providing a pre-solution of the phase ambiguity by aligning the phase
measurement to the code by an additional preprocessing.
On relatively short baselines the modeling of the troposphere and the relativistic
eccentricity correction is sufficient. The same is also valid on European baselines
when stations are equipped with commercial caesium clocks. On very long inter-
continental baselines the modeling of the solid Earth tides cannot be neglected if
frequency comparisons at the instability level of active hydrogen masers are desired.
The troposphere can in all cases be corrected with the help of the relatively
simple model (2.10). Improvements due to the model (4.3) are marginal. This is
surprising, because in the literature [4, 5, 32, 33] the importance of sophisticated
Figure 4.18.: One day maser comparison between NICT and PTB. The black
plot was calculated with PPP, the blue one with code-free AV, rel-
ativistic eccentricity correction and troposphere model (4.3). The
red plot represents the AV solution with the correction for the dis-
placement of the Moon applied, the green plot is AV with both
corrections for Moon and Sun applied
models is emphasized. The correlation between troposphere estimates and other

parameters is analyzed in the reference [72].
Unfortunately, an analytical solution of the phase ambiguity problem has not
yet been developed, but it has to be pointed out that the code free carrier-phase
method can be also used for calibrated time transfer. The time differences obtained
by integrating the frequency differences can be adjusted by comparing them to the
P3 code time differences measured at the same period. The double difference has
then to be subtracted from the code-free solution. The initial phase of the code-free
solution could also be determined by an other independent method, like PPP or
TWSTFT.
5. Hardware Experiments
In the previous Chapter the importance of the proper evaluation of the measure-
ment data and the accurate modeling of the error sources has been demonstrated.
However, the accuracy of GPS time and frequency comparison results particularly
depends on the performance of the receivers involved.
In this Chapter the performance of receivers used for time and frequency com-
parisons is analyzed with respect to the short, medium, and long term instability.
A comparison to state-of-the-art geodetic receivers that can only be used for fre-
quency comparisons, since they are not equipped with an 1 PPS input, show that
there is still room for improvements.
The common-clock difference (CCD) data of four receivers provided by the LUH
QUEST institutes involved (IfE, IQO) were analyzed. These receivers were used for
the first ”long baseline common-clock experiment” between LUH and PTB forming
double differences with measurement data taken from an optical telecommunication
fiber link between PTB and LUH at the same time. The term ”long baseline” is
used here, because the baseline is long from the point of view of a common-clock
experiment.
By comparing the different receivers in the quasi zero baseline setup at PTB, it
was decided which types of receivers were finally used for that experiment involving
the fiber.
Before that, the temperature stability of the receivers chosen for the long baseline
common-clock experiment is analyzed. It shows up that the internal delay changes
inside these receivers are highly correlated to the environmental temperature.
Based on the experiences with the QUEST receivers two new design concepts
for a highly accurate frequency receiver are proposed.
Finally, the results of the first common-clock long baseline experiment are out-
lined. Such experiments could be used in the future for detailed studies of the
correction models, because local clocks and other influences can be separated.
5.1. Performance of Receivers

In this Section, receivers which are typically used in time laboratories are analyzed
with respect to their stability on different averaging times. This is done by quasi
short baseline data evaluation at PTB.
67
68 5. Hardware Experiments
Receivers which were provided by the IfE and the IQO were installed and com-
pared with each other in order to figure out which type of receivers is most suitable
for highly accurate and stable frequency comparisons (cf. reference [73]). The tem-
perature stability of the IQO receivers is analyzed in detail.
5.1.1. Instability of Time and Frequency Receivers

In timing labs, such as PTB, receivers equipped with TICs as well as modified
geodetic receivers are operated (cf. Section 2.3.4). At PTB two Ashtech Z-12 T
receivers are continuously operated and providing data to the BIPM UTC compu-
tation as well as to the IGS network. These installations and the operation must
not be compromised by any experiment. The Ashtech receivers are based on the
design concept given in Figure 2.5. Additionally four receivers based on the con-
cept outlined in Figure 2.4 are installed. Among these are to GTR50s whose data
are analyzed.
In Figure 5.1 the results of a quasi zero baseline 5 day comparison of the Ashtech
Z-12 T receivers and of a zero baseline comparison of two GTR50 time and fre-
quency transfer receivers within the same period are shown. The GTR50s were
connected to the same antenna using a splitter. The RINEX data of receivers were
evaluated with the NRCan-PPP software using IGS products and the backward
smoothing option.
Figure 5.1.: Quasi zero baseline results of two Ashtech Z-12T and two GTR50
receivers. The frequency instability is expressed as the ADEV.
The performance of the GTR50 receivers is the same as in a short baseline

CCD experiment (cf. Figure 4.2 and Figure 4.3). The performance of this type of
receiver is limited by the performance of the internal time interval counter. The
phase plot as well as the ADEV show the domination white phase noise, which is
the typical limitation of time interval counters. In contrast, the Ashtech receivers
5.1. Performance of Receivers 69
are limited by another type of noise, namely by random walk phase noise (white
frequency noise). This is probably due to the locking of the receivers’ internal
crystal oscillator to the external reference frequency. Up to about 104 s their
stability is better than that of the GTR50.
In the framework of UTC computation and time scale comparisons especially
the long term performance of the receivers is of interest. Thus, the Ashtech re-
ceivers were compared with each other for 100 days in 2008 using the NRCan-PPP
software in the forward processing mode and without ambiguity resetting. The
question is how the random walk phase noise degrades the phase within time.
The daily averages of this comparison and the corresponding TDEV are shown
in Figure 5.2. Since these receivers are mainly used for the comparison of the
timescale UTC(PTB) to the timescales of other institutes in the framework of TAI
computation, the usage of the TDEV instead of the ADEV is preferable.
Figure 5.2.: Longterm comparison of two Ashtech Z-12 T receivers. The left plot
shows the daily averages, the right plot shows the corresponding
TDEV. The corresponding frequency instability at 105 s is 2 · 10−11 .
The variations show a maximum peak to peak difference of about 1.5 ns within
the 100 days. However, time comparisons with these receivers are possible with
a phase instability of less than 200 ps at an averaging time of 10 days. Unfortu-
nately no such long batch of continuous data exist for a comparison of two GTR50
receivers, because only one of the two PTB receivers of this type is continuously op-
erated and connected to UTC(PTB). The second one was periodically disconnected
from UTC(PTB) and used for other purposes. A comparison of the fixed GTR50
to one of the two Ashtech receivers shows the same instability as the comparison
of the two Ashtechs and does not need to be shown here.
In the framework of frequency comparisons mainly the short and medium term
stability of the receivers is of interest. Figure 5.3 depicts the short term comparison
of two GTR50s and two Javad Legacy receivers. The Javad Legacy receivers were
provided by the IfE and operated at PTB for a few hours in a zero baseline setup.
The output RINEX data are given in 1 s intervals. To compare both types of
receivers at the same level, also 1 s RINEX data were produced by the GTR50s’
internal processing software, in contrast to the 30 s files that are usually generated
in time and frequency laboratories. The data were processed with the NRCan
software’s backward mode and the satellite clocks were interpolated from the 5 min
IGS clock products.
Figure 5.3.: Short term zero baseline comparison of two GTR50s and two Javad
Legacy receivers. The frequency instability is expressed as the
ADEV.
Due to the interpolation of the satellite clocks additional noise is generated.

However, the GTR50s is about one order of magnitude less stable in short term
than the Javad Legacy receiver, since it is limited by the performance of its internal
time interval counter. In very short term the instability of the Javad legacy receiver
is limited by white noise, as that of the GTR50. Then the random walk noise is
dominating.
This comparison leads to the assumption that it should in principle be possible
to construct a receiver which can reach the 10−16 level at an averaging time of one
day. For this, a solution to overcome the random walk noise is necessary.
5.1.2. QUEST Experimental Receivers

In 2009 four receivers provided by the QUEST institutes IfE and IQO were oper-
ated at PTB in a zero baseline setup within the framework of an experiment with
using an optical fiber connecting PTB and IQO, which is explained later. One of
the IQO receivers is of the type Javad Legacy, as the receivers used for the analysis
given in Figure 5.3. The IQO receivers are based on the same GPS board that
is used inside the GTR50 receiver, a Javad receiver circuit, but without tempera-
ture stabilization. Also no time interval counter is used. Instead, the board can
be locked to an external reference frequency and the option of synchronizing the
5.1. Performance of Receivers 71
internal timescale to an external 1 PPS is available. This option was not used in
this analysis.
All receivers were recording the measurement data in 1 s RINEX files and the
phase results were again obtained with the NRCan-PPP software in backward
smoothing mode with IGS satellite clock interpolation. The results of the CCD
comparisons are depicted in Figure 5.4. The IfE receivers are labeled ”IFE2” (Javad
Legacy), ”IFE33”, and ”IFE82”, manufactured by Leica, respectively. The IQO
Javad receiver is named ”IQO2”. Four different CCD combinations are depicted
the Figure 5.4.
Figure 5.4.: Zero baseline CCDs of the QUEST receivers.
Both comparisons involving ”IFE2” are affected by a phase drift. This drift is
not present in the comparison of the two receivers ”IQO2” and ”IFE33”. Thus
this drift can be related to the ”IFE2” receiver, which was thus excluded from
additional experiments with the optical fiber. Unfortunately the ”IFE82” receiver
did not operate continuously during the complete measurement period. It was also
excluded from further examinations. Only the ”IQO2” and the ”IFE33” receiver
remain. The intercomparison of these two receivers indicates measurement inter-
ruptions, because the typical phase discontinuities due to a re-estimation of the
ambiguities of all satellites in view are visible. An analysis of the RINEX data has
shown that the interruptions are clearly related to the ”IFE33” receiver.
Since the four receivers were operated within an experiment with an optical fiber
between IQO and PTB, at the IQO four other receivers of the same types were
operated. Unfortunately only two receivers were correctly recording measurement
data. One of them was the second IQO receiver ”IQO1”. A comparison with the
other IFE receiver showed the same performance as the ”IQO2”-”IFE33” compari-
son. Again, interruptions were caused by the IfE receivers. This analysis is similar
to that depicted in Figure 5.4 and is not shown here.
In the framework of precise positioning one can probably accept discontinuities

and phase drifts, because these effects can be averaged out in the calculation of a
static position, but in the perspective of time and frequency comparisons continu-
ous and stable measurements are required. In terms of time and frequency transfer
receivers providing a continuous and stable measurement over long periods are
mandatory.
5.1.3. Temperature Sensitivity

As mentioned above the IQO receivers are based on the same GPS board as the
GTR50 receiver. Inside the GTR50 this board is temperature stabilized. This shall
avoid that the GPS measurements are highly affected by temperature variations
due to changes of the internal delays.
To figure out the sensitivity of this type of receivers to the temperature one of
the QUEST receivers was put in a metal box with regulated heating inside. The
box was heated up to a maximum temperature of 68◦ C and RINEX data were
recorded at the same time. Figure 5.5 shows the temperature inside the box as
well as the time differences between the receiver outside the box in the stabilized
measurement room and the receiver inside the temperature box.
Figure 5.5.: Time difference between two Javad receivers (blue plot). One of the
receivers is located inside a temperature regulated box. The red plot
depicts the temperature inside this box.
The temperature and the electrical delays of the GPS receiver are highly corre-
lated. A temperature sensitivity coefficient of about 20 ps/◦ C was estimated. The
operation of receivers used for time and frequency transfer under stable environ-
mental conditions is essential for highly precise comparison results.
5.2. Suggestions for Improvements in Frequency Transfer 73
5.2. Suggestions for Improvements in Frequency

Transfer
The performance of GPS time and frequency transfer is not only affected by tem-
perature variations influencing the electrical delays of the receivers, but also by
temperature variations changing the electrical lengths of the antenna cable and
the delays of the antenna itself, which are installed outdoor. In order to overcome
these effects and the noise due to locking the internal oscillator of a GPS receiver
to an external frequency the author has developed two ideas for improving the
frequency stability. These two new approaches are not yet realized and just some
preliminary experiments were performed.
5.2.1. Phase Comparator Measurements

To overcome the noise of the oscillator locking it would be desirable to construct a
GPS receiver that does not have an internal crystal oscillator, but instead directly
uses the reference frequency. Experiments with the GTR50’s receiver board were
performed but with no suitable result.
A second approach could be not to lock the internal oscillator to an external
frequency, but to measure the frequency of the internal free running crystal oscilla-
tor with a high resolution phase comparator with respect to a reference frequency,
similar to the receivers that measure 1 PPS signals with a time interval counter,
but in the frequency domain. If two of such receivers are operated at remote sites,
the GPS data can be combined with the frequency comparator data and exchanged
for highly precise frequency comparisons.
A first experiment was performed at PTB with the two IQO receivers. Here both
receivers were connected to the same frequency comparator such that no external
reference frequency was needed (Figure 5.6).
Figure 5.6.: Measurement of the output frequencies of the two IQO receivers. Rx
stands for receiver, DMTD depicts the frequency comparator.
Unfortunately the output frequency of the receivers was 20 MHz, in contrast to

the 10 MHz that are usually used in time laboratories. Thus, it was necessary to
convert the 20 MHz down to 5 or 10 MHz. Since the involved components, mainly
additional amplifiers, were not constructed for signal levels needed by the high
resolution frequency comparator at PTB, no suitable results were possible.
5.2.2. Stabilization of Cables

To cancel temperature effects on the antenna cable and the antenna a GPS receiver
could directly be integrated together with the antenna in a temperature stabilized
housing. In this case it is necessary to provide the reference frequency to the an-
tenna site. Inspired by the stabilization of optical frequencies in terms of frequency
comparisons with an optical fiber by a reflected signal, which is used to compenate
changes in the length of the fiber, the idea of stabilizing the 10 MHz reference
frequency came up. Figure 5.7 shows the proposed experimental setup.
Figure 5.7.: Proposed set-up for high precision GPS frequency comparisons. BP
denotes band-pass filter, DMTD is a high resolution frequency com-
parator. The elements shown in the box on the right handed side are
located outside together with the antenna. The elements shown on
the left handed side are located commonly inside a stabilized indoor
measurement room.
The parts of the schematic located inside the box are the outside installations.
It is connected to the other parts of the setup, which are located inside a stabilized
measurement room, by one cable transmitting frequencies in both directions. Ad-
ditional cables for electrical power and data transfer to the roof are also needed,
but are not depicted in the Figure. The antenna is also part of the stabilized
environment.
5.2. Suggestions for Improvements in Frequency Transfer 75
A 5 MHz signal is doubled to 10 MHz and connected to one input of a high

resolution frequency comparator. It is additionally send to the outside installations
and connected to the receiver which is integrated together with its antenna in a
temperature stabilized box. Inside this box the signal is doubled to 10 MHz and
send back to the measurement room. Bandpasses isolate the different signals from
each other. The reflected signal is connected to the second channel of the phase
comparator. The measurements of the phase comparator have to be applied to the
GPS measurement as corrections at each of the remote sites (if such a setup is used
at two remote sites).
First tests with common time laboratory equipment has shown that the reflection
method would work and that delay variations due to temperature changes on the
cable can be canceled out for the most part (Figure 5.8).
Figure 5.8.: Test of the proposed set-up without receiver and without electronic
components outside. Instead, the signal was connected to a two
channel high precision frequency comparator before and after the
reflection (see text). Phase data as the integral quantity are shown.
The red plot in the left handed picture shows the measurement of
the signal fed to the cable with respect to the input signal. The blue
plot depicts the double difference between these measurement and
the reflected signal, also measured with respect to the input signal.
The right handed plot depicts the related Allan deviations.
For the test the ”outside” installations were located inside a measurement room.
The connecting cable was a 50 m cable loop on the antenna roof of PTB. This
cable was not intended for outdoor use and shows significant delay variations due
to daily temperature variations. Instead of a receiver a second channel of a fre-
quency comparator was connected to the 5 MHz of the ”outdoor” installation and
measured with respect to the 5 MHz input signal. The double difference of both
frequency measurements shows that the delay variations of the cable be in princi-
ple compensated. The noise floor visible after averaging times of 4 · 103 is related
to the frequency doubling amplifiers used for this preliminary experiment. It is

expected that with the usage of modern low-noise amplifiers the instability is well
below 1016 for averaging times of 104 s.
5.3. The first Long Baseline Common-Clock

Experiment
In 2009 an optical fiber was operated between PTB and the IQO of the LUH.
The idea behind was to compare the experimental optical magnesium frequency
standard developed at the IQO to the optical frequency standards and to the
primary clocks at PTB with very high stability and accuracy [16]. A common
optical fiber originally intended for telecommunication signals was used for this
purpose. At PTB the 194 THz frequency of a laser was fed into the fiber and a
reflected signal was used to compensate changes of the length of the optical fiber.
By using erbium fiber laser frequency combs [62] at both sites the optical signal
can be compared to the frequencies of the experimental optical standards as well
as to microwave standards like the active masers at PTB and the passive maser
H7 at the IQO. The fiber link can be used as a reference to characterize the GPS
frequency link, because the optical link performance exceeds that of GPS on the
relatively short 60 km baseline.
In the double differences between GPS and fiber measurement data the con-
tributions of the masers cancel out and a common-clock characterization on the
60 km baseline is enabled. Similar experiments and measurements are planned for
the future, e.g. between PTB and the Max-Planck Institute of Quantum Optics
(MPQ) in Garching. They can be used for a detailed clock free analysis of the
errors in GPS frequency transfer and help to improve the modeling of systematic
GPS effects.
The measurement data of this experiment were evaluated by integrating the
frequency data of the 1 s fiber measurement and then comparing it to the GPS data.
The frequency reference at PTB was the active maser H6 (Figure 3.1). The 1 s
RINEX data of the two IQO Javad experimental receivers were used and analyzed
with the NRCan-PPP software in the backward smoothing mode. In order to get
GPS results which are also spaced in 1 s intervals the clock interpolation mode of
the software package was used. The results are depicted in Figure 5.9. The initial
phase was arbitrarily chosen to be zero.
In the left picture it is clearly visible that the GPS link is noisier than the fiber
frequency transfer. This could be partially induced by the analysis process due to
the interpolation of the satellite clocks from the 30 s IGS clock data. Thus, the
noise visible in the right picture can be mainly considered to be originating from
5.3. The first Long Baseline Common-Clock Experiment 77
Figure 5.9.: Results of the GPS - optical fiber experiment between PTB and LUH
The left picture shows the GPS and the fiber link separately and the
right picture depicts the double difference (see text).
the GPS link. The variations reach a maximum of 200 ps peak to peak and a
standard deviation of 65 ps during the shown measurement interval.
Unfortunately, measurement data of the optical fiber are only available for a
short period since the optical connection was in a very early experimental state.
The laser system and the other optical components were very sensitive to any
kind of mechanical vibrations. The measurement data were taken over a weekend
when these vibrations were at a minimum since only few persons were inside the
laboratories. With the given data set is was possible to calculate the frequency
instability only up to 104 s, as depicted in the MDEV plot (Figure 5.10).
For short averaging times the fiber connection is about half an order of magnitude
down to 0.8 orders of magnitude more stable than the GPS link. However, for very
short averaging times it was limited by a counter at the IQO. In future experiments
this can be avoided due to a more advanced experimental setup. For averaging
times exceeding 104 s the limitation is given by the performance of the passive
maser H7 (cf. Figure 3.4).
The instability of the stand-alone GPS link is at the same level as that of the
clock free double differences for the short averaging times. Due to the lack of more
fiber data a long term characterization was not possible in this framework. To avoid
noise due to the interpolation of the satellite clocks, the GPS data were processed
again with IGS 5 min clock products and the clock interpolation disabled in the
NRCan-PPP software package, as usual. The double differences are calculated by
comparing the GPS data to the interpolated fiber data every 300 s. The double
differences and the instabilities of the stand-alone GPS link, the fiber link, and the
double differences are depicted in Figure 5.11.
Figure 5.10.: Frequency instability of the GPS and optical fiber frequency com-
parison, expressed as the MDEV. In order to enable distinguish-
ing between all noise types, the MDEV instead of the ADEV was
calculated.
Figure 5.11.: Comparison between GPS and the optical fiber (left plot) and the
MDEV of the GPS link, the fiber link and the double differences,
respectively, evaluated on a 300 s basis.
The peak to peak divergence of the double differences is now at about 200 ps. In
very short term, the GPS receivers seem to be able to measure with a precision of a
few ps, but sometimes the solution is affected by instabilities. The MDEV indicates
no significant improvement compared to the 1 s evaluation for short averaging times
below 103 s. Then the canceling of the phase drift of the masers leads to a minor
improvement, but this can also be achieved by just removing the maser drift with
a linear interpolation.
5.3. The first Long Baseline Common-Clock Experiment 79
The variations in the GPS result could be due to temperature instabilities, lim-
itations of the error corrections, or limitations of the measurement process itself.
Since the PPP comparison is an AV process, also noise of the IGS timescale could
have an effect, because also on the relatively short 60 km baselines the two re-
ceivers were not always tracking the same satellites. Furthermore, especially the
signal of satellites with low elevation angles could be affected by multipath induced
by surrounding trees at PTB and surrounding buildings at the LUH, respectively.
To minimize the multipath effects the data were processed again, but the data of
all satellite tracks with a satellite elevation below 30◦ , and in a second step below
45◦ , were rejected (Figure 5.12).
Figure 5.12.: Comparison of GPS and optical fiber with all satellites with ele-
vation angles below 30◦ removed from the solution (red plot) and
with all satellites with elevation angles below 45◦ removed from the
elevation.
An improvement in the results is not visible. The double differences between

the red and the blue plot in Figure 5.12 and the plot in Figure 5.11 are at the
0.05 ns level. The variations are probably partly related to the external temper-
ature variations. The author hopes that with the suggestions described in the
previous Section such variations can be minimized and frequency comparisons at
the 10−16 level at an averaging time of one day become possible. Before future
experiments a detailed examination of the involved receivers and an analysis of the
source of the instabilities should be conducted.
6. Relative Calibration of TAI
links
6.1. Motivation
In the preceding chapters, all analysis has been done from the perspective of fre-
quency comparisons and phase offsets have been arbitrarily chosen. However, the
comparison of time scales, as done by the BIPM, and the evaluation of the corre-
sponding links requires the accurate knowledge of the internal delays of the time
transfer equipment involved or the delay of the complete link.
Up to now, operational links have been calibrated either by cost and labor in-
tensive TWSTFT calibration campaigns, reaching the uncertainty of 1 ns [56] at
the moment of the campaign, or by more easy to handle circulations of traveling
GPS receivers [57], as conducted by the BIPM. The uncertainty of the latter has
been estimated to be 5 ns for the operational link, considered valid for about 2
years until the next calibration campaign is conducted. This is based on BIPM’s
long-term experience on the performance of installed equipment during the past
decades.
Recently, with the use of a new receiver (type GTR50), an uncertainty of less
than 2 ns for a relative link calibration between ROA and PTB performed in 2008
has been reported [17]. The validity of this small uncertainty at the moment of the
campaign is supported by previous studies of delay changes in GPS receivers of this
type including switching on and off procedures [74]. In 2009 the author on behalf
of PTB made a calibration along these lines between the Swiss Federal Office of
Metrology (METAS) and PTB using a traveling receiver (TR) of the same type. At
METAS the work was supported by the local staff, André Stefanov, Laurent-Guy
Bernier, and Christian Schlunegger. The results are reported in the reference [75].
In 2010 PTB has conducted such a campaign involving USNO. For this purpose
a GTR50 time and frequency transfer receiver and a SR620 TIC together with a
monitor and a keyboard were integrated in a small transportable rack. The idea
behind this was to simplify operations and to reduce the systematic uncertainty:
In relative calibration campaigns the offset δt0 between the 1 PPS signal represent-
ing the local realization of UTC that is connected to the TR and the local UTC
reference point has to be measured in each laboratory. Usually TICs are used for
that purpose, but the internal delays between their two input channels vary from
80
6.2. Description of the Calibration Procedure 81
unit to unit. If a TIC is shipped together with the receiver and δt0 is measured
using exactly this device at all participating laboratories, the overall uncertainty
would be reduced. The software which automatically controls the operation of the
GPS receiver as well as of the TIC was developed by the author.
The USNO-PTB calibration campaign was organized by the author. At USNO
the installation and operation of the calibration set-up was performed by Blair
Fonville, Elizabeth Goldberg, and Stephen Mitchell, and supervised by Demetrios
Matsakis.
It was decided to integrate only the contributions to the uncertainty budget,
which are directly related to the calibration itself and not to consider the long
term performance of the links. Furthermore an improved data handling, making
use of the TDEV statistics was applied. Thus the overall uncertainty was reduced
to less than 1 ns for some links.
In this chapter the differential GPS link calibration procedure is briefly explained.
The uncertainty estimation is discussed and an introduction to PTB’s new calibra-
tion set-up is given. The results of the 2009 METAS-PTB calibration campaign and
of the 2010 USNO-PTB calibration campaign are presented. The METAS-PTB
data have been re-evaluated using the improved data processing and the newer
method of uncertainty estimation as for the USNO-PTB data.
The two campaigns between METAS and PTB and between USNO and PTB are
of special interest, because the timescales of METAS and USNO were additionally
connected to the PTB timescale by calibrated TWSTFT links which could thus
serve as a verification for the GPS based method.
6.2. Description of the Calibration Procedure

In a relative GPS link calibration campaign between two timing laboratories LAB1
and LAB2 a traveling receiver (TR) is first operated at LAB1, together with all
the fixed GPS receivers in a common-clock, very short baseline setup for several
days. Then it is shipped to LAB2 together with its antenna and antenna cable and
operated there again for several days in very short baseline setup together with the
equipment in LAB2. In order to ensure that the internal delays have not changed
during the travel the measurement at LAB1 is repeated after the TR was shipped
back to LAB1.
As depicted in Figure 6.1, the devices are connected to the local UTC realization
by 1 PPS signals with offsets relative to the local UTC reference point. Figure 6.1
shows the setup at two labs operating GPS as well as TWSTFT equipment. FR
denotes all fixed receivers at LAB1 and LAB2, respectively.
Normally, for the fixed equipment the 1 PPS signal delays with respect to the
local reference points are known with a low uncertainty, as well as equipment
internal delays and antenna cables. However, in terms of a relative calibration
82 6. Relative Calibration of TAI links
Figure 6.1.: Schematic of the link calibration setup at two labs equipped with
GPS and TWSTFT. δt0 is the delay of the local UTC signal at each
lab with respect to the TR’s reference point. It is different at each
lab and has to be determined by a measurement.
these delays do not have to be taken into account, if the goal is to calibrate the
entire link, including the complete chain of signal distribution, cables, and antennas
in both laboratories. The only important value is the offset δt0 of the temporary
connection between the TR and the local reference point in both labs. It should
be measured as accurately as possible.
To ensure that short term variations are averaged out in the common-clock
difference (CCD) between the TR and the FRs, the TR should be operated at
least about one week in each laboratory. The CCD values then represent the
difference of the sums of the delays in the chains comprising antenna, antenna
cable, receiver, and connection to the reference point. By differencing the CCD
results of both labs, the contributions of the TR cancel out and the calibration
value for the links between FRs (one pair or multiple combinations) is obtained.
This statement is valid to the extent that the properties of the TR are unchanged
during the campaign. In mathematical terms it reads as follows:
hTR@LAB1-FR(LAB1)i − hTR@LAB1-FR(LAB1)i = C1 − C2 = CGPS , (6.1)
where h...i stands for the mean value of measurements over a certain period.
It is noted that the sign of the calibration value CGPS is arbitrarily defined by
(6.1). With this convention the operational link between the two UTC realizations
UTC(LAB2) and UTC(LAB1) has to be corrected according to
[UTC(LAB2) − UTC(LAB1)]GPS = FR(LAB2) − FR(LAB1) − CGPS . (6.2)
The mean values of both CCDs at LAB1 before and after the trip to LAB2 are
used to calculate the value C1 in order to detect small changes of the TR’s internal
delays induced by the travel. The difference between the two CCDs obtained in
LAB1 is part of the uncertainty budget which is explained later.
If the two labs maintain a TWSTFT link, as shown in Figure 6.1, a calibration
value for the TWSTFT link can be calculated by comparing it to the calibrated
GPS link:
h[UTC(LAB2) − UTC(LAB1)]GPS
(6.3)
− TW(LAB2) − TW(LAB1)i = CTWSTFT .
The correction to the TWSTFT link is thus given by
[UTC(LAB2) − UTC(LAB1)]TWSTFT
(6.4)
= TW(LAB2) − TW(LAB1) + CTWSTFT .
In general, the TR can be shipped to more than one remote laboratory before
returning to LAB1, as done by the BIPM for the relative characterization of time
equipment (see e.g. [57]). Then also the links between the remote laboratories are
calibrated. However, in this work the results of two campaigns with one remote
laboratory involved are shown, and PTB acts as LAB1.
6.2.1. CV versus AV
As explained before in Chapter 2, a GPS C/A or P3 code link on an European
baseline, such as between METAS and PTB or ROA and PTB, can either be
evaluated in common-view (CV) or in all-in-view (AV) mode. AV is superior to
CV on intercontinental baselines, such as between USNO and PTB, and is currently
(October 2010) used for all the GPS-based TAI links by the BIPM [76]. One could
intuitively think that a CV link should be calibrated using CV CCD computation
and an AV link using AV CDD computation. In the following it is demonstrated
that CV CCD computation is always preferable in order to calculate the calibration
value, independent of the mode with which the link is operated later.
Two receivers of the same type (GTR50) and two receivers of different types
(GTR50 and TTS-3) were compared in CCD experiments. As depicted in Figure
6.2, with identical receivers the result of CV and AV are equal (the double difference
between AV and CV is zero) in 57.23 % of the epochs, while with different receivers
this is just the case in 3.36 % of all measurement epochs.
For the interpretation of this result the measurement performed by one single
receiver R with respect to one satellite at a certain epoch is expressed as
Rn (k) = UTC(L) − GPST + ∆(k) + εn (k) + δn , (6.5)
with the satellite number n, the receiver number k=1 or 2, the local UTC realization
UTC(L), the GPS time GPST, a receiver R specific offset ∆(k), a noise contribution
εn (k) and a satellite specific offset δn from the ideal GPS time. The sets of all
Figure 6.2.: CCD double differences CV-AV of two identical receivers (2 GTR50)
and two receivers of different types (GTR50 and TTS-3), each
equipped with its own antenna, but connected to the same clock.
satellite observations performed by two receivers in a very short baseline CCD

setup at a certain epoch τ can be defined as
S(1)τ = {R1 (1), R2 (1), ..., RN (1), RN+1 (1), ..., RP (1)} ,
(6.6)
S(2)τ = {R1 (2), R2 (2), ..., RN (2), RN+1 (2), ..., RQ (2)} .
The satellites are consecutively numbered and P and Q are the numbers of satellites
tracked by receiver R(1) and R(2), respectively. The observations numbered from
1 to N are done by both receivers simultaneously. Thus Pe = P − N is the number
e
of satellites tracked only by R(1), while Q = Q − N is the number of satellites
tracked only by R(2). A CV CCD comparison (CCC) can be written as
N N
1X 1X
CCC = [Rn (1) − Rn (2)] = ∆(1) − ∆(2) + [εn (1) − εn (2)]
N n=1 N n=1 (6.7)
= CL + ECV .
UTC(L) and the ideal GPS time GPST cancel out. An AV CCD (ACC) reads
P Q e
N+P N+Q e
1X 1X 1 X 1 X
ACC = Rn (1) − Rn (2) = Rn (1) − Rn (2) . (6.8)
P n=1 Q n=1 e
N+P N + e
Q
n=1 n=1
After using 1/(N + X) = 1/(N + X) + 1/N − 1/N = 1/N − X/[N(N + X)], where
e or Q,
X stands for P e and segmenting the sums, equation (6.8) can be written as
 
N N+ e
P N e
N+Q
1 X X X X
ACC =  Rn (1) + Rn (1) − Rn (2) − Rn (2)
N n=1 n=N+1 n=1 n=N+1
(6.9)
N+ e
P e
N+Q
Pe X e
Q X
− Rn (1) + Rn (2) .
N(N + P)e N(N + P)e
n=1 n=1
The two sums from 1 to N represent the CCC (6.7). The satellite independent
parts can be extracted from the other sums. Explicitly for receiver R(1) it reads
e
1 X
N+P e
P XPe
N+
Rn (1) − Rn (1)
N n=N+1 e
N(N + P) n=1
 
 N+ e
P 
1 e X
= P [UTC(L) − GPST + ∆(1)] + [εn (1) + δn ] (6.10)
N n=N+1

 
 N+ e
P 
e
P X
− e
(N + P) [UTC(L) − GPST + ∆(1)] + [εn (1) + δn ] .
e 
N(N + P) 
n=1
The satellite independent parts cancel out. The same is valid for the terms with
receiver R(2) measurements. Equation (6.9) can now be written as
 
 N+ e
P e
N+Q 
1 X X
ACC = CCC + [εn (1) + δn ] − [εn (2) + δn ]
N n=N+1 n=N+1

(6.11)
N+ e
P N+Qe
e
P X e
Q X
− [εn (1) + δn ] + [εn (2) + δn ] .
e
N(N + P) N(N + P)e
n=1 n=1
Both receivers are tracking a number of satellites in parallel, while some satellites
are only tracked by one of the receivers. An AV CCD is a CV CCD with additional
noise and offset contributions. In nearly 60% of all measurements two identical
receivers track exactly the same satellites, according to Figure 6.2. Achieving
nearly 100% would only be possible if the two receivers were connected to the same
antenna using a signal splitter. The other satellites add additional measurements
to the solution and can slightly shift the mean value. In the operation of very long
baseline links in AV mode the advantage of using many satellites, in particular at
high elevation, is larger than these negative effect [21, 22], but in the framework of
a calibration campaign they just increase the noise level, and thus the uncertainty.
The CV method for a code based link calibration is thus preferable. In reference [75]
the evaluation of the CCD data is done with both CV and AV method. The
results of the AV evaluation do not yield significantly different values than the CV
evaluation, but the standard deviation of the individual measurements around the
mean is higher.
6.2.2. Code Based Time Transfer and Position Errors

Time and frequency comparisons between two receivers are accomplished by com-
paring the 16 min spaced data of the CGGTTS files of two receivers. Since the
data in the RINEX files are calculated from fixed positions which are manually
entered to the processing software that generates the CGGTTS data, an error in
the position causes an offset in the time result and affects the data by an increased
noise level, because the geometric range is erroneous.
To study the impact of a wrong position information to the code based time
transfer two receivers at PTB were examined in a short baseline common-clock
setup. The P3 data of the two receivers are generated from the RINEX files
with the RINEX2CGGTTS software. The P3 data of one of the receivers were
recalculated several times and the position was fictitiously shifted in the three
geodetic coordinates in several steps between a few cm up to 30 m from the original
position. As processing period ten days were chosen. Then the new P3 data were
compared to the original P3 data of the other receiver using the CV method. The
mean values were subtracted from the true differences in order to get the delay due
to a wrong position. Then the standard deviations (SD), reflecting the noise level
of the data, were calculated. The results are depicted in Figure 6.3.
It is clearly visible that the error on the common-clock difference is linear within
an error of the position up to 30 m for all the three coordinates. The effect is
largest in the case of height errors, because the antenna is shifted virtually closer
to or farther from the satellite constellation. The errors on the angle coordinates
are smaller, because this coordinates benefit from symmetries. The offset in the
latitude coordinate is due to shifting the virtual position with respect to the ”north
hole”. In the longitude coordinate the offset is marginal, because the satellite
constellation is completely rotationally symmetric in this coordinate, assuming
perfect circular satellite orbits. The remaining divergence is due to the fact that
the troposphere corrections are not solved correctly with a wrong position and that
the orbits in reality are not perfect circles.
However, due to the incorrect positions the difference between the pseudoranges
and the calculated distance to the satellite (and thus the time difference results)
are different for each satellite, depending on the current satellite position. The
noise is increasing dramatically for all three types of errors introduced with the
same order of magnitude.
Figure 6.3.: Error on P3 CCD data due to wrong positions and the related stan-
dard deviation SD. ∆θ, ∆ϕ, and ∆h are the errors in geodetic lati-
tude, longitude, and height, respectively.
The results found here are only valid for the antenna site at PTB, because the
satellite constellation is different at each location. Since the offsets are linear for
position errors below 30 m it is sufficient to calculate the phase error only for one
fictitiously shifted point in order to get the impact of a wrong position at any
location.
In Figure 6.4 the fictive position shifting is applied to the comparison of the
timescales UTC(USNO)-UTC(PTB) with the two calibrated receivers USN3 at
USNO and PT02 at PTB. Here the positions of the USN3 receiver are shifted in
the height.
The comparison with the original position is in good agreement with the results
UTC(USNO)-UTC(PTB) taken from BIPM’S Circular T. With a position error of
3 meters the increase of the noise is marginal, but the offset with respect to the
original data is significant. From the noise level the error is not detectable before
the error is reaching 7.5 m.
This result shows the importance of proper position estimates for calibrated time
transfer. However, for small errors below a few meters the position error can not
be detected from the noise and the offset is absorbed by the calibration value (6.1)
in terms of a relative link calibration.
Figure 6.4.: Phase shift and noise on time transfer data due to errors of the
height h.
6.2.3. PPP and Antenna Phase Center Corrections

In order to calibrate the links which are evaluated using the PPP method in op-
eration the usage of PPP software is mandatory, even though the ambiguities,
and thus the absolute phase values of the time results, are estimated from the P3
code, because the P3 data are calculated using fixed positions that are manually
entered into the receiver’s internal processing software (see previous Section 6.2.2)
and could diverge from the positions calculated by the PPP software during the
calibration period.
The software which is used by the BIPM to evaluate GPS time links with PPP
is the NRCan-PPP package [15]. In Section 4.1 it was demonstrated that different
software solutions can yield different time offsets for the same RINEX input data,
also in a quasi zero baseline CCD setup. Thus, to calibrate a TAI link which is
evaluated with PPP, the same software which is used by the BIPM for the link
computation has to be used for the evaluation of the calibration values with exactly
the same settings and input data.
In Section 2.5 it was briefly noted that the phase center of the receiving an-
tenna depends on the elevation angle of the satellites. The IGS provides data files
which include corrections for a variety of commercial available antennas. These
corrections are obtained either by comparing the different antenna models to an
antenna which was considered to be absolutely calibrated or by using a calibration
robot [77]. These correction files can be used by the NRCan-PPP software. If the
antenna name in the RINEX observation file header (cf. Appendix D) coincides
with an antenna name in the IGS correction file, the correction is applied.
Although the effect of antenna phase center corrections does not show an effect
on the instability of a time link, it is applied by the BIPM, if antenna informations
are available in the RINEX header.
As shown in Figure 6.5 the corrections cause a general offset to the time results
of the PPP solution. The data of two receivers called ”Rx1” and ”Rx2” here to
simplify matters, operated at USNO in a common-clock quasi zero baseline setup
were evaluated with and without applying the antenna phase center corrections.
Then the possible four comparison combinations were calculated.
Figure 6.5.: PPP common-clock quasi zero baseline comparison of two receivers
Rx1 and Rx2 at USNO. Black plot: Antenna phase center correction
applied to Rx1, not to Rx2; red plot: No correction applied to both
receivers; green plot: Correction applied to both receivers; blue plot:
No correction applied to Rx1, correction applied to Rx2.
The red and the green plot, depicting the comparisons without antenna phase
center corrections and with corrections applied to both receivers, respectively, are
in close agreement, but the difference of the mean values is 0.19 ns. If the antenna
phase center corrections are only applied to one of the two receivers, the solutions
(black and blue plot) diverge from the red and green results by more than 3 ns.
In terms of calibration with an uncertainty at the nanosecond level it is absolutely
necessary to apply the corrections to all receivers involved, in order to get proper
results which can be used by the BIPM.
6.2.4. Data Processing

The standard 16 minute interval CGGTTS P3 and/or C/A code data of all FRs
involved in both laboratories are compared to the TR’s CGGTTS data using the
CV method to get the calibration values for the code based links. The standard
30 s interval RINEX data are processed with the NRCan-PPP software using the
same settings as done by the BIPM with 5 minutes IGS clock and 15 minutes orbit
data and applying the antenna correction according to the IGS. Then the time
results of the FRs are compared to the TR’s results.
As pointed out before, only few receivers apply the internal and cable delays DP1 ,
DP2 , DCab , and DRef to the RINEX data, according to equation (2.23). Of course,
these would be absorbed in the calibration values (6.1). However, it is useful to
correct the calibration values obtained with the PPP processing for the internal
delays taken from the P3 CGGTTS file header, in order to verify the consistency
between the P3 and the PPP solution. In this work, both the corrected and the
uncorrected values are given in the Tables 6.4, 6.5, 6.6, 6.9, and 6.11.
Subsequently, outliers in the code based data are removed by applying a 3σ filter
iteratively until it has no effect on the remaining data, with the limitation that not
more than 5% of the data are removed. In the two campaigns described here the
amount of removed code based data was marginal. In the case of the PPP data
a phase drift can exceed the level of the phase noise and a 3σ filter would remove
too many measurement data. Hence, obvious outliers are removed manually.
In the next step the TDEV of the data is calculated with the τ0 value chosen as
the average of the individual data spacing (see Chapter 2). From the minimum in
a double logarithmic diagram an averaging time for the individual data points is
estimated in order to remove the white phase noise. In case of a significant data
gap (more than 5% deviation from the original data spacing) the TDEV of the
data before the gap and after the gap is calculated separately. The TDEV with
its minimum at a shorter averaging time is used. Special attention has to be paid
to the last TDEV value. Especially if it does not follow the trend of the previous
values, it should be mistrusted and discarded.
The last step is to average the individual CCD data, to calculate the mean value,
and to calculate the SD of these averaged data around the mean.
To calibrate a TWSTFT link, the TWSTFT data are compared to the data of
one GPS link after applying the new calibration value (6.1) to this link, according to
(6.3). Since the TWSTFT data spacing interval is in general higher than that of the
GPS data, the GPS data are compared to values obtained by a linear interpolation
between two adjacent TWSTFT measurements. This double difference data could
be affected by significant outliers related to the TWSTFT as well as to the GPS
link. To clean the data a 3σ filter is applied iteratively, as for the code based
CCD data. Finally, the double difference data are also averaged by estimating the
averaging time with the help of the TDEV. The mean value of the averaged data
represents the TWSTFT calibration value, and the SD of the averages around the
mean contributes to the TWSTFT calibration uncertainty.
6.3. Uncertainty Estimation

The overall uncertainty of a relative GPS link calibration between two remote
laboratories is given by p
UGPS = ua 2 + ub 2 , (6.12)
6.3. Uncertainty Estimation 91
with the statistical uncertainty ua and the systematic uncertainty ub . The statisti-
cal uncertainty is related to the noise of the GPS measurement and changes of the
internal delays of the equipment, whereas the systematic uncertainty accounts for
effects and errors that affect all CCD measurement data in the same but unknown
way.
The statistical uncertainty is the geometric sum of the statistical uncertainty at
LAB1 and the remote LAB2
p
ua = ua,LAB1 2 + ua,LAB2 2 . (6.13)
At the remote laboratory LAB2 the SD of the averaged data around the mean
is directly the relevant contribution ua,LAB2 . Since CCD data are taken at LAB1
before and after the trip to LAB2 the higher SD is the relevant contribution. If
the difference of the mean values of the two CCD measurements (dCCD) is smaller
than this SD, the SD is taken as the uncertainty contribution ua,LAB1 . Otherwise
the absolute value of dCCD is used, because equipment delay changes apparently
exceed the other noise components.
The systematic uncertainty is given by
s
X
ub = ub,n 2 (6.14)
n
and explained in detail in Table 6.1 and the following text.

The instability of the connection to the local UTC signal (ub,1 , ub,2 ) is estimated
from long term experience at PTB, assuming that the environmental conditions at
other laboratories are equivalent, and related to long term variations of the signal
distribution system.
Since SR620 TICs were used in both calibration campaigns shown below to deter-
mine the offset δt0 , the errors related to this type of TIC have to be incorporated
to the uncertainty budget (ub,3 , ub,4 ). According to the manufacturer specifica-
tions [78] the trigger level timing error is given by
15 mV + 0.5% of trigger level

Trigger level timing error = (6.15)
1 PPS slew rate
for start and stop channel, respectively. With a trigger level of 1 V at one channel
and an estimated signal slew rate of 0.5 V/ns the error is 0.04 ns per channel and
0.06 ns for the measurement after adding the start and stop error in quadrature.
According to information given by the manufacturer [37], the trigger level timing
error of the TR’s internal TIC (ub,5 , ub,6 ) is estimated as 10 mV / (1 PPS slew rate)
per channel. The error of the stop channel cancels out, because this channel is
always provided with the signal of the receiver board and the error is the same at
each laboratory.
Table 6.1.: Systematic uncertainty contributions. The contributions identified

with an M are different in each calibration campaign and have to be
determined by additional measurements. The contributions marked
with an asterisk are only applied to special measurements (see text).
ub,n Value Description

ub,1 0.10 ns Instability of the connection to UTC(LAB1)
ub,2 0.10 ns Instability of the connection to UTC(LAB2)
ub,3 0.06 ns TIC trigger level timing error at LAB1 [78]
ub,4 0.06 ns TIC trigger level timing error at LAB2 [78]
ub,5 0.02 ns TR trigger level timing error at LAB1 [37]
ub,6 0.02 ns TR trigger level timing error at LAB2 [37]
ub,7 0.10 ns TIC nonlinearities at LAB1
ub,8 0.10 ns TIC nonlinearities at LAB2
ub,9 0.50 ns Systematic error of the TIC at LAB1 [78]
ub,10 0.50 ns Systematic error of the TIC at LAB2 [78]
ub,11 M Jitter of the TIC measurement at LAB1
ub,12 M Jitter of the TIC measurement at LAB2
ub,13 M Determination of the UTC reference point at LAB1
ub,14 M Determination of the UTC reference point at LAB2
ub,15 0.30 ns Multipath [79]
ub,16 0.18 ns Antenna cable and antenna [74]
ub,17 * M Position error at LAB1
ub,18 * M Position error at LAB2
ub,19 * 0.30 ns Uncertainty of the ambiguity estimation [64]
The uncertainty contributions ub,5 , ub,6 are related to imperfections in the TIC
in conjunction with the external reference frequency. When two pulses with stable
time offset between them are fed to the start and stop channel, the measurement
result depends on the relationship between the pulses and the zero-crossing of
the reference frequency. The traveling TIC was connected to different frequencies
generated by different clocks at PTB (masers, commercial caesium clocks) and
5 MHz and 10 MHz were used, respectively, while the start and stop channel were
provided with 1 PPS signals of one clock. The length of the reference cable was also
changed. After averaging 50 single measurements in each setup slightly different
results were obtained, but the difference was always below 0.1 ns. The results were
verified by repeating all measurements with other TICs available at PTB. This
effect has not yet been studied in detail, but it is probably related to interpolation
processes in the TIC. It is accounted for by 0.1 ns at each laboratory.
6.3. Uncertainty Estimation 93
The TR’s internal TIC uses a surface acoustic wave (SAW) filter as interpolator,
similar to the TIC analyzed in [80], but with less precision [37]. For TICs based
on this principle the imperfection (the so-called nonlinearity effect) is negligible,
because it is of the order of magnitude of a few picoseconds [37, 80].
According to the manufacturer’s specifications [78] an uncertainty of 0.5 ns has
to be taken into account at each laboratory (ub,9 , ub,10 ) if local TICs are used
to determine the delay δt0 , because the internal delays between their two input
channels vary from unit to unit.
The offset δt0 is obtained by averaging multiple single TIC measurements at each
lab. The jitter of the TIC is given by the standard deviation of these measurements
around the mean (ub,11 , ub,12 ). From the point of view of the TIC measurement this
is the statistical uncertainty of the TIC, while the other TIC related uncertainties
listed in Table 6.1 are contributions to the systematic uncertainty of the TIC.
However, in the framework of the GPS measurement it becomes a systematic error,
because it affects all CCD data in the same way.
Currently there is no standardized definition of the UTC reference point in a
time and frequency laboratory. It could be a fixed connector or the endpoint of
a cable. It could also be a virtual point somewhere inside the 1 PPS distribution
devices. In many cases the physical signal used for the δt0 measurement is defined
by an additional measurement with uncertainties ub,13 and ub,14 , respectively.
The propagation of the satellite signal is affected by multipath effects [79] (ub,15 ),
but in the quasi zero baseline setup, in which the CCDs are measured, the at-
mospherical and site displacement effects cancel out.
Since CCD data are taken for at least one week, diurnal variations due to tem-
perature variations over the day which affect the outside installations (cables, an-
tennas) should cancel out in the average, but increase the statistical uncertainty.
However, it cannot be assumed that the average outside temperature at both lab-
oratories is the same for the three measurement periods, and thus the length of
the antenna cable of the TR and the electrical delay of the antenna is exactly the
same. In both campaigns shown here a FSJ1 cable with an approximate electrical
delay of 224 ns was employed. The cable manufacturer indicates a temperature
sensitivity value of 400 parts per million (ppm) for the electrical delay change per
degree Celsius. With the assumption of a maximum average temperature change
of 20◦ C between each CCD measurement the delay change of the entire 50 m cable
is 0.09 ns. Unfortunately, the temperature sensitivity of the Novatel GPS 702GG
antenna is unknown. Therefore a temperature coefficient of less than 0.01 ns/◦ C
for the complete system cable and antenna is used here, according to CCD mea-
surements performed at ROA in 2008 between two GTR50 receivers connected to
the same type of antenna with the same type of cable with comparable length [74].
The order of magnitude is supported by the analysis presented in the reference [81].
The uncertainty contribution ub,16 is thus accounted for by 0.18 ns, indicating that
half of this sensitivity is related to the cable while the other half is related to the
antenna.
An evaluation of the temperature sensitivity of a GPS system using code-phase
residuals was proposed in [82], but this could not be done with a receiver of the
type GTR50, because the 1 PPS signal measured by the internal TIC versus the
external 1 PPS signal is derived from the code and applied to all data. Hence
the temperature sensitivity of the code measurement is transferred to the phase
measurement.
As demonstrated above, a position error is linear in the time results and domi-
nating in the height. According to that, the impact of the height error at each lab-
oratory can be estimated by fictitiously shifting the position of one of two receivers
taking CCD data in a quasi zero baseline setup by using the RINEX2CGGTTS con-
version software [31]. Within the framework of calibration, these errors eh (LAB1)
and eh (LAB2), given in units of ns/m, are multiplied with the absolute value of the
difference between the ”true”position and the position manually entered into the
TR’s processing software. The position error of the FRs is not subject of a relative
calibration, because the phase shift is included in the link calibration value and the
increased noise level of the CCD measurement is incorporated into the statistical
uncertainty. If the manually entered position of the TR’s antenna is obtained by
collecting RINEX data from one or two days and the position is estimated with the
NRCan-PPP software and then used for the CGGTTS processing, as done in the
METAS-PTB calibration campaign at METAS, the uncertainty can be neglected,
because the precision of the PPP software is expected to be at the sub-centimeter
level [47], if the transformation between the reference frames (A.5) is accounted
for correctly, such that the TR position and the positions used by the FRs are
consistent.
An additional uncertainty contribution ub,19 of 0.3 ns is applied to all PPP
calibrations, because the initial phase of the carrier frequency is a priori unknown
and has to be estimated by the PPP software from the P3 code. In reference [64]
a typical phase discontinuity of 0.15 ns per receiver was found for PPP batch
processing with the NRCan-PPP [15], independent of the length of the processed
batch. This adds up geometrically to 0.21 ns for a CCD comparison between a pair
of receivers and to 0.3 ns for the two CCD measurements at LAB1 and LAB2.
The uncertainty of the TWSTFT calibration (6.3) is given by geometrically
adding the uncertainty of the GPS link calibration to the SD of the averaged
double difference data around the mean.
The systematic error of the TICs is a dominating contribution to the fixed parts
of the statistical uncertainty. It can be remobved by using the same TIC at both
participating laboratories.
6.4. PTB’s Calibration Set-up 95
6.4. PTB’s Calibration Set-up

To measure the offset δt0 between the local UTC reference point and the 1 PPS
signal connected to the TR usually a TIC is used at each laboratory. The delay
between the start and stop channels of different TICs are unequal, also for TICs of
the same type. The time interval measurement with the Stanford Research System
SR620 TIC, which is well known and extensively used in the timing community, is
affected by a systematic error of 500 ps according to the manufacturers specifica-
tions [78]. This error has to be considered in the uncertainty budget (Table 6.1).
To reduce the overall uncertainty of a relative link calibration, a traveling TIC
can be shipped together with the receiver, and δt0 would then be measured using
exactly this device at all participating laboratories involved in the calibration cam-
paign. The systematic error of the traveling TIC cancels out and the uncertainty
contributions ub,9 and ub,10 vanish. In order to accomplish this, a GTR50 time and
frequency transfer GPS receiver has been integrated with a SR620 TIC in a small
transportable rack together with a monitor and a keyboard (Figure 6.6). This new
calibration set-up was used for the first time to calibrate the links between PTB
and USNO in 2010.
Figure 6.6.: Front view of the calibration set-up.
Beyond the reduction of the uncertainty the new calibration set-up offers further
advantages, because it is easy to use for the local staff at the involved laborato-
ries due to a text based software that controls the complete calibration procedure
including the TIC measurement as well as the GPS measurement, and thus mini-
mizes the risk of unintended errors. Furthermore the remote labs neither need to
allocate additional equipment like keyboard and monitor for the control of the TR
nor to care about documentation of the TIC measurement and its uncertainty.
Figure 6.6 shows the front side of the calibration set-up. All connectors are
easily accessible along the side, except the connection for power supply, which is
located on the back side. This avoids cable mess in the laboratory. The LAN
computer network connector is only designated for maintenance purposes at PTB
and deactivated in the remote laboratories.
In Figure 6.7 on the left photo the signal connectors are visible, on the right
scheme the cable layout plan inside the rack is depicted.
Figure 6.7.: Signal connectors (left picture) and internal cable layout (right
picture).
At the REF IN connector the set-up is provided with a 5 MHz or 10 MHz external
reference frequency. This frequency needs not to be the UTC frequency and can be
derived from any atomic standard. Effects due the TIC imperfections are included
in the uncertainty budget, as explained above. The 1 PPS M connector provides
the receiver with the 1 PPS UTC signal. The offset of this signal with respect to
the local UTC is measured by the TIC before the GPS measurement starts. For
this purpose a 1 PPS signal with known delay to UTC has to be connected to the
1 PPS REF connector. The delay induced by the internal cables, labeled red and
blue in Figure 6.7, cancels out in equation (6.2).
After setting up the calibration set-up in the laboratory’s measurement room
and connecting it to the power supply the operating system of the TR boots up
and the calibration software starts automatically. The required AC input voltage
can be adjusted to the anticipated conditions in each laboratory. In Figure 6.8
in the left picture the calibration set-up with pulled out monitor and keyboard is
shown.
The first program screen requests the user to wait until the internal SR620 TIC is
warmed-up. The right plot in Figure 6.8 shows that the TIC needs approximately
6.4. PTB’s Calibration Set-up 97
Figure 6.8.: The calibration set-up with pulled out monitor and keyboard (left)
and warm-up characteristics of the internal SR620 TIC (right).
one hour to reach a stable state after turning-on. For this test both the start and
stop channel were provided with the same 1 PPS signal by using a power splitter
and a length difference of the connecting cables. To ensure that the TIC is in the
stable state in any case under the unknown environmental conditions in a remote
lab, the waiting time is set to two hours. The remaining time is displayed on the
monitor.
After the warm-up period the operator is asked for the precise antenna position
in ITRF Cartesian coordinates. If the antenna position is unknown in a remote
laboratory an arrangement could be made in order to enable collecting RINEX
data for one or two days, to send it to PTB, and to input the position later after
estimating it from the RINEX data using the PPP method.
It is impossible to proceed with the procedure until both signals are correctly
detected by the TIC. The operator has to decide whether the reference frequency
is 5 MHz or 10 MHz. Normally a trigger level of 1 V is used for the 1 PPS M
signal, but if arrangements are made with the remote lab this level can be changed
between 0 V and 2 V. This level is then used by the TIC as well as by the TR.
After this, the SR620 TIC automatically starts to measure the delay between
1 PPS REF and 1 PPS M. The software takes 100 single measurements and cal-
culates the mean value, the jitter and the rate. The measurement can be repeated
in case of problems with the signals, e.g. if a significantly high jitter or rate is
observed. The measured delay of the reference signal and the internal cable delay
is applied to the GPS measurements automatically.
The reference cable can be removed after the delay measurement and used for
other purposes in the laboratory. Due to the internal TIC used inside the GTR50
no additional measurements are needed. The internal temperature stabilization
inside the GTR50 needs about 20 minutes to warm up the receiver board and
the TIC card to 45◦ C. Then the receiver starts to collect the first GPS data. At
the end, after about one week taking GPS measurement data, the reference delay
measurement can be repeated to check the stability of the environmental conditions
in the laboratory and to verify the correctness of the uncertainty estimates ub,1 or
ub,2 , respectively.
6.5. METAS-PTB Calibration

In June and July 2009 the time links between MEATS and PTB were calibrated
using a TR of the type GTR50. The TR was operated at PTB in June 2009 for
six days within the interval MJD 54985 - 54991. Then, after the TR was used
for another purpose, it was shipped to METAS and operated in July for ten days
between MJD 55016 and MJD 55026. After this, it was immediately sent back, so
that the second operation at PTB was done between MJD 55034 and MJD 55042.
Table 6.2 specifies the types of fixed receivers involved at PTB and METAS and
their data output. The nomenclature is according to the conventions used on the
BIPM data server (C/A, P3, RINEX) and by the IGS tracking network (RINEX),
respectively, if the data are provide to these institutes. Otherwise the laboratory’s
internal convention is used. An exception is the TTS-3 receiver. The TTS-3
RINEX data were found not useful for the purpose here [75]. In the following it
will be referred to the FRs as given in Table 6.2, which takes into account that one
receiver produces different kinds of data.
Table 6.2.: Receivers at PTB and METAS and the related data output.
Institute Receiver P3 RINEX

Ashtech Z12-T PT02 PTBB
PTB Ashtech Z12-T PT03 PTBG
AOS TTS-3 PT06 ---
Dicom GTR50 PT08 PT08
Ashtech Z12-T CH01 WAB2
MEATS
Septentrio PolaRx CH03 WAB4
As Table 6.2 shows, eight P3 links and six PPP links can be evaluated. Other
receivers at PTB that provide C/A code data have not been involved, because the
METAS receivers do not provide this kind of data.
In June 2009 lightning conductors were mounted nearby the antennas at PTB,
as visible in Figure 6.9, without notification of the laboratory. These installations
affected the measurements by multipath effects, which resulted in unstable carrier-
phase measurements. Thus, the lightning conductors were removed at the end of
6.5. METAS-PTB Calibration 99
July. At PTB the tracking of satellites with low elevation angles is limited by the
surrounding trees. In general, the roof of the METAS building is a better place for
GPS measurements, due to free sight to all directions.
Figure 6.9.: Antenna sites at PTB (left picture) and at METAS (right picture).
The TR antenna was not mounted at PTB site when the photo was
taken. CH05 did not collect data during the calibration campaign.
At PTB the offset due to a wrong position of the TR’s antenna was found
to be eh (PTB) = 2.22 ns/m. The absolute value of the manually entered a priori
position and the position estimated from the RINEX files of the TR during the CCD
measurement is 0.119 m and the related uncertainty contribution is ub,17 = 0.26 ns.
At METAS the TR’s antenna was installed on a temporary mount with unknown
a priori position. Thus, the TR was collecting one day RINEX data with arbitrary
offset δt0 and fictitious position before the CCD measurement started. The position
was estimated using the NRCan-PPP online version. The uncertainty contribution
ub,18 can be considered to be zero.
The offset of a 1 PPS signal with respect to the local UTC reference point at PTB
is measured using a reference cable connected to the 1 PPS UTC distribution. The
delay at the male BNC connector is periodically checked with respect to UTC(PTB)
with a fixed SR620 TIC which is used as the central measurement device to compare
all PTB clocks to UTC(PTB) and with each other. In normal operation the start
and stop channels are connected to a switching matrix that allows for measuring all
intercombinations of the connected clock signals. The true UTC(PTB) reference
point is a virtual point inside the signal distribution chain. If any changes of
the equipment are made, this point has to be reconstructed by aligning the data
after the change to the data before the change and adjusting delay constants in
the computer software which controls the switching matrix and reads the TIC
measurements.
The reference cable is checked by disconnecting the TIC’s stop channel from
the switching matrix and measuring the UTC signal, as provided by the switch-
ing matrix, with respect to these cable. The start channel is provided with the
UTC signal of the switching matrix. Before using the reference cable in order to
determine δt0 it was verified with the method described above. After averaging
20 single measurements the jitter was 0.02 ns. This has be taken into account as
uncertainty contribution ub,13 . The offset δt0 was measured with a second TIC in
the framework of the METAS-PTB calibration campaign.
The local UTC representation at METAS UTC(CH) is by definition the male
BNC connector of a cable connected to the start channel of a SR620 TIC. The
stop channel was directly used to determine δt0 at METAS. Thus the uncertainty
contribution ub,14 is zero.
Since different counters have been used for the δt0 measurement, ub,9 and ub,10
remain in the uncertainty budget. In both laboratories 20 single measurements
were averaged and the jitter was found to be 0.20 ns in both cases (ub,11 and ub,12 ).
In Figure 6.10 the results of the P3 and PPP CCD measurements at PTB are
depicted. The nomenclature here and in the following illustrations is to be under-
Figure 6.10.: Code (left) and PPP (right) CCD measurements at PTB before
and after the calibration trip to METAS.
stood as TR-FR, e.g. PT02 denotes the CCD measurement TR-PT02. For better
visibility arbitrary offsets have been applied to some of the data. No significant
differences of the results before and after the calibration trip to METAS are vis-
ible for the P3 measurements. Due to problems with the antenna, PT08 RINEX
data are not available for MJD 54985 and MJD 55042. Furthermore this data look
more unstable than expected from a comparison of two GTR50 [65,74], which was
probably caused by the lightning conductors.
The CCD results at METAS are shown in Figure 6.11. At MJD 55021 the
CH01/WAB2 receiver has not generated a daily RINEX file. The data were rebuilt
from hourly files, but it was not possible to reconstruct the corresponding P3 file
using the RINEX2CGGTTS conversion software [31]. The PPP results generated
with the rebuilt RINEX data show a phase jump indicating a complete loss of all
satellite tracks.
In Figure 6.11, as well as in Figure 6.10, phase drifts in the PPP results of the
Ashtech Z12-T receivers are clearly visible. These drifts cannot be found in the P3
data. They are only related to defects in the phase measurement and are proba-
bly intensified by the Kalman-Filter of the NRCan-PPP software (cf. Figure 4.5).
In reference [75] this effect was accounted for by an additional uncertainty con-
tribution. This was omitted in the analysis presented here, because the drift is
already increasing the statistical uncertainty and thus already taken into account.
A detailed analysis of the Ashtech Z12-T receiver can be found in reference [83].
Figure 6.11.: Code (left) and PPP (right) CCD measurements at METAS.
In Figure 6.12 the TDEV is exemplarily depicted for the second P3 CCD mea-
surement at PTB (CCD2) and the P3 CCD measurement at METAS. Due to the
data gap at MJD 55021 in the CH01 data the TDEV is calculated twice with the
data before and after the gap. In Table 6.3 the data averaging times estimated
from the TDEVs of all CCD measurements are listed. AVT denotes the averaging
time. In the PTB case AVT1 stands for the averaging times estimated for the
results obtained from the first CCD measurement (CCD1), while AVT2 denotes
the averaging time for the results of the second CCD measurement (CCD2).
Table 6.4 shows the results of the two CCD measurements at PTB, the corre-
sponding standard deviation (SD), and the difference between the CCD measure-
ments (dCCD) after averaging the data according to Table 6.3. The underlined
value is used as the statistical uncertainty contribution at PTB. Since PTBB and
PTBG do not apply the internal delays to the RINEX data, these values are added
to the PPP calibration values at the end of the data analysis according to equa-
Figure 6.12.: Double logarithmic plots of the time deviations of the second P3
CCD measurement at PTB (left plot) and the P3 CCD measure-
ment at METAS (right plot).
Table 6.3.: Estimated data averaging times.
PTB METAS
Type FR AVT1 AVT2 Type FR AVT
P3 PT02 1 day 1 day P3 CH01 3 · 104 s
P3 PT03 1.5 · 104 1 day P3 CH03 1 day
P3 PT06 7680 s 3840 s PPP WAB2 2400 s
P3 PT08 1 day 1 day PPP WAB4 2400 s
PPP PTBB 2400 s 2400 s
PPP PTBG 2400 s 2400 s
PPP PT08 2400 s 2400 s
tion (2.23) and stated in brackets. All P3 results agree with the PPP results within
a few 0.1 ns.
The P1 and P2 delays, DP1 and DP2 , and the antenna cable delay DCab of
the PT03/PTBG receiver have never been evaluated and are set to zero in the
RINEX2CGGTTS conversion software. The applied delay consists only of the
reference signal delay DRef . Hence the CCD values in Table 6.4 are at the level of
about -511 ns for PT02, as well as for PTBG after the correction for the reference
delay.
The results at METAS and the corresponding SD, which is the statistical un-
certainty at METAS, are depicted in Table 6.5. Here also the internal delays of
the WAB2 and WAB4 receivers are added after the analysis. The P3 and PPP
Table 6.4.: Results of the CCD measurement at PTB. Values are rounded to the
second decimal. The underlined value is chosen as the uncertainty
ua,PTB . The values in brackets are the calibration values after applying
the internal delays according to (2.23).
CCD1 CCD2 C1 dCCD SD1 SD2

Type FR
/ ns / ns / ns / ns / ns / ns
P3 PT02 -1.91 -1.92 -1.91 0.01 0.21 0.19
P3 PT03 -511.10 -511.76 -511.43 0.66 0.42 0.18
P3 PT06 11.95 11.48 11.71 0.47 0.83 1.04
P3 PT08 -3.64 -4.02 -3.83 0.38 0.31 0.24
-520.20 -520.26 -520.23
PPP PTBB 0.06 0.26 0.18
(-1.76) (-1.82) (-1.79)
-423.03 -423.75 -423.39
PPP PTBG 0.72 0.19 0.22
(-511.23) (-511.95) (-511.59)
PPP PT08 -3.66 -4.12 -3.89 0.45 0.06 0.04
Table 6.5.: Results of the CCD measurements at METAS. Values are rounded to
the second decimal. The values in brackets are the calibration values
after applying the internal delays according to (2.23).
Type FR C2 / ns SD / ns
P3 CH01 -8.08 0.25
P3 CH03 -2.11 0.14
-457.61
PPP WAB2 0.11
(-8.02)
-198.28
PPP WAB4 0.05
(1.56)
results are in good agreement for the receiver CH01/WAB2, but between CH03
and WAB4 the discrepancy is more than 3.5 ns. This is due to missing information
about the CH03/WAB4 antenna. The antenna type is not included in the RINEX
file header and thus no phase center correction is applied by the PPP software. As
demonstrated before (Chapter 4) this can lead to an offset with the observed order
of magnitude.
Table 6.6 shows the calibration values and the corresponding uncertainties for all
possible links between METAS and PTB. The 3 ns discrepancy between the links
Table 6.6.: Calibration values and corresponding uncertainties for all possible
GPS links between METAS and PTB. The values are rounded to
the second decimal. The values in brackets are the calibration values
Type Link CGPS / ns ua / ns ub / ns U / ns

P3 CH01-PT02 6.17 0.33 0.87 0.93
P3 CH01-PT03 -503.35 0.70 0.87 1.12
P3 CH01-PT06 19.79 1.07 0.87 1.38
P3 CH01-PT08 4.25 0.45 0.87 0.98
P3 CH03-PT02 0.20 0.25 0.87 0.90
P3 CH03-PT03 -509.32 0.67 0.87 1.10
P3 CH03-PT06 13.82 1.05 0.87 1.36
P3 CH03-PT08 -1.72 0.40 0.87 0.95
-62.62
PPP WAB2-PTBB 0.28 0.88 0.92
(6.23)
34.22
PPP WAB2-PTBG 0.73 0.88 1.14
(-503.56)
453.72
PPP WAB2-PT08 0.47 0.88 0.99
(4.13)
-321.95
PPP WAB4-PTBB 0.26 0.88 0.92
(-3.36)
-225.11
PPP WAB4-PTBG 0.72 0.88 1.14
(-513.15)
194.39
PPP WAB4-PT08 0.46 0.88 0.99
(-5.46)
involving CH03 and the links involving WAB4 is also visible here. All uncertainties
are at the level of about 1 ns and well below 2 ns, as in previous exercises [17, 75].
The results can be compared to the results of previous calibrations conducted
by the BIPM [84, 85], where a traveling Ashtech Z12T receiver was circulated
between the laboratories several times between 2002 and 2008. This traveling
receiver was considered to be absolutely calibrated and the CCD measurements
were used to calibrate the internal P1 and P2 delays of the fixed receivers. The
method of absolute receiver calibration is described in the reference [86]. PT02 has
been calibrated by the BIPM in April 2008 [84] and CH03 has been calibrated in
August 2008 [85]. The link calibration value of the link CH03-PT02 in Table 6.6
is nearly zero, confirming the stability of the equipment during more than 2 years.
The CH01 calibration was done in 2004. The link CH01-PT02 has to be corrected
by 6.17 ns.
Before this result is discussed further, it has to be noted that any changes of the
equipment result in different delays. This changes are modifications of the time
and/or frequency signal distribution system, modifications of the antenna and the
antenna cable, dislocations of the antenna, and variations in the environmental
conditions. Also the replacement of the receiver itself with an other one causes a
delay difference, because DP1 and DP2 are a-priori unknown.
The link calibration values can of course be readjusted, if a second calibrated
link is operated during the period in which the events described above have been
occurred. The data of the link which lost its calibration are therefore aligned to the
data of the calibrated link. But this practice obviously increases the uncertainty,
because the uncertainty of the readjusted calibration value consists of the geometric
sum of the uncertainty of the calibrated link and the SD of the comparison of the
two links. Furthermore, in case of the alignment of one link to the other the baseline
is not short and additional uncertainty contributions due to the atmospherical (2.8)
and site displacement effects have to be taken into account.
It can be assumed that over time changes occur to all the equipment at different
instants of time. It is thus necessary to perform periodical realignments on all
links. This leads to an increase of all calibration values over the years. This shows
that calibrations should be repeated periodically as often as possible.
Between 2008 and 2009 the operation of the receivers CH03 and PT02 was sta-
ble at the sub-nanosecond level. No changes of the equipment have been occurred
within this period. On the one hand this demonstrates that the calibration cam-
paign with the GTR50 receiver produced reliable results, on the other hand the
5 ns uncertainty applied by the BIPM to all GPS links used for TAI computation
becomes questionable in the case that verifiably no equipment changes have been
occurred after a calibration campaign. The 5 ns originates from BIPM’s long term
experience.
Unfortunately it is not possible to trace what happened to the CH01 receiver
between 2004 and 2009. Large steps typically point to changes in the installa-
tions rather than in the receiver itself. This supports the above statement that
calibration campaigns should be repeated periodically.
METAS and PTB maintain a TWSTFT link in the Ku-band, which was cal-
ibrated using a traveling TWSTFT earth station in 2008 [87]. The results of
the GPS calibration can be verified by comparing a calibrated GPS link to the
TWSTFT data. If the TWSTFT calibration value calculated according to equa-
tion (6.3) vanishes within the combined uncertainty, the proof is obtained that
relative GPS calibration campaigns can serve as an alternative to the cost and
labor intensive calibrations with traveling TWSTFT equipment and that the GPS
based calibration uncertainty is not underestimated.
For this purpose the now calibrated P3 link with the lowest uncertainty, which
is the link CH03-PT02, is compared to the operational TWSTFT link. The GPS
link is evaluated in the CV mode. The processed period is partially overlapping
with the CCD measurement at METAS and CCD2 at PTB. In Figure 6.13 the
results and the corresponding TDEV is depicted.
Figure 6.13.: GPS - TWSTFT double differences (left plot) and related TDEV
(right plot). The mean value of the averaged data is -1.19 ns, the
SD is 0.41 ns.
The left plot shows all data obtained with the interpolation method described in
Subsection 6.2.4, as well as the averages. The TDEV justifies an averaging period
of 1/2-day.
1.5
Differential correction / ns
1.0
0.5 GPS calibration
0.0
-0.5
-1.0
calibrated
-1.5 TWSTFT
-2.0
-2.5
Figure 6.14.: Differential correction between the TWSTFT calibration and the
new calibration value obtained with a relative GPS calibration.
Combining the SD of the averaged data around the mean with the GPS calibra-
tion uncertainty the TWSTFT calibration value is given by
CTWSTFT = −1.19 ns ± 0.99 ns. (6.16)

6.6. USNO-PTB Calibration 107
The uncertainty of the TWSTFT calibration was stated as 1 ns [87]. The GPS-
based result and the 2008 TWSTFT calibration result agree with each other within
the combined uncertainty.
6.6. USNO-PTB Calibration

The USNO-PTB calibration campaign was done with the new calibration set-up
described in Section 6.4. Thus the uncertainty contributions ub,9 and ub,10 vanish.
In June 2010 the set-up was operated at PTB for nine days between MJD 55350
and MJD 55358. Then it was shipped to USNO and operated there for twelve days
within the interval MJD 55400 to MJD 55411 in July and August 2010. After it
was sent back, the second operation at PTB was done between MJD 55448 and
MJD 55456 in August 2010.
In Table 6.7 the types of receivers involved at PTB and USNO and the related
data output are specified. The nomenclature is handled with the same principles as
explained in the METAS-PTB section. Data of the fixed GTR50 receiver at PTB
(PT08) were not available for this campaign. Hence, nine P3 links, six PPP links,
and one C/A code link between the TTS-3 at PTB (PT05) and an old TTS-2
at USNO (US01) can be evaluated. The TTS-2 is the precursor of the TTS-3,
providing only L1 C/A code CGGTTS data.
Table 6.7.: Receivers at PTB and USNO and the related data output.
Institute Receiver C/A P3 RINEX

Ashtech Z12-T --- PT02 PTBB
PTB Ashtech Z12-T --- PT03 PTBG
AOS TTS-3 PT05 PT06 ---
Ashtech Z12-T --- USNO USNO
AOS TTS-2 US01 --- ---
USNO
Ashtech Z12-T --- US03 USN3
NovAtel ProPak --- NOV1 NOV1
The ”USNO” receiver constitutes a special case. The antenna is not mounted at
the main antenna site of the laboratory, because this receiver is located at a different
building. The baseline between the TR and USNO is about 176 m. Furthermore it
is not connected to UTC(USNO). It is referenced to a different clock that is kept
close in frequency to UTC(USNO) but with a large phase offset.
The uncertainty contribution due to a wrong position was again found to be
ub,17 = 0.26 ns at PTB. At USNO the error was estimated as eh (USNO) =
2.13 ns/m. The absolute value of the difference between the a priori position and
the PPP estimates is 0.141 m, and thus ub,18 = 0.30 ns.
The δt0 measurement with the TIC integrated into the calibration set-up was
done at PTB before and after each CCD measurement and at USNO before and
after the CCD measurement. Exemplarily the measurement at PTB before CCD1
is depicted in Figure 6.15.
Figure 6.15.: δt0 measurement at PTB.
The blue dots are the individual measurements, the red line is the mean value
of N individual measurements. The jitter at PTB, and thus the uncertainty ub,11 ,
was calculated to be 0.01 ns. At USNO the uncertainty is ub,12 = 0.05 ns. This
significantly higher jitter is due to the fact that the 1 PPS M signal appeared before
the 1 PPS REF signal. The TIC measurements become more noisy for such long
time intervals (approaching 1 s).
Usually TICs are based on counting the zero crossings of a reference frequency
between start and stop signal. Interpolation processes are applied to minimize the
error which occurs due to the missing correlation between these signals and the zero
crossings [88]. However, the performance of the TIC is limited by the instability
of the clock at the measured time interval [78]. Since atomic clocks are dominated
by white frequency noise for averaging times of several seconds the product of the
measurement interval and the instability at this interval is increasing for longer
intervals (cf. the definition of the TDEV in Section 2.1).
In both laboratories the first measurement was in agreement with the second
measurement within 0.05 ns, supporting the estimate of 0.1 ns uncertainty of the
connection to the local UTC (ub,11 , ub,2 ).
In the USNO-PTB campaign the measurement of the offset δt0 required more
efforts, because the UTC(USNO) reference point is defined at the endpoint of a
female BNC connector. An additional cable with two male BNC connectors was
needed to establish the connection to the 1 PPS REF connector of the calibration
set-up. This cable was also used at PTB in order that its delay cancels out and
the uncertainty of the determination of the reference point at USNO (ub,14 ) can
be considered to be zero. At PTB the cable was interconnected between the set-
up and the UTC(PTB) reference cable using a male to male BNC adapter. The
delay of the adapter was determined to be 0.113 ns by an additional measurement
with an uncertainty less than 10 ps. After adding this uncertainty geometrically
to the uncertainty of the UTC(PTB) reference cable verification measurement,
the overall uncertainty of the UTC(PTB) reference point determination is again
0.02 ns, if rounded to the second decimal.
In Figure 6.16 the CCD measurements at PTB are depicted. Offsets are again
introduced for better visibility. As in the METAS-PTB campaign the Ashtech
Figure 6.16.: Code (left) and PPP (right) CCD measurements at PTB before
and after the calibration trip to USNO.
Z12-T phase measurements caused drifts in the PPP results. Figure 6.17 shows
the CCD measurements at USNO.
It becomes apparent that the Ashtech receivers do not inevitably show up the
drift behavior in the PPP results. The solution of the CCD measurement TR-
USN3 is very stable. A disturbance happened in the measurement at MJD 55407,
inducing the drift starting with a small phase shift on MJD 55407. It is visible in
both the TR-USN3 and TR-NOV1 solution and is thus related to the TR. Since
USNO is connected to a different clock this effect is below the noise level. Strictly
speaking, the comparison TR-USNO is not a CCD measurement, but the results
show that the coupling of USNO’s clock to UTC(USNO) is very tight and this
quasi CCD measurement can be treated as the others.
The TDEV plots are exemplarily shown for the USNO measurement in Fig-
ure 6.18. The right plot shows the typical TDEV for PPP CCD measurements.
In the majority of cases it decreases until about 2400 s and then it increases. The
averaging is done over 8 individual measurements.
All estimated averaging times are listed in Table 6.8.
Figure 6.17.: Code (left) and PPP (right) CCD measurements at USNO.
Figure 6.18.: Double logarithmic plots of the time deviations of the CCD mea-
surements at USNO for code (left plot) and PPP (right plot).
In Table 6.9 the results at PTB are listed. After adding the internal delays (2.23)
to the PPP results they are in good agreement with the P3 results.
The values obtained at USNO are shown in Table 6.10. In contrast to PTBB
and PTBG, at USNO the internal delays are applied to the RINEX data, as it is
done for receivers of type GTR50. Hence the P3 and PPP results are directly at
the same order of magnitude. The values for P3 and PPP are in good agreement
for US03/USN3 and NOV1. The discrepancy of about 3 ns for the receiver USNO
is due to an error of the a priori position used for the P3 generation of about 1.5 m.
If the position will be corrected in the future to decrease the noise, the calibration
value C2 will be no longer valid for this receiver.
Table 6.8.: Estimated data averaging times.
PTB USNO
Type FR AVT1 AVT2 Type FR AVT
P3 PT02 1 day 1 day P3 USNO 1 day
P3 PT03 1 day 7680 s C/A US01 1 day
C/A PT05 No avg. 1920 s P3 US03 1 day
P3 PT06 7680 s 3840 s P3 NOV1 1 day
PPP PTBB 1200 s 2400 s PPP USNO 2400 s
PPP PTBG 1200 s 2400 s PPP USN3 2400 s
PPP NOV1 2400 s
Table 6.9.: Results of the CCD measurement at PTB. Values are rounded to the
second decimal. The underlined value is chosen as the uncertainty
ua,PTB . The values in brackets are the calibration values after applying
the internal delays according to (2.23).
CCD1 CCD2 C1 dCCD SD1 SD2

Type FR
/ ns / ns / ns / ns / ns / ns
P3 PT02 -7.32 -7.65 -7.49 0.33 0.17 0.09
P3 PT03 -517.57 -518.36 -517.96 0.79 0.15 0.78
C/A PT05 -6.33 -6.18 -6.25 -0.15 1.05 0.82
P3 PT06 6.79 6.19 6.49 0.59 0.98 0.92
-525.65 -525.98 -525.81
PPP PTBB 0.33 0.25 0.22
(-7.21) (-7.53) (-7.37)
-432.58 -434.03 -433.30
PPP PTBG 1.45 0.28 0.33
(-516.78) (-518.23) (-517.50)
Table 6.11 shows the calibration values and the corresponding uncertainty for
all possible links between USNO and PTB.
The uncertainties are all well below 2 ns, as for the METAS-PTB calibration.
The link which is used by the BIPM for the TAI computation if no TWSTFT
data are available is the link USN3-PTBB. The small calibration value correction
indicates that the internal delays of the USN3 receiver had been aligned to a
TWSTFT calibration performed before, relying on links discussed subsequently.
All general statements made on link calibrations above in terms of the METAS-
PTB campaign are also valid for the USNO-PTB campaign.
Table 6.10.: Results of the CCD measurements at USNO. Values are rounded to
the second decimal.
Type FR C2 / ns SD / ns
P3 USNO -631.45 0.30
C/A US01 -8.59 0.11
P3 US03 -7.14 0.19
P3 NOV1 -6.85 0.12
PPP USNO -628.55 0.11
PPP USN3 -7.15 0.13
PPP NOV1 -6.88 0.14
Until June 2010 two TWSTFT links using two different communication satellites
were maintained between USNO and PTB. The first one was operating in the
X-band and turned off in June 2010, the second one in the Ku-band was still
operational at the time of writing this text (December 2010).
The advantage of the X-band link was that the same transponder was used at
USNO and PTB. Thus relative calibrations with a traveling X-band station were
possible, as for the European TWSTFT links in the Ku-band. In contrast the
USNO-PTB Ku-band link uses different transponders for Europe and the USA
and cannot be calibrated without involving GPS [56, 89].
Since the X-band link was periodically calibrated with a traveling TWSTFT
station [56] a comparison of a new calibrated GPS link can be used in order to
verify the GPS calibration (Table 6.11). The new calibration value is applied to
US03-PT02 P3 data taken during the last 25 days when the TWSTFT X-band link
was used for TAI computation. Due to the very long baseline between USNO and
PTB it was decided to evaluate the GPS data in the AV mode. After applying the
3σ filter to the GPS-TWSTFT double difference data about 5% of the data were
discarded. The TDEV (Figure 6.19) allows for daily averages. After averaging
the daily mean values, calculating the SD of the averages around the mean, and
adding it geometrically to the GPS calibration uncertainty the result for the X-
band calibration value is
CTWSTFT,X = −0.71 ns ± 0.92 ns . (6.17)
This demonstrates that the GPS calibration campaign was successful and that a
TWSTFT link can be calibrated with an uncertainty less than 1 ns in favorable
cases. Figure 6.20 depicts the agreement of the GPS calibration and the 2007 X-
band calibration within the combined uncertainty. The uncertainty of the X-band
Table 6.11.: Calibration values and corresponding uncertainties for all possible
GPS links between USNO and PTB. The values are rounded to the
second decimal. The values in brackets are the calibration values
Type Link CGPS / ns ua / ns ub / ns U / ns

C/A US01-PT05 2.27 1.06 0.58 1.21
P3 USNO-PT02 623.97 0.45 0.58 0.73
P3 USNO-PT03 113.49 0.84 0.58 1.02
P3 USNO-PT06 637.94 1.02 0.58 1.18
P3 US03-PT02 -0.35 0.38 0.58 0.69
P3 US03-PT03 -510.82 0.81 0.58 1.00
P3 US03-PT06 13.63 0.99 0.58 1.15
P3 NOV1-PT02 -0.63 0.35 0.58 0.68
P3 NOV1-PT03 -511.11 0.80 0.58 0.99
P3 NOV1-PT06 13.34 0.99 0.58 1.15
102.57
PPP USNO-PTBB 0.35 0.52 0.63
(621.18)
195.25
PPP USNO-PTBG 1.46 0.52 1.55
(111.05)
-518.83
PPP USN3-PTBB 0.35 0.52 0.63
(-0.22)
-426.16
PPP USN3-PTBG 1.46 0.52 1.55
(-510.36)
-519.09
PPP NOV1-PTBB 0.36 0.52 0.63
(-0.49)
-426.42
PPP NOV1-PTBG 1.46 0.52 1.55
(-510.62)
calibration was estimated as 0.88 ns. The complete history of X-band calibrations
since 2003 is outlined in the reference [56].
The calibration of the operational Ku-band link was done within a period after
the GPS calibration campaign. Due to the long period of 64 days an averaging
period of 4 days was chosen (Figure 6.21).
The new calibration value for the USNO-PTB Ku-band link was found to be
CTWSTFT,Ku = 3.67 ns ± 1.26 ns. (6.18)
Here the uncertainty is also well below 2 ns. However, although the Ku-band link
was aligned to the X-band calibration [56] the 3.67 ns are remarkable and neces-
Figure 6.19.: GPS - X-band double differences (left plot) and related TDEV
(right plot).
1.5
Differential correction / ns
1.0
0.5 GPS calibration
0.0
-0.5
-1.0
calibrated
-1.5 X-band
-2.0
Figure 6.20.: Differential correction between the 2007 X-band calibration value
and the new calibration value obtained with the relative GPS
calibration.
sitate an explanation: A long term comparison of the X-band and the Ku-band
link shows an increase of the noise and a reduced stability since the first quarter of
2010. This is also visible in the comparison between GPS PPP and the Ku-band.
The same effects are also visible in the link comparisons between European labora-
tories equipped with TWSTFT stations and PTB. They are probably caused by a
reduction of the bandwidth available for time comparisons on the communication
satellite connecting the US and European laboratories.
However, both the X-band and the Ku-band results shown here confirm that a
relative GPS calibration can be used as an alternative to cost effective TWSTFT
calibrations also on long intercontinental baselines, since the uncertainty at the
moment of the calibration is at the same low level as with TWSTFT equipment.
6.7. Suggestions for further Reductions of the Uncertainty 115
Figure 6.21.: GPS - Ku-band double differences (left plot) and related TDEV
(right plot).
6.7. Suggestions for further Reductions of the

Uncertainty
Due to the traveling TIC integrated into the calibration set-up the main part of
the uncertainty of the δt0 determination vanishes. However, other non-negligible
uncertainties related to 1 PPS time interval measurements are part of the uncer-
tainty budget. In order to cancel out more error contributions the set-up shall be
updated as depicted in Figure 6.22.
Figure 6.22.: Proposal for a future update of the calibration set-up.
If the 1 PPS measurement signal is generated with a traveling 1 PPS generator

integrated into the set-up, the trigger level timing error of the stop channel of the
traveling TIC as well as the trigger timing level error of the receiver’s internal
counter will completely cancel out, because the signal characteristic is the same in
all laboratories, as for the stop signal of the receiver’s internal TIC which is always
provided with the 1 PPS generated by the GPS board.
In this case no 1 PPS IN signal is needed. The 1 PPS provided to the TR is
derived from the external reference frequency. To ensure proper operation of the
receiver, initially this signal has to be synchronized with the 1 PPS REF signal, in
order to keep it in close agreement with the GPS time. The relationship to the local
UTC is then again established by a TIC measurement. If the reference frequency
signal is interrupted during the CCD measurement, the initial synchronization and
the TIC measurement have to be repeated. The reference frequency must be the
local UTC frequency.
The uncertainty contributions ub,3 and ub,4 will reduce to 0.04 ns and ub,5 and
ub,6 will be zero. Since the only reason of the initial synchronization of the 1 PPS
generator is to ensure that the 1 PPS signal provided to the GTR50 does not
deviate too much from the internal GPS signal, as stated above, the trigger level
timing error of the generator has no further influence on the measurements.
To calculate the overall uncertainty the classical root mean square method was
used here (6.12). By now, new rules for the uncertainty evaluation based on statis-
tical models have been formulated and are summarized in BIPM’s ”Guide to the
expression of uncertainty in measurement”(GUM) [90]. This rules could become
the mandatory standard in metrology in the future. Thus, the uncertainty of fu-
ture calibration campaigns should be evaluated according to the GUM. For this
purpose a detailed knowledge of the distribution of the measurement values and
the systematic error contributions is required [91]. In this framework, especially
the so-called nonlinearity effect of the TIC, related to the relationship between
the reference frequency’s zero crossing and the 1 PPS signals, has to be studied
in detail. The expected distribution function of the error contributions is used to
calculate weights for each of the uncertainty contribution.
7. Summary and Outlook
In this work the principles of the GPS technology for time and frequency compar-
isons and the relevant models for the correction of physical influences have been
explained. The improvement of the short term instability due to the usage of cor-
rection data calculated from a worldwide network of reference stations, namely
the IGS network, within a data analysis strategy called precise point positioning
(PPP) has been pointed out.
A highly precise mobile frequency reference based on a passive hydrogen maser
and a state-of-the-art time and frequency transfer GPS receiver was described. The
local evaluation of the equipment and the remote comparison of the passive maser
to PTB’s active masers has shown that the maser can be referenced to PTB’s
primary standards at an averaging time of 104 s using the PPP method.
The PPP method was subject of detailed research by comparing different soft-
ware packages with each other. A correlation was found between the clock offset
estimation and the position estimates. Furthermore, the accuracy of the position
estimates was also found to be related to the instability of the reference frequency
connected to the GPS receivers. This should be examined specifically in a future
analysis, by connecting a GPS receiver to different reference sources with different
short and medium term instability characteristics. A similar effect was not found
in the troposphere estimates. This correction seems more to be related to the
geographic location of the antenna. Also here future examinations are necessary.
By using only the carrier-phase measurements and a fixed position of the an-
tenna frequency transfer on different baselines involving different reference fre-
quency sources was established. The influence of the troposphere model and the
solid Earth tide correction was thus studied in detail. It was found that on relatively
short baselines the Earth tides can be neglected. This is also valid on European
baselines, if the instability is limited by the noise of a commercial caesium clock.
In case of a very long intercontinental baseline the modeling of the Earth tides
is essential for frequency comparisons at the instability level of an active hydro-
gen maser. Surprisingly, a relative simple model for the troposphere is sufficient
in all cases. Future research should verify this and examine the influences of the
models on shorter continental baselines by involving clocks with better instability
(masers).
Different receivers were analyzed in view of long and short term stability. Thereby
the requirements on the equipment for time and frequency transfer were outlined.
For time and frequency comparisons receivers which are capable to operate con-
117
118 7. Summary and Outlook
tinuously without interruptions and which are decoupled from the environmental
temperature are needed. New concepts for future improvements were proposed.
The first GPS common-clock measurement between two remote sites is shown.
The local clocks were canceled out by comparing the GPS data to the measurements
of an optical fiber operated for frequency transfer between the institutes PTB and
IQO at the LUH. The measurement period was relatively short and the receiver
operations were affected by effects which have caused variations at the 200 ps level.
In future experiments a longer measurement period is desirable. Additionally, all
effects which could influence the measurement should be detected, analyzed, and
removed as far as possible in advance.
The calibration of the operational time links between METAS and PTB and
USNO and PTB by means of a traveling GPS receiver was shown. These two
links were of particular interest, because calibrated TWSTFT links were operated
in parallel in both cases. An uncertainty at the nanosecond level at the moment
of calibration was achieved in both calibration campaigns. With a new traveling
calibration set-up, based on a GPS receiver and a time interval counter, systematic
effects in the local UTC reference point determination were canceled out. It has
also been demonstrated that relative GPS calibrations are a good alternative for
cost and labor intensive TWSTFT calibration campaigns. For further reduction
of the uncertainty an update of the calibration set-up is planned. Campaigns
involving the National Physical Laboratory (NPL) in Teddington, GB, the NICT,
and the National Institute of Metrology (NIM) in Bejing, China, are planned.
Appendix
A. Mathematical Description of the Coordinate
Systems
~ECI in the ECI system and a vector V
The transformation between a vector V ~ECEF
in the ECEF system is given by
V~ECEF = RM RS RN RP V
~ECI . (A.1)
Reading from left to right the matrices describe the polar motion, the nutation,
the Earth rotation, and the precession, respectively. Details can be found in the
references [5, 92].
The ECEF system can be expressed with the geodetic coordinates ϕ, θ and h,
representing the geodetic latitude, longitude and height, respectively, taking into
account the flattening of the Earth (Figure A.1).
Figure A.1.: Geometry of the curvature of the Earth ellipsoid.
119
120 Appendix
The transformation between the cartesian coordinates and the geodetic coordi-
nates is given by
x = (N + h) cos (ϕ) cos (θ) ,
y = (N + h) cos (ϕ) sin (θ) , (A.2)
2

z = N 1 − eE + h sin (ϕ) ,
and the inverse transformation reads
y
θ = arctan ,
x !
z
ϕ = arctan p , (A.3)
x2 + y 2 1 − eE 2 NN+h
p
x2 + y 2
h= −N .
cos (ϕ)
In (A.2) and (A.3) eE is the eccentricity of the Earth ellipsoid and the radius of
the curvature can be expressed as
aE
N=p , (A.4)
1 − eE 2 sin2 (ϕ)
with aE the semi-major axis of the Earth ellipsoid. The transformation (A.3) has
to be solved by iteration, because latitude and height are not independent from
each other.
Different realizations of the ECEF system with different parameters aE and eE
are possible and in use. The ECEF reference frame used by the GPS control
segment to predict the ephemeris data broadcasted with the navigation message
is the World Geodetic System 1984 (WGS 84), while the frame of the GLONASS
is the PZ-90 system [4]. To improve the precision ephemeris data provided by a
different source are often employed in post-processing applications. The reference
frames of this ephemeris data can differ from the reference frames broadcasted by
the satellites, e.g. the frame used by the International GNSS Service (IGS) is the
International Terrestrial Reference Frame (ITRF) which is in agreement with the
conventions of the International Earth Rotation Service (IERS).
The general coordinate transformation between reference frames (e.g. WGS84,
PZ-90, ITRF) is given by [93]
~r1 = T1,2 + Ψ1,2 R1,2~r2 , (A.5)
where r1 and r2 are the coordinates in a reference frame 1 and 2, respectively. T1,2
is a translation matrix, Ψ1,2 is a scale factor, and R1,2 is a rotation matrix.
To calculate the azimuth and the elevation of a satellite a transformation to the
local coordinates (xL , yL, zL ) is needed.
B. Mathematical Description of the Satellite Motion 121
The xL -axis is pointing to the north and the yL -axis is pointing to the east. The
transformation from the ECEF coordinates to the local coordinates is therefore
realized by one reflection and two rotations: RL (ϕ, θ) = Sx Ry (π/2 − ϕ)Rz (θ). The
matrices are given by
 
cos(π/2 − ϕ) 0 − sin(π/2 − ϕ)
Ry (π/2 − ϕ) =  0 1 0  ,
sin(π/2 − ϕ) 0 cos(π/2 − ϕ)
    (A.6)
−1 0 0 cos(θ) sin(θ) 0
Sx =  0 1 0 , Rz (θ) = − sin(θ) cos(θ) 0 .
0 0 1 0 0 1
Now any vector V ~ECEF in the ECEF coordinate system can be transformed to the
~L = RL V
receiver antenna’s local coordinates using V ~ECEF .
The vector pointing from the antenna to one satellite s in the EFCF system
at the time of signal emission ts and the receiver’s antenna at the time of signal
reception tr is
ρ~s = (xs (ts ) − xr (tr ), ys (ts ) − yr (tr ), zs (ts ) − zr (tr )) , (A.7)
with the satellite’s coordinates (xs , ys , zs ) and the receiver antenna’s coordinates
(xr , yr , zr ). After transforming (A.7) to the local coordinates the azimuth and
elevation angles are calculated by

ρL,y p
az = arctan , el = arctan ρL,z / ρL,y 2 + ρL,y 2 . (A.8)
ρL,x
B. Mathematical Description of the Satellite

Motion
A short mathematical description of the satellite motion in the orbital planes is
necessary to understand the derivation of relativistic effects (Section 2.4.2) which
have to be applied to the measurements in order to get correct results.
In each orbital plane the satellites are orbiting in the Newtonian potential V
on an ellipse with one focus being the center of mass of the Earth. The two
dimensional instantaneous position within the orbital plane can be described by
Cartesian coordinates (p, q) with the origin at the center of mass of the Earth:
p = rs cos(ν) , q = rs sin(ν) . (B.9)
The angle ν is measured from the perigee along the orbit to the satellite’s posi-
tion [38]. The radius rs is given by
a (1 − e2 )
rs = , (B.10)
1 + e cos(ν)
122 Appendix
where a is the semi-major axis of the ellipse and e is the orbit eccentricity. Due to
the conservation of the total energy the velocity vs can be expressed as
s
2 1
vs = GME − . (B.11)
rs a
G is the constant of gravitation and ME is the mass of the Earth.

The radius rs and the angle ν are connected via equation (B.10), but the time
dependence of this two variables is unknown. In order to relate the satellite position
to a time variable it is useful to define an auxiliary circle with radius a and its
center point coinciding with the center of the ellipse, as depicted in Figure B.2. By
a vertical projection of the satellite to the auxiliary circle a new angle E, called
the eccentric anomaly, can be defined [5].
Figure B.2.: The definition of the eccentric anomaly.
With this new variable the coordinates (p, q) and the radius rs can be rewritten
approximately as
√
p = a cos(E) − ae , q ≈ a 1 − e2 sin(E) , rs = a (1 − e cos(E)) . (B.12)
The equation for p is a direct consequence of the geometry shown in Figure B.2,
while the equation for q is obtained by substituting the equation for p in the ellipse
C. CGGTTS Data 123
√
equation p2 /a2 + q 2 /b2 = 1, with the semi-minor axis given by b = a 1 − e2 . Small
terms containing e have been neglected, because e ≪ 1 for a GNSS satellite.
The components of the velocity vector are given as the time derivative of (B.12)
dp dE dq √ dE
ṗ = = −a sin(E) , q̇ = = a 1 − e2 cos(E) . (B.13)
dt dt dt dt
With the help of the equations (B.12) and (B.13) and the conservation of the
angular momentum the Keplerian equation
r
GME (B.14)
E − e sin(E) = (t − tp )
a3
can be derived [5,38]. The integration constant tp is chosen as the time at which the
satellite is at perigee (E(tp ) = 0). The eccentric anomaly is now associated with
the coordinate time t. Differentiating both sides of equation (B.14) with respect
to the time yields r
dE GME 1
= , (B.15)
dt 3
a 1 − e cos(E)
which can be used in equation (B.13).
The position of the satellite in the orbital plane can be transformed to the
ECI coordinates by three succeeded three-dimensional rotations using the so-called
Keplerian elements, which are angles related to the inclination, the orientation of
the angular momentum, and the direction of the perigee. A detailed description is
given in the reference [5].
C. CGGTTS Data
Below a cutout of a typical CGGTTS file C/A-code file is shown. The header
includes the fixed position which is manually entered to the processing software that
generates the file. INT DLY, CAB DLY, and REF DLY represent the antenna’s
signal delay inside the receiver, the antenna cable delay, and the delay of the 1 PPS
reference signal, respectively. In case of a P3 CGGTTS file the internal delays for
P1 and P2 are listed separately. The relevant column is REFGPS, representing the
offset of the reference timescale with respect to the GPST for each satellite tracked
at a particular epoch.
GGTTS GPS DATA FORMAT VERSION = 01
REV DATE = 2009-06-08
RCVR = GTR50 0806091 1.6.1
CH = 20
IMS = GTR50 0806091 1.6.1
LAB = PTB
X = +3844056.75 m
Y = +709664.09 m
Z = +5023131.73 m
FRAME = ITRF2000
COMMENTS = NO COMMENTS
124 Appendix
INT DLY = -32.2 ns

CAB DLY = 211.4 ns
REF DLY = 45.1 ns
REF = UTC(PTB)
CKSUM = 42
PRN CL MJD STTIME TRKL ELV AZTH REFSV SRSV REFGPS SRGPS DSG IOE MDTR SMDT MDIO SMDI MSIO SMSI ISG CK
hhmmss s .1dg .1dg .1ns .1ps/s .1ns .1ps/s .1ns .1ns.1ps/s.1ns.1ps/s.1ns.1ps/s.1ns
3 FF 55038 000200 780 286 1658 -4249420 -55 +133 -3 9 057 169 +40 91 +14 65 -17 22 2A
11 FF 55038 000200 780 512 2813 +122052 +20 +190 +3 6 077 104 -11 62 -6 30 -18 21 E0
14 FF 55038 000200 780 404 930 +746559 -59 +184 -4 10 112 124 -7 73 -3 49 -7 34 F0
19 FF 55038 000200 780 581 1742 -176682 +10 +219 -5 6 023 95 +8 57 +5 39 +62 18 F8
22 FF 55038 000200 780 317 588 -1947103 +4 +134 -5 8 065 154 +29 86 +11 19 -119 48 2C
32 FF 55038 000200 780 397 2154 -2260066 +76 +211 +14 9 035 126 -21 74 -10 69 -6 29 30
3 FF 55038 001800 780 216 1661 -4249488 -75 +115 -24 11 057 220 +70 105 +16 64 +37 28 5A
11 FF 55038 001800 780 582 2810 +122079 +26 +201 +9 8 077 95 -8 57 -4 26 -8 24 E3
14 FF 55038 001800 780 421 833 +746507 -62 +185 -6 7 112 120 -2 71 -1 48 +5 29 D6
19 FF 55038 001800 780 504 1723 -176675 +21 +212 +6 7 023 105 +12 63 +7 57 -15 26 07
22 FF 55038 001800 780 252 588 -1947102 -7 +127 -17 10 065 189 +46 98 +14 35 -4 25 32
32 FF 55038 001800 780 473 2179 -2259998 +71 +218 +9 7 035 110 -14 65 -8 47 -3 29 2E
11 FF 55038 003400 780 653 2783 +122091 +20 +197 +3 6 077 89 -5 54 -3 42 -4 22 D7
14 FF 55038 003400 780 421 735 +746448 -58 +180 -2 7 112 120 +2 71 +1 45 -10 27 DA
19 FF 55038 003400 780 427 1715 -176664 +14 +208 -1 7 023 119 +18 70 +9 58 -72 35 1B
20 FF 55038 003400 780 293 2335 -763698 +7 +208 -3 10 002 165 -36 90 -13 63 +30 41 1A
32 FF 55038 003400 780 550 2205 -2259940 +63 +216 +0 7 035 99 -10 59 -5 59 -1 29 04
11 FF 55038 005000 780 719 2705 +122103 +14 +192 -3 5 077 85 -3 52 -1 30 +0 15 BA
14 FF 55038 005000 780 403 643 +746393 -52 +178 +4 7 112 125 +7 73 +3 23 -65 27 EA
19 FF 55038 005000 780 351 1713 -176659 +12 +199 -3 10 023 141 +27 80 +12 58 -19 39 35
20 FF 55038 005000 780 362 2368 -763694 +22 +202 +12 10 002 137 -23 78 -10 40 -18 27 31
32 FF 55038 005000 780 627 2231 -2259876 +61 +220 -2 5 035 91 -7 55 -4 55 -11 16 FB
11 FF 55038 010600 780 773 2514 +122113 +20 +186 +3 4 077 83 -1 51 -1 45 +21 17 CE
14 FF 55038 010600 780 370 564 +746327 -69 +166 -13 10 112 134 +13 77 +6 50 +7 35 0C
19 FF 55038 010600 780 276 1713 -176660 +3 +185 -11 10 023 174 +44 93 +14 41 -27 35 2D
20 FF 55038 010600 780 432 2403 -763687 -10 +199 -20 7 002 118 -16 70 -8 48 +23 22 1B
32 FF 55038 010600 780 704 2257 -2259820 +46 +216 -17 4 035 86 -4 52 -2 47 -42 20 0F
11 FF 55038 012200 780 789 2158 +122127 +21 +184 +4 6 077 82 +0 51 +0 45 +24 20 D4
• • •
D. RINEX Data
Below a cutout of a typical 30 s RINEX file is shown. The header contains an
approximate position estimated by the receiver using the C/A-code. Each data
block represents all measurements made at a particular epoch. The initiatory line
gives the year, the month, the day, the hour, and the minute in two digits, followed
by the second. 0.1 microsecond data spacing would be possible in terms of this
standard. The next entry is zero at almost all epochs, except in case of error events,
like loosing the track to all satellites, or power failure, etc. Then the number of
satellites is indicated, followed by the PRN numbers of each tracked satellite. The
measurements are listed one below the other for each satellite. The sequence of
the data is according to the header line ”# / TYPES OF OBSERV”. L1 and L2
are the carrier-phase measurements in cycles of the frequency, C1 is the C/A-code
pseudorange, and P1 and P2 are the P-code pseudoranges.
2.10 OBSERVATION DATA RINEX VERSION / TYPE
GTR50 PTB 2010-05-31 00:48:44 PGM / RUN BY / DATE
COMMENT
MARKER MARKER NAME
MARKER # MARKER NUMBER
OBSERVER PTB OBSERVER / AGENCY
0708522 GTR50 1.6.0 REC # / TYPE / VERS
NAE07020087 Novatel GPS-702 ANT # / TYPE
E. P3 linear combination 125
3844055.5780 709663.0059 5023130.2924 APPROX POSITION XYZ

0.0000 0.0000 0.0000 ANTENNA: DELTA H/E/N
1 1 WAVELENGTH FACT L1/2
5 L1 L2 C1 P1 P2 # / TYPES OF OBSERV
30.000 INTERVAL
2010 5 30 0 0 0.0000000 GPS TIME OF FIRST OBS
2010 5 30 23 59 30.0000019 GPS TIME OF LAST OBS
15 LEAP SECONDS
30 # OF SATELLITES
END OF HEADER
10 5 30 0 0 0.0000000 0 8 16 22 6 24 18 21 3 19
112790530.218 4 87888725.756 4 21463321.661 8 21463321.376 4 21463321.163 4
116637477.760 2 90886354.698 2 22195369.704 7 22195369.854 2 22195367.807 2
107203546.114 4 83535239.961 4 20400148.157 8 20400147.674 4 20400147.881 4
106115706.578 5 82687546.484 5 20193136.504 9 20193137.194 5 20193137.122 5
114193837.721 3 88982224.944 3 21730359.433 8 21730359.562 3 21730358.390 3
114011278.705 3 88839977.115 3 21695621.495 8 21695620.818 3 21695620.527 3
107782948.848 4 83986714.067 4 20510407.817 8 20510407.862 4 20510408.072 4
120201453.898 2 93663466.142 2 22873574.163 7 22873573.971 2 22873571.285 2
10 5 30 0 0 30.0000000 0 8 16 22 6 24 18 21 3 19
112859370.721 4 87942367.687 4 21476421.591 8 21476421.504 4 21476420.146 4
116557613.020 2 90824122.442 2 22180172.006 7 22180172.141 2 22180169.374 2
107174219.830 4 83512388.321 4 20394567.440 9 20394567.841 4 20394567.149 4
106128742.083 5 82697704.017 5 20195617.104 9 20195617.385 5 20195617.481 5
114173693.748 3 88966528.323 3 21726526.563 8 21726525.669 3 21726524.872 3
114070053.305 4 88885775.492 4 21706806.315 8 21706805.974 4 21706804.816 4
107728906.660 4 83944603.260 4 20500124.040 8 20500124.274 4 20500124.232 4
120113220.193 2 93594712.635 2 22856783.811 6 22856783.304 2 22856780.762 2
10 5 30 0 1 0.0000000 0 8 16 22 6 24 18 21 3 19
112928641.658 4 87996345.052 4 21489603.279 8 21489603.489 4 21489602.401 4
116478093.697 2 90762159.337 2 22165040.200 7 22165040.221 2 22165038.261 2
107145246.545 4 83489811.752 4 20389054.069 8 20389054.753 4 20389053.332 4
106142218.019 5 82708204.753 5 20198181.511 8 20198181.304 5 20198181.949 5
114154059.827 3 88951229.181 3 21722790.269 8 21722789.741 3 21722788.275 3
114129098.347 4 88931784.607 3 21718041.712 8 21718042.075 4 21718040.666 3
107675252.493 4 83902794.845 4 20489913.792 8 20489913.483 4 20489914.092 4
120025130.044 2 93526070.987 2 22840019.974 6 22840019.696 2 22840017.768 2
10 5 30 0 1 30.0000000 0 8 16 22 6 24 18 21 3 19
112998369.188 4 88050678.180 4 21502871.866 8 21502871.665 4 21502871.887 4
116398950.641 2 90700489.447 2 22149979.384 7 22149979.779 2 22149977.720 2
107116655.310 4 83467532.861 4 20383613.597 8 20383613.729 4 20383613.199 4
106156161.786 5 82719070.035 5 20200834.664 9 20200835.261 5 20200835.177 5
114134964.979 3 88936350.115 3 21719156.067 8 21719155.891 3 21719155.390 3
114188440.452 3 88978025.206 3 21729333.971 8 21729334.334 3 21729332.493 3
107622016.178 4 83861311.987 4 20479783.502 8 20479783.151 4 20479783.718 4
119937212.768 2 93457564.026 2 22823289.601 7 22823289.718 2 22823288.390 2
10 5 30 0 2 0.0000000 0 8 16 22 6 24 18 21 3 19
• • •
E. P3 linear combination
The ionosphere is a dispersive medium, group velocity vg and phase velocity vp of a
modulated signal are different and two different refractivities ng and np exist which
fulfill the conditions vg ng = c and vp np = c (c is the speed of light in vacuum).
126 Appendix
The relationship between ng and np is given by [5]

−1
dnp
ng = np 1−f /np , (E.16)
df
with f being the frequency. In first order this can be expressed as

dnp
ng = np + f , (E.17)
df
using the approximation formula
(1 − X)−1 = 1 + X − X 2 − ... . (E.18)
The phase refractivity can be represented in first order by

a1
np = 1 + . (E.19)
f2
a1 is related to the total electronic content (TEC) by a1 = −40.3 TEC [5]. Inserting
(E.19) into (E.17) yields
a1
ng = 1 − 2 . (E.20)
f
Together with equation (2.8) one gets
Z Z R
a1 a1 ds A1
δion,p = Ionp = (np − 1) ds = 2
ds = 2
= 2 ,
f f f
Z Z R (E.21)
a1 a1 ds A1
δion,g = Iong = (ng − 1) ds = − 2 ds = − 2
=− 2 ,
f f f
and thus Ionp = −Iong .
In case of a single frequency reception (C/A-code on f1 ) the electron density Ne
can be estimated using a model, and the ionospheric zenith path delay has to be
mapped to the received signal path direction, as in the case of the troposphere.
By using both frequencies f1 and f2 the ionospherical error can be removed from
the observation equations in first order. Two linear combinations consisting of the
square of the two frequencies and the phase and group ionospherical delays (E.21)
can be formed:
A1 A1
f1 2 Ionp (f1 ) − f2 2 Ionp (f2 ) = f1 22
− f2 2 2 = 0 ,
f1 f2
(E.22)
A 1 A1
f1 2 Iong (f1 ) − f2 2 Iong (f2 ) = −f1 2 2 + f2 2 2 = 0 .
f1 f2
F. Brief Derivation of Relativistic Corrections 127
Now a combination of the P1 and P2 code measurements with respect to the

satellite s according to equation (2.4) can be constructed

f1 2 P1s − f2 2 P2s = f1 2 − f2 2 [ρs (ts , tr ) − (δtr − δts )]
(E.23)
+ f1 2 ∆P 1 + f2 2 ∆P 2 ,
where the ionosphere error vanishes in first order. It is obvious that a scaling
with 1/(f1 2 − f2 2 ) yields a pseudorange formed by the measurements on the two
frequencies
f1 2 P1 − f2 2 P2
P3 = . (E.24)
f1 2 − f2 2
In the above equation the satellite index s has been omitted.
F. Brief Derivation of Relativistic Corrections

The proper time increment dτs of a satellite in the Newton gravitational potential
V = −GME /rs (F.25)
is approximately given by [38]

2 2 (V − Φ0 )
− (cdτs ) = − 1 + (cdt)2
c2
(F.26)
2V
+ 1− 2 drs 2 + rs 2 dϑs 2 + rs 2 sin2 (ϑs )dφs 2 .
c
c denotes the speed of light. (rs , ϑs , ϕs ) are the spherical polar coordinates of the
satellite in the ECI reference frame. The coordinate time t is measured on the
Earth geoid. Thus, the potential V − Φ0 is the effective potential on the rotating
geoid, with Φ0 including a term related to the angular velocity of the Earth. It can
be derived from a metric given in rotating ECI coordinates. This is extensively
explained in reference [38].
The square of the satellite’s velocity can be written as
drs 2 + rs 2 dϑs 2 + rs 2 sin2 (ϑs )dφs 2

vs 2 = . (F.27)
dt2
Equation (F.26) can be written as

2 2 (V − Φ0 ) 2V vs 2
dτs = 1 + − 1− 2 dt2
c2 c c2
(F.28)
2 (V − Φ0 ) vs 2
≈ 1+ − 2 dt2 .
c2 c
128 Appendix
The very small term including 1/c4 can be neglected and the square root can
be taken approximately with the help of the binomial expansion (A + X)n =
An +nAn−1 X+..., where the constant A = 1, the variable X = 2(V −Φo )/c2 −vs 2 /c2 ,
and n = 1/2, since the gravitational potential and the square of the satellite’s
velocity are small compared to the square of the speed of light:

V − Φ0 vs 2
dτs = 1 + − 2 dt . (F.29)
c2 2c
To solve for the coordinate time increment the approximation formula (E.18) is
used:
V − Φ0 vs 2
dt = 1 − + 2 dτs . (F.30)
c2 2c
By assuming Keplerian orbits, equation (B.11) can be used for the velocity. Af-
ter applying (F.25) and using 3GME /2ac2 − 4GME /2ac2 = −GME /2ac2 , equa-
tion (F.30) reads

3GME Φ0 2GME 1 1
dt = 1 − + 2 − − dτs . (F.31)
2ac2 c c2 a rs
To get the correction that has to be applied to the satellite clock by a user on
the Earth this has to be integrated along the path of the satellite parameterized
with its proper time τs . The first, the second, and the third term on the right hand
side are constants and 3GME /2ac2 − Φ0 /c2 = −4.467 · 10−10 [38].
The last term is related to the eccentricity of the orbit. It can be seen by using
the radius from equation (B.12):
Z Z
2GME 1 1 2GME e cos(E)
∆te = 2
− dτs = 2
dτs . (F.32)
c rs a c 1 − e cos(E)
Path Path
The integration can be done by taking into account that the proper time and the
coordinate time are approximately equal in the Newtonian limit: dτs ≈ dt [38].
The result of the integration is
√
2 GME a
∆te = e sin(E) + const . (F.33)
c2
The integration constant can be neglected, because it is a just constant clock offset
which can be removed by the clock steering of the master control station. With
the help of the equations (B.12), (B.13), and (B.15) the product of the position
and the velocity vector of the satellite can be written as
p
~rs · ~vs = pṗ + q q̇ = GME a e sin(E) , (F.34)
G. Estimation of the Ephemerides of Sun and Moon 129
and the eccentricity correction can be expressed as [4, 38]

2
∆te = ~rs · ~vs . (F.35)
c2
It can reach a maximum of 70 ns [4].
A second effect that has to be corrected by the user is the so-called Sagnac effect.
It is related to the rotation of the Earth. The signal propagation time is given by
∆t = tR − tT , with the reception time tR and the transmitting time tT . During this
interval the displacement of the receiver due to the Earth rotation is ~v∆t, with ~v
related to the angular velocity of the Earth. Neglecting all previously discussed
effects, the signal propagation time is given by [94]
c2 (∆t)2 = (~rR + ~v ∆t − ~rT )2 , (F.36)
with ~rR and ~rT the receiver’s and satellite’s position at reception and transmission
time, respectively. After defining R ~ = ~rR − ~rT and neglecting terms that are
quadratic in ~v /c is becomes
~ 2 + 2~v · R∆t
c2 (∆t)2 ≈ R ~ . (F.37)
The square root can be approximately solved by

~
~v · R∆t
c (∆t) ≈ R + , (F.38)
R
~ An approximate solution for ∆t leads to
with R being the absolute value of R.
~
R ~v · R
∆t ≈ + 2 . (F.39)
c c
With the angular velocity vector of the Earth ~ωE and ~v = ω
~ E × ~rR the final Sagnac
correction is given by [94]
1 ~ .
∆tS = ~ωE · ~rR × R (F.40)
c2
Although it does not look like a relativistic effect at first sight, it can also be derived
from a metric in a rotating reference frame [38].
G. Estimation of the Ephemerides of Sun and

Moon
The positions of the Moon and the Sun in their orbital planes are calculated with
the help of equation (B.14). Also the Sun is treated as a body orbiting the Earth
130 Appendix
in two dimensional coordinates according to the equations (B.12) and (B.13) [5].
Equation (B.14) can be rewritten as
E − e sin(E) = M . (G.41)
M is called the mean anomaly. According to IERS conventions [5] the mean
anomaly of the Moon can be estimated by
MMoon = 134.96340251◦ + 1717915923.2178′′ · T

(G.42)
+ 31.8792′′ · T 2 + 0.051635′′ · T 3 + 0.0002447′′ · T 4 .
It is given in angular dimensions. The time T is measured dimensionless in Julian

centuries with the origin at the noon of January 1st 2000. Each Julian century
consists of 36525 days. The relationship between T and MJD is
MJD − 51544.5
T = . (G.43)
36525
The mean anomaly of the Sun is also approximated with a polynomial of order
four:
MSun = 357.52910918◦ + 129596581.0481′′ · T
(G.44)
+ 0.5532′′ · T 2 + 0.000136′′ · T 3 + 0.0001149′′ · T 4 .
Equation G.41 has to be solved by iteration in order to get the eccentric anomaly
E at each epoch. This can be done e.g. by starting with E0 = M and inserting
this into the next step E1 = M + e sin(E0 ). E1 is then used as the function value
of the sinus function in the next step. In general, the iteration reads En+1 = M +
e sin(En ), with n being the numbers of previous iterations. This is repeated until
the difference |En+1 − En | reaches a lower bound or until E is solved with sufficient
numerical precision. The eccentricity of the Moon has to be set as eMoon = 0.05490
and the eccentricity of the Sun is given by
eSun = 0.016709114 − 0.000042052 · T . (G.45)
The eccentric anomalies of the Moon and the Sun, EMoon and ESun , are used
to calculate the coordinates of these bodies in their orbital plane (B.12). The
semimajor axis of the Moon and the Sun are aMoon = 384401 km and aSun =
1.495979005 · 108 km, respectively.
Subsequently, the position vectors of the Moon ~rMoon = (px,Moon , qy,Moon , 0) and
the Sun ~rSun = (px,Sun , qy,Sun , 0) are transformed to the ECI coordinates with the
help of so-called Keplerian elements
~rECI = Rx (−κ)Rz (−ΩMoon )Rx (−ζMoon )Rz (−η)~r . (G.46)

G. Estimation of the Ephemerides of Sun and Moon 131
Rx and Rz are rotation matrices around the x-axis and the z-axis, respectively. κ is
the angle between the ecliptic plane and the equatorial plane and can be estimated
in angular dimensions by
κ = 84381.448′′ − 46.815 · T − 0.00059 · T 2 + 0.001813 · T 3 . (G.47)
Ω is called the mean longitude of the ascending node of the Moon. The approxi-
mation formula reads
ΩMoon = 125.04455501◦ + 6962890.2665′′ · T
(G.48)
+ 7.4722′′ · T 2 + 0.007702′′ · T 3 + 0.00005939′′ · T 4 .
Since the Sun is ”orbiting” in the ecliptic plane, ΩSun is zero. The inclination of
the Moon is ζMoon = 5.145396◦. For the Sun it is also zero. η is called argument of
perigee. For the Moon it is calculated from the relation

tan(FMoon
ηMoon = arctan − νMoon , (G.49)
ζMoon
where FMoon is called mean longitude of the Moon and given by
FMoon = 93.27209062◦ + 173952762.8478′′ · T

(G.50)
+ 12.7512′′ · T 2 + 0.001037′′ · T 3 + 0.00000417′′ · T 4 ,
and the angle νMoon is calculated by comparing the equations (B.9) and (B.12).
The argument of perigee of the Sun is calculated from
ηSun = νSun − FMoon − ΩMoon + D , (G.51)
with the coefficient

D = 297.85019547◦ + 1602961601.209′′ · T
(G.52)
+ 6.3706′′ · T 2 + 0.006593′′ · T 3 + 0.00003169′′ · T 4 .
The transformation of (G.46) to the ECEF coordinates is done according to

equation (A.1). In this work, nutation and polar motion were neglected. The
precession matrix consists of three rotations
RP = Rz (α)Ry (χ)Rz (β) . (G.53)
The angles are approximated by
α = 2306.2181′′ · T + 1.09468′′ · T 2 + 0.018203′′ · T 3 ,

χ = 2004.3109′′ · T − 0.42665′′ · T 2 − 0.041833′′ · T 3 , (G.54)
β = 2306.2181′′ · T + 0.30188′′ · T 2 + 0.017998′′ · T 3 .
132 Appendix
The Earth rotation matrix is a rotation around the z-axis
RS = Rz (ξ) , (G.55)
with the angle
ξ = 280.46061837◦ + 360.98564736629◦ · 36525 · T

0.0000387933◦ (G.56)
+ · T4 .
38710000
H. BIPM’s Circular T
In the following a cutouts of the first section of one of BIPM’s Circular T are
shown. After the header information, on the left hand side the time laboratories
and their locations are given. The next five columns represent the UTC-UTC(k)
values. The last three columns are the statistical uncertainty ”uA”, the systematic
uncertainty ”uB”, and the overall uncertainty ”u’.
CIRCULAR T 273 ISSN 1143-1393
2010 OCTOBER 11, 12h UTC
BUREAU INTERNATIONAL DES POIDS ET MESURES
ORGANISATION INTERGOUVERNEMENTALE DE LA CONVENTION DU METRE
PAVILLON DE BRETEUIL F-92312 SEVRES CEDEX TEL. +33 1 45 07 70 70 FAX. +33 1 45 34 20 21 tai@bipm.org
1 - Coordinated Universal Time UTC and its local realizations UTC(k). Computed values of [UTC-UTC(k)]
and uncertainties valid for the period of this Circular.
From 2009 January 1, 0h UTC, TAI-UTC = 34 s.
Date 2010 0h UTC AUG 31 SEP 5 SEP 10 SEP 15 SEP 20 SEP 25 SEP 30 Uncertainty/ns Notes
MJD 55439 55444 55449 55454 55459 55464 55469 uA uB u
Laboratory k [UTC-UTC(k)]/ns
AOS (Borowiec) 10.4 8.0 8.5 8.1 7.6 5.5 2.6 0.5 5.2 5.2
APL (Laurel) 5.3 -4.9 - - - -8.8 -8.3 1.5 5.2 5.4
AUS (Sydney) 80.1 95.1 106.8 117.7 118.3 126.4 146.3 0.4 5.1 5.1
BEV (Wien) 11.7 8.3 12.2 5.0 3.0 -0.1 -0.6 1.5 3.2 3.6
BIM (Sofiya) -6594.3 -6583.0 -6576.8 -6564.9 -6558.5 -6542.4 -6540.6 2.0 7.1 7.4
BIRM (Beijing) -11355.3 -11396.1 -11437.0 -11482.7 -11526.1 -11563.6 -11601.2 2.0 20.0 20.1
BY (Minsk) 51.4 54.5 60.6 65.8 -33.7 -29.9 -27.8 2.0 7.1 7.4 (1)
CAO (Cagliari) -4596.0 -4598.4 -4601.9 -4615.7 -4625.7 -4643.0 -4660.6 1.5 7.1 7.2
CH (Bern) 13.5 12.0 9.2 9.1 9.0 9.4 7.5 0.6 1.6 1.7
CNM (Queretaro) -3.7 4.5 11.7 20.3 18.1 15.0 16.9 2.5 5.1 5.7
• • •
PL (Warszawa) 25.7 17.4 18.0 11.3 -5.1 -3.8 -12.6 1.5 5.1 5.3
PTB (Braunschweig) 2.1 0.9 2.0 1.2 0.5 -0.5 -2.3 0.2 1.2 1.3
ROA (San Fernando) -10.9 -6.5 -0.8 3.9 8.4 13.7 17.3 0.6 5.1 5.1
SCL (Hong Kong) -5.5 -3.1 -3.4 8.2 4.9 14.5 24.9 3.0 10.1 10.5
SG (Singapore) 21.3 12.2 5.5 -0.4 9.6 20.9 31.0 0.4 5.2 5.2
SIQ (Ljubljana) -189.7 -196.9 -222.1 -256.1 -238.8 -207.0 -231.6 5.0 20.0 20.6
SMD (Bruxelles) -2.8 -2.6 3.3 5.8 5.2 7.5 5.7 1.5 19.8 19.8
SMU (Bratislava) -188.9 -178.5 -165.8 -157.5 -145.9 -148.2 -139.5 1.5 20.0 20.1
SP (Boras) 15.6 18.7 20.8 17.1 11.6 5.3 -2.6 0.6 1.6 1.7
SU (Moskva) 4.5 1.3 2.0 0.4 -0.7 -2.7 -6.2 1.5 5.0 5.3
• • •
After the UTC-UTC(k) section information about the offset of GPS and GLONASS
with respect to UTC and the steering corrections derived from the primary fre-
quency standards are given.
H. BIPM’s Circular T 133
In the last section of the Circular T, the links, their uncertainties, the type of
calibration, and the date of calibration are listed:
The calibration type of the link is indicated as: GPS EC for GPS equipment
calibration; TW EC for two-way equipment calibration; LC (technique) for a link
calibrated using ’technique’; BC (technique) for a link calibrated using ’technique’
to transfer a past equipment calibration through a discontinuity of link operation.
DIC is used for direct internal calibration.
• • •
The calibration dates indicate: the most recent calibration results for the two laboratories in the case of EC and
the most recent calibration of the link in the case of LC and BC,
NA stands for not available, in this case estimated values are provided
Link Type uA/ns uB/ns Calibration Type Calibration Dates
AOS /PTB TWSTFT 0.5 5.0 BC(GPS MC) 2008 May

APL /PTB GPS MC 1.5 5.0 GPS EC/GPS EC 2003 Dec/2006 Sep
AUS /PTB GPSPPP 0.3 5.0 LC(GPS MC) 2009 Nov
BEV /PTB GPS MC 1.5 3.0 BC(TWSTFT) 2008 Jan
BIM /PTB GPS MC 2.0 7.0 GPS EC/GPS EC 2007 Nov
BIRM/PTB GPS MC 2.0 20.0 NA /GPS EC NA /2006 Sep
BY /PTB GPS MC 2.0 7.0 GPS EC/GPS EC 2008 Jun/2006 Sep
CAO /PTB GPS MC 1.5 7.0 GPS EC/GPS EC 2004 Nov/2006 Sep
CH /PTB TWSTFT 0.6 1.0 LC(TWSTFT)/BC(GPS PPP) 2008 Sep/2009 Aug
CNM /PTB GPS MC 2.5 5.0 BC(GPS SC) 2008 May
CNMP/PTB GPS MC 3.0 5.0 GPS EC/GPS EC 2004 May/2006 Sep
DLR /PTB GPSPPP 0.3 5.0 GPS EC/GPS EC 2007 Feb/2004 Aug
DMDM/PTB GPS MC 2.0 7.0 GPS EC/GPS EC 2007 Jan/2006 Sep
DTAG/PTB GPSPPP 0.3 10.0 LC(GPS MC) 2009 Jul
EIM /PTB GPS MC 3.5 5.0 GPS EC/GPS EC 2007 May/2003 Aug
HKO /PTB GPS MC 2.5 5.0 GPS EC/GPS EC 2004 Sep/2006 Sep
IFAG/PTB GPSPPP 0.3 5.0 GPS EC/GPS EC 2003 Jun/2004 Aug
IGNA/PTB NA
INPL/PTB NA
INTI/PTB GPS MC 4.0 20.0 NA /GPS EC NA /2006 Sep
Bibliography
[1] F. Riehle. Frequency standards: Basics and Applications. Wiley-VCH, Wein-
heim, 2005.
[2] F. Arias and A. Bauch. Metrology of Time and Frequency. Handbook of

Metrology, Wiley-VCH, Weinheim, 2010.
[3] J. Levine. Introduction to time and frequency metrology. Review of Scientific

Instruments, 70(6):2567–2596, 1999.
[4] E. D. Kaplan. Understanding GPS Principles and Applications. Artech House,

Norwood, MA, 1996.
[5] G. Xu. GPS Theory, Algorithms and Applications. Springer, New York, 2003.
[6] A. Bauch and T. Heindorff. Die SI-Basiseinheit ”Sekunde”. PTB-Mitteilungen,

112:291–298, 2003.
[7] A. Bauch, P. Hetzel, and D. Piester. Zeit- und Frequenzverbreitung mit

DCF77: 1959-2009 und darüber hinaus. PTB-Mitteilungen, 119(3):11–34,
2009.
[8] A. Bauch. Caesium atomic clocks: function, performance and applications.

Measurement Science and Technology, 14:1159–1173, 2003.
[9] D. W. Allan. Picosecond Time Difference Measurement System. In Proc. 29th

Annu. Symp. on Freq. Contrl., pages 404–411, May 1975.
[10] T. J. Quinn. The BIPM and the accurate measurement of time. Proceedings
of the IEEE, 79(7):894–905, 1991.
[11] H. Hellwig, D. W. Allan, S. R. Stein, and K. A. Prichard. Transcontinental

and Intercontinental Portable Clock Time Comparisons. IEEE Transactions
on Measurement and Instrumentation, 27(1):65–68, 1978.
[12] K. Horiai. Accuracy of time comparison derived from reception of LORAN-C

and Global Positioning System time signals. Radio Science, 27(5):545–551,
1992.
134
BIBLIOGRAPHY 135
[13] Yu. A. Fedorov and S. B. Pushkin. Comparing the national standards of time
and frequency of the USSR, Czechoslovakia, and East Germany. Measurement
Techniques, 22(10):1193–1195, 1979.
[14] J. Levine. Time and frequency distribution using satellites. Rep. Prog. Phys.,
65:1119–1164, 2002.
[15] J. Kouba and P. Heroux. Precise Point Positioning Using IGS Orbit and Clock
Products. GPS Solutions, 5(2):12–28, 2002.
[16] H. Schnatz, O. Terra, K. Predehl, T. Feldmann, T. Legero, B. Lipphardt,

U. Sterr, G. Grosche, T. W. Hänsch, R. Holzwarth, T. Udem, Z. Lu, L. Wang,
W. Ertmer, J. Friebe, A. Pape, E. M. Rasel, M. Riedmann, and T. Wübbena.
Phase-coherent frequency comparison of optical clocks using a telecommu-
nication fiber link. IEEE Transactions on Ultrasonics, Ferroelectrics, and
Frequency Control, 57(1):175–181, 2010.
[17] H. Esteban, J. Palacio, F.J. Galindo, T. Feldmann, A. Bauch, and

D. Piester. Improved GPS based time link calibration involving ROA and
PTB. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Con-
trol, 57(3):714–720, 2010.
[18] D. B. Sullivan, D. W. Allan, D. A. Howe, and F. L. Walls. NIST Technical

Note 1337, Characterization of Clocks and Oscillators, March 1990, online at
http://www.pfcosc.com/appnotes/Characterization of clocks and
oscillators.pdf.
[19] W. Riley. Handbook of Frequency Stability Analysis, NIST Special Publica-

tion 1065, online at tf.nist.gov/general/pdf/2220.pdf, as of November
2010.
[20] D. W. Allan. The statistics of Atomic Frequency Standards. Proc. IEEE,

45(2):221–230, 1966.
[21] M. A. Weiss, G. Petit, and Z. Jiang. A comparison of GPS common-view time

transfer to all-in-view. In Proc. Joint IEEE Intl. Freq. Control Symposium and
PTTI Systems and Applications Meeting, pages 324–328, Vancouver, Canada,
August 2005.
[22] T. Gotoh. Improvement GPS Time Links in Asia with All in View. In Proc.
Joint IEEE Intl. Freq. Control Symposium and PTTI Systems and Applica-
tions Meeting, pages 707–711, Vancouver, Canada, August 2005.
[23] M. Rost, A. Bauch, J. Becker, T. Feldmann, D. Piester, T. Polewka, D. Sibold,

and E. Staliuniene. PTB’s Time and Frequency Activities in 2008 and 2009.
136 BIBLIOGRAPHY
In Proc. 41th PTTI Systems and Applications Meeting, pages 197–213, Santa
Ana Pueblo, New Mexico, USA, November 16-19 2009.
[24] D. Piester, M. Fujieda, M. Rost, and A. Bauch. Time Transfer Through
Optical Fibers (TTTOF): First Results of Calibrated Clock Comparison. In
Proc. 41th PTTI Systems and Applications Meeting, pages 89–99, Santa Ana
Pueblo, New Mexico, USA, November 16-19 2009.
[25] Global Positioning System Standard Positioning Service Signal Specification,
2nd Edition, June 1995, online at
http://www.pnt.gov/public/docs/1995/signalspec1995.pdf.
[26] R. J. Meinhold and N. D. Singpurwalla. Understanding the Kalman Filter.
The American Statistician, 37(2):123–127, 1983.
[27] P. Defraigne, P. Uhrich, G. Petit, and W. Aerts. Requirements on GNSS re-
ceivers from the perspective of timing applications. In Proc. EFTF, Noordwijk,
NL, April 13-16 2010.
[28] J. Levine. Time transfer using multi-channel GPS receivers. IEEE Trans-
actions on Ultrasonics, Ferroelectrics, and Frequency Control, 46(2):392–398,
1999.
[29] D. W. Allan and C. Thomas. Technical Directives for Standardization of GPS
Time Receiver Software. Metrologia, 31:69–79, 1994.
[30] W. Gurtner. RINEX The Receiver Independent Exchange Format, Version
2.10. Astronomical Institute University of Berne, 2010.
[31] P. Defraigne and G. Petit. Time transfer to TAI using geodetic receivers.
Metrologia, 40:184–188, 2003.
[32] J. Saastamoinen. Atmospheric corrections for the troposphere and
stratosphere in radio ranging. Geophysical Monograph, 15:247–251, 1972.
[33] H. S. Hopfield. Improvements in the tropospheric refraction for range mea-
surements. Phil. Trans. R. Soc., 194:341–352, 1979.
[34] ASEN 5090 Error Models
http://www.colorado.edu/ASEN/asen5090/assignments/
ASEN5090 ErrorProject2010.pdf.
[35] W. Lewandowski, J. Azoubib, A. G. Gevorkyan, P. P. Bogdanov, W. J. Kle-
paynski, M. Miranian, J. Danaher, N. B. Koshelyaevsky, and D. W. Allan. A
Contribution to the Standardization of GPS and GLONASS Time Transfers.
In Proc. 28th PTTI Systems and Applications Meeting, pages 367–386, Reston,
VA, USA, December 03-05 1996.
BIBLIOGRAPHY 137
[36] S. Pireaux, P. Defraigne, L. Wauters, N. Bergeot, Q. Baire, and C. Bruyninx.

Higher-order ionospheric effects in GPS time and frequency transfer. GPS
Solutions, 14:267–277, 2010.
[37] P. Panek. private communication.
[38] N. Ashby. Relativity in the Global Positioning System. Living Reviews in
Relativity, 6(1), 2003.
[39] M. A. Lombardi and A. N. Novick. Effects of the rooftop environment on GPS
time transfer. In Proc. 38th PTTI Systems and Applications Meeting, pages
449–466, Reston, VA, USA, December 07-09 2006.
[40] A. J. van Dierendonck, P. Fenton, and T. Ford. Theory and Performance of
Narrow Correlator Spacing in a GPS Receiver. Navigation, 39(3):265–283,
1992.
[41] N. Penna, M. Bos, T. Baker, and H.-G. Scherneck. Assessing the accuracy of
predicted ocean tide loading displacement values. Journal of Geodesy, 82:893–
907, 2008.
[42] R. D. Ray. A global ocean tide model from topex/poseidon altimetry: Got99.2.
NASA Technical Memorandum, 209478, 1999.
[43] http://www.oso.chalmers.se/~loading.
[44] R. Pı́riz, G. Cerretto, A. Perrucca, and P. Tavella. Evaluation of the Time and
Frequency Transfer Capabilities of a Network of GNSS Receivers located in
Timing Laboratories. In Proc. 41th PTTI Systems and Applications Meeting,
pages 559–574, Santa Ana Pueblo, New Mexico, USA, November 16-19 2009.
[45] G. Beutler, M. Rothacher, S. Schaer, T. A. Springer, J. Kouba, and R. E.
Neilan. The International GPS Service (IGS): An interdisciplinary service in
support of Earth sciences. Advances in Space Research, 23(4):631 – 653, 1999.
[46] http://igscb.jpl.nasa.gov/network/netindex.html.
[47] J. Kouba. A Guide to using International GNSS Service (IGS) Products,
online at http://igscb.jpl.nasa.gov/components/usage.html, as of May
2009.
[48] http://igscb.jpl.nasa.gov/organization/centers.html.
[49] http://igscb.jpl.nasa.gov/components/formats.html.
[50] J. Ray and K. Senior. Geodetic techniques for time and frequency comparisons
using GPS phase and code measurements. Metrologia, 42:215–232, 2005.
138 BIBLIOGRAPHY
[51] W. Lewandowski, D. Matsakis, G. Panfilo, and P. Tavella. The evaluation of

uncertainties in [UTC-UTC(k)]. Metrologia, 43:278–286, 2006.
[52] W. Lewandowski and Z. Jiang. Use of GLONASS at the BIPM. In Proc. 41th
PTTI Systems and Applications Meeting, pages 197–213, Santa Ana Pueblo,
New Mexico, USA, November 16-19 2009.
[53] Z. Jiang and G. Petit. Combination of TWSTFT and GNSS for accurate UTC
time transfer. Metrologia, 46:305–314, 2009.
[54] G. Panfilo and E. Arias. Algorithms for International Atomic Time. IEEE
Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 57(1):140
–150, 2010.
[55] C. Thomas and J. Azoubib. TAI computation: study of an alternative choice

for implementing an upper limit of clock weights. Metrologia, 33:227–240,
1996.
[56] D. Piester, A. Bauch, L. Breakiron, D. Matsakis, B. Blanzano, and

O. Koudelka. Time transfer with nanosecond accuracy for the realization
of International Atomic Time. Metrologia, 45(2):185–198, 2008.
[57] W. Lewandowski and L. Tisserand. Relative characterization of GPS

time equipment delays at the OP, PTB, AOS, USNO and IT, online
at http://www.bipm.org/utils/common/pdf/rapportBIPM/2010/03.pdf,
March 2010.
[58] Vremya-Ch. Passive Hydrogen Maser Frequency and Time Standard VCH-
1006 Maintance and service manual.
[59] Dicom. GTR50 Time & Frequency Transfer Receiver, Operating Instructions.
[60] P. Panek and I. Prochazka. Time interval measurement device based on surface
acoustic wave filter excitation, providing 1 ps precision and stability. Review
of Scientific Instruments, 78(094701), 2007.
[61] J. Friebe, A. Pape, M. Riedmann, K. Moldenhauer, T. Mehlstäubler, N. Reh-

bein, C. Lisdat, E. M. Rasel, W. Ertmer, H. Schnatz, B. Lipphardt, and
G. Grosche. Absolute frequency measurement of the magnesium intercombi-
nation transition 1 S0 → 3 P1 . Phys. Rev. A, 78(3):033830, September 2008.
[62] S. Diddams. Optical frequency combs: Introduction, sources and applica-

tions. In Frequency Control Symposium, 2009 Joint with the 22nd European
Frequency and Time forum. IEEE International, page xv, April 2009.
BIBLIOGRAPHY 139
[63] T. Gotoh, M. Fujieda, and J. Amagai. Comparison Study of GPS Carrier

Phase and Two-Way Satellite Time and Frequency Transfer. In Frequency
Control Symposium, 2007 Joint with the 21st European Frequency and Time
Forum. IEEE International, pages 1188–1193, Geneva, Switzerland, May 29 -
June 1 2007.
[64] G. Petit. The TAIPPP pilot experiment. In Proc. EFTF - IEEE IFCS Joint
Conference, pages 116–119, Besançon, France, April 20-24 2009.
[65] T. Feldmann, D. Piester, A. Bauch, and T. Gotoh. GPS carrier phase time and
frequency transfer with different versions of precise point positioning software.
In Proc. 40th PTTI Systems and Applications Meeting, pages 403–414, Reston,
VA, USA, December 01-04 2008.
[66] T. Feldmann, D. Piester, H. Esteban, J. Palacio, F. J. Galindo, T. Gotoh,

H. Maeno, U. Weinbach, and S. Schön. GPS carrier phase and precise point
positioning timescale comparisons using different software packages. In Proc.
EFTF - IEEE IFCS Joint Conference, pages 120–125, Besançon, France, April
20-24, 2009.
[67] R. Dach, U. Hugentobler, T. Schildknecht, L.-G. Bernier, and G. Dudle. Pre-

cise continuous time and frequency transfer using GPS carrier phases. In Proc.
Joint IEEE Intl. Freq. Control Symposium and PTTI Systems and Applica-
tions Meeting, pages 329–336, Vancouver, Canada, August 2005.
[68] T. Gotoh, T. Hobiger, R. Ichikawa, T. Feldmann, and D. Piester. Application

of ray-traced troposphere delays to GPS time transfer. In Proc. 52nd Meeting
of the Japan Society for Aeronautical and Space Sciences, pages 1322–1326,
Hyogo, Japan, November 05-07 2008.
[69] F. Lahaye. private communication.
[70] F. Webb and J. Zumberge. An introduction to GIPSY/OASIS-II, Precision

Software for the Analysis of Data from the Global Positioning System, Jet
Propulsion Laboratory, Pasadena, CA, JPL D-11088, July 1997.
[71] R. Dach, U. Hugentobler, P. Fridez, and M. Meindl. Bernese GPS Software

Version 5.0, http://www.bernese.unibe.ch/docs/DOCU50.pdf.
[72] U. Weinbach and S. Schön. On the Correlation of Tropospheric Zenith Path

Delay and Station Clock Estimates in Geodetic GNSS Frequency Transfer. In
Proc. EFTF, Noordwijk, NL, April 13-16 2010.
[73] U. Weinbach and S. Schön. Stability analysis of the hardware delays of geodetic
GPS receivers for clock comparisons. In 2ndnd International Colloquium -
140 BIBLIOGRAPHY
Scientific and Fundamental Aspects of the Galileo Programme, Padova, Italy,

October 14-16 2010.
[74] H. Esteban, J. Palacio, F.J. Galindo, and J. Garate. GPS receiver performance
test at ROA. In Proc. 40th PTTI Systems and Applications Meeting, pages
403–414, Reston, VA, USA, December 01-04 2008.
[75] T. Feldmann, A. Bauch, D. Piester, A. Stefanov, L.-G. Bernier, C. Schluneg-

ger, and K. Liang. On improved GPS based link calibration of the time links
between METAS and PTB. In Proc. EFTF, Noordwijk, NL, April 13-16 2010.
[76] G. Petit and Z. Jiang. GPS All in View time transfer for TAI computation.
Metrologia, 45(1):35–45, 2008.
[77] G. Wübbena, M. Schmitz, G. Boettcher, and C. Schumann. Absolute GNSS

antenna calibration with a robot: repeatability of phase variations, calibration
of GLONASS and determination of carrier-to-noise pattern. In IGS 2006
Workshop Proceedings, IGS Central Bureau, 2008.
[78] Stanford Research Systems (SRS). SR620 Operating Manual and Program-
ming Reference.
[79] W. Lewandowski and C. Thomas. GPS Time transfer. Proceedings of the

IEEE, 79(7):991–1000, 1991.
[80] I. Prochazka and P. Panek. Nonlinear effects in the time measurement device
based on surface acoustic wave filter excitation. Review of Scientific Instru-
ments, 80(7):076102 –076102–3, July 2009.
[81] J. R. Ray and K. Senior. Temperature Sensitivity of Timing Measurements

Using Dorne Margolin Antennas. GPS Solutions, 5:24–30, 2001.
[82] D. Matsakis, K. Senior, and P. Cook. Comparison of continuously filtered

GPS carrier-phase time and frequency transfer with independent daily GPS
carrier-phase solutions and with two-eay satellite time transfer. In Proc. 33th
PTTI Systems and Applications Meeting, pages 63–88, Long Beach, CA, USA,
November 27-29 2001.
[83] G. Petit, C. Thomas, Z. Jiang, P. Uhrich, and F. Taris. Use of GPS ASHTECH
Z12T receivers for accurate time and frequency comparisons. IEEE Transac-
tions on Ultrasonics, Ferroelectrics, and Frequency Control, 46(4):941–949,
1999.
[84] http://www.bipm.org/utils/common/TimeCalibrations/GPS P3
/P3 calib PTB.pdf.
BIBLIOGRAPHY 141
[85] http://www.bipm.org/utils/common/TimeCalibrations/GPS P3
/P3 calib CH.pdf.
[86] J. Plumb, K. M. Larson, J. White, and E. Powers. Absolute calibration of

a geodetic time transfer system. IEEE Transactions on Ultrasonics, Ferro-
electrics, and Frequency Control, 52(11):1904–1911, 2005.
[87] A. Bauch, D. Piester, P. Whibberley, B. Blanzano, O. Koudelka, J. Achkar,

D. Rovera, E. Kroon, E. Dierikx, L. Lorini, F. Cordara, and C. Schlunegger.
Results of the 2008 TWSTFT calibration of seven european stations. In Fre-
quency Control Symposium, 2009 Joint with the 22nd European Frequency and
Time forum. IEEE International, pages 1209 –1215, april 2009.
[88] R. Stecher. Messung von Zeit und Frequenz. Verlag Technik, Berlin, 1990.
[89] D. Piester, T. Feldmann, A. Bauch, M. Fujieda, and T. Gotoh. Concept for

an Accurate Calibration of Long Baseline Two-Way Satellite Time and Fre-
quency Transfer (TWSTFT) Links via Two Separated Transponders on One
Telecommunication Satellite. In Proc. EFTF - IEEE IFCS Joint Conference,
pages 1076–1081, Besançon, France, April 20-24, 2009.
[90] Evaluation of measurement data - Guide to the expression of uncertainty in

measurement, 2008, online at
http://www.bipm.org/utils/common/documents/jcgm
/JCGM 100 2008 E.pdf.
[91] B. Pesch. Bestimmung der Messunsicherheit nach GUM. Books on Demand,

Alamogordo, New Mexico, USA, 2003.
[92] IERS Conventions Centre, http://tai.bipm.org/iers/.
[93] C. Boucher and Z. Altamimi. ITRS, PZ-90 and WGS 84: current realizations
and the related transformation parameters. Journal of Geodesy, 75:613–619,
2001.
[94] N. Ashby. The Sagnac Effect in the Global Positioning System, online at
http://digilander.libero.it/solciclos/, as of October 2010.
Acknowledgements
Primarily I wish to thank Ekkehard Peik (PTB) for supervising this work, and
Andreas Bauch (PTB) and Steffen Schön (IfE) for giving me the opportunity to
work in the field of time and frequency. All three are thanked for valuable sugges-
tions related to the present work. I am also grateful to my other colleagues in the
time and frequency department at PTB for the their support and the good working
atmosphere, namely Dirk Piester, Michael Rost, Jürgen Becker, Thomas Polewka,
and Egle Staliuniene. Special thanks go to Thomas Leder and Hans-Albert Krem-
ling for their help in constructing the mobile frequency reference, and to Christof
Richter and Sven Krenauer for help in constructing the calibration set-up.
André Stefanov (METAS), Laurent-Guy Bernier (METAS), Christian Schluneg-
ger (METAS), Héctor Esteban (ROA), Liang Kun (NIM), and Demetrios Matsakis
(USNO) and his team are acknowledged for their cooperation in the field of inter-
national time link calibrations. I also wish to thank Temmo Wübbena (IQO),
Ulrich Weinbach (IfE), and Tadahiro Gotoh (NICT) for their cooperation and for
providing equipment on loan.
I would like to thank Petr Panek and Alex Kuna from UFE for collaboration
in case of initial problems with the GTR50s and for continuously updating the
firmware.
I am grateful to Harald Schnatz (PTB) and his team in the length department
for supporting the use of the optical fiber between PTB and IQO.
This work was supported by the Centre for Quantum Engineering and Space-
Time Research (QUEST), Leibniz Universität Hannover.
Curriculum Vitae
Name: Thorsten Feldmann
Date of birth: 11th July 1977
Place of birth: Duisburg, Germany
1984 – 1988 Städt. Grundschule Hebbelstraße in Duisburg
1988 – 1998 Steinbart-Gymnasium in Duisburg

School leaving certificate: Abitur
1998 – 1999 Mandatory military service
1999 – 2005 Universität Duisburg-Essen

University degree: Diplom-Physiker
Nov. 2007 – Research assistant at the Physikalisch-Technische Bundes-

anstalt in Braunschweig
Publications
• T. Feldmann, D. Piester, A. Bauch, and T. Gotoh. GPS carrier phase time
and frequency transfer with different versions of precise point positioning
software. In Proceedings of the 40th PTTI Systems and Applications Meeting,
pages 403-414, Reston, VA, USA, December 01-04, 2008.
• T. Gotoh, T. Hobiger, R. Ichikawa, T. Feldmann, and D. Piester. Application

of ray-traced troposphere delays to GPS time transfer. In Proceedings of the
52nd Meeting of the Japan Society for Aeronautical and Space Sciences, pages
1322-1326, Hyogo, Japan, November 05-07, 2008.
• T. Feldmann, D. Piester, H. Esteban, J. Palacio, F. J. Galindo, T. Gotoh, H.

Maeno, U. Weinbach, and S. Schön. GPS carrier phase and precise point po-
sitioning timescale comparisons using different software packages. In Proceed-
ings of the EFTF - IEEE IFCS Joint Conference, pages 120-125, Besançon,
France, April 20-24, 2009.
• U. Weinbach, S. Schön, and T. Feldmann. Evaluation of state-of-the-art

geodetic GPS receivers for frequency comparisons. In Proceedings of the
EFTF - IEEE IFCS Joint Conference, pages 263-268, Besançon, France,
April 20-24, 2009.
• H. Esteban, J. Palacio, F.J. Galindo, T. Feldmann, A. Bauch, and D. Piester.

A GPS calibration trip experience between ROA and PTB. In Proceedings of
the EFTF - IEEE IFCS Joint Conference, pages 274-279, Besançon, France,
April 20-24, 2009.
• H. Schnatz, O. Terra, K. Predehl, T. Feldmann, T. Legero, B. Lipphardt,

U. Sterr, G. Grosche, T. W. Hänsch, R. Holzwarth, T. Udem, Z. Lu, L.
Wang, W. Ertmer, J. Friebe, A. Pape, E. M. Rasel, M. Riedmann, and T.
Wübbena. Phase-coherent frequency comparison of optical clocks using a
telecommunication fiber link. In Proceedings of the EFTF - IEEE IFCS
Joint Conference, pages 807-812, Besançon, France, April 20-24, 2009.
• D. Piester, T. Feldmann, A. Bauch, M. Fujieda, and T. Gotoh. Concept

for an Accurate Calibration of Long Baseline Two-Way Satellite Time and
Frequency Transfer (TWSTFT) Links via Two Separated Transponders on
One Telecommunication Satellite. In Proceedings of the EFTF - IEEE IFCS
Joint Conference, pages 1076-1081, Besançon, France, April 20-24, 2009.
• T. Feldmann, A. Bauch, and D. Piester. GPS Carrier-Phase Time and Fre-

quency Comparisons: Applications and Uncertainty Estimation. In Proceed-
ings of the 2nd International Colloquium - Scientific and Fundamental Aspects
of the Galileo Programme, Padova, Italy, October 14-16, 2009.
• M. Rost, A. Bauch, J. Becker, T. Feldmann, D. Piester, T. Polewka, D.

Sibold, and E. Staliuniene. PTB’s Time and Frequency Activities in 2008
and 2009. In Proceedings of the 41th PTTI Systems and Applications Meeting,
pages 197-213, Santa Ana Pueblo, New Mexico, USA, November 16-19, 2009.
• T. Feldmann, A. Bauch, D. Piester, A. Stefanov, L.-G. Bernier, C. Schluneg-

ger, and K. Liang. On improved GPS based link calibration of the time links
between METAS and PTB. In Proceedings of the EFTF, Noordwijk, NL,
April 13-16, 2010.
• K. Liang, T. Feldmann, A. Bauch, D. Piester, A. Zhang, and X. Gao. Perfor-

mance evaluation of NIM GPS receivers in use for time transfer with PTB.
In Proceedings of the EFTF, Noordwijk, NL, April 13-16, 2010.
• H. Schnatz, O. Terra, K. Predehl, T. Feldmann, T. Legero, B. Lipphardt, U.

Sterr, G. Grosche, T. W. Hänsch, R. Holzwarth, T. Udem, Z. Lu, L. Wang,
W. Ertmer, J. Friebe, A. Pape, E. M. Rasel, M. Riedmann, and T. Wübbena.
Phase-coherent frequency comparison of optical clocks using a telecommu-
nication fiber link. IEEE Transactions on Ultrasonics, Ferroelectrics, and
Frequency Control, 57(1):175-181, 2010.
• H. Esteban, J. Palacio, F.J. Galindo, T. Feldmann, A. Bauch, and D. Piester.

Improved GPS based time link calibration involving ROA and PTB. IEEE
Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 57(3):714-
720, 2010.
• T. Feldmann, A. Bauch, D. Piester, M. Rost, E. Goldberg, S. Mitchell, and

B. Fonville. Advanced GPS Based Time Link Calibration with PTB’s New
Calibration Set-Up. To be published in Proceedings of the 42th PTTI Systems
and Applications Meeting, Reston, VA, USA, 2010.
• D. Piester, M. Rost, M. Fujieda, T. Feldmann, and A. Bauch. Remote atomic

clock synchronization via satellites and optical fibers. Advances in Radio
Science, 9:1-7, 2011.

Feldmann 2011 Dissertation

Hochgeladen von

Copyright:

Verfügbare Formate

Feldmann 2011 Dissertation

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Feldmann 2011 Dissertation

Hochgeladen von

Copyright:

Verfügbare Formate

Advances in GPS based

Time and Frequency Comparisons

Von der Fakultät für Mathematik und Physik der

zur Erlangung des Grades

Doktor der Naturwissenschaften

Keywords: Time Comparisons, Frequency Comparisons, GPS

Stichworte: Zeitvergleiche, Frequenzvergleiche, GPS

2. State-of-the-Art in remote Time and Frequency Comparisons 5

3. A Highly Precise Mobile Frequency Reference 34

6. Relative Calibration of TAI links 80

7. Summary and Outlook 117

1 PPS one pulse per second signal

ADEV Allan Deviation

BIPM Bureau international des poids et mesures

C/A-code coarse/aquisition code

CCD common-clock difference

CCTF Consultative Committee for Time and Frequency

dCCD difference between two CCDs

DMTD dual mixer time difference technique

GLONASS Global Navigation Satellite System, Russian GNSS system

GNSS Global Navigation Satellite System

GPS Global Positioning System, US GNSS system

IERS International Earth Rotation and Reference Systems Service

IfE Institut für Erdmessung, geodetic institute at the LUH

IGS International GNSS Service

ITRF International Terrestrial Reference Frame

IQO Institute of Quantum Optics at the LUH

LUH Leibniz Universität Hannover

METAS Swiss Federal Office of Metrology, operating the timescale UTC(CH)

2.1. Characterization of Frequency Instabilities

x(t) − x(t − ∆t)

Noise ADEV MDEV TDEV

2.2. Today’s remote Time and Frequency

2.3. GPS and Time and Frequency Transfer

2.3.1. Brief GPS System Overview

correlation process performed by a receiver. All satellites transmit their signals

In principle the operation of GLONASS is similar to that of GPS, but GLONASS

2.3.2. Coordinate Systems and Satellite Motion

2.3.3. Observation Model

λΦs (ts , tr ) = ρs (ts , tr ) + (δtr − δts )c + λNs + ∆Φ + ε . (2.6)

Φs is measured in cycles of the carrier frequency. As in equation (2.4), it can

2.3.4. Time and Frequency Transfer Receivers

frequency comparisons must allow for referencing the measurements to an external

2.4. Signal Delays and Correction Models

2.4.1. Atmospherical Corrections

the variance of the P3 combination is

2.4.2. Relativistic Corrections

2.4.4. Site Displacement

2.4.5. Other Delay Sources

2.5. The International GNSS Service (IGS)

2.6. Precise Point Positioning

at the geodetic institute at Natural Resources Canada (NRCan). This software

2.7. GPS and UTC Generation

The calculation of UTC-UTC(k) and its uncertainties requires the knowledge of

3.3. Performance of the passive Hydrogen Maser

3.4. Performance of the GTR50 Time and

Figure 3.7.: A view inside the temperature stabilized box.

3.5. Installation of the Frequency Reference at

instability of a comparison of an active maser at the Max-Planck Institute for