Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO.

2, MARCH 2008 693


Measurement and Modeling of the Land Mobile
Satellite Channel at Ku-Band
Sandro Scalise, Member, IEEE, Harald Ernst, Member, IEEE, and Guy Harles
AbstractThis paper presents the results of a measurement
campaign conducted in the winter of 2002 around the city of
Munich in southern Germany. The primary goal was to evalu-
ate the link performance of the land mobile satellite channel at
Ku-band and to characterize the temporal behavior of the channel
to help in the evaluation of different fade mitigation techniques
for future mobile satellite systems that aim at employing this
frequency band. To achieve this objective, rst- and second-order
channel statistics will be presented, the outage probability of the
channel analyzed, and a Markov-chain-based narrowband chan-
nel model proposed.
Index TermsKu-band, land mobile satellite channel (LMSC),
radio propagation, satellite mobile communication, statistical
modeling.
I. INTRODUCTION
T
HE USAGE of the Ku-band (1012 GHz) in Europe is
traditionally mostly limited to stationary services, and
consequently, only a few publications on its usage for land
mobile applications exist, such as [1] and [2]. Although satellite
broadcasts in the Ku-band to xed (television) receivers already
has a successful history, today, L- and S-band dedicated high-
power satellites are preferred for mobile satellite radio recep-
tion (e.g., Worldspace, XM-Radio, and Sirius). One reason
is that the use of Ku-band satellites requires a higher gain
directional antenna. In the mobile case, this obliges to develop
terminals with satellite tracking capabilities and hinders the
possibility to employ satellite diversity technique as a coun-
termeasure against channel fades, as done in XM-Radio and
Sirius. The second drawback is the potential interference from
neighbor satellites. Nevertheless, the relatively high number
of existing Ku-band transponders and the large amount of
available bandwidth present an interesting opportunity that
has recently attracted the interest of several operators and
researchers of the satellite communications eld. Relevant ex-
amples are the recently nalized studies to provide satellite-
Manuscript received November 10, 2003; revised September 8, 2005 and
January 13, 2006; accepted May 16, 2006. This work was supported, in part,
by the European Space Agency as part of the European Space Research and
Technology Center under Contract 15593/01/NL/DS. The review of this paper
was coordinated by Dr. O. Ugweje.
S. Scalise is with the German Aerospace Center (DLR), 82230 Wessling,
Germany.
H. Ernst was with the German Aerospace Center (DLR), 82230 Wessling,
Germany. He is now with the European Space Agency (ESA), 2200 AG
Noodwijk ZH, The Netherlands.
G. Harles is with Socit Europenne des Satellites-Astra (SES Astra),
L-6815 Betzdorf, Luxembourg.
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TVT.2007.906338
based Internet and multimedia services to high-speed trains [3],
[4], the hybrid Ku/Ka band system being developed in Korea
[5] under the name of MoBISAT, and the satellite-based solu-
tion to provide wireless Internet access onboard aircraft using
Ku-band transponders (guaranteeing seamless connectivity also
for intercontinental ights) that are currently being tested by
the American company Connexion by Boeing on the rst
commercial ights [6].
Good knowledge of the behavior of the land mobile satellite
channel (LMSC) is of paramount importance to assess the
performance of a future communication system. The main
effect in an LMSC is signal shadowing, which is experienced
when no clear line-of-sight (LOS) between the satellite and
the user terminal is present. The average duration and depth
of this shadowing are key parameters that strongly depend on
the environment in which the mobile user is located. Statistics
of the LMSC based on measurements at L, S, Ka, and EHF
bands and using the Markov-chain-based model, which was
rst proposed in [7], are present in the literature (see, e.g.,
[8] and [9]) and can be theoretically extended to Ku-band;
but no model directly based on measurements in this band has
been proposed so far. For this reason, a measurement campaign
was performed within a feasibility study for a future Ku-band
mobile system, which was commissioned by the European
Space Agency (ESA), to validate the theoretical predictions
about channel behavior.
The measurement setup will be described in Section II, and
the considered scenarios will be presented in Section III. The
statistical analysis of the measured data presented in Section IV
will give an insight into the temporal behavior of the channel
and the possible countermeasures to be devised to counteract
its impairments. A parameter set for a Markov-chain-based
channel model [7], [10] will also be presented. Finally, our
conclusions will be drawn.
II. MEASUREMENT CAMPAIGN SETUP
The setup for the measurement campaign is depicted in
Fig. 2. A test signal using horizontal polarization was trans-
mitted by the geostationary satellite Astra, which is located at
19.2

E, and received by means of a low-gain 1010 cm at


antenna with 19 dBi gain. The antenna and the low-noise block
(LNB) were mounted on a mechanically steerable platform,
which was placed on top of the DLR measurement van and
covered with a plastic low-attenuation radome, as depicted in
Fig. 1.
The rst intermediate frequency (IF) in L-band at the LNB
output was downconverted to a second IF of 70 MHz and then
0018-9545/$25.00 2008 IEEE
694 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
Fig. 1. Measurement van of DLR.
split into I and Q baseband components, which were nally
sampled at 24 kHz and stored on a magnetic tape by means
of a digital audio tape recorder. Inter Range Instrumentation
Group (IRIG) time stamps were stored on the third track of the
tape. No Doppler precompensation was implemented since the
maximum estimated Doppler shift was roughly 1 kHz. This
frequency shift had no impact on the calculated power level,
since the receiver bandwidth of the I and Q demodulator was
larger than 4 kHz and its frequency response was additionally
equalized by means of a dedicated IIR digital lter.
The pointing and tracking of the antenna was achieved by
means of an open-loop algorithm: based on real-time navigation
data of the van (i.e., precise position, roll, pitch, and yaw an-
gles) and on the known satellite position, the effective azimuth
and elevation angles were computed, and the correspondent
correction was applied to the steering platform. The informa-
tion that concerns the position of the van was obtained from
the onboard GPS receiver and from a mechanical three-axis
gyroscope for what concerns the pitch, roll, and yaw angles (see
[11] for more details). The pointing error caused by eventual
blockages suffered by the GPS receiver in dense urban areas
is very low, since the last known position, together with the
gyroscope information, still gives a fairly precise azimuth and
elevation estimation for the position of the satellite.
In addition to the samples of the received signal and to
the recordings of the aforementioned navigation data, three
different video cameras have been used: 1) a directive camera
that points toward the satellite and is mounted on the antenna
platform; 2) a sh-eye camera recording a 360

view; and
3) a hand camera. Timing information, based on the GPS
receiver, has been added to all video streams, thus permitting
synchronization of themwith the samples of the received signal.
III. ENVIRONMENTS DEFINITION AND CAMPAIGN DETAILS
After some preliminary tests to calibrate the equipment,
the nal measurements were performed in winter 2002 in the
Munich area (southern Germany; satellite elevation was about
34

), following the itinerary depicted in Fig. 4. More than


10 GB of data were recorded, including navigation data plus
three video streams with more than 10 h of material. The van
moved at a normal speed, as detailed below, and the weather
conditions were mixed during the 2 days of trials. Therefore,
uctuations of 1 dB (with respect to the nominal power level)
due to rain and pointing misalignments were experienced at
some point.
The recorded data have been classied into the environments
described next. For each of these environments, a different set
of parameters has been extracted from the measured data.
1) Rural. The rural environment consists mainly of rela-
tively open areas, tree alleys, and forests. As the mea-
surement was performed in winter, deciduous trees had
no leaves. Although leaf-tree forests were included in our
itinerary, the results for this environment may be slightly
optimistic. The speed of the user terminal is normally
moderate (e.g., below 6070 km/h) because of the pres-
ence of many secondary roads, but sudden changes in its
direction can be experienced.
2) Suburban. The suburban environment can be found in
small villages and towns, as well as in the suburbs of large
cities. It is mainly composed of small obstacles such as
family houses or villas with gardens. The normal speed
should stay below 50 km/h, but sudden changes in the
direction of the motion are present (e.g., 90

turns at each
crossroad).
3) Urban. This environment is mainly characterized by
large and high buildings that produce severe blockage
effects. However, good visibility can still be achieved
when driving in the NS direction or in wide roads. For
this reason, both NS- and EW-oriented streets have
been roughly included in equal proportions in the trial
itinerary. The normal speed should stay below 50 km/h,
but sudden changes in the direction of the motion are
present (e.g., 90

turns at each crossroad).


4) Highway. The highway environment resembles an open
area apart from the presence of relatively frequent bridges
and tunnels. Furthermore, the higher terminal speed
(during our trial, a maximum speed of 120 km/h was ex-
perienced) will produce a greater Doppler effect, although
the direction of the motion can remain constant for longer
time intervals. Big national roads can also be considered
as part of this environment. Both NS- and EW-oriented
highways have been included in our itinerary.
IV. ANALYSIS OF THE CHANNEL RECORDINGS
A snapshot of the received signal power versus level time
is shown in Fig. 3. The vertical line in the lowermost plot
indicates the elapsed time from the beginning of the run, and it
is synchronized with the corresponding images taken from the
directive and sh-eye cameras shown in the upper part of the
picture. Fig. 3 refers to an EW-oriented street in Munich city.
By visual analysis, one can appreciate some interesting fading
events with an attenuation of approximately 12 dB relative to
LOS. They are due to the diffraction on the edge of the roof,
regularly interrupted by dormers (seen in the corners of the
picture), which completely block the signal. LOS occurs in this
run only in correspondence with crossroads.
SCALISE et al.: MEASUREMENT AND MODELING OF THE LAND MOBILE SATELLITE CHANNEL AT Ku-BAND 695
Fig. 2. Measurement campaign setup.
Fig. 3. Received signal power versus time and camera views.
Fig. 4. Measurement campaign itinerary.
A. Analysis of First- and Second-Order Statistics
The overall complementary cumulative distribution functions
(cdfs) and probability density functions (pdfs) relative to the
different environments are plotted in Figs. 5 and 6, respectively.
Fig. 5. Complementary cdfs for different environments.
Fig. 6. PDFs for different environments.
The latter have been estimated from the histograms of the
received power level with equally spaced bins (in linear unit),
normalized so that 0 dB in the plots corresponds to the nominal
received power level in LOS conditions, as obtained from the
696 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
Fig. 7. TSC for a highway environment and different link margins.
Fig. 8. TSC for an urban environment and different margins.
link budget. The cdfs can then be derived by direct integra-
tion. As the suburban and rural environments behave relatively
similarly, plots referred to the former case will, at some point,
be omitted in the rest of this paper.
The pdfs and complementary cdfs show the principal behav-
ior of the LMSC at the Ku-band. One can clearly identify the
Rice-fading component for the LOS state around 0 dB. Addi-
tionally, the region between 3 and 15 dB is characteristic for
a certain amount of echoes and refractions. The leftmost region
(below 15 dB) represents the totally blocked state, which is
particularly visible in the urban environment.
The second important aspect of the channel is its temporal
behavior. Time intervals where the received signal power is
greater than or equal to a certain threshold are dened as
connections. The time share of connections (TSC) is dened as
the probability that a given time instant belongs to a connection
that has a duration T
c
. This is presented in Figs. 79 for
the highway, urban, and rural environments, respectively. The
arrows indicate the direction of increasing link margins, which
are expressed in decibels and normalized with respect to the
mean power of the LOS state. For the urban and highway envi-
ronments, it can be seen that increasing the link margin by more
than 2.5 dB results only in small improvements of the TSC.
Fig. 9. TSC for a rural environment and different margins.
Fig. 10. TSF for a highway environment and different margins.
The equivalent statistic for the length of disruption of the
service, i.e., the time share of fades (TSF), is an important
indicator to evaluate the efciency of different fade mitigation
techniques. For the urban and highway environment (Figs. 10
and 11), an improvement due to higher margins is noticeable,
but mainly for fades with short durations. Longer fades are not
effectively counteracted by simply increasing the link margin.
For the rural environment (Fig. 12), the reduction in the TSF is
much more remarkable, even if it would require the usage of a
large margin. Such a margin would imply a drastic reduction of
the achievable data rate (unless the antenna size is increased),
bearing in mind that the maximum equivalent isotropic radiated
power per transponder is xed. In fact, the available margin is
determined by the receiver bandwidth and the net data rate since
the power spectral density of the thermal noise only depends on
the antenna temperature and on the receiver noise gure.
B. Analysis via a Channel-Capacity-Based Approach
One of the most important techniques in increasing the
capacity of communication system in a fading environment is
the usage of interleavers. An interleaver spreads a codeword
SCALISE et al.: MEASUREMENT AND MODELING OF THE LAND MOBILE SATELLITE CHANNEL AT Ku-BAND 697
Fig. 11. TSF for an urban environment and different margins.
Fig. 12. TSF for a rural environment and different margins.
over a longer interval to take advantage of the time-dependent
nature of the fading. An important question is the gain that can
be expected by increasing the size of the interleavers. Based
only on the statistics presented in Section IV-A, it is very hard
to gain an insight into this effect for longer periods, where there
is often a high correlation between consecutive fading events.
Therefore, a new approach based on the analysis of the channel
outage capacity is proposed.
For the following, a simple at-fading channel, as proposed
in [12], is assumed, where perfect state information is available
at the receiver. The discrete channel output y
k
is dened as
y
k
= a
k
x
k
+ n
k
(1)
where k stands for the discrete time index, a
k
is the resulting
value of the fading process, x
k
is the channel input, and n
k
is
the Gaussian noise samples. The capacity of the channel can
then be calculated as
C =
+
_
0
p(a)C(a)da (2)
where p(a) is the pdf of the ergodic fading process, and C(a)
is the capacity for a channel realization a.
If BPSK modulation is used, the channel capacity for a given
realization is
C(a) = 1 +
+
_

f (y|a, 1) log
2
f (y|a, 1)
f (y|a, 1) + f (y|a, 1)
dy
(3)
where f(y|a, x) is the pdf of the received signal y, given the
channel realization a and the transmitted symbol x = 1.
In a rst-order approximation, one could try to estimate
a capacity for a given interleaver window of length T by
generating a pdf based on the measured samples a(t
0
< t
i
<
t
0
+ T) during this time window and then calculating C as in
(2). In principle, this is what is done in the following, but some
restrictions and a special context have to be applied so that one
may speak about this as capacity.
The point is that the pdf used in the capacity calculation
implies that each fading amplitude of the pdf is indenitely
repeated often to ensure the ergodicity of (1). Therefore, the
following approach is used to motivate the usage of (2).
In the case of a very slowly changing channel, where a
k
is
constant for a time T during which a sufciently large number
k of samples is received, C(a
k
) could be interpreted as
an instantaneous capacity C
T
(a
k
) during time T.
By focusing on a time interval T = N T during which
the channel is characterized by a xed nite sequence of N
realizations a
i
, only changing at discrete time steps T, one is
then allowed to calculate the capacity C
N
for this time interval
T (see also, e.g., [13] or [14]) as
C
N
=
1
N
N

i=0
C
T
(a
i
). (4)
As a further motivation, one can see this channel also as an
approximation of a normal fading channel with the probability
density of the fading realizations of
p(a) =
1
N
N

i=0
(a
i
= a) (5)
where the innite number of samples N k are inter-
leaved and spread over time T.
This approach should now be used to help analyze the
channel recordings. First, it has to be veried that the channel
can be considered, at least approximatively, as slowly fading
in the previously mentioned sense. As shown in Fig. 13, where
the Doppler spectrum of the received signal level S(f) is given,
and in Table I, where the Doppler spread
=

_
+
_
0
f
2
S(f)df
+
_
0
S(f)df
(6)
has been estimated, this requirement can be considered fullled
for a large range of data rates, particularly since, also, an almost
698 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
Fig. 13. One-sided Doppler spectrum for different environments.
TABLE I
ESTIMATED DOPPLER SPREAD FOR DIFFERENT ENVIRONMENTS
frequency at fading can be assumed (see Section IV-C). One
can then calculate the evolution of the instantaneous capacity
C
T
(t) for a time window T and an initial observation
instant t.
Assuming that a codeword is transmitted with a certain
rate C
T
over a duration T, and depending on the initial time
instant t, the codeword experiences a sequence of channel
realizations. The corresponding approximated capacity C
N
(t),
as seen by the codeword of time length T, is then calculated
using (4).
Whenever the assumed rate C
T
was higher than the capacity
C
N
(t), we dene it in accordance with [12] as an outage, since
no reliable reception of the codeword can be accounted for. The
outage probability is, hence, dened as
P
out
= Pr {C
T
> C
N
(t)} . (7)
The probability that a reliable reception is possible is dened as
the complement of (7), i.e.,
P
suc
= P
out
= Pr {C
T
C
N
(t)} . (8)
Using the measured data, these probabilities using different
time windows T were estimated: the results for the case of
BPSK modulation can be seen in Figs. 1417, where the
arrows indicate the increasing length of the time window T. As
expected, the larger the time window T becomes, the steeper
the function gets. For T , the overall measured channel
would show its ergodic behavior, and P
suc
would be one for
values lower than the ergodic channel capacity C
erg
and zero
for higher rates. C
erg
can be directly derived from the pdf of
Fig. 14. Complementary outage probability for a rural environment.
Fig. 15. Complementary outage probability for a highway environment.
the measured channel using (2). In the following, the results for
the different environments will be highlighted.
The highway environment has the fastest convergences to
the ergodic capacity C
erg
and has the highest value C
erg
for
all environments as well. The gure shows not only that, not
surprisingly, a highway-like environment is very benign for
satellite reception but that, for time windows in the range of
10 s up to a minute and a code rate that is two thirds lower than
C
erg
, the probability of an outage can also be minimized. It is
interesting to note that in spite of the nature of a highway chan-
nel, time windows below 10 s do not signicantly change the
overall probability of outages. The reason is that the dominating
causes for outages are tunnels, which are, indeed, typically on
the order of some seconds to bypass. For a code rate of two
thirds of LOS, the shadowing event should then be lower than
SCALISE et al.: MEASUREMENT AND MODELING OF THE LAND MOBILE SATELLITE CHANNEL AT Ku-BAND 699
Fig. 16. Complementary outage probability for a suburban environment.
Fig. 17. Complementary outage probability for an urban environment.
one third of the time window, which explains the need for an
extended time in a range longer than 10 s.
For the other environments, it is interesting to note that, while
the rural environment has rather the same ergodic capacity as
the suburban case, its convergence is much slower due to the
presence of rather long tree alleys and forests obstructing the
satellite. Therefore, time windows in the range of 1 min have
a larger impact on the behavior in the suburban case, whereas
in the rural environment, the critical time is more in the range
of 5 min, which is still a bit off from achieving a vertical
line at C
erg
.
However, the most problematic environment is the urban one,
where even time windows in the range of 5 min show only a
very slow convergence toward the ergodic case.
As a general rule, it can be deduced that to reach probabilities
of reliable reception higher than 90%, the time window has to
be in the range of one to several minutes. At the same time,
as the function becomes steeper, the probability that a specic
realization of the channel for this time window has a capacity
larger than the ergodic capacity C
erg
decreases. Obviously, for
very long time windows, the nominal rate needs to be kept
lower than C
erg
to allowany successful reception of codewords.
Therefore, for long windows, one needs to dene up to which
environment they should work.
The opposite approach is also possible. Transmitting at the
highest possible rate, which can be received in LOS, and
sending the information as fast as possible can also have
advantages. If there are LOS conditions, it will be received,
if there are NLOS conditions during the time window, it will
not be received. Although unreliable, this scheme is robust to
changes in C
erg
due to different environments. This alternative
approach could be, e.g., interesting for data services that are
often updated. For general broadcast data services, a mixture of
both approaches is also possible: a window that is long enough
to ensure a relative high reliability for the codewords in good
environments but short enough that, even for bad environments,
at least some information gets through.
Getting back to the gures, it is interesting to note that the
curves for time windows in the range from 0.01 up to 1 s
behave very similarly, which means that spreading codewords
over this time range only has a negligible impact on the system
performance. On the other hand, this conrms the validity of
our assumption about the slow variation of the channel.
To conclude this section, it has been shown that codewords
with a length of up to several minutes should be used, if highly
reliable broadcast receptions need to be achieved in the land-
mobile context. For the practical implementation, interleavers
could be used with depth in the same time range and with
enough samples per codeword to approach the channel bounds,
or the spreading could be done on a higher OSI layer (see,
e.g., [15] for a practical system that makes use of this ap-
proach). Results for S-band data using a similar analysis can be
found in [16].
In the following section, a channel model will be derived,
which allows further insights into the statistical behavior and
characteristics of the channel.
C. Channel Model
A narrowband statistical model based on the Markov-chain
approach rst proposed in [7] has been adopted to elaborate
the measured data. Narrowband models assume that the signal
bandwidth is smaller than the coherence bandwidth of the
channel, with the latter being inversely proportional to the delay
spread [8]. For a xed measurement time t
0
, the delay spread
is dened as
=

_
+
_

(t m

)
2
|h(t
0
, )|
2
d
+
_

|h(t
0
, )|
2
d
(9)
700 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
where h(t, ) is the time-varying impulse response of the
channel, and m

is the mean echo delay given by


m

=
+
_

|h(t
0
, )|
2
d
+
_

|h(t
0
, )|
2
d
. (10)
Signal echoes are associated to the presence of indirect rays
that reach the receiver antenna and carry a signicant energy
with respect to the energy of the direct ray. Typical values
of the coherence bandwidth for outdoor environments are
in the range between 7 and 11 MHz at L-band (12 GHz)
and around 30 MHz at EHF band (40 GHz), as reported in
[8] after direct measurements with omnidirectional antennas.
In the latter case, only few echoes with strong attenuation
(22 to 27 dB) were observed. If directive antennas are used,
echoes with signicant delays are in most of the practical cases
ltered out by the antenna radiation pattern. Therefore, narrow-
band models can be normally employed for Ku-band signals
with a data rate of practical interest. Formally, the absence of
signicant echoes means that h(t, ) A(t)(), with A(t)
being a time-varying complex coefcient. Using the complex-
envelop notation, a generic passband received signal can, hence,
be written as
y(t) = Re
_
A(t) x(t t
0
)e
j2f
c
t
_
+ n(t) (11)
where x(t) is the complex envelop of the transmitted signal, t
0
is the propagation delay, f
c
is the carrier frequency, and n(t)
is the additive thermal (passband) noise. In these conditions,
the channel transfer function, which is dened as the Fourier
transform of h(t, ) with respect to the variable is almost
at for any observation time t
0
. For this reason, the term at
frequency fading channel is also adopted in the literature.
In narrowband models, the amplitude of the multiplicative
fading term R(t) = |A(t)| is divided into fast and slow fading.
Slow fading events, normally due to obstacles, are modeled as a
nite-state machine. Fast fading events due to the irregularity of
the obstacles (e.g., vegetative shadowing) and to the multipath
propagation phenomenon are additionally modeled as super-
imposed random variations that follow a given pdf for each
state. A qualitative example, clarifying the concept of channel
state, is provided in Fig. 4. At an arbitrary time instant t, and
assuming that the signal x(t) has unitary amplitude, the overall
pdf that describes the received signal amplitude R is, hence,
given by
p
R
(r) =
n

k=1
P
k
p
k,R
(r) (12)
where n is the number of states, P
k
is the absolute probability of
being in state k (which can be obtained from the state transition
matrix P, where the elements p
i,j
contain the probability of the
transition from state i to state j), and p
k,R
(r) is the pdf that
describes the fast fading associated with state k.
Following the approach in [10], a three-state model (see
Fig. 18), which has been developed by J. Kunisch and
Fig. 18. Three-state narrowband channel model.
J. Siemons from IMST GmbH and assuming a Rice pdf within
each state (reducing to a Rayleigh for the third state, where
no direct signal is present) has already been presented in [17].
The model presented here, although extracted from the same
channel recordings, is based on our independent analysis, and it
differs from [17] mainly for that which concerns the modeling
of the intermediate state, as detailed next.
The rst state, corresponding to LOS conditions, has been
modeled by means of a Rice distribution of the form
p
Rice,R
(r) =
r

2
r
exp
_

r
2
+ V
2
2
2
r
_
I
0
_
rV

2
r
_
, V 0
(13)
where I
0
() is the zeroth-order rst modied Bessel function.
By operating the randomvariable substitution S = (R/V )
2
and
by dening the Rice factor c (direct to multipath signal power
ratio) as V
2
/2
2
r
, the Rice distribution can be referred to the
normalized received signal power S as follows:
p
Rice,S
(s) = c exp (c(s + 1))) I
0
_
2c

s
_
. (14)
The Rice factor assumes values in the range between 17 and
18 dB as an obvious consequence of the usage of a directive
antenna.
The second state, corresponding mostly to shadowing
caused by single trees, is modeled here by means of the
Suzuki/lognormalRayleigh distribution of the normalized
received power S [7], which is described as follows:
p
Suzuki,S
(s) =

_
0
p
Rayleigh
(s|s
0
) p
lognormal
(s
0
)ds
0
(15)
p
Rayleigh
(s|s
0
) =
1
s
0
exp
_

s
s
0
_
(16)
p
lognormal
(s
0
) =
10

2
ln
ln 10

1
s
0
exp
_

(10 log s
0
)
2
2
2
ln
_
. (17)
SCALISE et al.: MEASUREMENT AND MODELING OF THE LAND MOBILE SATELLITE CHANNEL AT Ku-BAND 701
Fig. 19. Scan of different arrival angles, with LOS blocked by building,
elevation from 0

up to 60

.
Fig. 20. Fish-eye view of the scan, with the thick circle approximately
representing satellite position and the thin circle the position of the reection.
In the previous equations, s
0
is the time-varying short-term
mean received power, which follows a lognormal distribu-
tion. The superimposed Rayleigh term accounts for multipath
phenomena.
Finally, the third state corresponds to blockages mainly
due to buildings and bridges. In contrast to omnidirectional
measurements, reections from buildings are seldom measured,
since they come typically from angles that are different from
the satellite direction, and they are consequently suppressed
by the antenna pattern. An example can be seen in Fig. 19,
where the van was standing between two high buildings
and the antenna performed a conical scan of the horizon to
detect all relevant reections, thus emulating the behavior of
an omnidirectional antenna. As shown in Fig. 20, the major
reection comes from the opposite direction with respect to the
satellite position, with approximately the same elevation. In
the normal reception mode, where the directive antenna points
toward the satellite, such types of reections result in very deep
fades, which could no longer be correctly measured, due to the
noise oor of our setup, which is roughly 20 dB below LOS.
For this reason, no parameters to characterize the fast fading in
the blocked state are given.
Table II summarizes the parameters relative to the three-state
model earlier presented for the different environments listed in
TABLE II
CHANNEL STATE PARAMETERS
Fig. 21. Attenuation events: rst bald tree, second pine.
Section III: The rst three columns contain the probabilities P
k
,
which appear in (12); the fourth and the fth columns contain
the parameters and
ln
of (17), characterizing the shadowing
state; and the last one contains the Rice factor c, which appears
in (14), characterizing the LOS state.
With regard to the relatively low mean attenuation of the
shadowed state, which is mainly due to single trees, it should be
kept in mind that these measurements were made in the winter,
and most trees were without leaves: in Fig. 21, the difference
between the shadowing caused by a bald tree (visible in the
left-uppermost image), resulting in about 10 dB attenuation,
and that provoked by a pine (visible in the right-uppermost
image), resulting in up to 20 dB attenuation, can be appreciated.
Another interesting point is the relatively high probability of
being in the LOS state. This is partly due to the satellite
elevation of 34

, as well as to the height of more than 2 m above


ground of the antenna on our measurement van. Thus, typical
two-story buildings did not intercept the LOS to the satellite.
A further simplication of the model proposed so far can
be done by reducing it to a two-state GilbertElliot or Lutz
model. This is equivalent to assuming that only the LOS state
is used for transmission purposes, whereas the shadowed states
always results in erroneous reception. The calculation of the
theoretical channel capacity presented in [18] shows that the
improvements obtained by assuming correct reception also in
non-LOS states are negligible, even if one should also consider
the reduction in the throughput derived from synchronization
losses due to frequent fading events. In more general terms, the
choice of whether to use a simple onoff channel model, or
a more elaborated one, with three or even more states, is far
from obvious, and it is strictly related to the constraints derived
from the system design in which the satellite radio link to be
702 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008
Fig. 22. Two-state narrowband channel model protect [8].
Fig. 23. Probability of fade duration for an urban environment.
modeled will be used. For instance, in the case of mobile users
with a limited antenna gain as, e.g., in [19], no large link margin
can be accounted for (unless a dedicated high-power satellite is
used), and therefore, no particularly useful information can be
gained froma detailed knowledge of the behavior of the channel
in non-LOS states.
The Lutz model is presented in Fig. 22, where D
g
and D
b
are the mean durations of the good (LOS) and bad (shadowed)
states, respectively, and p
xy
denotes the probability of the state
transition x y.
The estimation of D
g
and D
b
was obtained by a minimum
mean square error criteria based on the durations of being con-
tinuously above or below the threshold for reliable signal recep-
tion, which is xed at 2.5 dB. This is rather straightforward,
with the exception that, for our purpose, particularly medium-
to long-term events should be well modeled. Therefore, we
excluded fading events with short durations. These events are
partly due to fast signal uctuations around the threshold rather
than to real state transitions, as well as to the effect of small in-
terruptions like single lamp posts, whereas we wanted to model
particularly the longer transitions, e.g., change from open street
to tree-shadowed parts and back. The emphasis has been done
by not taking into account the region of fading events smaller
than 0.25 of the mean duration. This approach is justied by
the intention of our model to be representative of the medium-
to long-term behavior of the channel. A typical comparison
between the measured and the estimated probability to be in
the bad state can be seen in Fig. 23, which shows a good match
TABLE III
LUTZ MODEL PARAMETERS
for long fading events. For short events, one can see that the
these parts are underestimated by the model.
As already shown in [7], due to the exclusion of the fast
changes, the estimated values for D
g
and D
b
no longer match
the well-known relationship of the overall shadowing, i.e.,
A =
D
b
D
g
+ D
b
(18)
where A is the time share of shadowing. Consequently, the
relationships shown in Fig. 22 no longer hold as well. Hence, a
correction factor had to be calculated by applying (14) in [7]
D

g
=
1
2
_
D
g
+
1 A
A
D
b
_
D

b
=
1
2
_
D
b
+
A
1 A
D
g
_
. (19)
The estimated values of D

g
and D

b
are reported in Table III.
The matching of these improved estimations D

g
and D

b
is less
perfect than in [7], where a space-based modeling approach
has been used. This is a direct consequence of our time-
based modeling, implicitly including both the length of the
obstacles and the speed variation of our measurement van.
Nevertheless, the given numbers give a rather robust estimation
of the channel. The change due to the correction term was
in the range of the rounding error of D
g
and D
b
. Typically,
D
b
needed only to be rounded off to comply with the mean
share of shadowing, whereas D
g
is often rounded to the nearest
natural number. Finally, regarding the precision of the modeling
parameters, one always needs to consider that the channel itself
exhibits variations of its parameters, depending on the region,
city, and, due to the trafc inuence, even the time of the day,
which are at least in the same order of magnitude than these
correction terms.
V. CONCLUSION AND OUTLOOK
A measurement campaign for the LMSC in the Ku-band
has been presented. A statistical analysis of the collected data
has been performed, showing the effect of increasing the link
margin by means of second-order fade statistics. It can be
concluded that, for highway and urban environments, limited
improvements can be gained by simply increasing the link
margin and different strategies, e.g., based on long interleavers
and time diversity like those presented in [18], should be used.
Finally, a three-state channel model has been extracted from
the data, and the fast fading within each state has been modeled
by means of suitable pdfs. For practical purposes such as link
performance evaluation, the parameters of a simplied two-
state model have also been presented.
SCALISE et al.: MEASUREMENT AND MODELING OF THE LAND MOBILE SATELLITE CHANNEL AT Ku-BAND 703
In addition to the initiative mentioned in the introduction,
the ESA is currently funding a research project that aims at
designing a future broadcast multimedia system for the ve-
hicular market that exploits Ku-band satellites [15][5], [19].
In the framework of this study, the results presented in this
paper have been widely used to assess the performance of the
system in terms of an achievable le transfer rate. Trials and
demonstrations are foreseen at the end 2005.
ACKNOWLEDGMENT
The authors would like to thank B. Oeste from DLR for his
essential contribution to the realization of the measurement
campaign and J. Hrle from ESA for fruitful discussions
and for helping us in making such an interesting project
possible. In addition, the authors would also like to thank their
colleagues J. Kunisch and J. Siemons from IMST GmbH, who
carried out an independent analysis of the data, DLR students
C. Frstl, A. Moragas, and S. Shabdanov, who helped in setting
up and processing the data of the measurement campaign,
and nally, a former colleague, R. Purvinskis, for his editorial
assistance.
REFERENCES
[1] J. Dutronc and J. N. Colcy, Land mobile communications in Ku-band.
Results of a test campaign on Eutelsat I-F1, Int. J. Satell. Commun.,
vol. 8, pp. 4363, Jan. 1990.
[2] G. Butt, B. G. Evans, and M. Richharia, Narrowband channel statis-
tics from multiband propagation measurements applicable to high eleva-
tion angle land-mobile satellite systems, IEEE J. Sel. Areas Commun.,
vol. 10, no. 8, pp. 12191226, Oct. 1992.
[3] S. Scalise, M. A. Daz, G. Sciascia, and R. Mura, Characterization of
the railroad satellite channel within the FIFTH project, in Proc. 12th IST
Mobile Wireless Commun. Summit, Aveiro, Portugal, Jun. 2003.
[4] F. Nagase, J. Mitsugi, M. Nakayama, and M. Ueba, Ku band mobile
multimedia satellite communications system for trains, in Proc. 21st
AIAA ICSSC, Yokohama, Japan, Apr. 2003.
[5] H. Lee, P. Kim, Y. Song, S. Jeon, and D. Oh, A Ka-band mobile broad-
band satellite Internet access system development, in Proc. 13th IST
Mobile Wireless Commun. Summit, Jun. 2004.
[6] M. de La Chapelle and E. Laase, Connexion by BoeingNew products
and services, in Proc. 10th Ka Broadband Commun. Conf., Oct. 2004,
pp. 6775.
[7] E. Lutz, D. Cygan, M. Dippold, F. Dolainsky, and W. Papke, The land
mobile satellite communication channelRecording, statistics and chan-
nel model, IEEE Trans. Veh. Technol., vol. 40, no. 2, pp. 375386,
May 1991.
[8] E. Lutz, M. Werner, and A. Jahn, Satellite Systems for Personal and
Broadband Communications. Berlin, Germany: Springer-Verlag, 2000.
[9] E. Kubista, F. Perez Fontan, M. Vazquez Castro, S. Buonomo,
B. R. Arbesser Rastburg, and J. P. V. Poiares Baptista, Ka-band propa-
gation measurements and statistics for land mobile satellite applications,
IEEE Trans. Veh. Technol., vol. 49, no. 3, pp. 973983, May 2000.
[10] F. Perez Fontan, M. Vazquez Castro, C. Enjamio Cabado, J. Pita Garcia,
and E. Kubista, Statistical modeling of the LMS channel, IEEE Trans.
Veh. Technol., vol. 50, no. 6, pp. 15491567, Nov. 2001.
[11] M. Holzbock, O. Luecke, and B. Oeste, Mobility requirements of land-
mobile antennas for broadband satellite communication, in Proc. 4th Eur.
Workshop Mobile/Pers. Satell. Commun., Sep. 2000, pp. 3443.
[12] E. Biglieri, J. Proakis, and S. Shamai, Fading channels: Information-
theoretic and communications aspects, IEEE Trans. Inf. Theory, vol. 44,
no. 6, pp. 26192692, Oct. 1998.
[13] L. Ozarow, S. Shamai, and A. Wyner, Information theoretic considera-
tions for cellular mobile radio, IEEE Trans. Veh. Technol., vol. 43, no. 2,
pp. 359378, May 1994.
[14] G. Caire and D. Tuninetti, The throughput of hybrid-ARQ protocols for
the Gaussian collision channel, IEEE Trans. Inf. Theory, vol. 47, no. 5,
pp. 19711988, Jul. 2001.
[15] H. Ernst, S. Scalise, G. Harles, C. Martin, and R. Midthassel, Personal-
ized radio concepts for land mobile satellite services, in Proc. 22nd AIAA
ICSSC, May 2004.
[16] A. Heuberger and M. Breiling, Prediction of QoS in mobile satellite
systems using FEC and interleavers, in Proc. 23rd AIAA ICSSC, Rome,
Italy, Sep. 2005.
[17] S. Scalise, J. Kunisch, H. Ernst, J. Siemons, G. Harles, and
J. Hoerle, Measurement campaign for the land mobile satellite channel
in Ku-band, in Proc. 5th Eur. Workshop Mobile/Pers. Satell. Commun.,
Sep. 2002, pp. 8794.
[18] H. Ernst, S. Scalise, R. Dietrich, G. Harles, and J. Hoerle, Mobile Ku-
band broadcast system, in Proc. 5th Eur. Workshop Mobile/Pers. Satell.
Commun., Sep. 2002, pp. 273280.
[19] J. Hrle, R. Midthassel, J. Sandberg, C. Loeillet, and G. Harles, Mo-
bile Ku-band receive only system, in Proc. 1st Int. Conf. ASMS,
H. Lacoste, Ed, Frascati, Italy, Jul. 2003.
Sandro Scalise (S00M06) was born in Utrecht,
The Netherlands, in April 1973. He received the
degree (with honors) in electronic engineering,
specializing in telecommunications, from the Uni-
versity of Ferrara, Ferrara, Italy, in July 1999.
Since 2001, he has been with the Institute
for Communications and Navigation, German
Aerospace Center (DLR), Wessling, Germany,
where he has been the leading the Mobile Satellite
Systems Group since October 2004. His research
activity deals with forward error correction and
synchronization schemes for mobile satellite applications, land mobile
satellite channel modeling, and link performance evaluation. He is the
Chairman of the R&D Working Group of the Integral SatCom Initiative
(ISI) European Technology Platform and is the Editor of the chapter
Satellite Channel Impairments in the book Digital Satellite Communications
(Springer, 2006).
Harald Ernst (S96M98) was born in Mannheim,
Germany, in August 1970. He received the de-
gree in electrical engineering, specializing in indus-
trial information, from the University of Karlsruhe,
Karlsruhe, Germany, in July 1996 and the Ph.D.
degree from the Technical University of Munich,
Munich, Germany, in 2006.
From 1996 to 2006, he was with the Institute
for Communications and Navigation at the German
Aerospace Center (DLR), Wessling, Germany,
where he worked on higher layer fade mitigation
schemes for land mobile satellite applications using error-correcting codes.
Since 2007, he has been with the European Space Agency, AG Noodwijk ZH,
The Netherlands.
Guy Harles was born in Luxembourg in June 1970.
He received the degree in electrical engineering,
specializing in telecommunications, from RWTH
Aachen, Aachen, Germany, in April 1996.
Since 1996, he has been with Socit Europenne
des Satellites-Astra (SES Astra), Betzdorf,
Luxembourg, as a Project Manager, working on
ground segment engineering. His activities include
the development and implementation of innovative
solutions for satellite operations and mobile Ku-band
satellite broadcasting. He is the holder of several
international patents in the eld of satellite operations and communication
systems.

You might also like