Escolar Documentos
Profissional Documentos
Cultura Documentos
Satellite Communications
Satellite Communications
Guest Editors: Ray E. Sheriff, Anton Donner,
and Alessandro Vanelli-Coralli
Copyright © 2007 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in volume 2007 of “EURASIP Journal on Wireless Communications and Networking.” All articles are
open access articles distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Editor-in-Chief
Luc Vandendorpe, Université Catholique de Louvain, Belgium
Associate Editors
Thushara Abhayapala, Australia David Gesbert, France Marc Moonen, Belgium
Mohamed H. Ahmed, Canada Fary Z. Ghassemlooy, UK Eric Moulines, France
Farid Ahmed, USA Christian Hartmann, Germany Sayandev Mukherjee, USA
Alagan Anpalagan, Canada Stefan Kaiser, Germany Kameswara Rao Namuduri, USA
Anthony Boucouvalas, Greece G. K. Karagiannidis, Greece Amiya Nayak, Canada
Lin Cai, Canada Chi Chung Ko, Singapore A. Pandharipande, The Netherlands
Biao Chen, USA Visa Koivunen, Finland Athina Petropulu, USA
Yuh-Shyan Chen, Taiwan Richard Kozick, USA A. Lee Swindlehurst, USA
Pascal Chevalier, France Bhaskar Krishnamachari, USA Sergios Theodoridis, Greece
Chia-Chin Chong, South Korea S. Lambotharan, UK George S. Tombras, Greece
Huaiyu Dai, USA Vincent Lau, Hong Kong Lang Tong, USA
Soura Dasgupta, USA David I. Laurenson, UK Athanasios V. Vasilakos, Greece
Ibrahim Develi, Turkey Tho Le-Ngoc, Canada Weidong Xiang, USA
Petar M. Djurić, USA Wei Li, USA Yang Xiao, USA
Mischa Dohler, France Yonghui Li, Australia Xueshi Yang, USA
Abraham O. Fapojuwo, Canada Tongtong Li, USA Lawrence Yeung, Hong Kong
Michael Gastpar, USA Zhiqiang Liu, USA Dongmei Zhao, Canada
Alex Gershman, Germany Stephen McLaughlin, Scotland Weihua Zhuang, Canada
Wolfgang Gerstacker, Germany Sudip Misra, Canada
Contents
Satellite Communications, Ray E. Sheriff, Anton Donner, and Alessandro Vanelli-Coralli
Volume 2007, Article ID 58964, 2 pages
Performance Analysis of SSC Diversity Receivers over Correlated Ricean Fading Satellite Channels,
Petros S. Bithas and P. Takis Mathiopoulos
Volume 2007, Article ID 25361, 9 pages
Advanced Fade Countermeasures for DVB-S2 Systems in Railway Scenarios, Stefano Cioni,
Cristina Párraga Niebla, Gonzalo Seco Granados, Sandro Scalise, Alessandro Vanelli-Coralli,
and Marı́a Angeles Vázquez Castro
Volume 2007, Article ID 49718, 17 pages
Capacity Versus Bit Error Rate Trade-Off in the DVB-S2 Forward Link, Matteo Berioli,
Christian Kissling, and Rémi Lapeyre
Volume 2007, Article ID 14798, 10 pages
A QoS Architecture for DVB-RCS Next Generation Satellite Networks, Thierry Gayraud and
Pascal Berthou
Volume 2007, Article ID 58484, 9 pages
Burst Format Design for Optimum Joint Estimation of Doppler-Shift and Doppler-Rate in Packet
Satellite Communications, Luca Giugno, Francesca Zanier, and Marco Luise
Volume 2007, Article ID 29086, 12 pages
Efficient Delay Tracking Methods with Sidelobes Cancellation for BOC-Modulated Signals,
Adina Burian, Elena Simona Lohan, and Markku Kalevi Renfors
Volume 2007, Article ID 72626, 20 pages
Analysis of Filter-Bank-Based Methods for Fast Serial Acquisition of BOC-Modulated Signals,
Elena Simona Lohan
Volume 2007, Article ID 25178, 12 pages
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 58964, 2 pages
doi:10.1155/2007/58964
Editorial
Satellite Communications
Copyright © 2007 Ray E. Sheriff et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We are delighted to bring to you this special issue on satel- ing: accent on polarization” looks at MIMO systems from the
lite communications, which we have prepared as part of the polarization diversity point of view and dwells on the satellite
spreading of excellence remit of the satellite communica- cooperative communication concepts.
tions network of excellence (SatNEx). The SatNEx project, Switch and stay combining (SSC) is a form of diversity
which began in 2004, is funded for five years under the Euro- technique used in digital receivers to compensate for fade
pean Union’s Sixth Framework Programme (FP6) Informa- events introduced by the mobile channel. The third paper
tion Society Technologies (IST) Thematic Area. Led by the “Performance analysis of SSC diversity receivers over corre-
German Aerospace Center, SatNEx brings together a network lated Ricean fading satellite channels” investigates the per-
of 24 partners, distributed throughout Europe, with mem- formance of dual-branch SSC receivers for different fading
bership drawn from ten countries. channel characteristics.
The philosophy underlying the SatNEx approach re- The next four papers deal with the emerging scenario
volves around the selection of focused actions under Joint of mobile digital video broadcasting (DVB-S2 and RCS mo-
Programmes of Activities, which are carried out collectively bile). Alternative approaches to counteracting fading chan-
by the partners and include research, integration, and dis- nels introduced when operating in a train environment re-
semination activities. Training represents an important part ceiving satellite DVB-S2 are presented in the paper “Ad-
of the SatNEx remit and is supported through a number of vanced fade countermeasures for DVB-S2 systems in railway
initiatives including the hosting of internship projects and an scenarios.” Here, as a result of simulation analysis, antenna
annual summer school. diversity and packet-level forward error correction mecha-
The call for papers resulted in a high number of submis- nisms are proposed and their impact is evaluated with respect
sions, from which we have been able to select 12 excellent to the receiver design and system complexity. The theme of
papers dealing with the different aspects of satellite commu- DVB-S2 is continued with the paper “Capacity versus bit er-
nications and navigation. ror rate trade-off in the DVB-S2 forward link,” which inves-
Multiple-input multiple-output (MIMO) techniques are tigates how satellite capacity can be optimised for DVB-S2
attracting a considerable amount of attention from within transmissions. The DVB return channel via satellite (DVB-
the terrestrial wireless community. The first paper of this spe- RCS) is then addressed in “Frequency estimation in iterative
cial issue, “Multisatellite MIMO communications at Ku band interference cancellation applied to multibeam satellite sys-
and above: investigations on spatial multiplexing for capac- tems,” which considers the application of interference cancel-
ity improvement and selection diversity for interference mit- lation on the reverse link of a multibeam satellite system, us-
igation,” considers the application of such technology over a ing DVB-RCS with convolutional coding as an example. The
satellite platform operating in the Ku band and above. The paper “A QoS architecture for DVB-RCS next-generation
paper considers how MIMO can be used to increase capac- satellite networks” proceeds to design and emulate a quality-
ity by using a satellite spatial multiplexing system and how of-service (QoS) architecture that demonstrates using real
antenna selection can be used to mitigate interference. The multimedia applications how QoS can be supported over a
next paper “Investigations in satellite MIMO channel model- DVB-RCS network.
2 EURASIP Journal on Wireless Communications and Networking
ACKNOWLEDGMENTS
It has been a pleasure for us to have put together this spe-
cial issue, which we hope you will find interesting. We would
like to thank the editorial staff at Hindawi for their sup-
port and assistance during the preparation of this special is-
sue. We would like to thank the contributing authors for the
excellent quality of their submissions and our SatNEx col-
leagues for their valuable assistance in the reviewing of pa-
pers. SatNEx is partially funded by the European Commis-
sion under the Sixth Framework Programme. Further in-
formation on SatNEx can be found on the project web site:
http://www.satnex.org/.
Ray E. Sheriff
Anton Donner
Alessandro Vanelli-Coralli
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 59608, 11 pages
doi:10.1155/2007/59608
Research Article
Multi-Satellite MIMO Communications at Ku-Band and
Above: Investigations on Spatial Multiplexing for Capacity
Improvement and Selection Diversity for
Interference Mitigation
Wireless & Satellite Communications Group, School of Electrical and Computer Engineering, National Technical University of Athens
(NTUA), 9 Iroon Polytechniou Street, Zografou, Athens 15780, Greece
Received 28 August 2006; Revised 2 March 2007; Accepted 13 May 2007
This paper investigates the applicability of multiple-input multiple-output (MIMO) technology to satellite communications at the
Ku-band and above. After introducing the possible diversity sources to form a MIMO matrix channel in a satellite environment,
particular emphasis is put on satellite diversity. Two specific different topics from the field of MIMO technology applications to
satellite communications at these frequencies are further analyzed: (i) capacity improvement achieved by MIMO spatial multi-
plexing systems and (ii) interference mitigation achieved by MIMO diversity systems employing receive antenna selection. In the
first case, a single-user capacity analysis of a satellite 2 × 2 MIMO spatial multiplexing system is presented and a useful analytical
closed form expression is derived for the outage capacity achieved. In the second case, a satellite 2 × 2 MIMO diversity system with
receive antenna selection is considered, adjacent satellite cochannel interference on its forward link is studied and an analytical
model predicting the interference mitigation achieved is presented. In both cases, an appropriate physical MIMO channel model is
assumed which takes into account the propagation phenomena related to the frequencies of interest, such as clear line-of-sight op-
eration, high antenna directivity, the effect of rain fading, and the slant path lengths difference. Useful numerical results obtained
through the analytical expressions derived are presented to compare the performance of multi-satellite MIMO systems to relevant
single-input single-output (SISO) ones.
Copyright © 2007 Konstantinos P. Liolis et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
are encountered. Thus, placing multiple antennas on a sin- antennas (see, e.g., [11]) which allow for compact MIMO
gle satellite does not seem a suitable choice in order to ex- setups. It has already been examined as a promising solu-
ploit the MIMO channel capabilities. In fact, the absence of tion to shape MIMO channels in S-band land mobile satellite
scatterers in the vicinity of the satellite leads to an inherent communications [7, 12–16]. Its main advantage over satellite
rank deficiency of the MIMO channel matrix. Therefore, at a diversity is the elimination of any additional cost associated
first glance, the applicability of MIMO technology to satellite with the utilization of multiple satellites. It also bypasses the
channels does not seem well justified. asynchronism problem associated with the distributed na-
The objective of this paper is in line with some other re- ture of satellite diversity. However, it can be disadvantageous
cent research efforts [4–8, 12–16] casting further light in this to satellite diversity especially in satellite networks operating
regard. These studies have been mainly concerned with the at high-frequency bands (i.e., Ku, Ka, and Q/V), which are
possible diversity sources that can be exploited in satellite affected by the highly correlated rainfall medium and, also,
communications to form a MIMO matrix channel. A cate- in case of large blockages resulting in hard system failures
gorization of these diversity sources follows. (i.e., on/off channel phenomena). Moreover, as concluded in
(i) Site diversity, where multiple cooperating terminal [13], polarization diversity can only increase the transmis-
stations (TSs), sufficiently separated from each other, are in sion rate of a satellite communication system by a factor of
communication with a single satellite. So far, it has only been two, whereas in multi-satellite systems, satellite diversity can
studied as an efficient rain fade mitigation technique at the result in m-fold capacity increase, where m is the number of
Ku (12/14 GHz), Ka (20/30 GHz), and Q/V (40/50 GHz) fre- satellites occupied.
quency bands because of its very low achievable spatial cor- This paper focuses particularly on dual-satellite MIMO
relation due to rain [3]. However, due to the enormous slant communication systems employing satellite diversity. More-
path lengths associated, the required separation distance be- over, emphasis is put on the less congested high-frequency
tween the multiple TSs to ensure ideally independent fading bands, such as Ku and above. At these frequencies, multi-
profile is of the order of several km, which rather hinders its path propagation is insignificant. However, by virtue of satel-
practical interest in MIMO applications. lite diversity, MIMO can be considered to effectively exploit
(ii) Satellite (or orbital) diversity, where multiple satel- the rainfall spatial inhomogeneity instead. A physical 2 × 2
lites, sufficiently separated in orbit to provide (ideally) in- MIMO satellite channel model is assumed taking into ac-
dependently fading channels, communicate with a single TS count the relevant propagation phenomena, such as clear
equipped with either multiple antennas or even a single mul- LOS operation, high antenna directivity, rain fading, and
tiple-input antenna. So far, it has been studied mostly as an rainfall spatial inhomogeneity [3, 17]. This model is flexi-
efficient rain fade mitigation technique in Ku-, Ka-, and Q/V- ble and can be applied on a global scale since it has physical
band satellite communications [3] and, also, recently, as a inputs obtained by regression fitting analysis on the ITU-R
candidate to form satellite MIMO matrix channels at high rainmaps [18] and is based on general assumptions about
(i.e., Ku, Ka, and Q/V) [4, 5] as well as at low frequency the rain process [17]. Moreover, it incorporates the general
bands, such as L (1/2 GHz) and S (2/4 GHz) [6–8]. Also, it case of an ordered MIMO satellite channel (due to the slant
is worthwhile noting that it is already successfully employed path lengths difference). To this end, the resulting propaga-
in the continental US digital audio radio services (DARS), tion delay offset is assumed to be properly taken into account
mobile systems, Sirius and XM satellite radio, operating at at the TS receiver. A possible practical solution to this prob-
the S-band [9]. Satellite diversity provides a rather practical lem might be the one implemented in [5] according to which
solution of reasonable complexity since the multiple received matched filters are first applied to the received signals for the
signals at the single TS can easily be combined due to the detection of the propagation delay offset, which is then fed to
colocation of the antennas. However, an inherent problem a timing aligner. Subsequently, the proposed timing aligner
of this scheme, apart from the costly utilization of multiple eliminates the delay offset by adjusting the timing of a signal
satellites, is the asynchronism of the multiple transmitted sig- parallel-to-serial converter. The study of more efficient solu-
nals at the TS receiver, which comes as a result of the prop- tions to the asynchronism problem associated with satellite
agation delay difference due to the wide separation between diversity, although rather challenging, is out of the scope of
the satellites. A similar problem is dealt with and solutions this paper and will be the subject of a future work.
are proposed in several papers mainly concerning distributed In the first part of this work, emphasis is put on a satellite
sensor networks, such as in [10]. To the authors’ knowledge, 2 × 2 MIMO spatial multiplexing system and on its possi-
for the more complicated satellite case—due to the much ble capacity improvement with respect to the relevant SISO
larger and variable delay difference—the only relevant solu- system. The term “spatial multiplexing” refers to the trans-
tion proposed so far is reported in [5]. mission of independent data streams from the multiple sep-
(iii) Polarization diversity, where a single dual-orthogonal arate satellites [1, 2]. Well-known results obtained from the
polarized satellite communicates with a single TS equipped MIMO literature [19, 20] are applied here for the capacity
with a dual-orthogonal polarized antenna. Its principle is analysis of such a 2 × 2 MIMO system. The figure of merit
based on the polarization sensitivity of the reflection and used to characterize the resulting MIMO fading channel is
diffraction processes, which causes random signal fading at the outage capacity [1], for which an analytical closed form
the TS receiver. It represents a solution of rather practical expression is provided. Note that such analytical expressions
interest due to the recent developments in MIMO compact are extremely hard to obtain even in the well-established field
Konstantinos P. Liolis et al. 3
S1 S2
To S1 To S2
d1 , AR1 d2 , AR2
Δθ
ϕ2
ϕ1
TS TS
(a) (b)
Figure 1: (a) Configuration of a dual-satellite 2 × 2 MIMO channel. Individual satellites S1 and S2 transmit either independent data streams
(MIMO spatial multiplexing system, Section 3) or the same signal over the multiple (ideally) independently fading paths (MIMO diversity
system, Section 4), (b) associated elevation angles.
of MIMO theory due to the intractability of the outage ca- analysis for the possible interference mitigation achieved by a
pacity distribution [2]. satellite 2 × 2 MIMO diversity system with receive antenna se-
In the second part, a satellite 2 × 2 MIMO diversity sys- lection is presented in Section 4. Useful numerical results ob-
tem employing receive antenna selection is examined, and tained for both the above satellite MIMO applications con-
issues specifically related to cochannel interference (CCI) are sidered are provided in Section 5. Section 6 concludes the
addressed from a propagation point of view. The term “di- paper.
versity” refers to the transmission of the same signal over the
multiple (ideally) independently fading paths [1, 2]. Receive
antenna selection is a low-cost, low-complexity approach to 2. MIMO SATELLITE CHANNEL MODEL
benefit from many of the advantages of MIMO technology
while, at the same time, bypassing the multiple RF chains Figure 1 depicts the configuration of a dual-satellite MIMO
associated with multiple antennas at the receiver, which are communication channel at the Ku-band and above. The TS
costly in terms of size, power, and hardware [21]. The inter- is equipped with two colocated highly directive antennas and
ference analysis presented here is quite different from con- communicates with two satellites, S1 and S2 , subtending an
ventional communication-oriented approaches followed in angle Δθ to the TS, large enough that the spatial correlation
standard MIMO theory [1]. Attention is paid to the CCI due to rain along the relevant slant paths is as low as possible.
problems arising on the forward link of such a 2 × 2 MIMO The normalized radiation pattern of each TS antenna, de-
satellite system due to differential rain attenuation from an noted by GR (·), is compatible with the ITU-R specifications
adjacent satellite [22]. To deal with the statistical behaviour [25] and is shown in Figure 2.2 The lengths of slant paths
of the signal-to-interference ratio (SIR) introduced by the Si -TS are denoted by di (i = 1, 2) and the random variables
rainfall spatial inhomogeneity, the concept of unacceptable (RVs) associated with the respective rain induced attenua-
interference probability1 [23, 24] is employed here. An ana- tions (in dB) are denoted by ARi (i = 1, 2). In general, the
lytical prediction model concerning the interference mitiga- two slant paths Si -TS have different elevation angles denoted
tion achieved by the proposed satellite 2 × 2 MIMO diversity by φi (i = 1, 2), respectively.
system is provided. Assuming that clear LOS between the TS and each satel-
The rest of the paper is organized as follows. Section 2 lite Si exists, that each TS antenna is at boresight with the
presents the channel model adopted for MIMO satellite com- corresponding satellite Si (i = 1, 2) and that rain attenuation
munications at the Ku-band and above. Section 3 provides a is the major fading mechanism, the path gain for each Si -TS
communication-based capacity analysis for a satellite 2 × 2 link is modeled as
MIMO spatial multiplexing system. A propagation-oriented
gi ∝ GR 0◦ · di−2 · 10−AR i /10 (i = 1, 2). (1)
1 Note that the concept of the “unacceptable interference probability
(UIP)” in this paper is exactly the same as that of the “acceptable interfer-
ence probability (AIP)” employed in [23, 24]. Their only difference con- 2 Note that the analyses presented hereafter are quite general and, therefore,
cerns their nomenclature. may incorporate other TS antenna radiation patterns, as well.
4 EURASIP Journal on Wireless Communications and Networking
0 1
−10
0.8
−15
0.7
−20
0.6
−25
0.5
−30
−35 0.4
−40 0.3
−100 −80 −60 −40 −20 0 20 40 60 80 100 0 20 40 60 80 100 120 140 160 180
Off-axis angle (deg) Angular separation, Δθ (deg)
Figure 2: Normalized radiation pattern of each TS antenna com- Figure 3: Spatial correlation coefficient due to rain ρ12 versus an-
patible with ITU-R specifications [25]. gular separation Δθ for a dual-satellite MIMO channel operating in
Atlanta, GA, at the Ka-band with satellite elevation angles φ1 = 45◦
and φ2 = 40◦ .
Hence, the total path loss along each Si -TS link (in dB) is
2
H = 11 12 PT P
h21 h22 C = log2 det I2 + HHH = log2 1 + T λi ,
⎡
⎤ 2N0 i=1
2N 0
√ j2πd1 f (3)
⎢ g1 exp 0 ⎥ (4)
=⎢ c
⎥
⎣ √ j2πd2 f ⎦.
0 g2 exp where I2 is the 2 × 2 identity matrix, PT the total average
c
power available at the transmit side,3 N0 the noise spectral
The diagonal structure of H is due to the high directivity of
the TS antennas and the large value of Δθ. In MIMO ter-
3 Note that PT is the sum transmit power of all transmitting satellites Si re-
minology, channels with diagonal H matrix are known as
parallel MIMO channels. Further details about such chan- gardless of their number. This means that in both the dual-satellite MIMO
case and the single satellite SISO case, the total available transmit power
nels can be found in [26]. Moreover, as opposed to standard is constant and equal to PT . This is ensured employing the normalization
MIMO theory [1, 2], H is not normalized here (i.e., ordered factor “2” in (4), which allows for a fair comparison between the relevant
MIMO channel) due to the different slant path lengths di MIMO and SISO cases.
Konstantinos P. Liolis et al. 5
density at the TS receiver input, and λi (i = 1, 2) the positive The quantities AmRi , SaRi (i = 1, 2), encountered in (8)–(11),
eigenvalues of the matrix HHH (the superscript H stands for are the statistical parameters of the lognormal RVs ARi (i =
conjugate transposition). 1, 2) given by [17]
Taking into account the channel modeling assumptions,
Hi
(4) is written as S2aRi = ln 1 + exp b2 S2r − 1 (i = 1, 2),
L2Di
2 2 (12)
b Sr − S2aRi
2
C= log2 1 + 0.5SNRCSi 10−ARi /10 , (5) AmRi = aRbm LDi exp (i = 1, 2),
i=1 2
where SNRCSi (i = 1, 2) are the nominal SNR values under where LDi (i = 1, 2) are the projections of the effective path
clear sky conditions. Based on the path gain model given in lengths Li (i = 1, 2) [17] on the earth surface, Hi (i = 1, 2) are
(1), the SNRCSi values (in dB) are related through spatial parameters related to each path of length LDi (i = 1, 2)
which may be found in [17], and a, b are constants depend-
d2
SNRCS1 − SNRCS2 = 20 log10 . (6) ing on the operating frequency f , the polarization tilt angle,
d1 the temperature, and the rainfall characteristics over the ser-
Equation (5) provides an expression for the instantaneous viced area. Rm , Sr are the lognormal statistical parameters of
capacity of a deterministic 2 × 2 MIMO channel H. How- the rainfall rate R (in mm/hr). A reliable database of rainfall
ever, since the rainfall introduces slow fading and stochastic statistics for any geographical location on earth is provided
behaviour over the channel H, the appropriate statistic mea- by ITU-R in [18] and is used throughout the present work as
sure to characterize the resulting fading channel is the outage an input to the simulations performed in order to determine
capacity defined by [1] the values of Rm , Sr .
P C ≤ Cout,q = q, (7)
4. SATELLITE MIMO DIVERSITY SYSTEM
where Cout,q is the information rate guaranteed for (1− WITH RECEIVE ANTENNA SELECTION:
q)100% of the channel realizations. INTERFERENCE ANALYSIS
Consider the RV transformation
In this section, the two satellites Si (i = 1, 2) depicted in
ln ARi − ln AmRi Figure 1 are assumed to transmit the same signal over the
ui = (i = 1, 2) (8)
SaRi (ideally) independently fading paths Si -TS (i = 1, 2) (i.e., di-
which relates the lognormal rain attenuation RVs ARi (i = versity is investigated). To alleviate the high cost and com-
1, 2) to the normalized normal RVs ui (i = 1, 2). Substituting plexity associated with multiple RF chains, the dual-antenna
(5) into (7) and after some straightforward algebra, the fol- TS receiver is equipped with only one RF chain and performs
lowing analytical closed form expression for the outage ca- antenna selection, that is, the 2 × 2 MIMO satellite system
pacity is obtained: assumed employs receive selection diversity [21]. Therefore,
the TS receiver detects the signal related to the path with the
P C ≤ Cout,q highest SNR. Under the constraint of only one RF chain at
+∞ the receiver, in order to know all SNRs simultaneously for
1 uB − ρn12 u1
= du1 fU1 u1 erfc = q, optimal selection, a training signal in a preamble to the trans-
2 uA 2
2 1 − ρn12 mitted data is assumed. During this preamble, the TS receiver
(9) scans the two antennas, finds that one with the highest SNR,
and selects it for reception of the next data burst. Thus, only
where erfc(·) is the complementary error function, fU1 (u1 )
a few more training bits are required instead of additional RF
the probability density function (pdf) of the normal distri-
chains.
bution, ρn12 the logarithmic correlation coefficient between
Particular emphasis is put on possible interference mit-
the normal RVs ui (i = 1, 2) [17] and uA , uB are analytically
igation offered by the proposed satellite 2 × 2 MIMO di-
given by
versity system. In this regard, a propagation-based analy-
uA = ln 10 log10 0.5SNRCS2 − 10 log10 2Cout,q − 1 sis is performed which is quite different from conventional
communication-oriented approaches followed in standard
− ln AmR2 SaR2 , MIMO theory [1]. Specifically, the effect of rainfall on the
interference analysis is taken into account and the differential
rain attenuation related to an adjacent satellite is considered
(10) as the dominant cause of the SIR degradation [22]. Such an
uB = ln 10 log10 0.5SNRCS1 interference problem is further aggravated due to the spa-
tial inhomogeneity of the rainfall medium. It constitutes a
+ 10 log10 1 + 0.5SNRCS2 10−AmR2 exp(u1 SaR2 )/10 typical interference scenario, especially over congested urban
− 10 log10 2Cout,q − 1 − 0.5SNRCS2 10−AmR2 exp(u1 SaR2 )/10 areas, where the increased demand for link capacity and ra-
dio coverage imposes the coexistence of many satellite radio
− ln AmR1 SaR1 . links over the same geographical and spectral area. In the fol-
(11) lowing, an analytical prediction model is presented, which
6 EURASIP Journal on Wireless Communications and Networking
S1 S3 S2
To S3
To S2
To S1
d3 , AR3
d1 , AR1 d2 , AR2
Δψ
Δθ
ϕ3
ϕ2
ϕ1
TS TS
(a) (b)
Figure 4: (a) Configuration of the satellite 2 × 2 MIMO diversity system assumed and the interference scenario on its forward link, (b)
associated elevation angles.
quantifies the adjacent satellite CCI mitigation achieved by is true. Assuming that
the proposed 2 × 2 MIMO system with respect to the corre-
sponding SISO one. Ωi = Ai < Md , Ai < A j (i, j) = (1, 2), (2, 1) (15)
Figure 4 depicts the configuration of the assumed inter-
ference scenario on the forward link of a satellite 2 × 2 MIMO denotes the event that “the TS is serviced by the correspond-
diversity system operating at the Ku-band and above and ing satellite Si (i = 1, 2),” it becomes clear that, due to selec-
employing receive antenna selection. The satellites S1 and tion diversity,
S2 constitute the dual-satellite transmit part of the MIMO Ω = Ω1 ∪ Ω 2 ,
system also depicted in Figure 1. Another cochannel satellite (16)
(denoted by S3 ), which may belong to either the same or to Ω1 ∩ Ω2 = ∅ .
another satellite network, is close in orbit to S1 . Thus, CCI
Therefore, the probability that the system is available (see the
problems may arise on the forward link of the 2 × 2 MIMO
appendix) can be expressed as
satellite system. S1 and S3 subtend an angle Δψ to TS. The
length of the slant path S3 -TS is denoted by d3 , while its ele-
P(Ω) = P Ω1 + P Ω2 . (17)
vation angle is φ3 . The RV associated with the rain induced
attenuation along the interfering path S3 -TS (in dB) is de- While the satellite 2 × 2 MIMO diversity system is avail-
noted by AR3 . able (i.e., when either Ω1 or Ω2 are true), it might suffer from
Due to selection diversity at the TS receiver, the antenna CCI originating from the adjacent satellite S3 . If SIRd and rd
with the maximum SNR is selected. In mathematical terms, denote the SIR and the minimum acceptable SIR threshold
the same statement is expressed as of the MIMO diversity system, respectively (both measured
at the output of the TS selection combiner), the probability
SNRout = max SNR1 , SNR2 ⇐⇒ Aout = min A1 , A2 , of the event that “the system is interfered while being avail-
(13) able” can be mathematically expressed based on the above
considerations as
where SNRi = SNRCSi − ARi (i = 1, 2) is the SNR at each TS
antenna under rain fades and Ai (i = 1, 2) the total path loss UIPd = P SIRd < rd , Ω
along each Si -TS link (i = 1, 2). SNRout corresponds to Aout
= P SIRd1 < rd , Ω1 + P SIRd2 < rd , Ω2 (18)
which determines the output of the selection combiner at ev-
ery instant. The proposed scheme requires only the knowl- = P1 + P2 ,
edge of the wanted signals’ channels at the receiver, whereas
where UIPd is the so-called unacceptable interference proba-
knowledge of the interferer’s channel is not necessary. More-
bility (UIP) [23, 24], and the quantities SIRdi (i = 1, 2) are
over, no CSI is required at the transmit side. If Md denotes
expressed (in dB) as
the diversity system margin associated with the system avail-
ability pavail (see the appendix), the satellite MIMO diversity SIRd = SIRdi = SIRCSi − ARi + AR3 (i = 1, 2). (19)
system is considered available when the probabilistic event
In (19), SIRCSi (i = 1, 2) is the nominal SIR value under
Ω = Aout < Md (14) clear sky conditions. In propagation terminology, ARi − AR3
Konstantinos P. Liolis et al. 7
(i = 1, 2) is known as the differential rain attenuation (DRA) the other two normal RVs u1 , u2 and can be expressed in
[22]. Based on (19), when DRA becomes sufficiently large terms of the logarithmic correlation coefficients ρni j ((i, j) =
due to the spatial inhomogeneity of the rainfall medium, se- (1, 2), (1, 3), (2, 3)) as [17, 27]
vere CCI problems may arise aggravating the SIRd distribu-
tion on the forward link of the proposed satellite 2 × 2 MIMO ρn13 − ρn12 ρn23 ρn23 − ρn12 ρn13
diversity system. To this end, UIPd is proposed as an efficient μ3/1,2 = 2 u1 + 2 u2 ,
metric to deal with the statistical behaviour of the SIRd and, 1 − ρn12 1 − ρn12
(23)
together with rd , they constitute a pair of design specifica- 2
1 − ρn12 2
− ρn13 2
− ρn23 + 2ρn12 ρn13 ρn23
2
tions concerning interference. Every user must comply with σ3/1,2 = 2 .
1 − ρn12
these specifications, given the QoS specified by the event Ω
related to the system availability (see the appendix).
The quantities SIRCSi (i = 1, 2) encountered in (19) are 5. NUMERICAL RESULTS AND DISCUSSION
given by
The previous analyses have been applied for the prediction
SIRCSi = SIR∗i − GR θi (i = 1, 2), (20) of possible capacity improvement and interference mitiga-
where θi (i = 1, 2) are the off-axis angles formed by the in- tion achieved by the proposed satellite 2 × 2 MIMO spa-
terfering link S3 -TS and the wanted links Si -TS (i = 1, 2) in tial multiplexing and diversity systems, respectively, and for
the radiation pattern of the TS antennas. From Figure 4, it comparison to the relevant SISO cases. To this end, the base-
follows that θ1 = Δψ and θ2 = Δθ − Δψ. Also, in (20), SIR∗i line configuration scenario considers a TS located in At-
(i = 1, 2) are the relevant SIR values of the interfered links lanta, GA, and communicating with geostationary satellites
Si -TS (i = 1, 2) when θi = 1◦ , and correspond to the nominal S1 (φ1 = 45◦ ) and S2 (φ2 = 40◦ ). The angular separation as-
CCI levels. Based on the channel model assumed, their inter- sumed is Δθ =40◦ , which results in a spatial correlation coef-
relationship is defined through (6) by simply substituting the ficient of rain attenuation ρ12 = 0.6 (see Figure 3). Moreover,
SNRCSi by SIR∗i . regarding the interference scenario, an adjacent geostation-
Extending the transformation given in (8) to include also ary satellite S3 (φ3 = 45◦ ), separated from S1 by Δψ =10◦ , is
the interfering link S3 -TS (i.e., for i = 1, 2, 3) and making the considered to cause CCI problems on the forward link of the
channel modeling assumptions, the probabilities Pi (i = 1, 2) satellite 2 × 2 MIMO diversity system.
encountered in (18) after some straightforward algebra are First, the validity of the proposed analytical model in (9),
evaluated, that is, predicting the outage capacity achieved by a satellite 2 × 2
MIMO spatial multiplexing system, is numerically verified.
uDi +∞
The effect of various geometrical and operational system pa-
Pi = du1 du2 fU1 U2 u1 , u2
uCi u1 rameters on the outage capacity distribution is also exam-
(21) ined.
1 uEi − μ3/1,2
× 1− erfc √ (i = 1, 2), Figure 5 shows the dependence of the 1% outage capac-
2 2σ3/1,2
ity of the assumed 2 × 2 MIMO satellite system on the SNR.4
where fU1 U2 (u1 , u2 ) is the pdf of the two-dimensional joint The baseline configuration scenario is adopted, whereas the
normal distribution. operating frequency band assumed is Ka (i.e., f = 20 GHz).
For i = 1, 2, the rest of the parameters encountered in For the sake of comparison, the capacity of the relevant SISO
(21) are system is also plotted. Together with the analytical results
obtained from the analytical closed form expression in (9),
ln xdi − ln AmRi Monte Carlo simulation results are also plotted for verifica-
uCi = ,
SaRi tion. The agreement observed between the analytical and the
⎧
⎪
⎪ rd > SIRCSi , simulation results is very good over the whole SNR range.
⎨0,
As can be seen, the difference between the relevant MIMO
xdi = ⎪ SIRCSi − rd cos φi , SIRCSi +FSLi − Md<rd ≤ SIRCSi ,
⎩M − FSL cos φ ,
⎪
rd ≤ SIRCSi + FSLi − Md ,
and SISO curves diminishes at very low SNR levels while
d i i it becomes significant as the SNR increases. As an illustra-
tion, for SNR = 10 dB, the spectral efficiency achieved by
ln Md − FSLi cos φi − ln AmRi
uDi = , the MIMO system is 4.84 bps/Hz, whereas the one achieved
SaRi
by the SISO system is 3.23 bps/Hz. This constitutes, approx-
exp ui SaRi AmRi imately, a 50% increase in user data rate obtained by MIMO
uEi = ln − SIRCSi + rd cos φ3
spatial multiplexing. For SNR = 20 dB, the respective per-
cos φi
formance figures obtained are 10.95 bps/Hz and 6.41 bps/Hz
− ln AmR3 SaR3 . corresponding to, approximately, a 71% increase in user data
(22)
4 Note that the clear sky SNR of strong eigenmode, SNRCS1 , has been par-
AmRi , SaRi (i = 1, 2, 3) are analytically given in (12). Fur- ticularly considered. However, due to the enormous slant path lengths as-
thermore, μ3/1,2 and σ3/1,2 are the statistical parameters of sociated, the resulting difference between SNR CSi (i = 1, 2) is minimum
the conditional distribution of the normal RV u3 given see (6) and, therefore, any of the two SNRCSi can be used as x-coordinates.
8 EURASIP Journal on Wireless Communications and Networking
100 100
10−1 10−1
10−2 10−2
Ka-band,
10−3 pavail = 99.9% 10−3 Δψ = 4◦ Δψ = 5◦
Ku-band,
pavail = 99.99%
10−4 10−4
Ka-band,
pavail = 99.99%
10−5 10−5
10−6 10−6
2 4 6 8 10 12 14 16 18 20 7 9 11 13 15 17 19 20
SIR (dB) SIR (dB)
SISO SISO
2 × 2 MIMO 2 × 2 MIMO
Figure 7: UIP versus SIR for a satellite 2 × 2 MIMO diversity sys- Figure 9: UIP versus SIR for a satellite 2 × 2 MIMO diversity sys-
tem employing receive antenna selection. Relevant SISO case is also tem employing receive antenna selection. Relevant SISO case is also
plotted for comparison. Effect of system availability pavail , operating plotted for comparison. Effect of angular separation ΔΨ.
frequency f , and rain climatic conditions over the serviced area.
2.5 ters assumed are the same as those in the baseline configura-
tion scenario, with the exception of a different angular sepa-
Atlanta, GA ration Δψ, that is, Δψ = 5◦ is now assumed. Operation of the
SIR improvement achieved through
2
system at the Ka-band and for a 99.99% availability is con-
sidered. To obtain the necessary QoS for UIP = 0.001%, sup-
MIMO diversity (dB)
Useful analytical results are presented for two different After straightforward algebra, (.2) yields
applications of MIMO technology:
(i) capacity improvement in a satellite 2 × 2 MIMO spatial pavail · 100%
multiplexing system, +∞
uF2 − ρn12 u1 (.4)
(ii) interference mitigation in a satellite 2 × 2 MIMO di- = 1 − 0.5 du1 fU1 u1 erfc .
2
versity system with receive antenna selection.
uF1 2 1 − ρn12
Research Article
Investigations in Satellite MIMO Channel Modeling:
Accent on Polarization
Péter Horváth,1 George K. Karagiannidis,2 Peter R. King,3 Stavros Stavrou,3 and István Frigyes1
1 Department of Broadband Infocommunications and Electromagnetic Theory, Budapest University of Technology and Economics,
H-1111 Budapest, Hungary
2 Division of Telecommunications, Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki,
Due to the much different environment in satellite and terrestrial links, possibilities in and design of MIMO systems are rather
different as well. After pointing out these differences and problems arising from them, two MIMO designs are shown rather well
adapted to satellite link characteristics. Cooperative diversity seems to be applicable; its concept is briefly presented without a de-
tailed discussion, leaving solving particular satellite problems to later work. On the other hand, a detailed discussion of polarization
time-coded diversity (PTC) is given. A physical-statistical model for dual-polarized satellite links is presented together with mea-
suring results validating the model. The concept of 3D polarization is presented as well as briefly describing compact 3D-polarized
antennas known from the literature and applicable in satellite links. A synthetic satellite-to-indoor link is constructed and its elec-
tromagnetic behavior is simulated via the FDTD (finite-difference time-domain) method. Previous result of the authors states that
in 3D-PTC situations, MIMO capacity can be about two times higher than SIMO (single-input multiple-output) capacity while a
diversity gain of nearly 2 × 3 is further verified via extensive FDTD computer simulation.
Copyright © 2007 Péter Horváth et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
In one class of cases, the ground terminals are located on- Compact antennas with low radiator spacing and dimensions
board large objects, such as trains, ships, or aircrafts. Large- as small as λ/20 or so are described, for example, in [12–
antenna distances are possible then, realizing diverse routes. 14]. These antennas were mainly developed for application
Multipath, on the other hand, is nonexistent or very sparse. in handheld terminals, in which the available space is very
Difference of LOS route lengths must be in such a case at limited. In the case of onboard antennas, the whole antenna
least λ/16 · · · λ/4. Site diversity might be applicable then, if need not be small, however, the radiator elements need to be
as a rough estimate, terminal antennas can be placed at a colocated, that is, their ports need to be very close to each
distance of b = 35 m from each other. (For that figure, an other. Note that polarization, and in many cases the 3D char-
LEO satellite and 30 GHz carrier frequency were assumed; acter of it, has a significant role in each of the known compact
note that b is proportional to the square root of satellite antennas.
distance × wavelength.) In this paper, the concept of cooperative satellite diversity
Satellite diversity for space-time processing would fulfill is briefly introduced, without, however, a detailed discussion;
the requirement of uncorrelated channels and so it would be this is done in Section 2. Polarization diversity and the appli-
applicable. There is a few papers dealing with this topic; for cation of space-time coding concepts in polarization diver-
example, [2] gives a physical-statistical model for satellite-to- sity are dealt with in Section 3. (In analogy to the name STC,
urban and satellite-to-highway channel and computes capac- we call that polarization time coding (PTC). Note that ac-
ity of a 2 × 2 MIMO system. In [3], a satellite-diversity MIMO cording to the authors’ understanding, the term STC is used
system and its system aspects are investigated. Further papers to distinguish a transmit-and-receive-space-diversity situa-
on satellite MIMO are, among others, [4, 5]. tion from a simple receive diversity. The same understanding
There exists, however, at least one problem not present is applied in this paper; so we will call our topic PTC even if
in terrestrial systems, that is, that of synchronization. In ter- particular coding problems are not at all dealt with but coded
restrial MIMO systems, both the group of transmit anten- signals are assumed.) Section 3.1 deals with dual-polarized
nas and that of receive antennas are at distances from each MIMO channels, stating a physical-statistical model, pre-
other in the order of a wavelength. Consequently, the path senting measuring results and validating the model; in this
lengths of the diversity routes are very closely identical, and discussion conventional dual-polarized antennas are applied.
thus signals arriving from the transmitter to the receiver are In Section 3.2, PTC antennas of 3-dimensional polarization
synchronous. This makes identification and decoding of the are dealt with, introducing the concept of 3D polarization,
received signals rather easy. In the case of satellite diversity, presenting a few compact MIMO antennas and showing
the satellites serving as diversity terminals are very far from the essential difference between terrestrial and satellite links
each other. Thus difference of path lengths and so delays be- from the point of view of 3D PTC. In Section 4, electro-
tween the satellites and the ground terminal can be very high magnetic simulation results are given; in these it is verified
and highly variable. (This variability is self-evidently existing that application of the FDTD method is suitable to investi-
in the case of LEO satellites but very likely also in the GEO gate MIMO channel characteristics of very complex environ-
case.) As a consequence, the arrival time of signals from two ments; capacity as well as diversity behavior are presented;
satellites (forming part of a single code word) can be shifted these verify (at least for the present example) the statements
by tens or hundreds of symbol times relative to each other. of Section 3.2 and of the authors’ references [15, 16]. Con-
Synchronization of the received signals is in this case rather clusions are drawn in Section 5.
complicated—both acquisition and tracking. Reference [2]
or [3] or other satellite/MIMO papers known by the authors 2. A FEW WORDS ON COOPERATIVE
do not deal with this problem. General aspects of it are dealt SATELLITE DIVERSITY
with, for example, in [6–8], taking explicitly, however, short-
range, that is, terrestrial situations only into account. In general, cooperative relaying systems have a source node
An alternative possible solution could be cooperative (e.g., TMT) multicasting a message to a number of cooper-
satellite diversity (CSD). In general, cooperative relaying sys- ative relays (SAT), which in turn resend a processed version
tems have a source node (e.g., a terrestrial mobile terminal to the intended destination node (another TMT). The des-
(TMT)) multicasting a message to a number of cooperative tination node combines the signal received from the relays,
relays (satellites (SAT)), which in turn resend a processed ver- possibly also taking into account the source’s original signal.
sion to the intended destination node (another TMT). The An example of a CSD system with two satellite relays is shown
destination node combines the signal received from the re- in Figure 1.
lays, possibly also taking into account the source’s original The idea of merging cooperation with space-time coding
signal. Recently, it has been shown that cooperative diversity resulted in the so-called distributed or cooperative space-time
systems provide an effective way of improving spectral and coding (CSTC). Compared to the conventional space-time
power efficiencies of the wireless networks without the ad- coding with collocated antennas, CSTC can be implemented
ditional complexity of multiple antennas [7–11]. However, a when transmitter and relays share their antennas to create a
study on CSD systems, where the relays are satellites, to the virtual transmit array.
best of the authors’ knowledge does not exist in the literature. A possible cooperation scenario is applied for the con-
A third possible method is to apply compact antennas, figuration of Figure 1, proposed in [9] as TMT1 communi-
in which case the synchronization problem is nonexistent. cates with SAT1 and SAT2 in a broadcasting mode during
Péter Horváth et al. 3
1 100
0.8 10−2
−20 −10 0 10 −45 −40 −35 −30 −25 −20
Power relative to FSL (dB) Power relative to FSL (dB)
1 100
P(fade depth < abscissa)
10−1
P(fade depth < abscissa)
0.9
10−2
0.8
−20 −15 −10 −5 0 −45 −40 −35 −30 −25 −20
Power relative to FSL (dB) Power relative to FSL (dB)
Figure 3: Comparison of modeled and measured cumulative distributions; upper figures: highway channel; lower figures: urban channel.
application. It is also mentioned that Stokes parameters form In the case of multipath propagation (or if the direction
a 4-vector in a Minkowskian space; their transformation, e.g., of propagation is unknown), wave polarization is a 3D phe-
by scatterers or polarization filters, is a Lorentz transforma- nomenon. In that case, the number of orthogonal polariza-
tion [26]; these properties, however, are not used in this dis- tion states is 3. This can increase the number of orthogo-
cussion.) nal channels to 3 if these are discriminated by polarization
6 EURASIP Journal on Wireless Communications and Networking
10−2
In [13], the so-called MIMO cube is dealt with. This con-
tains 12 electric dipoles arranged at the edges of a cube.
10−3
Cube-to-cube capacity and other parameters are computed,
showing surprisingly good performance; note, however, that
10−4 even very small cubes are investigated, (cube edges as short as
0.05λ) the problem of superdirectivity is not stressed in that
10−5 paper.
In [14], behaviors of three colocated monopole and
10−6 dipole antennas are investigated, versus their mutual angles,
0 10 20 30 40 50 60 via simulation. It is shown that their performance is very
Eb /N0 (dB) close to ideally orthogonal ones and also that the main cause
No diversity PTBC (2 Tx, 1 Rx)
of achieving that is their different polarizations rather than
MRRC (1 Tx, 2 Rx) PTBC (2 Tx, 2 Rx) different angular patterns.
Figure 4: Bit error rate curves for highway environment. 3.2.3. Compact antennas and 3D polarization in satellites
Window
t(t) r(t)
x = 2.8 m
Incident O1 O2 O3
wave
Scattering Scattering
medium medium
(a)
y = 4.5 m
Figure 5: Terrestrial multipath links with compact MIMO anten- In order to assess the performance of using three orthog-
nas in scattering media; (a) double-bounce scattering; (b) single onally polarized antennas in a satellite-to-indoor scenario,
bounce. some simulations were performed using full-wave electro-
magnetic tools. The FDTD method [30] was used to calculate
the time-dependent electromagnetic field inside a typical of-
t(t) fice room where the mobile terminal is assumed to be placed.
The office dimensions were 2.8 m × 4.5 m × 3.0 m (x, y, z),
where the floor and the ceiling are lying in and parallel to the
x-y plane, respectively, as seen in Figure 7. In the simulation,
r(t) the furniture and the walls of the room are modeled by re-
alistic material properties (brick walls, wooden and metallic
Plane wave furniture, and some plastic objects). These objects of vari-
ous geometries are nearly uniformly distributed in the room.
Aperture
Linear orthogonally polarized plane waves enter the room
Scattering through the window and through the external wall; one po-
medium larization during the first simulation run and the other one
during a subsequent run. This method allows us to split the
Figure 6: A satellite-to-mobile/indoor link. channel response according to the incoming polarizations.
The waveform is a modulated Gaussian pulse centered at
1.2 GHz, entering through the x-z plane at y = 0 m.
The electric field components (Ex , E y , and Ez ) are
This has the consequence that from the coding point of view, recorded at various spots in the room. We use these field
the system is not uniform. If as an example, space-time block components directly to draw conclusions about the signals
coding of the Alamouti type or orthogonal space-time block (voltages) which three antennas would produce if they would
coding (OSTBC) is chosen, RC = 1 can be applied downlink, be placed at a given observation point. Although this ap-
however in the uplink RC = 1/2 or at most RC = 3/4 can proach does not consider the current distribution on elec-
only be achieved. (RC designates the coding rate.) It is ques- trically long antennas, mutual coupling, scattering by the an-
tionable if this can be accepted from the frequency economy tennas, and so forth, previous FDTD studies demonstrated
point of view. If not, only two of the three or more antennas that only a very low crosstalk exists between three thin-wire
are used in the uplink transmitter. Note that other types of half-wave dipoles which are mounted parallel to the coor-
coding can give different results. dinate axes in an empty room [16]. Therefore, the results
On the other hand, the number of diversity routes is can be regarded as realistic, for short orthogonally mounted
increased—say up to 2 × 3. (This is valid if terminal antenna dipoles. The field components are recorded along various
is a tripole; with a vector element antenna, this is 2 × 6, with x-z cross-sections of the room, at three different observa-
a MIMO cube even 2 × 12.) tion planes (O1 at y = 1.5 m, O2 at y = 2.4 m, and O3
8 EURASIP Journal on Wireless Communications and Networking
signals). Their CDFs are shown in Figures 9 and 12. As ex- 0.7
pected, in Scenario 2, lower (even negative) correlation is to 0.6
be expected. Additionally, the relative received signal power 0.5
for the (1, 1), (2, 2), and (2, 3) systems and their CDF was also
determined, which results are shown in Figures 10 and 13 0.4
for the scenarios in consideration. Note that the confidence 0.3
for very low-probability (less than 0.01 or so) portions of the
0.2
curve might be low due to the relatively low number (about
2000) of observations, but still validates the claim based on 0.1
the higher probability portion of the curves. 0
−0.2 0 0.2 0.4 0.6 0.8 1
Envelope correlation
5. CONCLUSIONS
ρH y ρV y
ρHz ρVx
The main statement of this paper is that the generalized
coded form of polarization diversity is a very good—maybe
the best—way to apply the MIMO concept in multipath Figure 9: CDF of the envelope correlation (Scenario 1).
satellite links. Two main contributions are related to the
modeling of the conventional (2D) polarization diversity
channel and to the investigation via simulation of the 3D
MIMO channel, respectively. (The relevant signal processing acteristics are investigated. The main purpose of this study
is called here PTC.) was to verify (for this example) the findings of two of these
Concerning the first of these (modeling), a physical sta- authors [15] about the capacity and diversity characteristics
tistical model is given for the urban and the highway satellite of this type of channels. Results of this simulation are as fol-
mobile channels. Besides giving a validated model, it veri- lows.
fies once again the authors’ conviction that the best type of a From the capacity point of view, (i) the difference be-
multipath channel model is of the physical-statistical type. tween the 2 × 2 and the 2 × 3 cases is negligible (as stated in
Concerning the second of these (simulation), a very ex- [15]); and (ii) with high probability capacity of the MIMO,
tensive simulation study is carried out about the 3D polar- the situation is nearly exactly 2-times as high as that of the
ization characteristics of the satellite multipath channel. A SISO case, again in accordance with [15]. (Note that with low
synthetic satellite-to-indoor link is simulated and PTC char- probability, this difference is higher.)
Péter Horváth et al. 9
100 1
0.9
0.8
10−1 0.7
0.6
0.5
10−2 0.4
0.3
0.2
10−3 0.1
0
−50 −45 −40 −35 −30 −25 −20 −15 −10 −0.2 0 0.2 0.4 0.6 0.8 1
Combined received power (dBm) Envelope correlation
nT = 1; nR = 1 ρH y ρV y
nT = 2; nR = 2 ρHz ρVx
nT = 2; nR = 3
Figure 12: CDF of the envelope correlation (Scenario 2).
Figure 10: CDF of the received power (Scenario 1).
1 100
0.9
0.8
Probability (CEP < abscissa)
0.7 10−1
0.6
0.5
0.4 10−2
0.3
0.2
0.1 10−3
0
0 2 4 6 8 10 12 14 16 18 −50 −45 −40 −35 −30 −25 −20 −15 −10
Capacity (bits/s/Hz) Combined received power (dBm)
nT = 1; nR = 1 nT = 1; nR = 1
nT = 2; nR = 2 nT = 2; nR = 2
nT = 2; nR = 3 nT = 2; nR = 3
Figure 11: CDF of the equal-power capacity (Scenario 2). Figure 13: CDF of the received power (Scenario 2).
To characterize the diversity performance, CDF of the re- dealt with in Section 3, that is, the effect of extremely large
ceived power in the various situations is investigated; result and variable difference between the path-lengths of MIMO
shows that 3-fold (i.e., 3D) polarization diversity yields sig- branches must be taken into account.
nificantly higher received power than the 2-fold diversity (or
the nondiversity case).
From the simulation point of view, this study shows that ACKNOWLEDGMENTS
the FDTD method is very well applicable to investigate in an
exact way such extremely complex structures as the one here. This work was done in the framework of and is supported by
A statement of this paper (stated but not discussed in detail) the project SatNEx of the EU IST FP6 Program. Their sup-
talking about satellite-diversity-MIMO, the problems briefly port is gratefully acknowledged.
10 EURASIP Journal on Wireless Communications and Networking
Research Article
Performance Analysis of SSC Diversity Receivers over
Correlated Ricean Fading Satellite Channels
Institute for Space Applications and Remote Sensing, National Observatory of Athens, Metaxa and Vas. Pavlou Street,
15236 Athens, Greece
Received 3 October 2006; Revised 23 February 2007; Accepted 6 April 2007
This paper studies the performance of switch and stay combining (SSC) diversity receivers operating over correlated Ricean fading
satellite channels. Using an infinite series representation for the bivariate Ricean probability density function (PDF), the PDF of
the SSC output signal-to-noise ratio (SNR) is derived. Capitalizing on this PDF, analytical expressions for the corresponding cu-
mulative distribution function (CDF), the moments of the output SNR, the moments generating function (MGF), and the average
channel capacity (CC) are derived. Furthermore, by considering several families of modulated signals, analytical expressions for
the average symbol error probability (ASEP) for the diversity receivers under consideration are obtained. The theoretical analy-
sis is accompanied by representative performance evaluation results, including average output SNR (ASNR), amount of fading
(AoF), outage probability (Pout ), average bit error probability (ABEP), and average CC, which have been obtained by numerical
techniques. The validity of some of these performance evaluation results has been verified by comparing them with previously
known results obtained for uncorrelated Ricean fading channels.
Copyright © 2007 P. S. Bithas and P. T. Mathiopoulos. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
1. INTRODUCTION combined signals are correlated [2, 3]. A typical example for
such signal correlation exists in relatively small-size mobile
The mobile terrestrial and satellite communication channel terminals where typically the distance between the diversity
is particularly dynamic due to multipath fading propaga- antennas is short. Due to this correlation between the signals
tion, having a strong negative impact on the average bit er- received at the diversity branches there is a degradation in the
ror probability (ABEP) of any modulation scheme [1]. Di- achievable diversity gain.
versity is a powerful communication receiver technique used The Ricean fading distribution is often used to model
to compensate for fading channel impairments. The most propagation paths consisting of one strong direct line-of-
important and widely used diversity reception methods em- sight (LoS) signal and many randomly reflected and usually
ployed in digital communication receivers are maximal-ratio weaker signals. Such fading environments are typically en-
combining (MRC), equal-gain combining (EGC), selection countered in microcellular and mobile satellite radio links
combining (SC), and switch and stay combining (SSC) [2]. [2]. In particular for mobile satellite communications the
For SSC diversity considered in this paper, the receiver se- Ricean distribution is used to accurately model the mo-
lects a particular branch until its signal-to-noise ratio (SNR) bile satellite channel for single- [4] and clear-state [5] chan-
drops below a predetermined threshold. When this happens, nel conditions. Furthermore, in [6] it was depicted that the
the combiner switches to another branch and stays there re- Ricean K-factor characterizes the land mobile satellite chan-
gardless of whether the SNR of that branch is above or be- nel during unshadowed periods.
low the predetermined threshold. Hence, among the above- The technical literature concerning diversity receivers op-
mentioned diversity schemes, SSC is the least complex and erating over correlated fading channels is quite rich, for ex-
can be used in conjunction with coherent, noncoherent, and ample, see [7–13]. In [7] expressions for the outage probabil-
differentially coherent modulation schemes. It is also well ity (Pout ) and the ABEP of dual SC with correlated Rayleigh
known that in many real life communication scenarios the fading were derived either in closed form or in terms of
2 EURASIP Journal on Wireless Communications and Networking
single integrals. In [8] the cumulative distribution functions duced. In Section 3, the SSC received signal statistics are pre-
(CDF) of SC, in correlated Rayleigh, Ricean, and Nakagami- sented, while in Section 4 the capacity is obtained. Section 5
m fading channels were derived in terms of single-fold in- contains the derivation of the most important performance
tegrals and infinite series expressions. In [9] the ABEP of metrics of the SSC output SNR. In Section 6, various numer-
dual-branch EGC and MRC receivers operating over corre- ical evaluation results are presented and discussed, while the
lated Weibull fading channels was obtained. In [10] the per- conclusions of the paper can be found in Section 7.
formance of MRC in nonidentical correlated Weibull fad-
ing channels with arbitrary parameters was evaluated. In
2. SYSTEM MODEL
[11] an analysis for the Shannon channel capacity (CC) of
dual-branch SC diversity receivers operating over correlated
By considering a dual-branch SSC diversity receiver operat-
Weibull fading was presented. In [12], infinite series expres-
ing over a correlated Ricean fading channel, the baseband re-
sions for the capacity of dual-branch MRC, EGC, SC, and
ceived signal at the th ( = 1 and 2) input branch can be
SSC diversity receivers over Nakagami-m fading channels
mathematically expressed as
have been derived.
Past work concerning the performance of SSC operat-
ing over correlated fading channels can be found in [14– ζ = sh + n . (1)
17]. One of the first attempts to investigate the performance
of SSC diversity receivers operating over independent and In the above equation, s is the transmitted complex sym-
correlated identical distributed Ricean fading channels was bol, h is the Ricean fading channel complex envelope with
made in [14]. However, in this reference only noncoher- magnitude R = |h |, and n is the additive white Gaus-
ent frequency shift keying (NCFSK) modulation was con- sian noise (AWGN) having single-sided power spectral den-
sidered and its ABEP has been derived in an integral rep- sity of N0 . The usual assumption for ideal fading phase esti-
resentation form. In [15] the performance of SSC diversity mation is made, and hence, only the distributed fading enve-
receivers was investigated for different fading channels, in- lope and the AWGN affect the received signal. Moreover, the
cluding Rayleigh, Nakagami-m and Ricean, and under dif- AWGN is assumed to be uncorrelated between the two diver-
ferent channel conditions but dealt mainly with uncorre- sity branches. The instantaneous SNR per symbol at the th
lated fading. For correlated fading in this reference only the input branch is γ = R2 Es /(2N0 ), where Es = E|s|2 is the
Nakagami-m distribution was studied. In [16] the moments transmitted average symbol energy, where E· denoting ex-
generating function (MGF) of SSC was presented in terms of pectation and | · | absolute value. The corresponding average
a finite integral representation for the correlated Nakagami- SNR per symbol at both input branches is γ = ΩEs /N0 , where
m fading channel. In [17] expressions for the average output Ω = ER2 . The PDF of the SNR of the Ricean distribution
SNR (ASNR), amount of fading (AoF) and Pout for the cor- is given by [2, Equation (2.16)]
related log-normal fading channels have been derived.
All in all, the problem of theoretically analyzing the per- 1+K (1 + K)
fγ (γ) = exp − K − γ
formance of SSC over correlated Ricean fading channels has γ γ
not yet been thoroughly addressed in the open technical lit- (2)
erature. The main difficulty in analyzing the performance of K(K + 1) 1/2
× I0 2 γ ,
diversity receivers in correlated Ricean fading channels is the γ
complicated form of the received signal bivariate probability
density function (PDF), see [14, Equation (17)], and the ab- where K is the Ricean K-factor defined as the power ratio
sence of an alternative and more convenient expression for of the specular signal to the scattered signals and I0 (·) is the
the multivariate distribution. An efficient solution to these zeroth-order modified Bessel function of the first kind [21,
difficulties is to employ an infinite series representation for Equation (8.406)]. The CDF of γ is given by [14, Equation
the bivariate PDF, such as those that were proposed in [18] (8)]
or [19]. Such an approach was used in [20] to analyze the per-
formance of MRC, EGC, and SC in the presence of correlated
√ 2(1 + K)
Ricean fading. Similarly here the most important statistical Fγ (γ) = Q1 2K, γ , (3)
metrics and the capacity of SSC diversity receivers operat- γ
ing over correlated Ricean fading channels will be studied. In
particular, we derive the PDF, CDF, MGF, moments and the where Q1 (·) is the first-order Marcum-Q function [2, Equa-
average CC of such receivers operating over correlated Ricean tion (4.33)].
fading channels. Furthermore, analytical expressions for the The joint PDF of γ1 and γ2 , presented in [14, Equation
average symbol error probability (ASEP) of several modula- (17)], can be expressed in terms of infinite series by follow-
tion schemes will be obtained. Capitalizing on these expres- ing a similar procedure as for deriving [18, Equation (9)].
sions, a detailed performance analysis for the Pout , ASNR, Hence, substituting I0 (·) with its infinite series representa-
AoF, and ASEP/ABEP will be presented. tion [21, Equation (8.445)], expanding the term [γ1 + γ2 +
√
The remainder of this paper is organized as follows. Af- 2 γ1 γ2 cos(θ)]i using the multinomial identity [22, Equa-
ter this introduction, in Section 2 the system model is intro- tion (24.1.2)], using [21, Equation (3.389/1)] and after some
P. S. Bithas and P. T. Mathiopoulos 3
mathematical manipulations the joint PDF of γ1 , γ2 can be Hence, by substituting (4) in (7) and using [21, Equation
expressed as (3.351/2-3)], these integrals can be solved and rssc (γ) can be
expressed as
∞
∞
fγ1 ,γ2 γ1 , γ2 = A exp − β1 γ1 + γ2
i,h=0
rssc (γ) = A exp − β1 γ γβ2 −1/2
v1 +v2 +v3 =i i,h=0
v1 +v2 +v3 =i
β2 −1 β3 −1 β −1/2 β3 −1/2
× Bγ1 γ2 + Cγ−1 γ1 2 γ2 Bγ β3 , β1 γτ Cγ β3 + 1/2, β1 γτ
× √ β + β +1/2 ,
(4) γβ1 3 γβ1 3
(8)
with
where γ(·, ·) is the lower incomplete Gamma function [21,
2v3 +2h−1 (1 + K)1+β4 ρ2h K i exp − 2K/(1 + ρ) Equation (8.350/1)].
A= √ 1+2h ,
πγ1+β4 1 − ρ2 v1 !v2 !v3 !i!(1 + ρ)2i
3.2. Cumulative distribution function (CDF)
1 + (−1)v3 Γ h + 1 + v3 /2 Similar to [23, Equation (20)], the CDF of γssc , Fγssc (γ), is
B= ,
Γ h + 1 + v3 /2 Γ 1 + 2h given by
Fγssc (γ) = Pr γτ ≤ γ1 ≤ γ + Pr γ2 < γτ ∧ γ1 < γ (9)
− 1 + (−1)v3 2ρ(1 + K)Γ 1 + h + v3 /2
C=
,
ρ2 − 1 Γ(2 + 2h)Γ h + 3 + v3 /2 which after some manipulations can be expressed in terms of
CDFs as
⎧
(1 + K) v3 ⎪
⎨Fγ1 ,γ2 γ, γτ , γ ≤ γτ ,
β1 = , β 2 = v1 + + h + 1,
1 − ρ2 γ 2 Fγssc (γ) = ⎪
⎩Fγ (γ) − Fγ γτ + Fγ ,γ γ, γτ , γ > γτ .
1 2
v (10)
β3 = v2 + 3 + h + 1, β4 = i + 2h + 1,
2 γ γτ
(5) Hence, by substituting (4) in Fγ1 ,γ2 (γ, γτ ) = 0 0 fγ1 ,γ2 (γ1 ,
γ2 )dγ1 dγ2 using [21, Equation (3.351/1)], Fγ1 ,γ2 (γ, γτ ) can be
where Γ(·) is the Gamma function [21, Equation (8.310/1)] derived as
and ρ is the correlation coefficient between γ1 and γ2 . It can ∞
A
be proved that the above infinite series expression always Fγ1 ,γ2 γ, γτ = β +β
converges [18]. i,h=0 β1 2 3
v1 +v2 +v3 =i
3. RECEIVED SIGNAL STATISTICS × Bγ β2 , β1 γ γ β3 , β1 γτ
Using [21, Equation (3.381/4)], I1 can be expressed in terms where BW is transmission bandwidth of the signal in Hz.
of infinite series as Hence, substituting (6) in (18), C becomes
∞
∞ ∞
Γ β2 −β3
I1 = A β2 Bβ1 γ β3 , β1 γτ C= log2 (1 + γ)rssc (γ)dγ + log2 (1 + γ) fγ (γ)dγ
β1 + s 0 γτ
i,h=0 (19)
v1 +v2 +v3 =i
= I5 + I6 .
−β3 −1/2 Γ β2 +1/2 1
+Cβ1 β2 +1/2 γ β3+ , β1 γτ . By representing ln(1 + γ) = G1,2 1,1
β1 +s 2 2,2 γ | 1,0 , [27, Equation
(13) (01.04.26.0003.01)], and exp(−γ) = G1,00,1 γ | − , [27, Equa-
0
erfc(·) is the complementary error function [21, Equation In (25), Γ(·, ·) is the upper incomplete Gamma function [22,
(8.250/1)] and D, E are constants the values of which depend Equation (6.51)].
on the specific modulation scheme under consideration. Us- (ii) For noncoherent binary frequency shift keying
ing this expression, by substituting (6) in (22), yields (BFSK) and binary differential phase shift keying (BDPSK),
∞ ∞ Pe (γ) = D exp(−Dγ). Similar to the derivation of (12), that
P se = D erfc Eγ rssc (γ)dγ + D erfc Eγ fγ (γ)dγ is, using [21, Equation (3.381/4)] and [2, Equation (4.33)],
0 γτ P se can be expressed as
= I7 + I8 . ∞
(23) P se = AD
i,h=0
√ v1 +v2 +v3 =i
Expressing erfc( Eγ) = π −1 G2,0 1
1,2 Bγ | 0,1/2 , [27, Equation
Γ β2 B
(06.27.26.0006.01)], and exp(−γ) = − , [27, Equa- G1,0
0,1 γ|0 × β2 β γ β3 , β1 γτ
tion (01.03.26.0004.01)], using [28] and after some straight- β1 + E β1 3
forward mathematical manipulations I7 can be expressed as
CΓ β2 + 1/2 1
+ β2 +1/2 β3 +1/2 γ β3 + , β1 γτ
∞
ADΓ β2 + 1/2 β1 + E β1 2
I7 = √
β
πβ1 3 Eβ2
i,h=0
v1 +v2 +v3 =i 2K(1 + K) 2(1 + K + γE)γτ
+ Q1 ,
1 + K + γE γ
BΓ β2
× γ β3 , β1 γτ K(1 + K) (1 + K) exp(−K)
Γ β2 + 1 × exp .
1 + K + γE 1 + K + γE
1 β1 (26)
× 2 F1 β2 , β2 + ; β2 + 1; −
2 E
(iii) For Gray encoded M-ary PSK and M-ary DPSK,
Λ
Cγ β3 + 1/2, β1 γτ Γ β2 + 1 Pe (γ) = D 0 exp[−E(θ)γ]dθ, where Λ is constant. Thus, P se
+
1/2 3 can be expressed as
β1 E Γ β2 +
2 ∞
P se = AD
1 3 β1 i,h=0
× 2 F1 β2 + , β2 + 1; β2 + ; − v1 +v2 +v3 =i
2 2 E
(24) Λ
Bγ β3 , β1 γτ Γ β2
× β
β2 dθ
with 2 F1 (·, ·; ·; ·) being Gauss Hypergeometric
function [21, β1 3 0 β1 + E(θ)
∞
Equation (9.100)]. Moreover, I8 = 0 D erfc( Eγ) fγ (γ)dγ −
γτ Cγ β3 + 1/2, β1 γτ
+
0 D erfc( Eγ) fγ (γ)dγ = I8,a − I8,b . Hence, substituting β +1/2
β1 3
again I0 (·) with its infinite series representation [21, Equa-
Λ
tion (8.445)], I8,a can be solved with the aid of [28] and I8,b Γ β2 + 1/2)
using [27, Equation (06.27.21.0019.01)]. Thus, using these ×
β2 +1/2 dθ
0 β1 + E(θ)
solutions of I8,a and I8,b and after some mathematical ma-
nipulations, I8 can be expressed as in (25): Λ
2K(1 + K) 2g(θ)γτ
∞ k + Q1 ,
D(1 + K) exp(−K) K(K + 1) 0 g(θ) γ
I8 = (k!)−2
γ k=0
γ K(1 + K) (1 + K) exp(−K)
× exp dθ,
g(θ) g(θ)
Γ(k + 1)Γ(k + 3/2) (27)
× √ k+1
πE Γ(k + 2)
where g(θ) = 1 + K + γE(θ). The above finite integrals can be
3 1+K easily evaluated via numerical integration.
× 2 F1 k + 1, k + ; k + 2; −
2 γE
√ ∞
ρ 5.3. Average output SNR (ASNR) and
2 E/π − (1 + K)/γ Eρ
−
k+3/2 amount of fading (AoF)
β1 1 − ρ 2 ρ=0
(2ρ + 1)ρ!
The ASNR, γssc , is a useful performance measure serving as
3 (1+K)γτ Γ k+1, (1+K)γτ /γ an excellent indicator for the overall system fidelity and can
× Γ k+ +ρ, −
k+1 . be obtained from the first-order moment of γssc as
2 γ 2 β1 1 − ρ 2
(25) γssc = μγssc (1). (28)
6 EURASIP Journal on Wireless Communications and Networking
1.3
0.7
1.25
0.6
Normalized average output SNR
1.15
0.4
1.1
0.3
1.05
0.2
1
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8
Ricean K-Factor Ricean K-Factor
ρ = 0.1 ρ = 0.7 ρ = 0.1 ρ = 0.7
ρ = 0.3 ρ = 0.9 ρ = 0.3 ρ = 0.9
ρ = 0.5 ρ = 0.5
Figure 1: Normalized average output SNR (ASNR) versus the Figure 2: Amount of fading (AoF) versus the Ricean K-factor for
Ricean K-factor for several values of the correlation coefficient ρ. several values of the correlation coefficient ρ.
Δ
The AoF, defined as AoF = var(γssc )/γ2ssc , is a unified mea- show that as K increases, that is, the severity of the fading de-
sure of the severity of the fading channel [2] and gives an creases, and/or ρ increases, the normalized ASNR decreases,
insight to the performance of the entire system. It can be ex- resulting in a reduced diversity gain. We note that similar ob-
pressed in terms of first- and second-order moments of γssc servations have been made in [3, 30]. Furthermore, the re-
as sults presented in Figure 2 indicate that as K increases and/or
ρ decreases, AoF is degraded.
μγssc (2) Next the ABEP has been obtained using (23)–(27). In
AoF = − 1. (29) Figures 3 and 4 the ABEP is plotted as a function of the av-
μγssc (1)2
erage input SNR per bit, that is, γb = γ/ log2 M, for several
values of K. Figure 3 considers the performance of DBPSK,
6. PERFORMANCE EVALUATION RESULTS BPSK, and M-ary PSK signaling formats and ρ = 0.5. As
expected, when K increases, the ABEP improves and BPSK
Using the previous mathematical analysis, various perfor- exhibits the best performance. Figure 4 presents the ABEP
mance evaluation results have been obtained by means of of 16-QAM for different values of ρ and K. For comparison
numerical techniques and will be presented in this section. purposes, the performance of an equivalent single receiver,
Such results include performances for the ASNR, AoF, Pout , that is, without diversity, is also included. Similar to the pre-
ABEP1 , and C and will be presented for different Ricean vious cases, it is observed that the ABEP improves as K in-
channel conditions, that is, different values for K and ρ, as creases and/or ρ decreases, while significant overall perfor-
well as for various modulation schemes. mance improvement, as compared to the no-diversity case,
In Figures 1 and 2 the normalized ASNR (γssc /γ) and AoF is also noted.
are plotted as functions of the Ricean K-factor for several val- In Figure 5, Pout is plotted as a function of the normalized
ues of the correlation coefficient ρ. These performance eval- outage threshold per bit, γth /γb , for several values of K and
uation results have been obtained by numerically evaluating ρ. These performance results have been obtained by numer-
(15)–(17), (28), and (29). The results presented in Figure 1 ically evaluating (10), (11), and (21) and for ρ = 0 they are
identical to the ones obtained by using [2, Equation 9.273].
It is observed that Pout decreases, that is, the outage perfor-
1 For the consistency of the presentation from now on instead of the ASEP mance improves, as K increases and/or ρ decreases.
the ABEP performance will be used. As it is well known [2] for M-ary Finally, the normalized average CC can be obtained as
(M > 2) modulation schemes, assuming Gray encoding, the ABEP can
be simply obtained from the ASEP as P be ∼ = P se / log2 M, since Es =
C! = C/BW (in b/s/Hz) by employing (19) and (20). In
Eb log2 M, where Eb denotes the transmitted average bit energy. Figure 6, C! is plotted as a function of γb for several values
P. S. Bithas and P. T. Mathiopoulos 7
10−1
K =1
Average bit error probability (ABEP)
10−1 K =1
K =4
K =8
10−3
10−3
10−4 10−4
−5 0 5 10 15 20 −10 −7.5 −5 −2.5 0 2.5 5
Average input SNR per bit (dB) γth /γb
DBPSK 8-PSK ρ=0
BPSK 16-PSK ρ = 0.4
ρ = 0.8
Figure 3: Average bit error probability (ABEP) versus average in- Figure 5: Outage probability (Pout ) versus the normalized average
put SNR per bit for DBPSK, BPSK, and M-PSK (M = 8 and 16) input SNR per bit for several values of the Ricean K-factor and the
signaling formats, for different values of the Ricean K-factor. correlation coefficient ρ.
10−1 3
No diversity
Normalized average channel capacity (b/s/Hz)
Average bit error probability (ABEP)
2.5
10−2
10−3
ρ = 0.6 1.5
10−4 1
ρ = 0.2
0.5
10−5
−5 0 5 10 15 20 −4 −2 0 2 4 6 8 10
Average input SNR per bit (dB) Average input SNR per bit (dB)
K =1 ρ = 0.1 ρ = 0.9
K =4 ρ = 0.4 No diversity
K =8 ρ = 0.7
Figure 4: Average bit error probability (ABEP) versus average input Figure 6: Normalized average channel capacity (C/BW) versus the
SNR per bit for 16-QAM signaling format for different values of the average input SNR per bit for several values of the correlation coef-
Ricean K-factor and the correlation coefficient ρ. ficient ρ.
8 EURASIP Journal on Wireless Communications and Networking
of ρ and for K = 1. These results illustrate that as ρ increases, Transactions on Wireless Communications, vol. 4, no. 3, pp.
C! decreases, as expected [12], and the receiver without diver- 841–846, 2005.
sity has always the worst performance. [10] M. H. Ismail and M. M. Matalgah, “Performance of dual
maximal ratio combining diversity in nonidentical correlated
Weibull fading channels using Padé approximation,” IEEE
7. CONCLUSIONS Transactions on Communications, vol. 54, no. 3, pp. 397–402,
2006.
In this paper, the performance of dual branch SSC diversity [11] N. C. Sagias, “Capacity of dual-branch selection diversity re-
receivers operating over correlated Ricean fading channels ceivers in correlative Weibull fading,” European Transactions
has been studied. By deriving a convenient expression for on Telecommunications, vol. 17, no. 1, pp. 37–43, 2006.
the bivariate Ricean PDF, analytical formulae for the most [12] S. Khatalin and J. P. Fonseka, “Capacity of correlated
important statistical metrics of the received signals and the Nakagami-m fading channels with diversity combining tech-
capacity of such receivers have been obtained. Capitalizing niques,” IEEE Transactions on Vehicular Technology, vol. 55,
on these formulas, useful expressions for a number of per- no. 1, pp. 142–150, 2006.
[13] C.-D. Iskander and P. T. Mathiopoulos, “Performance of
formance criteria have been obtained, such as ABEP, Pout ,
dual-branch coherent equal-gain combining in correlated
ASNR, AoF, and average CC. Various performance evalua- Nakagami-m fading,” Electronics Letters, vol. 39, no. 15, pp.
tion results for different fading channel conditions have been 1152–1154, 2003.
also presented and compared with equivalent performance [14] A. A. Abu-Dayya and N. C. Beaulieu, “Switched diversity on
results of receivers without diversity. microcellular Ricean channels,” IEEE Transactions on Vehicular
Technology, vol. 43, no. 4, pp. 970–976, 1994.
ACKNOWLEDGMENTS [15] Y.-C. Ko, M.-S. Alouini, and M. K. Simon, “Analysis and opti-
mization of switched diversity systems,” IEEE Transactions on
This work has been performed within the framework of Vehicular Technology, vol. 49, no. 5, pp. 1813–1831, 2000.
[16] C. Tellambura, A. Annamalai, and V. K. Bhargava, “Unified
the Satellite Network of Excellence (SatNEx-II) project (IST-
analysis of switched diversity systems in independent and cor-
027393), a Network of Excellence (NoE) funded by European related fading channels,” IEEE Transactions on Communica-
Commission (EC) under the FP6 program. tions, vol. 49, no. 11, pp. 1955–1965, 2001.
[17] M.-S. Alouini and M. K. Simon, “Dual diversity over corre-
REFERENCES lated log-normal fading channels,” IEEE Transactions on Com-
munications, vol. 50, no. 12, pp. 1946–1959, 2002.
[1] T. S. Rappaport, Wireless Communications, Prentice-Hall PTR, [18] D. A. Zogas and G. K. Karagiannidis, “Infinite-series repre-
Upper Saddle River, NJ, USA, 2002. sentations associated with the bivariate Rician distribution
[2] M. K. Simon and M.-S. Alouini, Digital Communication over and their applications,” IEEE Transactions on Communications,
Fading Channels, John Wiley & Sons, New York, NY, USA, vol. 53, no. 11, pp. 1790–1794, 2005.
2nd edition, 2005. [19] M. K. Simon, Probability Distributions Involving Gaussian
[3] N. C. Sagias and G. K. Karagiannidis, “Gaussian class mul- Random Variables: A Handbook for Engineers and Scientists,
tivariate Weibull distributions: theory and applications in Kluwer Academic Publishers, Norwell, Mass, USA, 2002.
fading channels,” IEEE Transactions on Information Theory, [20] P. S. Bithas, N. C. Sagias, and P. T. Mathiopoulos, “Dual diver-
vol. 51, no. 10, pp. 3608–3619, 2005. sity over correlated Ricean fading channels,” Journal of Com-
[4] G. E. Corazza and F. Vatalaro, “A statistical model for land munications and Networks, vol. 9, no. 1, pp. 67–74, 2007.
mobile satellite channels and its application to nongeostation- [21] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series,
ary orbit systems,” IEEE Transactions on Vehicular Technology, and Products, Academic Press, New York, NY, USA, 6th edi-
vol. 43, no. 3, part 2, pp. 738–742, 1994. tion, 2000.
[5] H. Wakana, “A propagation model for land mobile satellite [22] M. Abramowitz and I. A. Stegun, Handbook of Mathemati-
communications,” in Proceedings of IEEE Antennas and Propa- cal Functions with Formulas, Graphs and Mathematical Tables,
gation Society International Symposium, vol. 3, pp. 1526–1529, Dover, New York, NY, USA, 9th edition, 1972.
London, Ont, Canada, June 1991. [23] A. A. Abu-Dayya and N. C. Beaulieu, “Analysis of switched di-
[6] E. Lutz, D. Cygan, M. Dippold, F. Dolainsky, and W. Papke, versity systems on generalized-fading channels,” IEEE Transac-
“The land mobile satellite communication channel-recording, tions on Communications, vol. 42, no. 11, pp. 2959–2966, 1994.
statistics, and channel model,” IEEE Transactions on Vehicular [24] A. Papoulis, Probability, Random Variables, and Stochastic Pro-
Technology, vol. 40, no. 2, pp. 375–386, 1991. cesses, McGraw-Hill, New York, NY, USA, 2nd edition, 1984.
[7] M. K. Simon and M.-S. Alouini, “A unified performance anal- [25] A. Nuttal, “Some integrals involving the Q-function,” Tech.
ysis of digital communication with dual selective combin- Rep. 4297, Naval Underwater Systems Center, New London,
ing diversity over correlated Rayleigh and Nakagami-m fad- Conn, USA, April 1972.
ing channels,” IEEE Transactions on Communications, vol. 47, [26] W. C. Y. Lee, “Estimate of channel capacity in Rayleigh fad-
no. 1, pp. 33–43, 1999. ing environment,” IEEE Transactions on Vehicular Technology,
[8] Y. Chen and C. Tellambura, “Distribution functions of selec- vol. 39, no. 3, pp. 187–189, 1990.
tion combiner output in equally correlated Rayleigh, Rician, [27] “The Wolfram functions site,” http://functions.wolfram.com/.
and Nakagami-m fading channels,” IEEE Transactions on Com- [28] V. S. Adamchik and O. I. Marichev, “The algorithm for cal-
munications, vol. 52, no. 11, pp. 1948–1956, 2004. culating integrals of hypergeometric type functions and its
[9] G. K. Karagiannidis, D. A. Zogas, N. C. Sagias, S. A. Kot- realization in REDUCE system,” in Proceedings of Interna-
sopoulos, and G. S. Tombras, “Equal-gain and maximal-ratio tional Symposium on Symbolic and Algebraic Computation (IS-
combining over nonidentical Weibull fading channels,” IEEE SAC ’90), pp. 212–224, Tokyo, Japan, August 1990.
P. S. Bithas and P. T. Mathiopoulos 9
Research Article
Advanced Fade Countermeasures for DVB-S2 Systems in
Railway Scenarios
Stefano Cioni,1 Cristina Párraga Niebla,2 Gonzalo Seco Granados,3 Sandro Scalise,2
Alessandro Vanelli-Coralli,1 and Marı́a Angeles Vázquez Castro3
1 ARCES, University of Bologna, Via Toffano 2, 40125 Bologna, Italy
2 German Aerospace Center (DLR), Institute of Communications and Navigation, Postfach 1116, 82230 Wessling, Germany
3 Department of Telecommunications and Systems Engineering, Universitat Autònoma de Barcelona, Campus Universitari, s/n,
This paper deals with the analysis of advanced fade countermeasures for supporting DVB-S2 reception by mobile terminals
mounted on high-speed trains. Recent market studies indicate this as a potential profitable market for satellite communications,
provided that integration with wireless terrestrial networks can be implemented to bridge the satellite connectivity inside railway
tunnels and large train stations. In turn, the satellite can typically offer the coverage of around 80% of the railway path with existing
space infrastructure. This piece of work, representing the first step of a wider study, is focusing on the modifications which may
be required in the DVB-S2 standard (to be employed in the forward link) in order to achieve reliable reception in a challenging
environment such as the railway one. Modifications have been devised trying to minimize the impact on the existing air interface,
standardized for fixed terminals.
Copyright © 2007 Stefano Cioni et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
However, the aforementioned standards have not been As summarized in Table 1, none of these alternatives
designed for mobile use. Collective terminals installed in a alone represents a satisfactory solution. As a matter of fact,
mobile platform, such as train, ship, or aircraft, are exposed deployed or upcoming commercial services are based on
to a challenging environment that will impact the system per- combinations of different access technologies. In this light,
formance considering the current standard in absence of any a satellite access based on an open standard can have very
specific provision. significant benefits in terms of interoperability (achieved for
Mobile terminals will have to cope in general with strin- DVB-S2/RCS through SatLabs Qualification Program) and
gent frequency regulations (especially in Ku band), Doppler competition, thus benefiting from availability of fully com-
effects, frequent handovers, and impairments in the synchro- patible terminals from multiple vendors and reducing the
nization acquisition and maintenance. Furthermore, the rail- cost of terminals.
way scenario is affected by shadowing and fast fading due However, the aforementioned DVB standards have been
to mobility, such as, for example, the deep and frequent designed for fixed terminals. To cope with these new market
fades due to the presence of metallic obstacles along electri- opportunities, DVB TM-RCS has investigated how the cur-
fied lines providing power to the locomotive1 [6] and long rent DVB-RCS standard could be applied to mobile applica-
blockages due to the presence of tunnels and large train sta- tions. A white paper on the applicability of DVB-RCS to mo-
tions. This suggests that hybrid networks, that is, interwork- bile services was prepared and a technical annex was added
ing satellite and terrestrial components, are essential in order to the implementation guidelines document [4]. This annex
to keep service availability. states the boundary conditions and limitations under which
In this context, this paper is focused on proposing and the existing standard could be used in mobile environment,
evaluating fade countermeasures to compensate the impact considering the impact of mobility in terminal synchroniza-
of fade sources in the railway scenario, that is, shadowing, tion and demodulator performance in forward and return
fast fading, and power arches, excluding tunnels which will links. Furthermore, a survey on applicable regulations and a
be address at a later stage. In particular, antenna diversity and brief analysis on DVB-RCS features that can be used for mo-
packet level forward error correction (FEC) are investigated. bility management are provided, the latter referring to inter-
The rest of the paper is organized as follows: Section 2 beam handover only.
discusses the potential of opening the current DVB-S2/RCS Thus, the DVB-RCS guideline cannot support the full
standards to provide mobile services efficiently. Section 3 adaptability to mobile environments and hence the applica-
presents the peculiarities of the trains’ scenario and discusses ble services and scenarios happen to be very limited. Fur-
the different aspects that can impact the system performance. thermore, additional issues related to mobility are not fully
Section 4 describes the fade countermeasures proposed in solved, such as handling of nonline-of-sight (nLOS) channel
this paper. Section 5 introduces the simulation platforms in conditions, which will require the interworking with terres-
which the proposed fade countermeasures are evaluated and trial gap fillers in the railway scenario due to the presence of
Section 6 presents and discusses the obtained results. Finally, tunnels. In addition, even if DVB-RCS features to be applied
Section 7 draws the conclusions of this work. for mobility management are analyzed, a determined mech-
anism or protocol should be specified in order to ensure in-
teroperability. Finally, the impact of control signals loss (due
2. THE VISION: A NEW DVB-S2/RCS STANDARD FOR to deep fades or handover) is not negligible. For instance, the
MOBILE COLLECTIVE TERMINALS loss of terminal burst time plan (TBTP) tables damages the
operation of the resource management, essential in the re-
turn link for a coordinated access to the radio resources.
The large capacity of DVB-S2/RCS systems can efficiently ac- As a matter of fact, mobile services could be more effi-
commodate broadcast services (e.g., digital TV) and unicast ciently supported if the present standards could be improved
IP multimedia interactive services to fixed terminals. How- for mobile scenarios. The reopening of the standard2 would
ever, the increasing interest on broadband mobile services allow for the specification of methods for improving the link
suggests that the natural evolution of DVB-S2/RCS standard reliability in mobile environments (e.g., packet level FEC),
to cover new market needs goes towards the support of mo- handover protocols, interfaces to terrestrial gap fillers (even
bile terminals. using terrestrial mobile technologies), improved mobility-
In particular, the required antenna performance in Ku aware signalling and resource management, and so forth.
(10–12 GHz) and Ka (20–30 GHz) bands focuses the mar- In this context, a number of R and D initiatives are on-
ket opportunities of DVB-S2/RCS onto mobile terminals in going with the aim at investigating enhancements of the
collective transportation means. Actually, transport opera- DVB-S2/RCS standards for the efficient support of mobil-
tors are starting to announce the provision of TV services in ity. Among those, the SatNEx network of excellence has set
ships, trains, buses, and aircrafts, and broadband IP connec- up a dedicated working group investigating different aspects
tivity, for passengers. For the specific case of trains, broad- related to mobility in DVB-S2/RCS. The first results of this
band services can provided using satellite systems, cellular activity in the field of forward link reliability for the rail-
connectivity or dedicated trackside installations. way scenario are presented in this paper. For the return link,
1 Hereafter referred to as “power arches,” for the sake of brevity. 2 Envisaged at the time of writing.
Stefano Cioni et al. 3
Table 1: Pros and cons of different solutions for providing broadband services on trains.
Type of
Examples Pros Cons
technology
(i) No new trackside (i) Available tracking antennas and
infrastructure—quick to efficient satcom modems expensive
deploy, project costs may be
lower on long distance routes
DVB-S2/RCS
Satellite Proprietary systems, (ii) Dedicated bandwidth available (ii) High variable cost per MB
for example, ViaSat
(iii) Performance easy to predict (iii) Return bandwidth constrained
depending on satellite visibility by antenna size
(iv) Not affected by borders—good (iv) Satellite visibility seriously
for international trains restricted on some rail routes
(i) Equipment is small and cheap (i) Geographic coverage of UMTS
limited for years to come
(ii) Usage is cheap (50–75 C per month (ii) Coverage of railway lines often
flat rate) worse than roads
GPRS
EDGE (iii) Data rates improving year on year (iii) GPRS/EDGE not really fast enough
Cellular UMTS (iv) Competitive supply—3 or 4 network (iv) Inverse relationship between
HSUPA/HSDPA operators in most countries throughput and train speed
(EV-DO)
(v) No QoS guarantees—affected by
network congestion at peak times
(vi) Organized country by country—data
roaming charges are punitive
(i) High data rates possible (i) Existing standards not designed to
support fast-moving terminals
(ii) Can control bandwidth and QoS (ii) Proprietary equipment is more
expensive
Flash OFDM (iii) On-train equipment relatively (iii) No suitable public services yet in
Trackside IEEE 802.11 inexpensive licensed bands—will licence-holders be
IEEE 802.16 (WiMAX) allowed to provide mobile services?
(iv) No volume-related usage costs (iv) Licence-exempt bands are low power,
thus limited range
analogue solutions have to be devised, which are however not more than 10 years ago in the north of Spain. These results
in the scope of the present work. represent a very interesting reference, although no specific
channel model has been extracted from the collected data.
3. THE RAILWAY SCENARIO, A CHALLENGING After an initial qualitative analysis, the railway environment
ENVIRONMENT appears to differ substantially with respect to the scenarios
normally considered when modelling the LMSC. Excluding
3.1. Overview railway tunnels and areas in the proximity of large railway
stations, one has to consider the presence of several metallic
The land mobile satellite channel (LMSC) has been widely obstacles like power arches (Figure 1, left uppermost), posts
studied in the literature [7]. Several measurement campaigns with horizontal brackets (Figure 1, left lowermost), which
have been carried out and several narrow and wideband may be often grouped together (Figure 1, rightmost), and
models have been proposed for a wide range of frequencies, catenaries, that is, electrical cables, visible in all the afore-
including Ku [8] and Ka [9] bands. Nevertheless, for the spe- mentioned figures.
cific case of the railway environment, only few results are The results of direct measurements performed along the
presented in [10] as a consequence of a limited trial cam- Italian railway and aiming to characterize these peculiar ob-
paign using a narrowband test signal at 1.5 GHz, performed stacles are reported in [6] and references herein. In summary,
4 EURASIP Journal on Wireless Communications and Networking
1E + 0
1E − 01
h-d h
1E − 03
E2 /E0 a E1 /E0 v
α1
α2 1E − 04
1 3 5 7 9 11 13 15 17 19 21
Eb /N0 (dB)
1/2 - QPSK (LOS, FAST, noPA)
2/3 - 8PSK (LOS, FAST, noPA)
(a)
3/4 - 16APSK (LOS, FAST, noPA)
5/6 - 16APSK (LOS, FAST, noPA)
d = 0.4 m, a = 2.5 m 1/2 - QPSK (LOS, FAST, PA)
5
2/3 - 8PSK (LOS, FAST, PA)
0 3/4 - 16APSK (LOS, FAST, PA)
5/6 - 16APSK (LOS, FAST, PA)
−5
−10
Figure 3: BBFRAME error rate for DVB-S2 in the presence of
power arch blockage events. LOS propagation conditions and train
Attenuation (dB)
−25
antennas, and assuming perfect compensation of time delays
−30 of the two replicas, the combined signal can be written as
−35
rc (t) = w1 r1 (t) + w2 r2 (t), (2)
−40
−45
where wi and ri (t), i = 1, 2, are the combing weights and
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 the received signals, respectively. The received signals at each
h (m) antenna is
0.6 m ri (t) = αi s0 (t) + ni (t), (3)
0.4 m
0.2 m
where s0 (t) is the transmitted signal, αi is the time variant
(b) fading envelope over the ith antenna, and ni (t) is the thermal
noise.
Figure 2: Knife-edge diffraction model applied to the railway sce- The simplest combining scheme is the signal selection
nario and possible attenuation caused by power arches at Ku band Combining (SC), in which the branch-signal with the largest
for different antenna diameters. amplitude or signal-to-noise ratio (SNR) is the one selected
for demodulation. In this case, wi will be 1 or 0 if the
ith power branch is the largest or the smallest, respectively.
Clearly, SC is bounded by the performance of the single re-
ceiving antenna in absence of fading, that is, there is no di-
4.1. Antenna diversity versity gain when the two antennas experience good chan-
nel conditions at the same time. Maximum-ratio combin-
The adoption of multiple receiving antennas to counteract ing (MRC), although requiring a larger complexity at the
power arch obstructions in railway environment has been re- receiver, allows for the exploitation of the diversity gain. In
cently proposed and investigated in [13, 14]. Antenna diver- fact, MRC scheme provides for the maximum output SNR.
sity is used to provide different replica of the received signal According to the optimum combination criterion, the signal
to the detector for combination or selection. If the receiving weights are directly proportional to the fading amplitude and
antennas are sufficiently spaced, the received signals fade in- inversely proportional to the noise power, Ni , as follows:
dependently on each antenna thus providing multiple diver- αi
sity branches that can be linearly or nonlinearly combined to wi = . (4)
Ni
improve detection reliability. There are mainly three types of
linear diversity combining approaches: selection, maximal- Another technique, often used because it does not require
ratio, and equal-gain combining. Considering two receiving channel fading strength estimation, is equal gain combining
6 EURASIP Journal on Wireless Communications and Networking
(EGC) in which the combination weights are all set to one, 4.2. Packet level FEC
thus leading to a simpler but suboptimal approach. Clearly,
SC and MRC (or EGC) represent the two extremes in diver- 4.2.1. The concept of packet level FEC
sity combining strategy with respect to the complexity point
of view and the number of signals used for demodulation Reliable transmission occurs when all recipients correctly re-
process. Furthermore, the classical combining formula can ceive the transmitted data. This target can be achieved by op-
be generalized for nonconstant envelope modulations such erating at different layers of the protocol stack and in dif-
as 16-APSK or 32-APSK (amplitude and phase shift keying) ferent ways. Retransmission techniques allow that lost pack-
and integrated with the soft demodulator that computes the ets are retransmitted to the receivers, while packet level FEC
channel a posteriori information to feed the low density par- schemes create redundant packets that permit to reconstruct
ity check (LDPC) FEC decoder. The maximum likelihood a the lost ones at the receiver side, with a very beneficial in-
priori information for a single receiver antenna given by put on the final end-to-end delay. In fact, as detailed in [15],
the additional delay introduced by packet level encoding and
decoding is always lower than the delay deriving from any
Pr bi = 0 | rk retransmission scheme.
log
Pr bi = 1 | rk Regarding the retransmission schemes, efficient proto-
2 (5) cols should limit the use of acknowledgement- (ACK-) based
si ∈S0 exp −rk − αk si /N0 mechanisms because they introduce heavy feedback traffic
= log
2
si ∈S1 exp −rk − αk si /N0 towards the sender, thus increasing the congestion of reverse
link that, typically, has a reduced capacity with respect to
forward link. Negative acknowledgement- (NACK-) based
can be extended for L receiving antennas, according to the approaches are hence particularly interesting. In combina-
MRC principle, as follows: tion with (or in alternative to) the traditional retransmission
schemes, packet level FEC can be added on top of physical
layer FEC, in order to achieve the same level of reliability with
Pr bi = 0 | rk a reduced number of retransmissions. This might be partic-
log
Pr bi = 1 | rk ularly useful if resources on the return link need to be saved
L
p
r p 2 p (smaller number of NACKs or no NACKs are needed at all),
si ∈S0 exp −
p=0 k k si /N0
−α
= log or when multiple lost packets are recovered with the retrans-
L p
r p 2 p ,
si ∈S1 exp − p=0 k k si /N0
−α mission of a lower number of redundant packets. Basically,
(6) h redundancy packets are added to each group of k informa-
tion packets, thus resulting in the transmission of n = k + h
packets. These packets are finally transferred to the physi-
where rk is the received sample at time k, αk is the true or cal layer, which adds independent channel coding to each of
the estimated channel coefficient, and S0 and S1 are the sets them. This principle is described in Figure 5.
of symbols which have “0” or “1” in the ith position, respec- At the physical layer, the bits affected by low noise lev-
tively. els can be corrected by the physical layer FEC, so that the
In the configuration proposed in this work, we adopt related packets are passed to the higher layer as “correct.” If
MRC combining with two antennas. The antennas are placed the noise level exceeds the correcting capability of the phys-
on the same coach so as to reduce the costs of installa- ical layer, the received bit cannot be properly decoded, but
tion and the connection length. The antenna spacing is cho- the failure to decode can be usually detected with a very high
sen as a function of the distance between two consecutive reliability. Since erroneous packets are not propagated to the
power arches so as to guarantee that only one antenna at higher layers, we have an erasure channel. The system can use
a time can be obscured. Accordingly, the distance between the redundancy packets to recover these erasures. By using
the two antennas is about 15 m. Considering the maximum maximum distance separable (MDS) codes, like the Reed-
train speed (about 300 km/h), this translates into the fact Solomon, it is possible to reconstruct the original informa-
that power-arch blockage on a single antenna lasts for about tion if at least k out of n packets are correctly received. There-
7 msilliseconds, and it hits the second antenna after about fore, the receiver can cope with erasures, as long as they result
180 msilliseconds. Therefore, it is reasonable to assume that in a total loss not exceeding h packets, independently from
there is enough time for the combining circuit to react and where the erasures occurred. LDPC codes and their deriva-
maintain constant signal connection. A drawback of this ap- tions might be also used because of their low complexity and
proach is that the receiving chain will be duplicated in or- greater flexibility, thus permitting to encode larger files, al-
der to maintain connection and avoid frequent reacquisitions though a small inefficiency, depending on the code design
process with the consequent loss of packet. As proposed in and typically around 5%–10%, will be taken into account.
[14], the solution which considers the presence of a second If packet level FEC is implemented at IP or data link layer,
receiving antenna is depicted in Figure 4. The gray blocks very near to the physical channel, no change in the trans-
represent the subsystems that need to be duplicated in the port and network layers protocols and in the physical layer
two antenna case. Further details on the digital receiver are are necessary. This solution presents the additional advantage
described in Section 5.1. that it can be adapted to the propagation channel conditions
Stefano Cioni et al. 7
Frame
synch
Received
signal Matched Symbol Data
from sampling DeMUX Buffer
filter
antenna no. 1
Preamble / Noise level N01
Frequency From second
Timing pilots estimation antenna
acquisition recovery θ01 Signal
Digital α1k combiner
AGAC
θ01 Lock
Buffer detector Hard/soft
demodulator
θk1
Freq./phase
tracking
De-
interleaver
LDPC/BCH
decoder
n packets
Channel coding
Transmission
by choosing n, so that the interleaver size is long enough to (iii) Different QoS classes with different redundancy pro-
compensate the channel outages. However, different protec- files can be supported. Furthermore, redundancy
tion for individual transfers (e.g., specific files) is not possi- packets for low-priority traffic can be put in a special
ble (although different QoS classes may be supported), extra queue, which is served only if free capacity is available
memory is required, and additional delays must be properly and, in turn, increased redundancy can be sent during
handled. handovers, minimizing the overall probability of lost
For the forward link, the usage of packet level FEC is packets.
especially powerful in allowing online variable coding ap- (iv) Different IP-based access methods can be used in par-
proaches, which can be fine tuned in a closed-loop approach. allel, improving the link reliability if different redun-
Based upon the “history” of the link, appropriate redun- dancy is sent via different access methods.
dancy can be easily added. Packet level FEC has then impact
on different layers.
4.2.2. The GSE-FEC method
(i) The requirements on control loops can be lessened, for
example, power control and or adaptive coding and When moving to the concrete applicability of this scheme to
modulation control, if a loss of up to h packets can tol- the scenario under consideration, even though the fact that
erated. IP packets have three sizes that are the most common ones,
(ii) The typical fade structure of a link can be measured the fact that IP packet size can actually take any value up
and accordingly coding with the correct profile added. to a maximum value (typically 64 Kbytes) represents a clear
8 EURASIP Journal on Wireless Communications and Networking
difficulty in applying packet level FEC (PL-FEC). The funda- while the second process employs MPE-MPEG. The whole
mental difficulty comes from the fact that most codes take as implementation is called MPE-FEC in DVB-H. Our proposal
input a fixed amount of data, from which they compute the for DVB-S2 is based on keeping the same first process as in
redundancy bytes. As a given number of IP packets corre- DVB-H, whereas it employs GSE in the second process. This
spond to a variable amount of data depending of their sizes, proposal for applying PL-FEC in DVB-S2 is named GSE-
codes needing a fixed amount of data cannot be directly ap- FEC.
plied. One possible solution is to use codes that can be eas- A block diagram of GSE-FEC is depicted in Figure 6. The
ily adapted to different input sizes; however, this comes at incoming IP packets are arranged in the so-called FEC ma-
the price of a much more complex encoding and decoding trix, where also the packet-level redundancy is added. The
process. Another solution has been proposed in the DVB-H filling of the FEC matrix and the encoding are done in the
standard [16]. In this case, units of constant length are built same way as in DVB-H. For the sake of completeness, this
by interleaving IP packets and, therefore, codes with fixed in- will be briefly described below. Next, each IP packet is en-
put size can be easily applied. It is worth noting that those capsulated using GSE, and this represents one of the novel
units are not built by concatenating IP packets but by inter- aspects of our proposal. Each IP packet may be fragmented
leaving them. However, interleaving is this case must not be into several GSE units or it may also be sent unfragmented.
understood as it is typical in physical layer coding, where it Subsequently, the maximum number of GSE units that can
means that data is written in one direction in a matrix and be fitted inside a BBFRAME is concatenated and introduced
it is read in the orthogonal direction for transmitting. In PL- in the BBFRAME. The size of the BBFRAME depends on the
FEC, we understand interleaving as computing the redun- combination of coding rate and modulation scheme (MOD-
dancy in an orthogonal direction to the writing direction of COD) adopted by the DVB-S2 modem, so the number of
the data; however, in this case the writing and reading direc- GSE units that can be concatenated also depends on the
tions coincide. This kind of interleaving is advantageous be- MODCOD. By making the GSE units small enough to have
cause the redundancy is computed across a large number of the required flexibility, but large enough in order not to pe-
packets. Thus, a fade event may destroy one or several pack- nalize encapsulation efficiency, this method provides an easy
ets but not the majority of them, assuming that the system mechanism to adapt the output of the packet-level FEC to the
is well dimensioned, so the added redundancy can effectively variations of the physical layer. Moreover, note that padding
help in recovering the destroyed packets. is not applied inside the GSE unit but only at BBFRAME level
DVB-H also provides a solution for encapsulating the if the size of the BBFRAME does not coincide with that of the
coded IP packets for transmission over DVB-T. The solution concatenation of the GSE units.
is based on the use of multiprotocol encapsulation (MPE) The IP packets are placed one after another along the
combined with MPEG. Although it would be possible to columns of the FEC matrix, see Figure 7. Each IP packet may
adapt the same approach for DVB-S2, it presents a number be split among two or more columns. Only the first block of
of drawbacks, such as lack of flexibility, low encapsulation the matrix, from column 1 to 191, can be filled in with IP
efficiency, delay constraints. A new encapsulation protocol packets. The second block of the matrix, from column 192 to
call generic stream encapsulation (GSE) has been recently de- 255, carries the redundancy information, which is computed
fined [17]. It is a very flexible protocol applicable to several by a Reed-Solomon (255,191) code applied to the first block
physical layer standards. It overcomes most of the limitations on a row basis. Each column in the second block is encap-
of MPE-MPEG. GSE is especially suitable for transmitting IP sulated individually using GSE, whereas in the first block the
packets through the generic stream interface mode of DVB- GSE encapsulation is performed on an IP packet basis. In the
S2, and it has been proposed for the second generation of baseline operation, padding is only applied in the first block
Terrestrial digital video broadcasting (DVB-T2) as well. GSE to account for the fact that an additional IP packet may not
also efficiently supports the ACM functionalities of DVB-S2 be fitted without overrunning the 191 columns and all 64 re-
and facilitates the provision of QoS guarantees because it re- dundancy columns are transmitted. The code can be made
duces the constraints on the scheduling operation. weaker (i.e., with higher rate) by puncturing some of the re-
It can be deducted from the previous discussion that the dundancy columns, which are then not transmitted and are
implementation of PL-FEC consists of two main processes: considered as unreliable bytes in the decoding process. The
the encoding the IP packets and, second, the encapsulation code can also be made more robust (i.e., with lower rate)
of the result of the encoding process in order to adapt it to by padding with zeros columns in the first block and, hence,
the underlying transmission system. In DVB-H, the first pro- leaving less space for IP packets. The padded columns are not
cess consists in arranging the IP packets in a matrix (here- transmitted but they are used in the encoding process. In the
after called FEC matrix) and applying a Reed-Solomon code, decoding process, they are considered as reliable.
Stefano Cioni et al. 9
Coding direction
FEC matrix
Writing direction
1 2 3 188 189 190 191 192 193 254 255
IP packet 1 (cont.)
Punctured column
Punctured column
Column size
IP packet 1
Padding
Padding
Padding
··· ···
IP packet 2
Padding
IP packet 3
After GSE encapsulation, the GSE packets are introduced merger/slicer that, depending on the applications, allocates
in BBFRAMEs and transmitted. On the receive side, erro- a number of input bits equal to the maximum data field ca-
neous BBFRAMEs are detected by checking the CRC. The pacity. In this way, user packets are broken in subsequent
receiver reconstructs the FEC matrix and marks any column data fields, or an integer number of packets are allocated in
that is totally or partially received by means on an erroneous it. Then, a fixed length base-band header (BBHEADER) of
BBFRAME as unreliable. Finally, if the reconstructed FEC 80 bits is inserted in front of the data field, describing its for-
matrix has no more than 64 unreliable columns, the code mat. For example, it reports to the decoder the input streams
can correctly compute all bytes in the matrix. If there are format, the mode adaptation type and the roll-off factor.
more than 64 unreliable columns, the code cannot correct The efficiency loss introduced by this header varies from
anything, and only those columns received by means of cor- 0.25% to 1% for long and short codeword lengths, respec-
rect BBFRAMEs will be correct. tively. The role of stream adaptation is to provide padding
when needed, in order to complete a constant length frame,
5. SIMULATION SCENARIOS and scrambling. Padding is applied when the user data avail-
able for transmission are not sufficient to completely fill a
In the following, the simulation platforms used to evaluate BBFRAME, or when more than one packet have to be allo-
the performance of DVB-S2 with advanced fade countermea- cated in a BBFRAME. The built frame is randomized using a
sures in the railway environment as described in Section 3 are scrambling sequence generated by the pseudorandom binary
duly detailed. sequence described by the polynomial (1 + X 14 + X 15 ). After
this scrambling, each BBFRAME is processed by the forward
5.1. Advanced physical layer simulation platform error correction (FEC) encoder which is carried out by the
concatenation of a Bose-Chaudhuri-Hocquenghem (BCH)
To cover a rather large set of spectral efficiency, four MOD- outer code and an LDPC inner code. Available code-rates
CODs have been considered: 1/2-QPSK, 2/3-8PSK, 3/4- for the inner code are 1/4, 1/3, 2/5, 1/2, 3/5, 2/3, 3/4, 4/5,
16APSK, and 5/6-16APSK. The LOS channel condition 5/6, 8/9, and 9/10. Depending on the application area, code-
(Rice factor equal to 17.4 dB) and the train speed equal to words can have length NLDPC = 64800 bits or 16200 bits. In
300 km/h have been simulated. Equally spaced power arches the following, the case of 64800 bits is considered. Regard-
with a separation of 50 m have been included in some sce- ing the modulation format, each coded BBFRAME can be
narios, with a duty cycle of 1%, corresponding to a width of mapped onto QPSK, 8PSK, 16APSK, or 32APSK constella-
0.5 m in accordance with Figure 2. The symbol rate was fixed tions. Modulated streams enter in the physical layer framing
to 27.5 Mbaud. where physical layer signalling and pilot symbols are inserted.
The considered DVB-S2 physical layer transmitter [2] is For energy dispersal, another scrambling sequence is applied
depicted in Figure 8. A continuous stream of MPEG pack- to the entire physical layer frame (PLFRAME). The system
ets passes through the mode adaptation which provides has been designed to provide a regular PLFRAME structure,
input stream interfacing. This data flow is passed to the based on slots of M = 90 modulated symbols, which allow
10 EURASIP Journal on Wireless Communications and Networking
Single/multiple
input data streams
Figure 8: DVB-S2 physical layer transmitter block diagram (taken from [2]).
reliable receiver synchronization on the FEC block struc- which allows match filtering with minimal intersymbol in-
ture. The first slot, PLHEADER, is devoted to physical layer terference regrowth; then the subsequent block deals with
signalling, including start-of-frame (SOF) delimitation and clock recovery for timing adjustment, performed by a digi-
MODCOD definition. Receiver channel estimation is facil- tal interpolator. The demultiplexer is used to separate pilots
itated by the introduction of a set of P = 36 pilot sym- from data symbols in a PLFRAME. The pilot symbol stream
bols, that are inserted every 16 slots. In addition, a pilot- is used by the following four subsystems: the noise level esti-
less transmission mode is also available, ensuring greater sys- mator, the digital automatic gain and angle control (AGAC),
tem capacity. Finally, for shaping purposes, a squared-root the block in charge of tracking the residual frequency offset
raised cosine (SRRC) filter with variable roll-off factors (0.2, and carrier phase, and finally the coarse frequency acquisi-
or 0.25, or 0.35) is considered. To cope with the intrinsic tion loop (not performed). On the other path, the data sym-
nonlinearity of the on-board high power amplifier (HPA), bols, softly combined with the last equation of Section 4.1,
a purposely designed predistortion technique is considered. feed the hard/soft demodulator. The demodulator provides
In particular, a fractional predistortion technique based on the hard decisions on data symbols as a feed-back for car-
a lookup table (LUT) approach is considered which operates rier frequency and phase tracking, and computes the soft ini-
right after the shaping filter [18]. The fractional predistorter, tial a posteriori probability (APP) on the received informa-
which is a digital waveform predistorter, acts on the signal tion bits. Finally, the APPs are deinterleaved and given to the
samples for precompensating the HPA AM/AM and AM/PM LDPC-BCH decoder. As far as frame synchronization and
characteristics and mitigating the impact of non linear dis- frequency acquisition are considered, that is, dashed white
tortion. In particular, the signal is processed by means of blocks in Figure 4, they are not considered in the simula-
the LUT, which stores the inverted HPA coefficients com- tion chain because the receiver behaviour is assessed during
puted offline through analytic inversion of a proper HPA steady state.
model. The steps needed to obtain LUT coefficients are the
following: HPA model selection, parameter extrapolation, an- 5.2. Packet level coding simulation platform
alytical model inversion, and LUT construction. Regarding the
first step, a simple yet robust empirical model is the clas- A simulation platform to analyze the performance of GSE-
sic Saleh model [18]. Given the measured HPA character- FEC has been developed. Given that this performance as-
istics, the second step can be performed by minimizing the sessment entails many layers, in particular, from the physical
energy of the difference between the modelled and the ex- to the network layers, of the protocol stack, a modular ap-
perimental HPA curves (MMSE criterion). These parameters proach has been considered as the only feasible way to de-
are then applied to the analytically inverted characteristics, velop the platform. The physical-layer simulator described
so as to obtain the analytical predistortion transfer function. in the previous section interfaces with the packet-level sim-
The last step is the quantization of the analytical curve in ulator shown in Figure 9. This takes as input a stream of
order to store it into the LUT. The adopted strategy is lin- IP packets and applies the GSE-FEC encoding technique as
ear in power indexing, that is, table entries are uniformly described above, generating a sequence of BBFRAMEs. At
spaced along the input signal power range, yielding denser this point, the output of the physical-layer simulator is used
table entries for larger amplitudes, where nonlinear effects to mark the BBFRAMEs as correctly or wrongly received.
reside. Next, the GSE-FEC decoding process is applied. The effect
The proposed digital receiver architecture is depicted in of the BBFRAMEs on the GSE units and subsequently on the
Figure 4. In particular, several subsystems are present in or- columns of the reconstructed FEC matrix is calculated. Then,
der to coherently demodulate and combine the received sig- the correction capability of the Reed-Solomon code is taken
nals. The first coarse correction regards the carrier frequency, into account to eliminate, if possible, the unreliable columns
Stefano Cioni et al. 11
1E + 0
IP PER
calculation
1E − 01
Mapping to
Power arches floor
PER
correct/wrong 1E − 02
IP packets
Traffic generation
1E − 03
Corrected FEC
IP packets matrix
1E − 04
−2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
GSE-FEC FEC decoding
Eb /N0 (dB)
1/2 - QPSK (LOS, FAST, noPA)
Mapping to 2/3 - 8PSK (LOS, FAST, noPA)
BBFRAMEs correct/wrong 3/4 - 16APSK (LOS, FAST, noPA)
FEC matrix 5/6 - 16APSK (LOS, FAST, noPA)
columns
1/2 - QPSK (LOS, FAST, MRC)
2/3 - 8PSK (LOS, FAST, MRC)
Mapping to 3/4 - 16APSK (LOS, FAST, MRC)
correct/wrong 5/6 - 16APSK (LOS, FAST, MRC)
Selective BBFRAME
corruption GSE units
Figure 10: MRC performance in LOS channel condition and train
speed equal to 300 km/h.
Probability
Probability
0.5 0.5 0.5
0 0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Error burst length Error burst length Error burst length
(a) (b) (c)
Figure 11: Histogram of the BBFRAME error burst length for two different MODCOD modes and target BBFRAME error rate equal to
0.02.
6.2.1. Dimensioning the FEC matrix FRAMEs and hence BBFRAMEs are affected by each power
arch.
In order to present the procedure to compute the col-
First of all, it is worth remarking that the appropriate size
umn size of the FEC matrix, we consider a numerical exam-
of the FEC matrix depends on the length of the bursts of
ple. We use for instance the least efficient MODCOD, that
erroneous BBFRAMEs. It is clear that longer bursts will re-
is, 1/2-QPSK. It can be seen in Figure 11 that the maximum
quire larger FEC matrices to avoid that the number of wrong
error burst length due to power arches is 7 BBFRAMEs. In
columns exceeds the correction capability of the code. There-
this MODCOD, each BBFRAME has a data field of length
fore, the design of the height of the FEC matrix should be
32128 bits [2], which is equal to 4016 bytes. Therefore, a
derived from an analysis of the length of the error bursts.
burst of 7 BBFRAMES corresponds to 28112 bytes. We con-
Figure 11 shows the histogram of the length of the bursts for
sider that this amount of bytes should correspond to less
some particular MODCOD modes for the scenario described
than 30 columns in the FEC matrix. The value of 30 has
above. In all modes besides the two shown in Figure 11,
been chosen arbitrarily. It is nevertheless a reasonable num-
it is observed that the distribution is bimodal. The bursts
ber since the objective is to leave a margin with respect to the
of short length (typically between 1 and 4 BBFRAMEs) are
64 columns that the code can correct (assuming no punctur-
due to random errors caused by noise, whereas the rest of
ing) so as to be able to cope with errors caused by noise as
bursts are caused by the power arches. Second, the higher
well. Therefore, the column size of the FEC matrix should
the modulation order, the longer the error bursts produced
fulfil
by power arches are. This is justified by the fact that ac-
cording to the DVB-S2 standard, BBFRAMEs are coded and
converted into FECFRAMEs, which have constant length 30Lc ≥ 28112 =⇒ Lc ≥ 938 bytes, (7)
in bits regardless of the used modulation [2]. The bits in
the FECFRAME are transformed by the modulator into where Lc is the number of rows (i.e., the length of each col-
bytes in the PLFRAMEs. Higher modulations need fewer umn) of the FEC matrix in bytes. In the previous compu-
symbols and, hence, less time to transmit an FECFRAME. tation, we have not taken into account the overhead intro-
The duration of the fade event caused by a power arch duced by GSE since it is small and we are only interested
only depends on the speed of the train, which we have in obtaining an approximate value for the column size. If
considered to be 300 km/h throughout the rest of the pa- the same calculation is repeated for the most efficient MOD-
per. Therefore, the shorter the PLFRAME, the more PL- COD, that is, 5/6-16APSK, the result is Lc ≥ 2912 bytes. The
Stefano Cioni et al. 13
1/2-QPSK 2/3-8PSK
0.12 0.12
0.1 0.1
IP packet error rate
0.06 0.06
0.04 0.04
0.02 0.02
0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Column size (bytes) Column size (bytes)
(a) (b)
3/4-16APSK 5/6-16APSK
0.12 0.14
0.1 0.12
0.1
IP packet error rate
0.08
0.08
0.06
0.06
0.04
0.04
0.02 0.02
0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Column size (bytes) Column size (bytes)
(c) (d)
Figure 12: Comparison of the IP packet error rate for different ACM modes in a channel with BBFRAME error rate equal to 12% (circles →
results without any kind of PL-FEC, squares → results with GSE-FEC).
results for the intermediate MODCODs, 2/3-8PSK and 3/4- 6.2.2. Performance analysis
16APSK, are 1790 and 2618 bytes, respectively.
We conclude from this discussion that the appropriate
Dependence on the size of the FEC matrix
size of the FEC matrix strongly depends on the error burst
length caused by the power arches, which in its turn depends
on the train speed. The lower the train speed is, the longer The IP packet error rate as a function of the column size for
the bursts are and the taller the FEC matrix must be. How- different MODCODs is shown in Figures 12 and 13. The con-
ever, the size of the FEC matrix cannot be increased arbitrar- sidered columns sizes and the corresponding number of GSE
ily because it has an impact on the delay of GSE-FEC process units used to encapsulate each RS redundancy column are
and, on top of that, because more errors due to noise appear listed in Table 2. The number of GSE units per column has
inside the FEC matrix. These errors may risk the correction been selected in such a way that the size of the units is small
capability of the code, as will be seen below. Therefore, the enough to limit the amount of padding in the BBFRAMEs,
performance of GSE-FEC may be limited for low train speeds but large enough not to penalize encapsulation efficiency
since it is not possible to combat simultaneously very long er- (encapsulation efficiency is out of the scope of this work and
ror bursts due to power arches and a large amount of random will be analyzed in a follow-on paper). A fixed IP packet
errors due to noise. length equal to 576 bytes has been considered.
14 EURASIP Journal on Wireless Communications and Networking
1/2-QPSK 2/3-8PSK
0.025 0.025
0.02 0.02
IP packet error rate
0.01 0.01
0.005 0.005
0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Column size (bytes) Column size (bytes)
(a) (b)
3/4-16APSK 5/6-16APSK
0.02 0.03
0.025
0.015
IP packet error rate
0.01 0.015
0.01
0.005
0.005
0 0
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Column size (bytes) Column size (bytes)
(c) (d)
Figure 13: Comparison of the IP packet error rate for different ACM modes in a channel with BBFRAME error rate equal to 2% (circles →
results without any kind of PL-FEC, squares → results with GSE-FEC).
Figures 12 and 13 also compare the results obtained when IP packets were correctly received in spite of the fact that the
GSE-FEC is used and when no packet-level FEC is applied. BBFRAME error rate is higher than 10%.
The baseline GSE-FEC is employed, that is to say, no ad- For small column sizes, the IP PER decreases as the col-
ditional padding has been used in the first 191 columns umn size increases. This behaviour is in line with the discus-
and no puncturing of the last 64 columns has been per- sion at the beginning of this section: when the FEC matrix
formed. The case of no packet-level FEC follows the same is too small, a power arch causes errors in a portion of the
architecture as for GSE-FEC, depicted in Figures 6 and 7. matrix that is too large to be corrected by the code. The IP
The difference is that the 255 columns of the FEC ma- PER decreases until it reaches a minimum, which is attained
trix are filled with IP packets and no redundancy is intro- at a column length that is well approximated by the previ-
duced into it. Figure 12 was obtained when the physical- ous back-of-the-envelope calculations. If the column length
layer simulator was tuned to provide a BBFRAME error rate is increased further, the IP PER increases because the correc-
around 0.12, whereas Figure 13 was obtained for a value tion capability of the code is fixed and equal to 64 columns,
of 0.02. but the size of the FEC matrix becomes larger and, hence,
In the case of no packet-level FEC, the IP PER is almost the number of errors due to noise increases. This behaviour
insensitive to changes in the column size and its value is very is visible in Figure 12, but not in Figure 13. The reason is
close to the BBFRAME error rate, as expected. It is very in- that the later figure corresponds to a scenario with very high
teresting to observe that the proposed scheme, GSE-FEC, ef- signal-to-noise ratio, and BBFRAME errors are almost only
fectively reduces the IP PER and, in many configurations, the caused by power arches.
IP PER is exactly zero.3 This means that, in those cases, all
Dependence on the IP packet length
3 Note that the simulation duration was equal to 5000 BBFRAMEs, so we The effect of different IP packet lengths is shown in Figure 14.
can only say that the IP PER is not worse than 2 × 10−5 . In this case, the column size of the FEC matrix is fixed
Stefano Cioni et al. 15
1/2-QPSK 5/6-16APSK
0.16 0.16
0.14 0.14
0.12 0.12
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0 0
0 500 1000 1500 0 500 1000 1500
IP packet length (bytes) IP packet length (bytes)
No PL-FEC No PL-FEC
Column size: 1024 bytes Column size: 1024 bytes
Column size: 4096 bytes Column size: 4096 bytes
(a) (b)
Figure 14: Dependence of the IP packet error rate with the IP packet size for two column sizes (1024 and 4096 bytes) and two MODCOD
modes (1/2-QPSK and 5/6-16APSK).
and equal to 1024 or 4096 bytes. The general trend is that it is not possible to propose a single value appropriate for
the IP PER slightly increases as the IP size increases. There all scenarios. We consider that the column size must be an
are however some lengths, such as 576 bytes, that are espe- adaptive parameter, which is changed in response to vari-
cially favourable. This happens because for those lengths an ations of the propagation conditions, train speed, and so
integer number of IP packets fit in an integer number of forth. This adaptation would constitute an example of cross-
columns of the FEC matrix. For instance, it is fulfilled that layer optimization, whereby a link layer parameter (i.e., the
576 × 16 = 1024 × 9, which means that 16 IP packets of column size of the FEC matrix) is adapted as function of
length 576 bytes fit in 9 columns of length 1024 bytes. As this the physical-layer conditions. The padding and puncturing
perfect fitting reduces the ratio of IP packets that are split of columns in the FEC matrix are other degrees of freedom
across two columns, the number of IP packets corrupted by that can be exploited in the parameterization of GSE-FEC.
a wrong column is also reduced on average. If the length of A detailed analysis of these aspects is a subject for further
IP packets follows a certain distribution, as it happens with research.
real traffic, the IP PER can be obtained by computing an av-
erage of the values shown in Figure 14. This average would 6.3. Comparative analysis
be computed by weighting the IP PER for a given length by
the frequency of occurrence of that length. As it can be seen from the results presented in the last two
sections, very satisfactory results to ensure reliable reception
Conclusions on GSE-FEC results can be obtained with both techniques. In the case of antenna
diversity, this does not penalize the overall system efficiency,
The analysis of the GSE-FEC and the corresponding numer- although some additional complexity in the receiver imple-
ical results has shown that the column size is a key design menting the MRC scheme will be accounted for. However,
parameter. Long columns appropriate to obtain low IP PER the main issue to be addressed in the practice is represented
when the duration of the fade events caused by power arches by the installation of two antennas. Many experiments and
is large (e.g., when the train is moving slowly) or when very trials have shown that this is a very critical point, since anten-
spectrally efficient MODCODs are used; but this comes at nas suitable for installation on trains are subject to very strict
the price of a large encoding and decoding delay, and an in- requirements in terms of pointing accuracy, size, and ro-
creased sensitivity to random BBFRAME errors caused by bustness against mechanical vibrations, wind, pressure gra-
noise and interference. Therefore, the column size must be dients when entering or exiting a tunnel, and so forth. With
selected as the result of a tradeoff between competing goals; current antenna technologies, a relatively high failure rate
16 EURASIP Journal on Wireless Communications and Networking
REFERENCES
of mechanical components included in the antenna plat- [1] EN 300 421 v1.1.2: Digital Video Broadcasting (DVB); Fram-
form has to be expected. Furthermore, train operators are ing structure, channel coding and modulation for 11/12 GHz
extremely keen on keeping the installation and maintenance satellite services.
procedures as simple as possible. For all these reasons, addi- [2] ETSI EN 302 307 v1.1.1: Digital Video Broadcasting (DVB):
tional countermeasures must be also investigated as possible Second generation framing structure, channel coding and
complement to the presence of two antennas (e.g., in case modulation system for Broadcasting, Interactive Services,
one antenna suddenly breaks and no immediate replacement News Gathering and other broadband satellite applications.
is possible). [3] ETSI EN 301 790 v1.4.1: Digital Video Broadcasting (DVB):
Although it has been shown that the dimensioning of Interaction channel for satellite distribution systems.
[4] ETSI TR 101 790 v1.3.1: Digital Video Broadcasting (DVB):
packet level FEC is a complex task, that will be carried out
Interaction channel for satellite distribution systems; Guide-
following a cross-layer approach, the results presented in the lines for the use of EN 301 790.
previous section confirm that also this technique, if properly [5] ETSI EN 302 304 v1.1.1: Digital Video Broadcasting (DVB);
designed, can guarantee reliable reception at the expenses of Transmission System for Handheld Terminals (DVB-H).
a limited increase in the system complexity and overhead. [6] S. Scalise, R. Mura, and V. Mignone, “Air interfaces for satellite
The concrete solution presented in this paper has been es- based digital TV broadcasting in the Railway environment,”
pecially devised taking into account the architectural con- IEEE Transactions on Broadcasting, vol. 52, no. 2, pp. 158–166,
straints introduced by the latest encapsulation scheme (GSE) 2006.
currently being proposed for future DVB systems. Clearly, [7] E. Lutz, M. Werner, and A. Jahn, Satellite Systems for Per-
packet level FEC results in a reduction of the overall spectral sonal and Broadband Communications, Springer, New York,
efficiency of approximately 33% with the adopted RS code, NY, USA, 2000.
partially compensated by the migration to a more efficient [8] S. Scalise, H. Ernst, and G. Harles, “Measurement and mod-
elling of the land mobile satellite channel at Ku-band,” to ap-
encapsulation scheme such as GSE.
pear in IEEE Transactions on Vehicular Technology.
[9] E. Kubista, F. P. Fontan, M. A. V. Castro, S. Buonomo,
7. CONCLUSIONS B. R. Arbesser-Rastburg, and J. P. V. Polares Baptista, “Ka-
band propagation measurements and statistics for land mobile
To conclude, two countermeasures are thoroughly analyzed satellite applications,” IEEE Transactions on Vehicular Technol-
in this paper: antenna diversity and a packet-level forward ogy, vol. 49, no. 3, pp. 973–983, 2000.
[10] A. Benarroch and L. Mercader, “Signal statistics obtained form
error correction mechanism especially tailored to DVB-S2,
a LMSS experiment in Europe with the MARECS satellite,”
named GSE-FEC. Simulations have shown the excellent per-
IEEE Transactions on Communications, vol. 42, no. 2–4, pp.
formance of both approaches, while they have complemen- 1264–1269, 1994.
tary features in terms of hardware complexity, delay, and [11] G. Sciascia, S. Scalise, H. Ernst, and R. Mura, “Statistical char-
bandwidth efficiency. Generally speaking, the results in this acterization of the railroad satellite channel at Ku-band,” in
paper show that effective countermeasures to compensate the Proceedings of the International Workshop of Cost Actions 272
impairments of the railroad satellite channel are possible and and 280, Noordwijk, The Netherlands, May 2003.
can be integrated into the existing DVB-S2 standard with a [12] S. Scalise, O. Lücke, and E. V. Torralbo, “A link availability
limited to moderate impact on the receiver design and on the channel model for the railroad satellite channel,” in Proceed-
system complexity. In fact, to support antenna diversity, the ings of 24th AIAA International Communications Satellite Sys-
receiver structure will be modified as depicted in Figure 4, tems Conference (ICSSC ’06), vol. 1, pp. 305–317, San Diego,
whereas for packet level FEC a software implementation may Calif, USA, June 2006.
[13] S. Cioni, G. E. Corazza, and A. Vanelli-Coralli, “Antenna di-
be considered.
versity for DVB-S2 mobile services in Railway environments,”
Further topics to be addressed in order to conclude the to appear in Journal of Satellite Communications and Networks,
analysis of the forward link are the following: special issue on ASMS Conference.
(i) cross-layer optimization of all the relevant parameters [14] S. Cioni, M. Berdondini, G. E. Corazza, and A. Vanelli-Coralli,
(MODCODs and GSE-FEC), taking also into account “Antenna diversity for DVB-S2 mobile services in Railway en-
vironments,” in Proceedings of the 3rd Advanced Satellite Mobile
nLOS channel conditions and the usage of ACM to
Systems (ASMS) Conference, Herrsching am Ammersee, Ger-
compensate for slower fades due to atmospherical ef- many, May 2006.
fects, [15] S. Cioni, A. Vanelli-Coralli, C. Párraga Niebla, S. Scalise, G.
(ii) inclusion of mechanizm(s) to support QoS and study Seco Granados, and M.A. Vázquez Castro, “Antenna diver-
of their integration and interaction with the proposed sity and GSE-based packet level FEC for DVB-S2 systems
GSE-FEC scheme. in Railway scenarios,” in Proceedings 25th AIAA International
Stefano Cioni et al. 17
Research Article
Capacity Versus Bit Error Rate Trade-Off in
the DVB-S2 Forward Link
German Aerospace Center (DLR), Institute of Communications and Navigation, Oberpfaffenhofen, 82234 Wessling, Germany
The paper presents an approach to optimize the use of satellite capacity in DVB-S2 forward links. By reducing the so-called safety
margins, in the adaptive coding and modulation technique, it is possible to increase the spectral efficiency at expenses of an
increased BER on the transmission. The work shows how a system can be tuned to operate at different degrees of this trade-off,
and also the performance which can be achieved in terms of BER/PER, spectral efficiency, and interarrival, duration, strength of
the error bursts. The paper also describes how a Markov chain can be used to model the ModCod transitions in a DVB-S2 system,
and it presents results for the calculation of the transition probabilities in two cases.
Copyright © 2007 Matteo Berioli et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Other modern media codecs (e.g., MPEG-4 [5]) have been the channel impairments to provide low BER. A destination
designed to be highly resilient to residual errors in the in- with a bad channel state can thus use a very robust modula-
put bit-stream, to detect and localize errors within the packet tion and coding pair (ModCod) while other terminals with a
payload, and to employ concealment techniques, like for in- very good channel state can still transmit in highly efficient
stance interframe interpolation, that hide errors from a hu- ModCods. The adaptive selection of the best suited ModCod
man user. These codecs offer acceptable quality at a resid- results in an increased net data throughput while terminals
ual BER poorer than 10−3 , and some at poorer than 10−2 in bad channel conditions are still able to receive their data
[6]. In order to support these error-tolerant codecs, the IETF since they can use ModCods with lower-order modulation
has also standardized a new multimedia transport proto- and higher coding (but at the cost of lower spectral efficiency
col, UDP-Lite [7], that allows to specify the required level of and thus lower throughput).
payload protection, while maintaining end-to-end delivery As can be seen in Figure 1 the system architecture of
checks (verification of intended destination, IP header fields DVB-S2 is subdivided into six main components [1]. The
and overall length). mode adaptation subsystem provides an interface to the ap-
When these services are operating over the satellite con- plication specific data stream formats and also contains a
nection, it is convenient to reduce the quality of the trans- CRC-8 error detection coding scheme. It is possible to merge
mission in DVB-S2 forward links, by allowing higher BERs, different input streams together and to segment them into
in order to increase the precious capacity and the through- the so-called data fields which are the payload part of the so-
put. The first motivation for this is to make use of cross-layer called baseband-frames (BBFRAME) created at the output
mechanisms by voluntarily allowing higher bit error rates of the consecutive stream adaptation module. Buffers store
which can be compensated with error correction at higher data until they are processed by the merger/slicer and in case
layers. A second motivation to allow for higher BERs is that not enough data is available to fill a data field or if it is re-
not all applications have the same stringent BER require- quired to have only an integer number of packets in a frame
ments. This represents a natural trade-off between errors and (in general integer number of packets will not perfectly fit
capacity. The present work analyzes this trade-off, proposes into a frame but their payload sum will always be slightly
a way to tune the system parameters in order to work in smaller or larger than the data field), the unused space can be
optimal conditions, and investigates the performance of the padded, this operation is accomplished by the stream adap-
system in this situation. The work is organized as follows: tation subsystem. In order to complete the baseband frame
Section 2 presents the background and the scenario of the (BBFRAME) additional header information (BBHEADER)
subject, Section 3 describes the main ideas of the paper and is added in front of the data field and scrambling of header
the original approach to the problem, Section 4 evaluates the and payload is applied. The final BBFRAME structure is il-
performance of a system operating in the suggested condi- lustrated in Figure 2.
tions, and Section 5 drives the conclusions of the paper. The consecutive FEC encoding block performs outer and
inner coding and bit interleaving. The coding scheme which
2. BACKGROUND AND SCENARIO is used is selected based on the channel measurements re-
ceived from the terminals the data of which is contained in
2.1. Overview of DVB-S2 the frame. The outcome of this module, called forward error
correction frame (FECFRAME), is shown in Figure 3. The
The second generation of DVB-S provides a new way of fade FECFRAMEs can either have a length of 16200 bits for short
mitigation by means of adaptation of the coding and modu- frames or 64800 bits for normal frames. Since the length
lation (ACM) to the different channel states. This of course of the encoded frame is fixed, this means that the length
implies the need for every terminal to signal its perceived of the payload in the underlying BBFRAME changes with
channel state back to the gateway which can then make a the applied coding. For applying higher-order modulation
frame-by-frame decision of the modulation and coding com- schemes the subsequent mapping block performs a serial-to-
bination (ModCod) to be applied based on these measure- parallel conversion. The mapper chooses the applied mod-
ments. DVB-S2 offers a broad range of modulations and cod- ulation schemes again based on the channel measurements
ings for ACM. The supported modulation schemes comprise for the destination(s) of the data contained in the frame. The
QPSK, 8-PSK, 16-APSK, 32-APSK and considered coding outcome of the mapping of the data into symbols is called
rates are 1/4, 3/4, 1/3, 2/5, 3/5, 4/5, 1/2, 5/6, 8/9, 9/10. The an XFECFRAME which is afterwards formed into a physical
possibility to select the modulation and coding for an indi- layer frame (PLFRAME) after pilots and PL signalling have
vidual destination allows to make a more efficient use of the been inserted and after final scrambling for optimization of
system capacity since transmission in a higher-order modu- energy dispersal. In case no XFECFRAMES are provided by
lation in combination with a low coding rate (e.g., for clear the preceding subsystems, the PLFRAMING module inserts
sky conditions) allows to transmit more bits per symbol than the so-called DUMMY PLFRAMES to provide a continuous
a low-order modulation with high coding rate (e.g., for rainy TDM stream on the link. To allow every terminal indepen-
channels). In this way it is possible to use individually for ev- dent of its channel state to receive the PLHEADER informa-
ery ground terminal (or for every group of terminals in the tion (which also contains the used modulation and coding
same spot beam) the highest possible modulation scheme scheme for the underlying frame) this header is always mod-
and the lowest coding rate which still allows to cope with ulated with BPSK.
Matteo Berioli et al. 3
Mode adaptation
Data BB
Single signalling
input Input Null-packet Dotted subsystems are
Input stream CRC-8
stream interface synchroniser deletion encoder Buffer not relevant for
ACM (ACM, TS) single transport stream
command Merger broadcasting
slicer applications
Multiple
input Null-packet
Input Input stream CRC-8 Buffer
streams deletion
interface synchroniser encoder
(ACM, TS)
QPSK,
8PSK, PL signalling & α = 0.35, 0.25, 0.2
Rates 1/4, 1/3, 2/5 16APSK, pilot insertion
1/2, 3/5, 2/3, 3/4, 4/5, 32APSK
5/6, 8/9, 9/10
I PL
Bit scram BB filter
BB BCH LDPC Bit mapper Q BLER and
Padder scram encoder encoder inter- into quadrature
BLER (nbch , kbch ) (nldpc , kldpc ) constellations Dummy
leaver modulation
PLFRAME
Stream insertion
adaptation FEC encoding Mapping PLFRAMING Modulation
80 bits DFL Kbch -DFL-80 eral hundreds milliseconds (250 milliseconds). This means
that though the order of magnitude for the propagation de-
BBHEADER Data field Padding lay allows for a compensation of very slow changing channel
BBFRAME (Kbch bits) effects, like rain attenuation, it is too long to compensate fast,
high-frequent changes in the SNR as those caused by scintil-
Figure 2: Structure of a BBFRAME [1]. lation, this will be explained in the next section.
Power spectral density of rain attenuation and scintillation Scintillation and attenuation time series
104 of useful user (forward downlink)
0.25 4.5
0.2 4
102 0.15
∼ f −2 3.5
0.1
Scintillation (dB)
Attenuation (dB)
3
PSD (dB2 /Hz)
100 0.05
2.5
0
2
−0.05
10−2 1.5
−0.1
−0.15 1
∼ f −8/3
10−4 −0.2 0.5
−0.25 0
0 1000 2000 3000 4000 5000 6000 7000 8000
10−6
10−4 fa 10−3 10−2 10−1 fs 100 101 Time (s)
Frequency (Hz)
Figure 5: Example of scintillation and rain attenuation.
Rain attenuation
Scintillation
Figure 4: Attenuation and scintillation spectrum (typical values: ing can be mitigated by mode adaptation whereas counter-
fa ≈ 10−4 Hz, fs ≈ 0.1 − 0.65 Hz). measures for scintillation require a different compensation.
For every combination of modulation and coding a threshold
thrdem (ModCod) exists which is needed to be able to decode
of the scintillation process can be calculated according to (1) the frame with a quasi-zero BER. The decision of the gateway
corresponding to the theory of Tatarskii [9] and the model of on which ModCod will be used is thus driven by thresholds.
Matricciani [10], For switching among ModCods these thresholds could theo-
σ = σ0 · A5/12 . (1) retically be used directly for the decision about which Mod-
Cod will be used, but in practice this would result in frequent
The value σ0 is the standard deviation of the scintillation for transmission errors since the high frequent variations of the
a rain attenuation of A[dB]. [10] suggests a typical value of channel (due to scintillation) would cause a frequent crossing
0.039 for σ0 in the frequency range of 19.77 GHz. According of the threshold. On the one hand, this high frequent cross-
to (1) the resulting scintillation standard deviation σ is then ing cannot be compensated by signalling to the gateway, on
in the order of tenths of a dB for rain attenuations smaller the other hand such signalling would also mean a high fre-
than 20 dB. quent change of the ModCod which is as well undesirable.
Within this work the main focus is on the scintillation To provide more reliability the minimal needed demod-
effects since these cannot be compensated by signalling of ulation thresholds thrdem (ModCod) can be replaced by
the channel states via the return channel because of the thresholds which have a certain safety margin. This means
long propagation delay of the GEO satellite. Nevertheless the that a lower ModCod is selected already before the critical
channel simulations used in the rest of this work consider threshold (the threshold below which a strong increase in
spatial correlated rain attenuation as well since the magni- bit error rate occurs) is reached. The size of the safety mar-
tude of the scintillation also depends on the intensity of the gin does thus determine the robustness against fast occurring
rain attenuation (see (1)). Similar to the generation of the scintillation fades. On the other hand, this size of the safety
scintillation, also the rain attenuation is created via a normal margin also influences the system performance since trans-
distributed random variable whereas its spectrum has a dif- mission in a higher ModCod would result in a higher net
ferent corner frequency of fa (see also Figure 4). data rate. Since fast oscillations between neighboring Mod-
Figure 5 shows a channel example for the attenuation Cods are also possible when safety margins are used, an ad-
caused by scintillation and rain for a user located at longi- ditional hysteresis margin is introduced. Figure 6 illustrates
tude 8.6◦ E and latitude 52.7◦ N, in the area around Hamburg the different thresholds and margins.
(Germany). It can be seen here that scintillation effects occur Within Figure 6 the terms thrdem (N − 1) and thrdem (N)
with a much higher frequency than regular rain attenuation denote the minimum SNR values which are just enough to
events and how rain attenuation and scintillation are corre- provide quasi error free decoding. If, for example, ModCod
lated. N is used and the signal strength falls below the thrdem (N)
threshold, the BER will drastically increase. These thresholds
2.3. ModCod switching strategies have also been called critical in [11] for this reason. If on the
other hand the signal strength increases, for example, while
While the rain attenuation occurs on a larger time scale scin- using ModCod N − 1, the next higher ModCod is not selected
tillation effects occur very rapidly. For this reason rain fad- as soon as the demodulation threshold of the next higher
Matteo Berioli et al. 5
SNIR
The threshold for switching to a smaller ModCod is denoted No switching
here because
as thrdown (ModCod) and the size of the hysteresis margin as of hysteresis
thrdown (N − 1)
(Δthrhyst (N)). The distance between the critical demodula- Δthrsafety (N − 1)
tion threshold thrdem (ModCod) and the threshold that trig- thrdem (N − 1)
gers a downswitching thrdown (ModCod) is called the safety
margin Δthrsafety (ModCod).
ModCod N − 1 ModCod N
The safety margin(Δthrsafety (ModCod)) can be seen as
an additional security for high frequent oscillations which Time
cannot be countervailed due to the long satellite propagation
delay. If the signal strength oscillates within this area no in- Figure 6: Illustration on the different thresholds.
crease in BER will occur since the thrdem (ModCod) thresh-
old is not crossed. The values for the safety margin and the
hysteresis margin can be varied and they can also be different
for every ModCod. Wörz et al. [11] presented a calculation is important to carefully describe the assumptions on which
method for all aforementioned parameters which provides a the analysis is based, this is what is presented in this sec-
quasi error free system performance. The calculation of the tion. Though the obtained results have a quantitative mean-
parameters in [11] mainly depends on estimated values of ing only considering these assumptions, it is worth stating
the scintillation standard deviations and a numerically de- that their qualitative relevance has a general importance, as it
rived function which accounts for the fact that the standard will be later explained.
deviation of the scintillation is also dependent on the inten- Existing systems compliant with the DVB-S2 standard
sity of the rain attenuation. can provide the higher-layers protocols with a quasi-error-
Within the remaining parts of this work the influence of free underlaying physical layer (PER = 10−7 ). For this pur-
the size of the safety margins with respect to gain or loss in pose regular 8-bytes CRC (cyclic redundancy check) fields
channel net efficiency and increase/decrease of BER is inves- are used to identify errors in the BBFRAME, which were not
tigated. The term “Wörz-Schweikert safety margins” denotes corrected by the coding schemes (LDPC and BCH) at re-
the safety margins calculated according to the algorithm pre- ception. In case an error is detected in a frame, it has to be
sented in [11] while “zero-safety margin” denotes the fact considered that the wrong bit(s) cannot be singularly identi-
that no safety margin is used. fied in the frame, so one of the two following choices can be
made:
2.4. Investigated environment (1) the whole frame is discarded (this is what is normally
done);
Within the examined scenarios, a set of user terminals has (2) the packets in the frame are passed to the higher proto-
been located in a geographical region close to the city of cols with uncorrected failures (this can be done in case
Hamburg, Germany (longitude 9.5◦ −10.5◦ E, latitude 52.5◦ − the higher protocols are able to cope with errors).
54◦ N) within the aforementioned channel simulator. The set These cases are very rare when high safety margins are
comprises 38 different terminal locations whereby the chan- adopted, and systems are normally dimensioned to avoid
nel states are sampled with 10 Hz. The investigated duration them, but they become more frequent if the system works
is 7200 seconds per simulation run. In order to get statistical closer to the demodulation thresholds (as we defined them
significant results the simulation duration of 7200 seconds in the previous section), for the reasons already explained.
per simulation have been extended to 60 hours. The 60 hours If a system is dimensioned also to operate in these con-
channel simulation results for the 38 terminals can be seen as ditions, it is important to evaluate the statistical properties
2280 hours of simulated channel states for a single terminal and characteristics of these situations, that is, how often they
what in turn means that all results are based on channel in- occur and what failures they bring in comparison to the ca-
formation which corresponds to roughly a quarter of a year. pacity gain. In order to do that we performed three levels
of analysis. They are theoretically described in this sections,
3. SYSTEM MODELLING whereas the results obtained for each of them are shown and
discussed in the next one.
The main idea behind this study is that by reducing the safety
margin in the ModCod switching strategy it is possible to 3.1. Markov model
gain in spectral efficiency, and thus to increase the net data
throughput, at the expenses of an increased BER (and con- The first analysis is a comparison of the new approach with
sequently a higher PER). In order to investigate this and to a classical one existing in literature (the already mentioned
derive a detailed quantitative estimation of this trade-off, it Wörz et al. [11]), in terms of ModCod switching statistics.
6 EURASIP Journal on Wireless Communications and Networking
The best way to show the difference between the two ap- versus SNR characteristic, but to use one single function for
proaches is to model the system according to a Markov chain, all ModCods already seems an excellent approximation to the
where each state represents the system operating with one real case, so this is how it was implemented in the simulator.
particular ModCod. The chain presents two states for each PERs are derived from these BERs under consideration
ModCod N: the good one, NG , and the bad one, NB ; so the of the payload length of each BBFRAME also regarding the
overall number of states is twice the total amount of allowed applied ModCod. A BBFRAME is considered as erroneous if
ModCods (56). In the good state the SNR measured at the re- at least one of the payload bits is erroneous. For the rest of
ceiver is above the demodulating threshold for that ModCod, this paper the term PER denotes the BBFRAME packet error
and so no failures are expected, in the bad state the system rate. Thanks to this definition of the states of Markov chain,
SNR is below the demodulating threshold for that ModCod, this model allows to derive the properties of the communi-
and so failures may occur with probabilities that are not neg- cation in terms of PER and BER statistics, and by knowing
ligible. the spectral efficiency associated to each ModCod it is easy to
A similar Markov chain is an excellent model, because it derive an average resulting capacity.
summarizes very well the properties of a ModCod switch-
ing approach. So once the transition probabilities for one 3.3. Error bursts analysis
particular ModCod switching criterion have been calculated
(normally by simulations), the Markov chain can be used The third and last level of analysis goes into the details of the
as a basis for all types of analysis without the need of run- failures introduced with this novel approach. In the previous
ning again computationally heavy simulations, which might section we explained how to derive a measure of the trade-
be very long in order to gather statistically meaningful data. off between average capacity and average BER (or PER). An
In this sense the calculated Markov chain (i.e., the ModCod average measure of the BER (or PER) does not seem a very
transition probabilities) can be considered independent from precise information, since these failures come in bursts. The
the simulated channel conditions, only if the simulation is errors are mainly due to the ModCod switchings, and they
long enough to represent general channel statistics. On the are mostly introduced by reduced safety margins. So we want
other hand, it should be mentioned that the same Markov to investigate three main properties: (i) how often the error
chain depends on some parameters which might be charac- bursts arrive (interarrival times statistics), (ii) how long the
teristic of particular cases, for example, the link budget in bursts last (duration statistics), and (iii) how deep the fades
clear sky, and consequently the system availability. So even if are (i.e., how high are BER and PER during one error burst).
the resulting numbers are only meaningful bearing in mind These three properties can be estimated thanks to the Markov
these assumptions, the quantitative conclusions which can be model, and this analysis produces interesting information,
derived have general relevance, and this will be clearer in the which will be presented in the next section.
next section.
4. RESULTS EVALUATION
3.2. Error rate versus capacity trade-off
4.1. Markov model
The second level of analysis describes the details of each state
of the Markov chain. The good states present quasi-error-free A software simulator was developed in order to derive the
conditions according to the DVB-S2 recommendations, so Markov model presented in the previous section. Once the
PER = 10−7 . Since the BER versus SNR characteristics for ModCod switching criterion has been specified the software
all ModCods are very steep, the BER values increase quite simulates the evolution over time of the system; from these
rapidly when the SNR level goes below the demodulating simulations we can derive statistics about the permanence in
threshold. In particular they change of several orders of mag- the different ModCods for each ModCod switching criterion,
nitude within a few tenths of dB, going from BER ≈ 10−10 this was done by computing transition matrices and solving
when SNR is close or bigger than the demodulating thresh- them. In the following we present two full transition matrices
olds, up to BER ≈ 10−2 when the SNR is just 0.3 dB be- for two different ModCod switching criteria.
low the threshold. Each bad state represents a set of different Simulations equivalent to 3 months of SNR time series
BERs, the proper BER is selected at each time step according have been carried out, one using Wörz-Schweikert safety
to the received level of SNR with respect to the demodulat- margins, the other one using zero-safety margin bounds with
ing thresholds. The exact characteristics for the BER-versus- Wörz-Schweikert hysteresis bounds.
SNR functions, which were used in the simulations and to The matrices in Figures 7 and 8 represent the transition
derive the Markov chain parameters, were taken from [12]. probabilities for those two approaches, where position (i, j)
In that work, end-to-end performances of the BER versus is the probability in each time step (0.1 second) to move from
the SNR are presented for the DVB-S2 system, the whole state i to state j; the first line and the first column of each
communication chain is modelled and simulated, including ModCod represent the bad state (iB and jB ), the second one
coding, modulation with predistortion techniques, satellite the good state (iG and jG ). Figures 7 and 8 show the transi-
transponder impairments, downlink, demodulation with the tion matrices for zero-safety margin and the Wörz-Schweikert
synchronization, and the final LDPC and BCH decoders. In safety margins. The cells marked black indicate that their
the Markov chain, each ModCod might have its own BER content is unequal to zero. In Figure 8, we can see that the
Matteo Berioli et al. 7
1 2 3 4 5 6 7 8 9 12 11 13 14 18 19 20 21
1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
1 0, 01 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 13 0, 74 0, 13 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
2 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 14 0, 74 0, 12 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
3 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 14 0, 73 0, 13 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
4 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 14 0, 72 0, 14 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
5 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 13 0, 71 0, 16 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
6 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 13 0, 71 0, 16 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
7 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 12 0, 71 0, 16 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
8 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 13 0, 71 0, 17 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
9 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 01 0, 12 0, 73 0, 15 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
12 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 12 0, 70 0, 18 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
11 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 01 0, 11 0, 71 0, 17 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
13 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 11 0, 70 0, 19 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
14 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 10 0, 66 0, 23 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
18 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 09 0, 65 0, 26 0, 00 0, 00 0, 00 0, 00
19 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 07 0, 57 0, 36 0, 00 0, 00
20 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 07 0, 57 0, 37
21 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00
1 2 3 4 5 6 7 8 9 12 11 13 14 18 19 20 21
1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
1 0, 01 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
2 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
3 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
4 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
5 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
6 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
7 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
8 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
9 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
12 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
11 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 99 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
13 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
14 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
18 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
19 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
20 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 1, 00 0, 00 0, 00
0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0 ,00 0 ,0 0
21 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 00 0, 01 0, 00 0, 99
more “stable” states are the good states. This makes sense if ity. Transitions only occur between good states and this con-
we look at the SNR time series, because switchings between firms that this approach is designed to work only in good
ModCods are quite spaced in time compared to the time step states.
of 0.1 second. Simulations estimate an average number of 4.5 For the novel approach, the zero-safety margin one, it
ModCod switchings per hour. may be interesting to derive the probability to be in each
Moreover on the diagonal, the probabilities of remaining ModCod (bad or good state). Once the transition matrix
in a bad state are not so high; this is also correct, since when for the Zero-Safety margin is solved [13], we end up with
we are in a bad state, we know that a down-switch should Figure 9 which shows a stacked probability density graph for
occur. The only bad state which has a higher stability is the good and bad states of each ModCod. This is the result of
bad state for ModCod 1B . This results from the fact that when a simulation of an equivalent of 4.5 years of SNR evolution
the SNR goes below the last demodulation threshold, the sys- over time. What we can see is that the most used ModCods
tem cannot switch to a lower ModCod, so it remains in bad are those whose demodulation threshold is just below the
state until the SNR rises again. This is basically an outage SNIR in clear sky conditions. That makes sense because most
where the DVB-S2 receiver is not available; the simulator was of the time we are in clear sky conditions, so we use the high-
designed to give a system availability of 99.96% of the time, est ModCods. We can also notice the high value of the bad
for both approaches. state in ModCod 1B , because of system unavailability. Some
With the Wörz-Schweikert scheme, shown in Figure 8, the ModCods are never used due to overlapping with other ones,
matrix is far more sparse, and no bad states are ever ac- some ModCods achieve a better spectral efficiency requiring
tive, except for ModCod 1B because of system unavailabil- less SNR.
8 EURASIP Journal on Wireless Communications and Networking
100 100
10−1
10−1 10−2
10−3 6 × 10−3
4 × 10−3
10−2 10−4
Error rate
10−5
Probability
10−3 10−6
10−7
10−4 10−8
10−9
10−5 10−10
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5
Multiplying factor on Schweikert-Wörz safety margin
10−6
2 4 6 8 10 12 14 16 18 20 22 24 26 28 Packet error rate
ModCods Bit error rate
Bad states
Good states Figure 10: Packet error rate (PER) versus Safety margin.
3.5
Max @ − 1.5 = 3.12 b/Hz/s Max @ − 0.4 = 3.01 b/Hz/s
Average spectral efficiency (b/s/Hz)
This section presents the main results which are obtained 2.5
when reducing the safety margins, in terms of increase spec-
tral efficiency and increase errors. The starting point is the set 2
of threshold selected by Wörz-Schweikert; this set guarantees 1.5
a quasi-error-free system operation. We try to proportionally
reduce those margins and even to have negative margins, to 1
see how the system performs. The x axis in Figures 10 and 11
0.5
represents factors to be multiplied to the Wörz-Schweikert
set to get the tested thresholds. This means that for multiply- 0
ing factor 1 we have the Wörz-Schweikert set, for the factor −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1
0, we have the zero-safety margin approach, and for negative Multiplying factor on Schweikert-Wörz safety margin
values of the factors we are testing thresholds which are be- Gross efficiency
low those thresholds recommended by the DVB-S2 standard. Net efficiency (without bit aggregation)
This may seem strange, but it will appear clear how useful Net efficiency (with bit aggregation)
this is to show that there is a trade-off between errors and
increase in capacity. Figure 11: Average spectral efficiency versus safety margin.
Figure 10 shows (as expected) that the PER objective of
10−7 is achieved already before the Wörz-Schweikert bounds.
This is not surprising since the model has been designed to
do so. As expected as well, PER and BER are fast-growing up and −1.5 at the bit level. Corresponding values of PER/BER
to 1 when the safety margin becomes negative. A surprising at these maxima are 6 · 10−3 and 4 · 10−3 . The two curves
fact here is that there are possibilities to achieve the goal PER represent the two ways of operating described in Section 3:
even for margins which are 0.4 times the Wörz-Schweikert bit aggregation is when failures cause BBFRAME discard, no
safety margins. That means that those Wörz-Schweikert mar- bit aggregation means when the frame is passed to the higher
gins may not be the optimum selection. layers with failures. It should be noted that for bit error ag-
Figure 11 shows the core result of this work. A trivial gregation (see Figure 11) the PER (see Figure 10) is the rele-
thing is that the gross capacity (total amount of received bits vant result since in case of a bit error the complete BBFRAME
with failures) is still increasing when we go for lower and is discarded. Without consideration of bit error aggregation,
lower bounds, because of course we are using less and less the BER is the relevant result since erroneous bits within the
robust ModCods that provide better spectral efficiency. The BBFRAME are expected to be corrected by the higher layers.
very interesting point comes with the fact that the net ca- This means that a system which wants to have the indicated
pacity (throughput of correct bits) shows a maximum in the throughput with or without bit aggregation, is operating at
negative part of the scaling factor: −0.4 at the packet level those PER/BER.
Matteo Berioli et al. 9
SNIR at receiver and ModCod selection
100
10−1 0.8
10−2
PE
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 1e − 3
R
×104
Time in samples (0.1 s) 1e − 2
lev
> 0.5
0.5
e
1e − 1 0.4
l
0.2 0.3
0.1 tion (s)
Figure 12: PER with SNR and ModCod selection for zero-safety Fade dura
margin.
Figure 14: State probabilities.
ACKNOWLEDGMENTS
This work was partly supported by EC funds SatNEx under
the FP6 IST Programme, Grant number: 507052. This work
was supported by the European Satellite Network of Excel-
lence (SatNEx).
REFERENCES
[1] ETSI EN 302 307 V1.1.2, “Digital Video Broadcasting (DVB);
second generation framing structure, channel coding and
modulation systems for broadcasting, interactive services,
news gathering and other broadband satellite applications,”
June 2006.
[2] ETSI EN 301 790 V1.4.1, “Digital Video Broadcasting (DVB);
interaction channel for satellite distribution systems,” April
2005.
[3] G. Fairhurst, M. Berioli, and G. Renker, “Cross-layer control
of adaptive coding and modulation for satellite Internet multi-
media,” International Journal of Satellite Communications and
Networking, vol. 24, no. 6, pp. 471–491, 2006.
[4] ETSI TS 126 102, “AMR Speech Codec,” 2001.
[5] ISO/IEC 14496-2, “Coding of audio-visual objects (MPEG-
4)—part 2: visual,” 2004.
[6] ETSI TR 126 975, “Performance Characterisation of the Adap-
tive Multi-Rate (AMR) Speech Codec,” 2004.
[7] L.-A. Larzon, M. Degermark, S. Pink, L.-E. Jonsson, and G.
Fairhurst, “The Lightweight User Datagram Protocol (UDP-
Lite),” IETF, RFC 3828, 2004.
[8] S. Datta-Barua, P. H. Doherty, S. H. Delay, T. Dehel, and J.
A. Klobuchar, “Ionospheric scintillation effects on single and
dual frequency GPS positioning,” in Proceedings of the 16th In-
ternational Technical Meeting of the Satellite Division of the In-
stitute of Navigation (ION GPS/GNSS ’03), pp. 336–346, Port-
land, Ore, USA, September 2003.
[9] V. I. Tatarskii, Wave Propagation in a Turbulent Medium,
McGraw-Hill, New York, NY, USA, 1961.
[10] E. Matricciani, M. Mauri, and C. Riva, “Relationship between
scintillation and rain attenuation at 19.77 GHz,” Radio Science,
vol. 31, no. 2, pp. 273–280, 1996.
[11] T. Wörz, R. Schweikert, A. Jahn, and R. Rinaldo, “Physical
layer efficiency of satellite DVB using fade mitigation tech-
niques,” in Proceedings of the International Communication
Satellite Systems Conference (ICSSC ’05), Rome, Italy, Septem-
ber 2005.
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 62310, 12 pages
doi:10.1155/2007/62310
Research Article
Frequency Estimation in Iterative Interference Cancellation
Applied to Multibeam Satellite Systems
This paper deals with interference cancellation techniques to mitigate cochannel interference on the reverse link of multibeam
satellite communication systems. The considered system takes as a starting point the DVB-RCS standard with the use of convolu-
tional coding. The considered algorithm consists of an iterative parallel interference cancellation scheme which includes estima-
tion of beamforming coefficients. This algorithm is first derived in the case of a symbol asynchronous channel with time-invariant
carrier phases. The aim of this article is then to study possible extensions of this algorithm to the case of frequency offsets af-
fecting user terminals. The two main approaches evaluated and discussed here are based on (1) the use of block processing for
estimation of beamforming coefficients in order to follow carrier phase variations and (2) the use of single-user frequency offset
estimations.
Copyright © 2007 J. P. Millerioux et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Interference cancellation
Initial phase Estimation Interference
y1 [n] Decoding
recovery of g1,. cancellation The interference cancellation block output at the mth itera-
tion (or the decoding block input at the (m + 1)th iteration)
is for the nth symbol of the kth user
dl
(m)
[n]
From beam l, for l
interfering on beam k ∗ (m)
(m)
yk(m+1) [n] = g
k,k
(m)
yk [n] − g
k,l dl [n] . (8)
(m)
yk [n]
l =/ k
Initial phase Estimation Interference (m+1)
yk [n]
recovery
Decoding
of gk,. cancellation
yk [n] In the case of perfect channel estimation and interfering
symbol decisions, we get
dk [n]
(m) To beam l, for k 2
interfering on beam l yk(m+1) [n] = gk,k dk [n] + gk,k
∗
nk [n], (9)
interference is entirely removed, and the carrier phase is per-
fectly compensated.
Initial phase Estimation Interference
yK [n] Decoding
recovery of gK,. cancellation
Decoding
Figure 2: Block diagram of the receiver (synchronous case). Decoding is performed by the Viterbi algorithm, by assimi-
lating the residual interference plus noise after deinterleaving
at the decoder input to AWGN.
where
Initialization
T
G= g1T · · · gKT = gk,l = H diag ρk exp jϕk , For the kth user, an initial carrier phase is estimated from
T pilot symbols on the kth beam. After phase compensation,
d[n] = d1 [n] · · · dK [n] ,
the signal received on the kth beam is sent to the decoding
T block to initialize the iterative process.
y[n] = y1 [n] · · · yK [n] with yk [n] = yk (t) ∗ s(−t)|t=nT ,
T
n[n] = n1 [n] · · · nK [n] with nk [n] = nk (t) ∗ s(−t)|t=nT , 3.2. Asynchronous case
E n[k]n[l] = δ(k − l)Rn . We now consider a symbol-asynchronous time-invariant
(6) channel, that is, τk =/ τl for k =/ l, and ϕk (t) = ϕk for all k.
We introduce
A synoptic of the receiver is given in Figure 2, where inter- −1
N
leaving and deinterleaving operations are omitted for sim-
uk (t) = dk [n]s t − nT − τk ,
plicity. All operations are performed in parallel on the dif- n=0
ferent beams, with exchange of information from one to an- (10)
−1
N
other. The main steps are described in the following. For any
parameter c, c
(m) denotes an estimate or a decision on c at the u
(m)
k (t) = d
k(m) [n]s t − nT − τk ,
n=0
mth iteration.
and vectors u(t) = [u1 (t) · · · uK (t)]T and u
(m) (t) =
(m) (m)
Channel estimation [u
1 (t) · · · u
K (t)] .
T
We get
The channel estimation on the kth beam is processed by
a least-square estimator using currently estimated symbols y(t) = Gu(t) + n(t), (11)
(and including pilot symbols). At the mth iteration, we get
for the kth beam where G is defined in Section 3.1. We refer to u
(m)
k (t) as the
estimated kth signal at the mth iteration.
N −1 N −1 −1
The algorithm on the asynchronous channel is then very
g
k(m) =
(m) [n]H
yk [n]d
(m) [n]d
d
(m) [n]H . similar to the one on the synchronous channel. For the kth
n=0 n=0 beam, at the mth iteration:
(7)
(i) channel estimation is processed by a least square ap-
We only use for estimation (and consequently for interfer- proach using the estimated signals at the matched fil-
ence cancellation in (8)) estimated symbols of the useful sig- ter output u
(m) (t) ∗ s(−t) and yk (t) ∗ s(−t), syn-
nal and of adjacent interfering ones (see Section 2.2. assump- chronously sampled, with 2 samples per symbol (sam-
tion (iii)), which is not specified in the equations for the sake ples of u
(m) [n] and sam-
(m) (t) ∗ s(−t) corresponds to d
of simplicity. ples of yk (t) ∗ s(−t) corresponds to yk [n] in (7));
4 EURASIP Journal on Wireless Communications and Networking
Number of
1 2 3 Cell number C/I [dB]
interferers
1, 3 3 5
4 5 6 7
2 4 4
4, 7 3 5
8 9 10 5, 6 6 2
8, 10 5 3
11 12 13 14
9 6 2
11, 14 2 6
12, 13 4 4
(b)
(a)
(ii) interference cancellation is processed at 1 sample per reaches the Cramer-Rao bound (CRB). This bound is more
symbol, at optimal sampling instants. precisely the phase single-user modified CRB [9], given with
our notations by
More details on the implementation can be found in [1].
−1
1
hk,l 2 Es
3.3. Simulation results CRB Arg gk,l = Rd2 . (12)
2N N0
We use for the evaluation the fictitious configuration de-
scribed in Figure 3 (which is interference configuration 2 in Notice that these simulation results and all the following ones
[1]). We consider 14 cochannel beams. The 14 users have correspond to at least 20 packet errors and 200 binary errors
an equal SNR. For each cell, assumption (iii) of Section 2.2 for each user. Consider as an example the results at iteration 3
is perfectly respected, and interference is equally distributed for Eb /N0 = 2.5 dB, our evaluation of confidence intervals at
among the interfering cells: for example we have for cell 1 95% leads to [4.8, 5.9] · 10−3 for the BER of cell 5, [1.2, 12.1] ·
h1,1 = 1, h1,2 = h1,4 = h1,5 = (3 · C/I |1 )−1/2 , and other coef- 10−3 for the modulus bias of coefficient g5,1 , and [4.61, 4.89]◦
ficients of the first row of H are set to zero. We consider the for the phase error standard deviation of coefficient g5,1 .
following simulation parameters.
(i) Rate 1/2 nonrecursive nonsystematic convolutional 4. EXTENSION TO THE CASE OF
code with constraint length 7 and generators (133, FREQUENCY OFFSETS
171) in octal.
(ii) Packets of 53 information bytes (ATM cell), or 430 in-
formation symbols (with closed trellis). In geostationary systems, frequency offsets between the emit-
ter and the receiver are mainly due to frequency instabilities
(iii) 32 pilot symbols, leading finally to N = 462 transmit-
of local oscillators. Considering the use of the Ka-band with
ted symbols in a burst.
low-cost user terminals, they are inevitable. In order to help
Users timings τk are independent and uniformly distributed the receiver to recover these frequency offsets, synchronisa-
on [0, T]. Carrier phases ϕk are independent and uniformly tion bursts, which are periodically transmitted, are defined
distributed on [0, 2π]. Additive noises are uncorrelated. New in the DVB-RCS standard. However, it always remains resid-
random interleavers and training sequences are generated at ual frequency offsets on the traffic bursts. In case of short
each burst. bursts and low SNR, frequency and phase recovery become
We consider a target bit error rate (BER) equal to 2 · 10−4 , a challenging task, especially with a reduced number of pilot
which is reached on AWGN channel with perfect synchroni- symbols.
sation for Eb /N0 equal to 3.2 dB. Some results for cells 5 and In the following, we study possibilities of adaptation of
6, which are symmetric, are given in Figure 4. The algorithm the interference cancellation algorithm to the case of fre-
exhibits a degradation with respect to single-user reference quency deviations affecting user terminals. We first evaluate
of 0.15 dB after 3 iterations. At first iterations, the modulus the algorithm sensitivity to frequency offsets in Section 4.1.
estimate of g5,9 and g6,9 (which are symmetric) is widely bi- We find that it is only suited to very low frequency offsets. We
ased: it is underestimated due to imperfect symbol decisions. then evaluate in Section 4.2 the use of block processing for
As the algorithm converges, this bias is removed. In the same estimation of beamforming coefficients in order to cope with
way, the unbiased phase estimate of g5,9 and g6,9 shows an higher frequency offsets. As this approach is shown to lead
error standard deviation decreasing with iterations, until it to possible significant degradations, we finally propose and
J. P. Millerioux et al. 5
10−1
10−2
BER
10−3
10−4
10−5
0 0.5 1 1.5 2 2.5 3 3.5 4
Eb /N0 (dB)
No MUD PIC 3
PIC 1 Reference
PIC 2
(a)
Modulus estimate of g5,9 and g6,9 Phase estimate of g5,9 and g6,9
12
0.4
Error standard deviation (◦ )
10
Normalized bias ()
0.3
8
0.2
0.1 6
0
4
−0.1
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
evaluate in Section 4.3 different schemes based on a single- define in a multiuser context, we choose here to evaluate a
user frequency estimator. mean case. We model carrier phases ϕk (t) as
Notice the following:
ϕk (t) = ϕk + Δ fk t, (13)
(i) we possibly consider the use of pilot symbols dis- for all k, where the ϕk are independent and uniformly dis-
tributed within the burst (which is not possible while tributed on [0, 2π], and the Δ fk T follow independent zero-
strictly following the DVB-RCS standard); mean Gaussian distributions with standard deviation σΔ f T .
(ii) all numerical values of frequency offsets are given for No change is performed on the algorithm, which assumes
a burst of 462 symbols (430 information symbols and time-invariant phases, but pilot symbols are set in the mid-
32 pilot symbols). dle of the bursts (to avoid too biased initial phase estimates).
Other simulation parameters are those of Section 3.3.
Some results in term of degradation with respect to
4.1. Algorithm sensitivity to reduced frequency offsets single-user reference to reach the target BER are shown in
Figure 5. Notice that the BER is independent of the sym-
We evaluate in this section the algorithm sensitivity to re- bol locations in the burst due to the use of interleavers. The
duced frequency offsets. As a worst case (which is the clas- algorithm appears maintainable with σΔ f T = 10−4 , but the
sical approach for single-user phase recovery) is difficult to degradations with σΔ f T = 2 · 10−4 are very large.
6 EURASIP Journal on Wireless Communications and Networking
1
section a classical block processing: the channel is no more
considered invariant on the whole burst, but is considered
invariant on windows of reduced length. The algorithm is
modified in this way: channel estimation (7), which includes
0.5 carrier phase estimations, is performed on reduced windows.
Interference cancellation and phase compensation (8) is then
performed on each window using the corresponding esti-
mated coefficients gk,l .
0 Channel estimation sensitivity to frequency offsets de-
0 1 1.5 1.75
creases when the length of estimation windows decreases, be-
Standard deviation of 104 ·Δ f ·T cause the constellation rotations on a window are reduced.
Single user PIC 2 cells 5 and 6 However, sensitivity to additive noise increases when the
PIC 2 cells 4 and 7 PIC 3 cells 5 and 6 length of estimation windows decreases, because noise is av-
PIC 3 cells 4 and 7 eraged on shorter windows. The optimal length of estimation
windows then results from a tradeoff between frequency off-
Figure 5: Degradation with frequency offsets. sets and noise.
We evaluate in this section the effect of reduced estima-
tion windows without frequency offsets. Pilot symbols for
initialization are uniformly distributed on the burst. Some
1.5 results in term of degradation are shown in Figure 6. The
degradation increases when the length of windows decreases.
This is partially due to the fact that CRB for estimation of gk,l
increase while the length of windows decreases, leading to a
less-efficient interference cancellation and phase compensa-
Degradation (dB)
1
tion in (8). However, the degradation is much more impor-
tant for cells 5 and 6 than for cells 4 and 7, whereas the CRB
for channel estimation are equal in both cases (as we have
|g5,2 | = |g5,6 | = |g5,9 | = |g5,8 | = |g5,4 | = |g5,1 | = |g4,1 | =
0.5 |g4,5 | = |g4,8 |). In fact, it can be seen in Figure 7 that similarly
to single-user phase estimation, our channel estimator takes
down from the CRB with short estimation windows and low
SNR. It appears much more critical for cells 5 and 6 than for
0 cells 4 and 7, as the least square estimation is performed on
32 64 128 256 462 7 (6 + 1) coefficients in the first case, and only 4 (3 + 1) in
Length of windows for estimation (symbol) the second case. This effect also appears for longer channel
PIC 2 cells 4 and 7 PIC 2 cells 5 and 6 estimation windows, but it is less obvious to see it.
PIC 3 cells 4 and 7 PIC 3 cells 5 and 6 Notice that in order to optimize the length of windows
for a given σΔ f T , we would consequently have to consider dif-
Figure 6: Degradation with reduced estimation windows. ferent lengths of windows for the different cells: the optimal
length would be shorter for cells 4 and 7 than for cells 5 and
6.
The main conclusion is that the use of reduced estima-
tion windows to cope with higher frequency deviations can
By comparing the degradations in single-user and mul- lead to a significant loss (let us recall that evaluations have
tiuser cases, we can see that they are similar for σΔ f T = 10−4 been performed in this section without frequency offsets),
and for σΔ f T = 0 (i.e., without frequency offsets). We can particularly for cells with a high number of interferers.
conclude that the degradation in the multiuser case with
σΔ f T = 10−4 is mainly due to imperfect user phase recovery.
Beyond σΔ f T = 10−4 , it can be observed that the degradation 4.3. Approach with single-user frequency estimations
in the multiuser case increases more quickly than the degra-
dation in the single-user case: interference cancellation effi- As the previous approach does not appear sufficient to cope
ciency is limited. The considered algorithm is consequently with higher frequency offsets without a significant degrada-
limited to about σΔ f T = 10−4 for a burst length equal to 462 tion, we study in this section another approach. It is based on
symbols. the use of single-user frequency estimations.
J. P. Millerioux et al. 7
40 40
Phase error standard deviation (◦ )
30 30
25 25
20 20
15 15
10 10
5 5
2 2.5 3 3.5 4 4.5 5 2 2.5 3 3.5 4 4.5 5
Eb /N0 (dB) Eb /N0 (dB)
Figure 7: Channel estimation errors for different coefficients and lengths of window.
b y y n
Windows for channel estimation
c up to IT n n y
Pilot symbols
c beyond IT — y n Information symbols
(a) (b)
Figure 8: Approach with frequency estimations: (a) operations performed, (b) distributions of pilot symbols.
k (t) ∗ s(−t)) exp( j2πΔ fk t) are potentially very close following sections.
(ideally identical if Δ f
k = Δ fk ), it is then possible to keep
large estimation windows to perform estimation in (7): us- 4.3.2. Architectures with single user
ing the whole burst allows obtaining the minimum degra- frequency estimations
dation. Clearly, this approach requires “accurate” single-user
frequency estimations, which become the hard task. Two modes are considered for single-user frequency esti-
A first possibility is to use initial frequency estimations mation: the pilot aided mode (PA), based on pilot sym-
before interference cancellation. In this case, the estimation bols, and the decision directed mode (DD), based on symbol
8 EURASIP Journal on Wireless Communications and Networking
10−4
10−5
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
No MUD PIC 3 No MUD
PIC 1 Reference
PIC 2
(a) (b)
Modulus estimates of g5,9 and g6,9 Phase estimates of g5,9 and g6,9
12
0.4
Error standard deviation (◦ )
10
Normalized bias ()
0.3
8
0.2
0.1 6
0
4
−0.1
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
decisions. For the PA mode, pilot symbols are distributed The CRB considered for frequency estimation in DD
within the burst into 3 blocks (see Figure 8(b), cases a and mode is the single-user frequency modified CRB [9], given
b). We follow the approach of [10]. First, a mean phase by
is computed on each block of pilot symbols. Then, a least
square estimation based on these mean phases is used to −1
estimate the frequency. For the DD mode, the principle 3 Es
CRB Δ fk T = . (14)
is the same: the burst is divided into adjacent blocks, on 2π 2 N 3 N0
which mean phases are computed using symbol decisions.
For the DD mode, frequency estimations are performed For PA frequency estimation, the CRB is different from (14)
after interference cancellation, that is, Δ f
k(m) are used to with N replaced by the number of pilot symbols (because
obtain g
k(m+1) . pilot symbols are not consecutive).
J. P. Millerioux et al. 9
10−4
10−5
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
No MUD PIC 3 No MUD PIC 2
PIC 1 Reference PIC 1 CRB
PIC 2
(a) (b)
Modulus estimates of g5,9 and g6,9 Phase estimates of g5,9 and g6,9
12
0.4
Error standard deviation (◦ )
10
Normalized bias ()
0.3
8
0.2
0.1 6
0
4
−0.1
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
10−4
10−5
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
No MUD PIC 3 PIC 1
PIC 1 PIC 4 PIC 2
PIC 2 Reference CRB
(a) (b)
Modulus estimates of g5,9 and g6,9 Phase estimates of g5,9 and g6,9
12
0.4
Error standard deviation (◦ )
10
Normalized bias ()
0.3
8
0.2
0.1 6
0
4
−0.1
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
PIC 3 PIC 3
PIC 4 PIC 4
CRB
(c) (d)
4.3.3. Results with σΔ f T = 2 · 10−4 In case b (Figure 10), DD frequency reestimations allow
to get a frequency error standard deviation close to the CRB.
We first consider in this section a target σΔ f T equal to 2 · 10−4 . Hence, the phase estimate error standard deviation of g5,9
Some results are given in Figures 9, 10, and 11 (with and g6,9 is much closer to the CRB than in case a. The BER
IT = 2) for cells 5 and 6. degradation is the same as that in the case without frequency
In case a (Figure 9), after initial frequency estima- offsets in Section 3.3.
tion, the frequency error standard deviation is about 10−4 . In case c (Figure 11), interference cancellation is efficient
Iterative interference cancellation works, but leads to a but converges slower than in cases a and b. Four iterations
degradation in term of BER, as in Section 4.1. The er- are necessary in case c to get the BER reached with three iter-
ror standard deviation on the phase of g5,9 and g6,9 is far ations in case b.
from the CRB, clearly because of imperfect frequency esti- With σΔ f T = 2 · 10−4 , the most efficient architecture is
mates. consequently architecture b. However, if architecture c leads
J. P. Millerioux et al. 11
10−4
10−5
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
No MUD PIC 4 PIC 3
PIC 1 PIC 5 PIC 4
PIC 2 Reference CRB
PIC 3
(a) (b)
Modulus estimates of g5,9 and g6,9 Phase estimates of g5,9 and g6,9
12
0.4
Error standard deviation (◦ )
10
Normalized bias ()
0.3
8
0.2
0.1 6
0
4
−0.1
2 2.5 3 3.5 4 2 2.5 3 3.5 4
Eb /N0 (dB) Eb /N0 (dB)
PIC 4 PIC 4
PIC 5 PIC 5
CRB
(c) (d)
to a slower convergence of the algorithm, a significant advan- appears to work. After optimization, we use IT = 3 with
tage is that it appears more suited to high-frequency offsets, window lengths for gk estimation from 60 to 100 symbols
as we will see in the following section. (depending on the number of interferers of the considered
cell, Section 4.2). Some results are given in Figure 12. For
Eb /N0 equal to 3.2 dB, the block processing approach allows
4.3.4. Results with σΔ f T = 5 · 10−4 obtaining a BER equal to about 8 · 10−3 at iteration 3, which is
sufficient to obtain reliable frequency estimates at the follow-
We now consider a target σΔ f T equal to 5 · 10−4 . ing iterations. The degradation in terms of BER at iteration 5
For this range of frequency deviations, it is very difficult is then similar to the case without frequency offsets.
to obtain reliable initial frequency estimates without a huge Finally, notice that we have considered average BER along
number of pilot symbols. On the contrary, architecture c the paper. Actually, this average BER can hide some complete
12 EURASIP Journal on Wireless Communications and Networking
ACKNOWLEDGMENT
Research Article
A QoS Architecture for DVB-RCS Next Generation
Satellite Networks
The standardization of a return channel via satellite (DVB-RCS) and satellite community efforts in term of interoperability over
the last few years leads to quite a positive outcome: geostationary satellite networks are intended to provide broadband access to
interactive multimedia services in low-infrastructure areas. However, in order to take in account real-time multimedia traffic, an
efficient resource management scheme is still necessary to maximize the scarce uplink capacities usage. To address this capacity
issue, this paper proposes a complete DVB-RCS QoS architecture that is implemented, thanks to an emulation platform, and
evaluated with real multimedia applications. This paper first gives an overview of the QoS architecture usually used in DVB-
S/RCS satellite system, especially in layers 2 and 3. It then introduces the satellite system emulation used in the experimentation
and its calibration. The main contribution of this work focuses on the signaling principle designed to allow applications to take
benefit from the QoS features of the satellite system even if they are non-QoS aware. It is then shown how signaling in such QoS
architecture allows the user to change dynamically the QoS of his application using QoS agent and QoS server applications even
if the application is not QoS-aware. It is also given quantitative results related to such a dynamic QoS change in the experiments
done on the satellite emulation system.
Copyright © 2007 T. Gayraud and P. Berthou. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
based on either a regenerative or transparent satellite that will network. The end-to-end delay decreases to only one
be the future of satellite networks. single hop.
This paper proceeds in the following way. Section 2 gives
Furthermore, DVB-RCS requires a medium access con-
an overview of new trends in next generation satellite sys-
trol (MAC) protocol because satellite terminals (ST) are able
tems and sums up the principle of the DVB-RCS standard.
to simultaneously access the return channel capacity. The
Section 3 describes the features of our QoS architecture. The
standard method relies on a multifrequency time division
QoS signalling principle is explained here. Then, Section 4
multiple access (MF-TDMA). It basically relies on the avail-
shows our satellite emulation platform and an evaluation of
ability of several TDMA channels (corresponding to different
the new services provided by the QoS architecture, demon-
carrier frequencies), each subdivided into frames and further
strating especially the dynamic QoS change features.
into timeslots of fixed length (bursts) during which the STs
are able to transmit data through MPEG2-TS or ATM traffic
2. NEW TRENDS IN SATELLITE SYSTEMS bursts.
The entity responsible for this timeslot allocation within
2.1. Forward link the superframe shared by competing STs is the NCC (network
control center) that centralizes the satellite resources man-
The first DVB norm described a transmission scheme based agement. Thus it periodically broadcasts a signaling frame,
on MPEG-2 (Motion Picture Expert Group) video compres- the TBTP (terminal burst time plan) that contains the infor-
sion and transmission schemes, using MPEG-TS (MPEG- mation on which STs relies to know when to transmit their
transport stream). This latter was adapted for satellite sys- bursts.
tems through DVB-S (DVB transmission via satellite) that This allocation can be dynamically modified by STs re-
defines series of options to send MPEG-TS packets over satel- quests so as to prevent from wasting satellite resources that
lite links and that is currently used for digital TV. The suc- would be otherwise statically allocated. The implementation
cess of this standard has caused its adoption for Internet ser- of such a mechanism is generally known as bandwidth on-
vices over satellite. Then, the encapsulation of IP over MPE demand (BoD) algorithm.
(multiprotocol encapsulation) or more recently ULE (ultra
lightweight protocol) is needed. This leads to a complex net-
work stack. DVB-S2 standard [2] is intended to be a suc- 2.2.1. Bandwidth on-demand mechanisms
cessor of DVB-S with the same applications (TV, Internet,
etc.). It offers new coding techniques that can increase per- In order to dynamically manage the bandwidth allocation,
formance by 25% over that of DVB-S, but is still compati- a bandwidth on demand protocol called demand assignment
ble with encapsulation layers as MPE or ULE. An alternative multiple access (DAMA) is defined. It relies on the STs ability
known as GS (generic stream) intends to gain direct access to to request frequently “capacities” to the NCC which enables a
the physical layer, avoiding the MPEG2-TS packet overhead, regular update of the TBTP to fit to the STs respective traffic
but this protocol remains a work in progress. load. The latter provides signaling schemes as well as MAC
The satellite terminals could therefore only receive DVB- QoS classes and their mapping on capacity types.
S/S2 frames from the satellite, but did not have the ability to Thus, the norm defines 4 capacity categories to fit the ap-
send any traffic towards the satellite. plications needs:
(i) continuous rate assignment (CRA) which is static ca-
2.2. Return link pacity, not subject to dynamic requests;
In 1999, the ETSI proposed a standard for a return channel (ii) rate-based dynamic capacity (RBDC), which is dy-
via satellite, the DVB-RCS [3, 4], which supplements the STs namic rate capacity (in slots/frame), upper-bounded
with the ability to transmit traffic towards the satellite. by MaxRBDC, granted in response to dynamic re-
According to this basis, two types of satellite can be de- quests from the STs to track their instantaneous traffic
fined. rate;
(iii) (absolute) volume-based dynamic capacity (VBDC and
(i) Transparent satellite simply forwards the signal re- AVDBC), which is also dynamic rate capacity (in slots),
ceived with no additional processing. A gateway (GW) granted in response to dynamic requests from the STs
is needed on the ground to convert DVB-RCS frames to track their traffic queue state;
into DVB-S one. Each communication goes through
the gateway with a “star” topology. The delay to cross (iv) free capacity allocation (FCA), which is assigned to STs
the satellite network is about half a second and a dou- on an “as available” basis from unused capacity.
ble hop (at least 1 second) is needed to connect two Capacity types are vital to return path QoS support at MAC
satellite users. layer; therefore, they are described in detail in the following.
(ii) Regenerative satellite with onboard switching payload Any given ST can be assigned one or a mix of the four capac-
is able to demodulate, process, and remodulate the ity types. Generally, higher priority classes of service are asso-
traffic that goes though it and therefore to multiplex ciated with guaranteed capacity (CRA, RBDC), while lower
several DVB-RCS signals into a single DVB-S one. The priority classes are predominantly given best effort capacity
associated topology could be a “star” or a “meshed” (VBDC, FCA).
T. Gayraud and P. Berthou 3
Even if the service classes are properly defined, the allo- Application
QoS Agent
cation algorithms implemented in the NCC to fulfill the ser- QoS signalling
vices requirements are not specified. IP downstream
QoS Server
from user terminal
MF-classifier IP classes
3. QoS ARCHITECTURE
This section describes the QoS architecture we propose for Traffic shaping/policing /
DVB-S/RCS satellite systems. The main contribution is built
on return link management. Thus the downlink is generally
EF AF BE
considered not to be a bottleneck and classical traffic engi-
IP Layer
neering techniques are enough to managed the network.
Scheduling
EDF EDF
3.1. Basis of QoS in satellite systems
Transmission
To reach an optimal exploitation of uplink resources, at least allowed/denied
PQ
three functions must be implemented to provide QoS guar-
EF AF + BE
antees.
IP DVB/RCS
(i) QoS admission control consists, before the application Segmentation
interface
sends its traffic, to check that the network has enough
resources. This prevents some applications from send-
ing traffic that would otherwise lead to congestion
MAC layer
Threshold
among high priority traffic.
(ii) QoS enforcement consists in checking that the admit-
ted traffic respects its contracts, that is, that it does RT DVB NRT DVB
not use more resources than requested. This is done frames frames
by policing and shaping. Framing DAMA client
(iii) QoS differentiation consists in having several classes of
DVB-RCS frames Capacity
traffic, each class provides different behavior adapted TBTP requests
to a given service. This task is complex and needs dy- To satellite To/from NCC
namic management during the connections lifetime
and must be performed at two layers: the DVB-RCS DAMA server
and IP layers.
Figure 1: QoS architecture.
Thanks to the 5 bandwidth allocation mechanisms in-
cluded in DVB-RCS standard, the traffic differentiation is
made easily in introducing several MAC queues in the ST
stack and mapping the capacity requests over the MAC The user is able to classify its own flows in any available
queues. Then, IP DiffServ-based router architecture can be service through a dedicated agent (the QoS agent) that
setup over these new MAC services. However, this cannot be communicates with the QoS server to deliver the clas-
done without cross-layer mechanisms that ensure perfect re- sification. The goal is to exploit the capabilities offered
sources use. by the IP QoS capabilities.
An overview of this QoS architecture within the ST is given
3.2. Cross-layer architecture in Figure 1.
(i) Satellite terminal resources: is medium access control QoS management at the MAC layer aims at sharing with op-
level, where the DVB-RCS DAMA allocates the band- timal way the global uplink resources among the STs. Thus,
width on a fixed basis for real time applications and on the MAC layer must be able to
demand for other flows (nonreal-time traffic).
(i) provide strict guarantees in terms of delay and jitter;
(ii) Class of service resources: a specific IP level module im-
(ii) preserve these resources through fitting their alloca-
plements a queue management system aiming at pro-
tion to the effective ST traffic load.
viding a differentiated service with regards to three ser-
vice classes. These service classes are deemed to exploit Within the ST MAC layer, the traffic is split into 2 classes
the capabilities offered by the MAC level QoS capabil- of service (CoS), DVB-RT for “real-time service” and DVB-
ities. NRT for “nonreal-time” service, which are associated to
(iii) User level resources: this level is related to the share of 2 different ATM permanent virtual channels (PVC). One
previous services resource between the different users. (DVB-RT) benefits from static resource assignment through
4 EURASIP Journal on Wireless Communications and Networking
CRA; on the contrary, (DVB-NRT) relies on a dynamic re- With reference to IntServ/DiffServ traffic classes, the
source allocation scheme also called BoD algorithm which best-effort (BE) traffic category supports the traditional ser-
will be further detailed in this section. vice offered by the Internet by default without any specific
QoS measure and whose performance are strongly impacted
(i) Real-time queue: the CRA consists in a fixed capacity
by network congestion states. Real-time IP data category in-
that is set at the ST log-on and is not subject to renego-
cludes both IntServ guaranteed service class and DiffServ ex-
tiation during the ST connection lifetime. Each super-
pedited forwarding (EF) PHB (per-hop behavior) while the
frame contains one or more slots assigned to this con-
nonreal-time IP category is used for IntServ controlled load
nection. This reserved static rate is entirely dedicated
service class and DiffServ assured forwarding (AF) PHB [6].
to the DVB-RT traffic, since its high delay sensitive re-
quirements hardly tolerates throughput fluctuations.
(ii) Nonreal-time queue: the request category retained for 3.4.1. QoS enforcement
DVB-NRT traffic class is VBDC and FCA. Delay and
jitter tolerant traffic is supplied by the MAC scheduler The fundamental component of the architecture is the
to the DAMA controller that computes the adequate EDF scheduler preceded by token buckets (RC-EDF, rate-
dynamic volume to request to the NCC. These requests controlled earliest deadline first) which allows fixing and up-
are sent out of band, not in traffic slots assignments, per bound to queuing delay and a minimum bandwidth for
but signaled in each SYNC slots broadcasted periodi- separate IP flows. Namely, the presence of token buckets is
cally by the NCC. a guarantee that each IP flow will receive a minimum band-
This architecture uses an original DAMA protocol that aims width, given sufficient demand, equal to the relevant token
to reduce the allocation delays without reducing the network rate, while the EDF scheduler will guarantee to each packet
use. of an IP flow, once suitably regulated by a token bucket to be
As soon as an application produces a data, a free slot in served within a deadline equal to its associated static param-
the next super frame should be available to send it. However, eter.
the allocation done with the DAMA protocol takes at least In Figure 1, the RC-EDF components are gathered under
600 milliseconds (minimum scheduling latency—MSL). To the appellation “traffic shaping/policing.” The traffic polic-
reduce its impact on the end-to-end delay, application needs ing and shaping are then realized, thanks to single-rate token
are anticipated in monitoring the DVB-NRT queue length buckets.
that grows conjointly. If the queue grows, the requested ca-
pacity is increased by a factor α otherwise only the minimum 3.4.2. Layer 3/layer 2 mapping
is requested as. This algorithm is detailed in [5]. As shown
later, by properly setting α, the latency introduced by the BoD The 3 traffic categories are served by a scheduler using a sim-
algorithm can be effectively reduced. ple priority queuing (PQ) discipline. This means that
In addition to the VDBC requests, the MAC scheduler in
the NCC distributes extra capacities to the logged STs if the (i) packets from NRT queues are served only when RT
network is not congested. This last capacity category (FCA) queues get completely emptied,
comes out to enhance the ST performance especially on low (ii) packets in the BE queue are extracted only when RT
loading conditions, preventing the ST from waiting at least and NRT queues are empty.
for the MSL to be able to transmit.
3.4.3. Application mapping
3.4. QoS at IP layer
The EF traffic includes a number of real-time applications
In order to achieve a complete traffic control framework, a with stringent time and bandwidth requirements such as
classifier separates IP traffic into 3 categories: telephony or video conferencing. IP signaling which has very
(i) real-time: such an IP flow should be guaranteed a min- stringent delay requirements but which is characterized by
imum bandwidth, an upper bound on queuing delay, low-data rates should use this service class too.
a mean queuing delay of a few tens of milliseconds; The AF traffic should include a number of traditional
(ii) nonreal-time: such an IP flow should be guaranteed a Internet applications to be served with a satisfactory level
minimum bandwidth, a mean queuing delay of a few of service and transported over TCP. They include telnet,
hundreds of milliseconds; HTTP, SMTP, FTP. Such applications can greatly vary in
terms of bandwidth and delay requirements. This means that
(iii) best-effort: all IP packets not recognized as belonging
applications such as telnet or HTTP should be assured small
to a particular IP flow are treated without any guaran-
queuing delay though with limited bandwidth.
tee on bandwidth or delay.
The BE class is designed to manage all traffic which is
The classifier then maps the packets to the 2 MAC categories. not recognized as belonging to a particular user entitled to
The overall goal of the architecture is to enforce the con- receive better QoS or to applications with no particular delay
straints for the IP categories as defined above while maxi- or bandwidth requirement. SMTP or FTP should belong to
mizing utilization of the available time-varying capacity. this class.
T. Gayraud and P. Berthou 5
Satellite
IP
Hub Router
ST
IPv4 , IPv6 NCC
ST: satellite terminal QoS
MAC: medium access control MAC
segregation
NCC: network control center
4. EXPERIMENTAL MEASUREMENTS The network elements that belong to a classical satellite net-
work (Satellite, NCC, STs) are emulated individually on a
4.1. Emulation principle dedicated computer. A gigabit Ethernet interconnects them
and emulates the satellite carrier emulation. Ethernet was
Evaluating performances over real data links or networks chosen for its native broadcast abilities and also for its high
is expensive, even impossible for systems in development bandwidth capacities. Each satellite channel is mapped on a
phase. single Ethernet multicast address.
6 EURASIP Journal on Wireless Communications and Networking
4.2.1. Satellite link emulation Table 1: Basic physical and MAC layer configuration.
The satellite link emulator (SLE) simulates satellite link char- ST information peak rate 2048 Kbps
acteristics in term of delay (and distribution); bit-error rate Physical Superframe 10 Trames
(error burst frequency distribution, error burst length dis- layer Frame duration 50 ms
tribution), computed according to precalculated distribution Global DVB-RCS resource 2048 Kbps
and based on real measurements. Basic FCA None
Each channel crosses the link emulator to simulate the DAMA α 1
effects of the two-way satellite link in real-time. The packets
CRA 96 Kbps
sent from an ST to the SLE are delayed and are also subject to SLA ST
a sequence of bit errors at random positions before they are VBDC [FCA;1760 Kbps]
forwarded to the emulated “downlink” (a multicast address
per spot).
pacity segmentation scheme. The satellite emulator delay is
4.2.2. Network control center set to 250 milliseconds, the jitter is equal to ±1 milliseconds,
The NCC is the core of the satellite network management. It and the loss pattern is typical of a nice weather. Please note
deals with allocating radio resources to the STs according to that last notion which could sound subjective corresponds,
their subscriber profile and available satellite resource. It cer- in the satellite emulator, to real satellite measurement traces.
tainly implements a DAMA controller, but provides also an
address resolution protocol to map IP addresses over under- 4.3.2. MAC layer
lying protocols and a QoS admission mechanism.
The main parameters are closely linked to resource sharing
assignment from the NCC that distinguished two CoS at the
4.2.3. Satellite terminals
MAC layer in the ST. The ST maximum transmission rate is
The satellite terminals are based on Linux systems. They act shared by CRA and VBDC. Therefore, the peak transmission
as an access router interconnecting a LAN to an Internet ser- rate is defined at first, then the CRA amount and finally the
vice provider over a satellite link. Its DVB-S/DVB-RCS in- DAMA configuration through the α anticipation parameter
terfaces allow the data emission and reception and it imple- and the FCA threshold (Table 1). The MAC queue sizes have
ments the corresponding network layers. The proposed QoS to respect minimum thresholds so as to prevent congestion
architecture is mainly located in the terminals. from occurring in the MAC Layer.
The accuracy of the ST implementation is close to a pro-
totype version and makes the emulation very realistic but 4.3.3. IP layer
also critical for its configuration. The calibration of the ser-
vices given to the users has been a touchy part of this work. Considering that EF and AF services are implemented strictly
according to the single-rate token buckets and that there is no
4.3. Platform calibration BE service conditioning, the main parameters of traffic con-
ditioning blocks (TCB) and IP scheduling are summarized in
As we explained in the former section, the available platform Table 2.
will provide us with a prototype as close as possible to a real
DVB-S/RCS system behavior. So the quantitative results per- 4.4. Measurements
formed on it are rather significant. To reach this result, the
platform has to be calibrated, that means that the right pa- The measurements given in this section aims at evaluating
rameters have to be set to the right value. If this stage is not the dynamic QoS change mechanism. First, the experimenta-
properly achieved, then the results offer no interest, even if tion done on DAMA results that have already been presented
all the parts of the emulation testbed are very accurate. So, in in [5] are used to prove the right calibration of the emula-
this section, the basic platform configuration chosen in order tion testbed, so that the result obtained further is realistic.
to carry out our experimental measurements is detailed. The second part of the measurements has been performed on
This configuration stands for the reference scenario from a multiflow scenario involving several applications. The ob-
which all the calibration adjustments are done in order to tained figures and tables show that the QoS architecture im-
ensure the significance of platform performance. Through plementing the QoS server allows the user to change dynam-
this nonexhaustive list of the main platform configuration ically the QoS of one application thanks to the QoS agent.
parameters, we emphasize the huge possibility to customize
the platform which still remains, even for simplified current 4.4.1. Impact of FCA and α
commercial DVB-RCS deployments, a vital and difficult task.
The following study is linked to a VBR traffic source: a DIVX
4.3.1. Physical layer streaming session. The DAMA influence cannot be neglected
in these experiments when the throughput variations can be
The satellite diffusion properties are configured through the absorbed by the DAMA algorithm. Thus the basic DAMA
SE (delay, Jitter and Losses patterns) and the return link ca- performances can be enhanced by increasing the anticipation
T. Gayraud and P. Berthou 7
100 100
90 90
80 80
70 70
60 60
CDF (%)
CDF (%)
50 50
40 40
30 30
20 20
10 10
0 0
0 0.5 1 1.5 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Delay (s) Delay (s)
α=0 α=0
α = 0.5 α = 0.5
α=1 α=1
Figure 4: DAMA impact (VBR flow without FCA). Figure 5: DAMA impact on a VBR flow with FCA.
factor α which enables to bring 50% of the packets delay un- 800
der 1 seconds (Figures 4 and 5).
α=1
Under 0.5, the anticipation factor does not improve the 700
Queuing delay (ms)
4.4.2. Global architecture under different loading throughput and protects the AF traffic from losses in con-
conditions gestion state. Finally, the BE service is the most affected by
congestion while the satellite overload implies less capacities
The purposes of different tests are to measure the SATIP6 for overall BE traffic and therefore higher delays and losses.
QoS performances under heterogeneous traffic flows (GSM In Table 3, we can notice that voice service is not affected
VoIP sessions, DIVX VoD, FTP, and web browsing) which by the loading conditions when the delay experimented by
are mapped onto the three different SATIP6 services. As seen FTP and BE traffic increases. Inside the DVB-NRT traffic
previously, the voice service offers strict guarantees in terms class, FTP traffic is protected from losses at the expense of
of delay and jitter. The AF service (FTP) ensures a minimum the BE class delay and loss ratio in congestion states.
8 EURASIP Journal on Wireless Communications and Networking
800
when it is downgraded, still longer in these two cases than in
600
the simple addition of a flow.
400 These results conclude the section dedicated to experi-
0 20000 40000 60000 80000 100000 120000 mental measurements by putting the stress on the proper
Time (ms) traffic differentiation carried out by our QoS architecture and
Throughput (kbits/s) the ability to change from one class of service to the other one
thanks to the QoS agent GUI.
Figure 7: Dynamic change of class of service using the QoS agent.
5. CONCLUSION
Table 3: SATIP6 QoS performance under different loading condi-
tions. It was proved in this paper that it was possible to specify and
implement QoS architecture for DVB-S/RCS satellite system
Network Average delay (Jitter) [ms] Losses [%] in order to provide the user with QoS guarantees even if a
load [%] BE FTP Voice BE FTP Voice non-QoS-aware application is used. MAC algorithms (such
25 293 (26) 293 (26) 283 (23) 0 0 0 as DAMA) were proved to be efficient. It was also explained
50 291 (23) 291 (24) 283 (23) 0 0 0 how to proceed to evaluate such an architecture. Using mea-
75 290 (24) 289 (24) 283 (23) 0 0 0 surement tools on a well-calibrated testbed, the global QoS
100 919 (23) 948 (24) 283 (23) 0 0 0 of the satellite system may be evaluated accurately.
125 6753 (23) 1783 (24) 283 (23) 33 0 0
Using other capacity category than VBDC (RBDC for in-
stance) is improving resource utilization especially if appli-
150 6755 (28) 1783 (24) 283 (23) 37 0 0
cations throughputs are known. Unfortunately, this is not
usual today in the Internet. In that case, we propose to use
rate-based signaling protocols (SIP) in order to set up the
right capacity requests.
4.4.3. Dynamic change of QoS
Other future work may also be done related to DVB-S2
The change of multimedia stream QoS is done thanks to QoS and new admission control mechanisms.
agent.
The different following steps are easy to find in Figure 7. ACKNOWLEDGMENTS
(i) Step 1: the scenario begins as a UDP video stream The authors wish to thank all the partners of the SATIP6
starts. This stream is sent in the BE class of service. [8] consortium: Alcatel Space (France), which is the project
(ii) Step 2: 25 seconds later, another flow is sent on the coordinator, Telecom Italia Lab (Italy), France Telecom SA
same uplink; it is done so that the uplink is now con- (France), University of Rome “La Sapienza” (Italy), Sintef
gested. The throughput of the first stream is then re- (Norway), LAAS-CNRS (France), and Alliance Qualité Logi-
duced to 800 kbps. ciel (France).
(iii) Step 3: 25 seconds later, the user decides to upgrade the
class of service of this stream and set it up to “voice.”
After traffic burst, due to the addition of the EF service REFERENCES
of 1 Mbps, and the traffic buffered before the resource [1] ETSI TR 102 157, “Satellite Earth Stations and Systems (SES);
reservation, the throughput is around 1 Mbps. Broadband Satellite Multimedia; IP Internetworking over satel-
(iv) Step 4: at t = 80 s, the stream is downgraded back to BE lite; Performance, Availability and Quality of Service,” July
and the stream throughput is then around 800 kbps. 2003.
[2] ETSI Standard TR 102 376 V1.1.1, “Digital Video Broadcasting
In Figure 7, the time needed to change from one Cos to an- (DVB); User guidelines for the second generation system for
other one for a flow could also be evaluated when the link is Broadcasting, Interactive Services, News Gathering and other
congested by other data flows. broadband satellite applications (DVB-S2)”.
The delays given in Table 4 show as usually that the delay [3] ETSI EN 301 790 V1.3.1, “Digital Video Broadcasting (DVB);
is around 3 seconds. It is less when a new flow is added on a Interaction channel for Satellite Distribution Systems,” March
link. It is longer when the QoS of a flow is upgraded and less 2003.
T. Gayraud and P. Berthou 9
Research Article
Maximum Likelihood Timing and Carrier Synchronization in
Burst-Mode Satellite Transmissions
This paper investigates the joint maximum likelihood (ML) estimation of the carrier frequency offset, timing error, and carrier
phase in burst-mode satellite transmissions over an AWGN channel. The synchronization process is assisted by a training sequence
appended in front of each burst and composed of alternating binary symbols. The use of this particular pilot pattern results into
an estimation algorithm of affordable complexity that operates in a decoupled fashion. In particular, the frequency offset is mea-
sured first and independently of the other parameters. Timing and phase estimates are subsequently computed through simple
closed-form expressions. The performance of the proposed scheme is investigated by computer simulation and compared with
Cramer-Rao bounds. It turns out that the estimation accuracy is very close to the theoretical limits up to relatively low signal-to-
noise ratios. This makes the algorithm well suited for turbo-coded transmissions operating near the Shannon limit.
Copyright © 2007 M. Morelli and A. A. D’Amico. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
that they can operate correctly even in the presence of carrier quency, timing, and phase estimates. Simulation results are
frequency offsets (CFOs) as large as 20% of the symbol rate. presented in Section 5 while some conclusions are drawn in
Frequency estimation is usually performed by exploiting Section 6.
the received time-synchronized samples. A large number of
schemes proposed in the past operate in either the frequency 2. SYSTEM MODEL AND PROBLEM FORMULATION
or time domain. The Rife and Boorstyn (R&B) algorithm
[10] belongs to the former class and provides maximum like- 2.1. Statement of the problem
lihood (ML) estimates of frequency and phase errors by look-
We consider the reverse link of a satellite communication
ing for the peak of a periodogram. Interpolation techniques
system and assume a time-division multiple-access (TDMA)
may be employed to find an explicit expression of the peak
scheme where each earth station transmits bursts of data. The
location [11]. In the time-domain approach, suitable corre-
structure of a burst is detailed in Figure 1. Essentially, it con-
lations of the received samples are exploited to compute the
sists of two parts: a header section followed by a payload. The
frequency estimates. A representative selection of schemes
header is further divided in two portions, namely, a synchro-
derived along this line of reasoning can be found in [12–15].
nization preamble and a unique word (UW). The preamble is
These methods attain the Cramer-Rao lower bound (CRB) at
made of a sequence of training symbols which are exploited
intermediate/high SNRs, but exhibit different performance
by the receiver for carrier and symbol timing recovery. The
in terms of estimation range and threshold, that is, the SNR
UW is located just after the preamble and is used for burst
below which large estimation errors are likely to occur.
identification as well as to establish the start of the payload.
A possible drawback of conventional frequency estima- The first task of the receiver is the start of burst (SoB) de-
tion schemes as those discussed in [10, 12–15] is that they all tection, that is, the recognition of the time-of-arrival (ToA)
assume ideal timing synchronization. Their performance is of a generic burst. This is normally performed through a sim-
thus limited by the accuracy of the timing estimator. A DA ple noncoherent energy-detection scheme which provides a
algorithm for the joint estimation of the carrier phase, fre- coarse estimate of the position of each burst. Once the SoB
quency offset, and timing error has been proposed in [16] by has been identified, the preamble is exploited for carrier and
resorting to ML arguments. In order to work properly, how- symbol timing synchronization. This is the second task of the
ever, the demodulated signal must incur negligible phase ro- receiver and represents the focus of our paper. In order to ex-
tations during the preamble duration. This poses a stringent plain how synchronization can be achieved, we concentrate
limit to the maximum tolerable CFO, which may prevent the on a single burst and assume that the SoB detection algo-
application of this method to many practical situations. rithm has provided a ToA estimate with an error τ, as shown
In the present paper we are concerned with the joint esti- in Figure 2. The offset τ can be decomposed as follows:
mation of all synchronization parameters for a burst-mode
satellite system operating over AWGN channels. Since one τ = ηT − εT, (1)
distinct feature of packetized transmissions is that synchro-
nization must be achieved as fast as possible, in the follow- where T is the symbol period, η is an integer (integer delay),
ing we only focus our attention to a feedforward structure. and ε ∈ [−0.5, 0.5) is a real-valued parameter (fractional de-
Also, we assume that a preamble of alternating binary sym- lay). During the preamble we are interested in the estimation
bols is transmitted at the beginning of each burst to facili- of the fractional delay, because the integer delay is recovered
tate the timing estimation task [17]. Our approach is based later by searching for the location of the UW within the burst.
on ML methods and leads to a three-step procedure. In The estimation of the synchronization parameters (fractional
the first step frequency recovery is accomplished through a delay, carrier phase, and frequency offsets) is performed by
mono-dimensional grid search. The estimated CFO is then observing a portion of the preamble of length NT (N is a de-
exploited in the second step to obtain a closed-form expres- sign parameter) at the right of the assumed SoB, as shown in
sion of the timing estimate. The final step is devoted to phase Figure 2. Clearly, the total duration of the preamble has to be
estimation and can be skipped in case of differential data larger than τ + NT. Since τ is a random variable, this condi-
detection. Surprisingly, no complicated multidimensional tion can be practically met by a proper design of the preamble
searches are needed to jointly estimate all the unknown syn- length. Since we are not concerned with the estimation of the
chronization parameters. Simulations indicate that the pro- integer delay, in the following we set η = 0.
posed estimator is well suited for turbo-coded transmissions
since its accuracy approaches the relevant CRBs even at low 2.2. Signal model
SNR values. However, it should be observed that this advan-
tage is achieved at the price of a higher computational com- We consider a linearly modulated digital signal transmitted
plexity as compared to other existing alternatives. over an AWGN channel. The complex envelope of the re-
ceived waveform is modeled as
The paper is organized as follows. In Section 2 we in-
troduce the signal model and formulate the synchronization r(t) = e j(2π fd t+ϕ) s(t − εT) + w(t), (2)
problem. Section 3 illustrates the joint ML estimation of the
unknown parameters and discusses in some detail the prac- where s(t) is the useful signal, ϕ and fd are the carrier phase
tical implementation of the frequency estimator. In Section 4 and frequency offset, respectively, and w(t) is thermal noise
we derive CRBs to characterize the ultimate accuracy of fre- with independent real and imaginary components, each with
M. Morelli and A. A. D’Amico 3
Guard Sync.
interval preamble UW
Header Payload
with 1.2
−1
N Es /N0 = 10 dB
Ye ( ν) = (−1)k x(2k)e− j2πk ν , 1 N = 64
k=0
−1
N
(10)
ν
− j2πk
0.8
Yo ( ν) = k
(−1) x(2k + 1)e .
k=0
P(ν)
0.6
To ease the search for the maximum of Ψ( ν, ε , ϕ),
we rewrite
(9) in the form
0.4
= Z(
Ψ( ν, ε , ϕ) ν, ε ) cos ψ( ν, ε ) − ϕ , (11)
0.2
where Z( ν, ε ) is a function of ν and ε defined as
P( ν) as obtained by simulation with N = 64, Es /N0 = 10 dB, of N and Es /N0 . For this reason, it is of interest to derive
and ν = 0.1. As discussed in [10], the global maximum can the CRB for the joint estimation of the set of parameters
be sought in two steps. The first one (coarse search) calculates η = (ν, ε, ϕ) based on the model (6).
P( ν) for a set of ν values, say { νn }, covering the uncertainty We begin by computing the entries of the Fisher infor-
range of ν and determines the location νM of the maximum mation matrix F. They are defined as [20]
over this set. The second step (fine search) makes an inter-
polation between the samples P( νn ) and computes the local ∂2 Λ(η)
[F]i, j = −E , 1 ≤ i, j ≤ 3, (25)
maximum nearest to νM . It should be noted that the shape ∂ηi ∂η j
of P( ν) is occasionally so badly distorted by noise that its
highest peak may be far from the true ν. When this happens, where Λ(η) is the log-likelihood function in (7) and η de-
the MLE makes large errors (outliers) and the system perfor- notes the th entry of η. Substituting (7) into (25), after some
mance is highly degraded. The SNR below which the outliers manipulations we obtain
start to occur is referred to as the threshold of the estimator. ⎡ ⎤
π 2 (2N − 1) 4N − 1 − 3 cos(2πε) 0 3π 2N − 1 − cos(2πε)
(3) In practice the coarse search can be efficiently per- N ⎢ ⎥
F= 2 ⎢ 0 6π 2 0 ⎥.
formed using fast Fourier transform (FFT) techniques, as it 6σ ⎣ ⎦
3π 2N − 1 − cos(2πε) 0 6
is now explained. Starting from the observed samples {x(k)},
(26)
we first compute the following zero-padded sequences of
length KN: The CRB for the estimation of η is given by [F−1 ], . Skip-
⎧
⎨(−1)k x(2k), 0 ≤ k ≤ N − 1, ping the details, it is found that
ye (k) = ⎩
−1
0, N ≤ k ≤ NK − 1, 12 Es /N0
⎧ (23) CRB(ν) = , (27)
⎨(−1)k x(2k + 1), π 2 N 4N 2 − 4 + 3 sin2 (2πε)
0 ≤ k ≤ N − 1,
yo (k) = ⎩
−1
0, N ≤ k ≤ KN − 1, Es /N0
CRB(ε) = , (28)
π2N
where K is a design parameter called pruning factor. Next, the
−1
FFTs of { ye (k)} and { yo (k)} are evaluated at the points (2N − 1)(4N − 1 − 3 cos ε) Es /N0
CRB(ϕ) = 2 . (29)
n KN KN N 4N 2 − 4 + 3 sin2 (2πε)
νn =
, − ≤n≤ . (24)
KN 2 2
Interestingly, for large data records we can approximate (27)
This produces the quantities {Ye ( νn )} and {Yo ( νn )}, which as
are next exploited to get {P( νn )} as indicated in (19). Finally,
the largest P( νn ) is sought and this provides the coarse fre- 3(Es /N0 )−1
CRB(ν) ≈ (30)
quency estimate. π2N 3
(4) Collecting (10) and (19), it is seen that P( ν) is peri- which represents the CRB for the estimation of the frequency
odic of unit period. This means that MLE gives unambigu- of a complex sinusoid embedded in AWGN [10].
ous frequency estimates as long as ν is confined within the
interval [−1/2, 1/2). This is the frequency estimation range of
5. SIMULATION RESULTS
MLE.
(5) Compared to the R&B algorithm [10], the MLE is In this section we report on simulation results illustrating the
more complex to implement as it requires the computation performance of MLE over an AWGN channel. Unless other-
of two FFTs instead of a single FFT. In addition to carrier syn- wise specified, the synchronization parameters vary at each
chronization, however, the MLE also provides timing recov- new simulation run and are modeled as statistically indepen-
ery. Actually, from (16) and (21) we see that computing the dent random variables with a uniform distribution. In par-
timing estimate only requires knowledge of Ye (ν) and Yo (ν). ticular, ν and ε are confined within [−0.5, 0.5) while ϕ takes
Since these quantities can easily be obtained by {Ye ( νn )} and values in the interval [−π, π). A pruning factor K = 4 is
{Yo ( νn )} through interpolation, the timing estimation task is used to compute the quantities {Ye ( νn )} and {Yo ( νn )}. Also,
accomplished with a relatively low computational effort once a parabolic interpolation is chosen in the implementation of
ν is available. However, it should be observed that the com-
the fine search. This yields a frequency estimate in the form
plexity associated to the synchronization process is negligible
compared to that of iterative data decoding [19]. So, the re- δ P νM −1 − P νM+1
ν = νM + ν ·
, (31)
quirement for an additional FFT has only a marginal impact 2 P νM −1 − 2P( νM ) + P νM+1
on the overall receiver complexity.
where δν = 1/KN is the distance between two adjacent sam-
4. CRB ANALYSIS ples P( νn ) while νM is the output of the coarse search. The
observation length is fixed to either N = 32 or 64. For
By invoking the asymptotic efficiency property of the MLE, comparison, in the ensuing discussion we also consider a
we expect that the accuracy of the estimates (20)–(22) ap- synchronization scheme in which timing recovery is first ac-
proaches the corresponding CRBs for relatively large values complished by resorting to the Oerder and Meyr (O&M)
6 EURASIP Journal on Wireless Communications and Networking
0.5 10−1
0.4 Es /N0 = 10 dB
N = 64
0.3
E{ν} = ν 10−2
0.2
0.1
E{ν}
MSEε
10−3
−0.1
−0.2
−0.3 10−4
−0.4
N = 32
−0.5
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
10−5
ν 0 5 10 15 20
Es /N0 (dB)
Figure 4: Mean normalized frequency estimates versus ν.
O&M
MLE
0.5 CRB
0.4 Es /N0 = 10 dB
N = 64
Figure 6: MSE performance of MLE and O&M estimator with N =
0.3 32.
E{
ε} = ε
0.2
10−1
0.1
ε}
E{
−0.1
10−2
−0.2
−0.3
MSEε
−0.4 10−3
−0.5
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
ε
10−4
Figure 5: Mean timing phase estimates versus ε.
N = 64
10−5
algorithm [8] and carrier synchronization is next achieved 0 5 10 15 20
by applying the R&B method [10] to the time-synchronized Es /N0 (dB)
samples. The O&M operates with four samples per symbol
period. O&M
MLE
Figures 4 and 5 illustrate average frequency and timing CRB
estimates, E{ν} and E{ ε}, provided by MLE as a function of
ν and ε, respectively. The observation length is N = 64 while
Figure 7: MSE performance of MLE and O&M estimator with N =
the SNR is fixed to Es /N0 = 10 dB. The ideal lines E{ν} = ν 64.
and E{ ε} = ε are indicated as references. These results show
that ν and ε are unbiased over the full range [−0.5, 0.5).
Figures 6 and 7 compare the mean square error (MSE)
of the timing estimates, E{(ε − ε)2 }, as obtained with MLE marks. We see that MLE has the best accuracy, especially at
and O&M. The observation length is N = 32 in Figure 6 low SNRs where a significant gain is observed with respect
and N = 64 in Figure 7. Marks indicate simulation results to O&M. In particular, for N = 64 the MLE is close to the
while the thin solid lines are drawn to ease the reading of the CRB down to Es /N0 values of 0 dB, while O&M approaches
graphs. The corresponding CRBs are also shown as bench- the bound only for Es /N0 > 10 dB. This feature of the MLE
M. Morelli and A. A. D’Amico 7
10−2 100
8
6
4
10−3
2
10−4 N = 32 10−1
8
MSEϕ (rad2 )
6
MSEν
4
10−5
2 N = 32
10−6 10−2
8
6
4
10−7
N = 64 N = 64
2
10−8 10−3
0 5 10 15 20 0 5 10 15 20
Es /N0 (dB) Es /N0 (dB)
R&B R&B
MLE MLE
CRB CRB
Figure 8: MSE performance of MLE and R&B estimator with N = Figure 9: Mean square error MSEϕ versus Es /N0 with N = 32 and
32 and N = 64. N = 64.
the other hand, at intermediate/high SNR values the conven- [17] Y. Jiang, F.-W. Sun, and J. S. Baras, “On the performance limits
tional scheme is preferable as it achieves similar performance of data-aided synchronization,” IEEE Transactions on Informa-
with reduced complexity. tion Theory, vol. 49, no. 1, pp. 191–203, 2003.
[18] H. Meyr, M. Oerder, and A. Polydoros, “On sampling rate,
analog prefiltering, and sufficient statistics for digital re-
ceivers,” IEEE Transactions on Communications, vol. 42, no. 12,
REFERENCES
pp. 3208–3214, 1994.
[19] S. Benedetto, R. Garello, G. Montorsi, et al., “MHOMS: high-
[1] U. Mengali and A. N. D’Andrea, Synchronization Techniques speed ACM modem for satellite applications,” IEEE Wireless
for Digital Receivers, Plenum Press, New York, NY, USA, 1997. Communications, vol. 12, no. 2, pp. 66–77, 2005.
[2] H. Meyr, M. Moeneclaey, and S. Fechtel, Digital Communica- [20] S. M. Kay, Fundamentals of Statistical Signal Processing: Estima-
tion Receivers: Synchronization, Channel Estimation, and Signal tion Theory, Prentice-Hall, Englewood Cliffs, NJ, USA, 1993.
Processing, John Wiley & Sons, New York, NY, USA, 1997.
[3] K. H. Mueller and M. Müller, “Timing recovery in digital syn-
chronous data receivers,” IEEE Transactions on Communica-
tions, vol. 24, no. 5, pp. 516–531, 1976.
[4] F. M. Gardner, “A BPSK/QPSK timing-error detector for sam-
pled receivers,” IEEE Transactions on Communications, vol. 34,
no. 5, pp. 423–429, 1986.
[5] A. N. D’Andrea and M. Luise, “Optimization of symbol timing
recovery for QAM data demodulators,” IEEE Transactions on
Communications, vol. 44, no. 3, pp. 399–406, 1996.
[6] A. A. D’Amico, A. N. D’Andrea, and R. Reggiannini, “Efficient
non-data-aided carrier and clock recovery for satellite DVB at
very low signal-to-noise ratios,” IEEE Journal on Selected Areas
in Communications, vol. 19, no. 12, pp. 2320–2330, 2001.
[7] M. Moeneclaey and G. de Jonghe, “Tracking perfor-
mance comparison of two feedforward ML-oriented carrier-
independent NDA symbol synchronizers,” IEEE Transactions
on Communications, vol. 40, no. 9, pp. 1423–1425, 1992.
[8] M. Oerder and H. Meyr, “Digital filter and square timing re-
covery,” IEEE Transactions on Communications, vol. 36, no. 5,
pp. 605–612, 1988.
[9] E. Panayirci and E. K. Bar-Ness, “A new approach for evaluat-
ing the performance of a symbol timing recovery system em-
ploying a general type of nonlinearity,” IEEE Transactions on
Communications, vol. 44, no. 1, pp. 29–33, 1996.
[10] D. C. Rife and R. R. Boorstyn, “Single-tone parameter estima-
tion from discrete-time observations,” IEEE Transactions on
Information Theory, vol. 20, no. 5, pp. 591–598, 1974.
[11] D.-K. Hong and S.-J. Kang, “Joint frequency offset and car-
rier phase estimation for the return channel for digital video
broadcasting,” IEEE Transactions on Broadcasting, vol. 51,
no. 4, pp. 543–550, 2005.
[12] M. P. Fitz, “Planar filtered techniques for burst mode carrier
synchronization,” in Proceedings of IEEE Global Telecommuni-
cations Conference and Exhibition (GLOBECOM ’91), vol. 1,
pp. 365–369, Phoenix, Ariz, USA, December 1991.
[13] B. C. Lovell and R. C. Williamson, “The statistical perfor-
mance of some instantaneous frequency estimators,” IEEE
Transactions on Signal Processing, vol. 40, no. 7, pp. 1708–1723,
1992.
[14] M. Luise and R. Reggiannini, “Carrier frequency recovery
in all-digital modems for burst-mode transmissions,” IEEE
Transactions on Communications, vol. 43, no. 234, pp. 1169–
1178, 1995.
[15] U. Mengali and M. Morelli, “Data-aided frequency estimation
for burst digital transmission,” IEEE Transactions on Commu-
nications, vol. 45, no. 1, pp. 23–25, 1997.
[16] Y. Fan and P. Chakravarthi, “Joint carrier phase and symbol
timing synchronization for burst satellite communications,” in
Proceedings of the 21st Century Military Communications Con-
ference (MILCOM ’00), vol. 2, pp. 1104–1108, Los Angeles,
Calif, USA, October 2000.
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 29086, 12 pages
doi:10.1155/2007/29086
Research Article
Burst Format Design for Optimum Joint
Estimation of Doppler-Shift and Doppler-Rate
in Packet Satellite Communications
This paper considers the problem of optimizing the burst format of packet transmission to perform enhanced-accuracy estimation
of Doppler-shift and Doppler-rate of the carrier of the received signal, due to relative motion between the transmitter and the
receiver. Two novel burst formats that minimize the Doppler-shift and the Doppler-rate Cramér-Rao bounds (CRBs) for the joint
estimation of carrier phase/Doppler-shift and of the Doppler-rate are derived, and a data-aided (DA) estimation algorithm suitable
for each optimal burst format is presented. Performance of the newly derived estimators is evaluated by analysis and by simulation,
showing that such algorithms attain their relevant CRBs with very low complexity, so that they can be directly embedded into new-
generation digital modems for satellite communications at low SNR.
Copyright © 2007 Luca Giugno et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. INTRODUCTION tains pilot symbols known at the receiver. The most common
burst format is the conventional preamble-payload arrange-
Packet transmission of digital data is nowadays adopted ment, wherein all pilots are consecutive and they are placed
in several wireless communications systems such as satel- at the beginning of the data burst. Other formats are the mi-
lite time-division multiple access (TDMA) and terrestrial damble as in the GSM system [8], wherein the preamble is
mobile cellular radio. In those scenarios, the received sig- moved to the center of the burst, or the so-called pilot sym-
nal may suffer from significant time-varying Doppler dis- bol assisted modulation (PSAM) paradigm [9], where the
tortion due to relative motion between the transmitter and set of pilot symbols is regularly multiplexed with data sym-
the receiver. This occurs, for instance, in the last-generation bols in a given ratio (the so-called burst overhead). Data-
mobile-satellite communication systems based on a con- aided (DA) algorithms, which exploit the information con-
stellation of nongeostationary low-earth-orbit (LEO) satel- tained in the pilot symbols, are routinely used to attain good
lites [1] and in millimeter-wave mobile communications for performance with small burst overhead. The recent intro-
traffic control and assistance [2]. In such situations, car- duction of efficient channel coding with iterative detection
rier Doppler-shift and Doppler-rate estimation must be per- [10] has also placed new and more stringent requirements
formed at the receiver for correct demodulation of the re- for receiver synchronization on satellite modems. The car-
ceived signal. rier synchronizer is requested to operate at a lower signal-to-
A number of efficient digital signal processing (DSP) al- noise ratio (SNR) than it used to be with conventional coding
gorithms have already been developed for the estimation of [11].
the Doppler-shift affecting the received carrier [3] and a few Therefore, it makes sense to search for the ultimate ac-
algorithms for Doppler-rate estimation are also available in curacy that can be attained by carrier synchronizers. It turns
the open literature [4, 5]. The issue of joint Doppler-shift out that the Cramér-Rao bounds (CRBs) for joint estima-
and Doppler-rate estimation has been addressed as well, al- tions are functions of the location of the reference symbols
though to a lesser extent [6, 7]. In all the papers above, the in the burst. The issue to find the optimal burst format that
observed signal is either an unmodulated carrier, or con- minimizes the frequency CRB has been already addressed in
2 EURASIP Journal on Wireless Communications and Networking
-3P format-
N/3 M/2 N/3 M/2 N/3 -1st 2P subburst- -2nd 2P subburst-
b P Payload P Payload P d P Payload P P Payload P
L L
[12–14], but only for joint carrier phase/Doppler-shift es- the total number of symbols within the burst:
timation. The novelty of the paper is to extend the anal-
ysis to the joint carrier phase/Doppler-shift and Doppler- N N 1
η= = = . (1)
rate estimation. It is known [12–15] that the preamble- L N +M 1 + M/N
postamble format (2P format) described in the sequel min- We also assume BPSK/QPSK data modulation for the pilot
imizes the frequency CRB with no Doppler-rate, and with fields, and additive white Gaussian noise (AWGN) channel
constraints on the total training block length and on the with no multipath. Filtering is evenly split between transmit-
burst overhead of the signal. We demonstrate here that such ter and receiver, and the overall channel response is Nyquist.
format is optimal in the presence of Doppler-rate as well, Timing recovery is ideal but the received signal is affected by
and that the Doppler-rate CRB is minimized by estima- time-varying Doppler distortion. Filtering the received wave-
tion over three equal-length blocks of reference symbols that form with a matched filter and sampling at symbol rate at
are equally spaced by data symbols (3P format). We also the zero intersymbol interference instants yields the follow-
show that other formats are very close to optimality (4P for- ing discrete-time signal:
mat).
In addition to computation of the burst, we also in- L−1 L−1
z(k) = ck e jϕk + n(k), k=− , . . . , 0, . . . , ,
troduce new high-resolution and low-complexity carrier 2 2
Doppler-shift and Doppler-rate DA estimation algorithms (2)
for such optimal burst formats. where
The paper is organized as follows. In Section 2, we
first outline the received signal model affected by Doppler ϕk = θ + 2πνkT + παk2 T 2 (3)
distortions. Next, in Section 3 we present and analyze a
low-complexity DA Doppler-shift estimator for the optimal is the instantaneous carrier excess phase, {ck } are unit-energy
2P format. Extensions of this algorithm for joint carrier (QPSK) data symbols and L (odd) is the observation (burst)
phase/Doppler-shift and Doppler-rate estimation for the 2P length. Also, 1/T is the symbol rate, θ is the unknown initial
format, the 3P format, and the sub-optimum 4P format, are carrier phase, ν is the constant unknown carrier frequency
introduced in Sections 4 and 5, respectively. Finally, some offset (Doppler-shift), and finally α is the constant unknown
conclusions are drawn in Section 6. carrier frequency rate-of-change (Doppler-rate). For signal
model (2) to be valid, we assumed that the value of the
2. SIGNAL MODEL Doppler-shift ν is much smaller than the symbol rate, and
that the value of the Doppler-rate α is much smaller than
In this paper, we take into consideration three different data the square of the symbol rate. The noise n(k) is a complex-
burst formats as depicted in Figure 1. valued zero-mean WGN process with independent compo-
In all cases, the total number of pilot symbols that are nents, each with variance σ 2 = N0 /(2Es ), where Es /N0 repre-
known to the receiver is equal to N, and the total length of sents the ratio between the received energy-per-symbol and
the “data payload” fields that contain information symbols is the one-sided channel noise power spectral density.
equal to M. The formats differ for the specific pilots arrange- Estimation of ν and α from the received signal z(k) re-
ment in two/three/four groups of N/2, N/3, N/4 consecutive quires preliminary modulation removal from the pilot fields.
pilot symbols equally spaced by data symbols. Hereafter we Broadly speaking, it is customary to adopt BPSK or QPSK
will address them as “2P,” “3P,” “4P” formats as in Figures modulation for pilot fields, so that modulation removal is
1(a), 1(b), 1(c), respectively. We denote also with L = N + M easily carried out by letting r(k) = ck∗ z(k). The result is
the overall burst length, and with η the burst overhead, that
is, the ratio between the total number of pilot symbols and r(k) = e jϕk + w(k), k∈K = NPi , (4)
Luca Giugno et al. 3
MEV
The signal r(k) will be considered from now on as our ob- 0
served signal that allows to carry out the carrier synchro-
nization functions. We show in Appendix B that the burst −0.5
formats in Figure 1 are optimum so far as the estimation of
parameters ν and α is concerned. To keep complexity low, we −1
will not take into consideration here “mixed,” partially blind,
methods to perform carrier synchronization that use both the −1.5
−1.5 −1 −0.5 0 0.5 1 1.5
known pilot symbols and all of the intermediate data sym- νT (Hz × s) ×10−3
bols of the burst, like envisaged in [16] for the case of channel
estimation. Ideal Es /N0 = 20 dB
Es /N0 = 0 dB Es /N0 = 100 dB
Es /N0 = 10 dB
3. DOPPLER-SHIFT ESTIMATOR: FEPE ALGORITHM
We momentarily neglect the effect of the Doppler-rate α in Figure 2: MEV of FEPE estimator for different values of ES /N0 —
(4), to concentrate on the issue of Doppler-shift estimation simulation only. Preamble + postamble DA ML phase estimation,
N = 44, M = 385.
only. Under such hypothesis, (4) can be rewritten as follows:
r(k) = e j(θ+2πνkT) + w(k), k ∈ K. (5)
The 2P format minimizes the CRB for Doppler-shift esti- fixed-point arithmetic on a digital hardware. It is easy to ver-
mation for joint carrier phase/Doppler-shift estimation [12– ify that such estimator is independent of the particular ini-
15]. Conventional frequency offset estimators for consecu- tial phase θ, that vanishes when computing the phase dif-
tive signal samples [3] are not directly applicable to a burst ference at the numerator of (7). It is also clear that the
format encompassing a preamble and a postamble. In addi- operating range of the estimator is quite narrow. In order
tion, straightforward solution of a maximum-likelihood es- not to have estimation ambiguities, we have to ensure that
timation problem for ν appears infeasible. We introduce thus −π ≤ |θ2 |2π − |θ1 |2π < π, and therefore the range is bounded
a new low-complexity algorithm suitable for the estimation to
of the Doppler-shift ν in (4) with the burst format as above.
The key idea of the 2P frequency estimator is really a naive 1
|ν | ≤ . (8)
one: we start by computing two phase estimates, the one on 2(M + N/2)T
the preamble section, and the other on the postamble, us-
ing the standard low-complexity maximum-likelihood (ML) This relatively narrow interval does not allow to use the FEPE
algorithm [17]: algorithm for initial acquisition of a large frequency offset at
−(M −1)/2 (N+M −1)/2 receiver start-up. Its use is therefore restricted to fine esti-
mation of a residual offset after a coarse acquisition or com-
θ1= arg r(k) , θ2 = arg r(k) ,
k=−(N+M −1)/2 k=(M −1)/2
pensation of motion-induced Doppler-shift. Figure 2 depicts
(6) the normalized mean estimated value (MEV) curves of the
FEPE algorithm (i.e., the average estimated value E{ν} as a
where arg{·} denotes the phase of the complex-valued ar- function of the true Doppler-shift ν) for different values of
gument. Then we associate the two phase estimates to the Es /N0 as derived by simulation. In our simulations we use
two midpoints of the preamble and postamble sections, re- the values N = 44 and M = 385 taken from the design de-
spectively, whose time distance is equal to (M + N/2)T scribed in [11], so that the overhead is η = 10% (typical for
(Figure 1(a)). After this is done, we simply derive the fre- short bursts). MEV curves show that the algorithm is unbi-
quency estimate as the slope of the line that connects the two ased in a broad range around the true value (here, ν = 0). It
points (−(M − 1)/2 − N/4, θ1 ) and ((M − 1)/2 + N/4, θ2 ) on can be shown that this is true as long as ν2NT 1, so that
the (time, phase) plane: the “ancillary” estimates θ2 and θ1 are substantially unbiased
as well. Such condition is implicitly assumed in (8) since in
θ2 − θ1 2π 2π
ν=
2π
. (7) the practice M N/2. The curve labeled Es /N0 = 100 dB
2π(M + N/2)T (which is totally unrealistic) has the only purpose of showing
This simple algorithm is known as frequency estimation the bounds of the unambiguous estimation range.
through phase estimation (FEPE) [15]. The operator |x|2π re- It is also easy to evaluate the estimation error variance of
turns the value of x modulo 2π, in order to avoid phase am- the FEPE estimator. It is known in fact that θ1 and θ2 in (7)
biguities, and is trivial to implement when operating with have an estimation variance σθ2 that achieves the Cramér-Rao
4 EURASIP Journal on Wireless Communications and Networking
1 1
σθ2 = CRB(θ) = . (9)
2 · N/2 Es /N0
Therefore, considering that the two phase estimates in (7) are
independent, we get
10−3
2 · σθ2 1 1
2
σFEPE (ν) = = .
4π 2 (M + N/2)2 T 2 4π 2 T 2 N/2(M + N/2)2 Es /N0
Normalized RMSEE
(10) VCRB (ν)
The vector CRB [18] for the frequency offset estimate in the
10−4
joint carrier phase/Doppler-shift estimation with the 2P for-
mat is derived in Appendix A and reads as follows:
3 1
VCRB2P (ν) =
.
4π 2 T 2 (N/2) 4(N/2)2 + 3M 2 +3MN − 1 Es /N0
(11) VCRB2P (ν)
10−5
Both from the expression of the bound (11) and of the
variance (10), it is seen that the estimation accuracy has an
inverse dependence on (N/2)3 , and this is nothing new with
respect to conventional estimation on a preamble only. The
important thing is that we also have inverse dependence on
M 2 , due to the 2P format that gives enhanced accuracy (with 10−6
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
small estimation complexity) with respect to the conven-
Es /N0 (dB)
tional estimator. From (1), we also have M = N(1/η − 1),
so that the term 3M 2 dominates (N/2)2 as long as η < 1/2, 2
σFEPE (ν)
which is always verified in the practice. νT = 1 × 10−3
Therefore, the ratio between the CRB (11) and the vari- νT = 1 × 10−4
ance of the FEPE estimator is very close to 1. With N = 44
2
and M = 385, we get, for instance, σFEPE /VCRB2p = 0.99. Figure 3: RMSEE of FEPE estimator for different values of
The enhanced-accuracy feature is also apparent in the com- ES /N0 and relevant bounds—solid lines: theory—marks: simula-
parison of the VCRB2p (ν) as in (11) with the conventional tion. Preamble + postamble DA ML phase estimation, N = 44,
M = 385.
VCRB(ν) [18] for frequency estimation on a single preamble
with length N, that is obtained by letting M = 0 in (11). The
reverse of the coin is of course the reduced operating range
(8) of the estimator. 13] is presented in Table 1. It is clear that the strength of the
Figure 3 shows curves of the (symbol-rate-normalized) FEPE algorithm is its very low complexity as compared to
RMSEE (root mean square estimation error) of the FEPE conventional algorithms.
algorithm (i.e., T E{(v − v)2 }) as a function of Es /N0 for
various values of the true offset ν. In particular, marks are 4. DOPPLER-RATE ESTIMATORS IN 2P FRAME:
2
simulation results for σFEPE , whilst the lowermost line is the FREPE AND FREFE ALGORITHMS
VCRB2p (ν). We do not report the curve for (10) since it
would be totally overlapped with (11). We take now back into consideration the presence of a non-
Performance assessment of the FEPE estimator is con- negligible Doppler-rate in the received signal, modeled as in
cluded in Figure 4 with the evaluation of the sensitivity of the (3)-(4). We focus again on the 2P format (Figure 1(a)), since
RMSEE to different values of an uncompensated Doppler- it is the optimal format for Doppler-shift estimation in joint
rate α. Just to have an idea of practical values of αT 2 to be en- carrier phase/Doppler-shift and Doppler-rate estimation too,
countered in practice, we mention that the largest Doppler- as demonstrated in Appendix B. A new simple estimator for
rates in LEO satellites are of the order of 200 Hz/s [1, 19] for α in the 2P format is found by a straightforward general-
a carrier frequency of 2.2 GHz, and assuming a symbol rate ization of the FEPE approach. Assume that we further split
of 2 Mbaud, we end up with the value αT 2 = 5.10−11 . From both the preamble and the postamble into two subsections of
simulation results, we highlight that the performance of this equal length, and we compute four (independent) ML phase
algorithm is affected by α, but only in the case of a normal- estimates on the two subsections. We know in advance that
ized Doppler-rate αT 2 ≥ 10−7 , that is larger than those that the time evolution of the phase is described by a parabola.
are found in the practice. The four phase estimates can thus be used to fit a second-
Finally, the complexity of the FEPE estimator with re- order phase polynomial according to the Minimum Mean
spect to conventional methods of frequency estimation [3, Squared Error (MMSE) criterion; taking the origin in the
Luca Giugno et al. 5
VCRB2P (ν) αT 2 = 1 × 10−7 and solving for a we get the following so-called frequency rate
αT 2 = 0 αT 2 = 2 × 10−7
αT 2 = 2 × 10−8
estimation through phase estimation (FREPE) algorithm [15]:
αT 2 = 2.5 × 10−7
αT 2 = 5 × 10−8
a θ4 − θ3 − θ2 − θ1
αFREPE = 2 = (15)
Figure 4: Sensitivity of FEPE estimator to different values of the T πN/2(N/2 + M)T 2
Doppler-rate αT 2 . Preamble + postamble DA ML phase estimation,
N = 44, M = 385, vT = 1.0 × 10−3 . (all differences to be intended modulo-2π). This extremely
simple approach can be viewed as a generalization of the
FEPE introduced in the previous section. In particular, by us-
first section of the preamble, we obtain the phase model ing (7), the terms
M − 1 3N 2 θi − θi−1
ϕP (n) = aπ n + + , i = 2, 4, (16)
2 8 2π(N/4)T
(12)
M − 1 3N
+ 2πb n + + + c, represent two Doppler-shift estimations, the first on the
2 8
preamble and the second on the postamble, respectively,
where the regression coefficients a and b directly repre- which are spaced M + N/2 symbols apart. The Doppler-rate
sent estimates for the (normalized) carrier Doppler-rate and estimate is thus simply the difference between the two fre-
Doppler-shift, respectively, and c is an estimate for the initial quency estimates, divided by their time distance (M +N/2)T.
phase (that we are not interested into). The coefficients are The considerations above allow us to also introduce
found after observing that the MSE is written as the frequency rate estimation through frequency estimation
(FREFE) algorithm [15]
4
2
4
ε(a, b, c) = ϕP ni − θi = ei2 , (13) ν2 − ν1
i=1 i=1
αFREFE = , (17)
(M + N/2)T
where θi , i = 1, . . . , 4, are the above-mentioned ML phase wherein the two frequency estimates ν1 and ν2 can be ob-
estimates on N/4 pilots each, and n1 = −[(M − 1)/2 + 3N/8], tained by any conventional algorithm [3] operating sepa-
n2 = −[(M − 1)/2 + N/8], n3 = [(M − 1)/2 + N/8], and rately on the preamble and on the postamble, respectively.
n4 = [(M − 1)/2 + 3N/8] are the four time instants that we We can choose for instance the L&R algorithm [20] or the
conventionally associate to the four estimates (the midpoints R&B algorithm [21]. Assuming that the selected algorithm
of the four subsections). Equating to zero the derivatives of operates close enough to the CRB (as is shown in [3]), the
6 EURASIP Journal on Wireless Communications and Networking
MEV
(18) 0
where we have used σν2 = 3 · (Es /N0 )−1 /[2π 2 T 2 N/2((N/2)2 − −0.5
1)] [17]. This can be compared to the variance of the FREPE
algorithm that is easily found to be −1
4 · σθ2 −1.5
2 −1.5 −1 −0.5 0 0.5 1 1.5
σFREPE (α) = ×10−4
π 2 (N/2)2 (M + N/2)2 T 4 αT 2 (Hz/s × s2 )
(19)
4 1 Ideal Es /N0 = 20 dB
= 2 4 ,
π T (N/2)3 (M + N/2)2 Es /N0 Es /N0 = 0 dB Es /N0 = 100 dB
Es /N0 = 10 dB
where now σθ2 = (Es /N0 )−1 /(N/2). The relevant vector CRB
for Doppler-rate estimate is (see Appendix B): Figure 5: MEV of FREPE estimator for different values of ES /N0 —
VCRB2P (α) simulation only. Preamble + postamble DA ML phase estimation,
N = 44, M = 385, vT = 1.0 × 10−3 .
45 1
= .
π2T 4 (N/2)3− N/2 16(N/2) +15M +30MN/2 − 4 Es /N0
2 2
(20) ×10−4
1.5
All expressions inversely depend on (N/2)5 as in conven-
tional preamble-only estimation of the Doppler-rate [6], but 1
they also bear again inverse dependence on M 2 that gives en-
0.5
hanced accuracy. For sufficiently large values of N and M,
MEV
M N, we have 0
2
σFREFE (α) ∼ 3 VCRBPP (α) ∼ −0.5
2 = , 2 = 1. (21)
σFREPE (α) 4 σFREFE (α)
−1
Figure 5 shows the MEV curves (i.e., E{ α}) of the FREPE al-
gorithm for different values of Es /N0 , in the case of N = 44, −1.5
M = 385, and Doppler-shift vT = 10−3 . The estimator is −1.5 −1 −0.5 0 0.5 1 1.5
αT 2 (Hz/s × s2 ) ×10−4
unbiased with an operating range equal to
1 Ideal νT = 5 × 10−3
αFREPE ≤ . (22) νT = 0 νT = 1 × 10−2
N/2(M + N/2)T 2
νT = 1 × 10−3
The sensitivity of FREPE to different uncompensated val-
ues of vT is illustrated in Figure 6 in terms of MEV. Figure 6: MEV of FREPE estimator for different values of the
The same simulations have been run also for the FREFE Doppler-shift vT—simulation only. Preamble + postamble DA ML
algorithm. In particular, Figure 7 illustrates the MEV curves phase estimation, N = 44, M = 385, Es /N0 = 10 dB.
for different values of Es /N0 and with vT = 10−3 . By using
the L&R algorithm to estimate ν1 and ν2 , the operating range
of FREFE is roughly twice that of FREPE:
5. OPTIMUM DOPPLER-RATE ESTIMATION
1
αFREFE ≤ . (23)
(N/4 + 1)(M + N/2)T 2 5.1. Odd number of pilot fields: FRE-3PE algorithm
In particular, the term [(N/2 + 1)T]−1 represents the fre- We turn now to the issue of optimum burst configuration
quency pull-in range of L&R on N/2 pilots [20]. for the estimation of the Doppler-rate. We demonstrate in
Figure 8 demonstrates that FREPE is also less sensitive Appendix B that the 3P format (Figure 1(b)) minimizes the
than FREFE to an uncompensated Doppler-shift. Finally, CRB for Doppler-rate estimation, with the usual constraints
Figure 9 shows the curve of the Doppler-rate RMSEE of on the total training block length and on the burst over-
FREPE and FREFE as a function of Es /N0 , for νT = 10−3 and head (1). In the following, we develop a new low-complexity
αT 2 = 10−6 . The FREPE estimator loses only 10 log10 (4/3) = algorithm suitable for Doppler-rate estimation with the 3P
1.25 dB in terms of Es /N0 with respect to the performance of format. We know in advance that the time evolution of the
the more complex FREFE when N 1. phase is described by a parabola. As was done for the FREPE
Luca Giugno et al. 7
×10−4 10−3
2.5
2
1.5
1
10−4
0.5
MEV
0
−0.5
−1 VCRBP (α)
−1.5
Normalized RMSEE
10−5
−2
−2.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
αT 2 (Hz/s × s2 ) ×10−4
10−6
Ideal Es /N0 = 20 dB VCRB2P (α)
Es /N0 = 0 dB Es /N0 = 100 dB
Es /N0 = 10 dB
Figure 7: MEV of FREFE estimator for different values of ES /N0 — 10−7 VCRB4P (α)
simulation only. Preamble + postamble Luise and Reggiannini, N =
VCRB3P (α)
44, M = 385, vT = 1.0 × 10−3 .
×10−4
2.5 10−8
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
2
Es /N0 (dB)
1.5
1 FREPE, αT 2 = 1 × 10−6 FRE-2FEPE, αT 2 = 1 × 10−6
0.5 FREFE, αT 2 = 1 × 10−6 FRE-3PE, αT 2 = 1 × 10−6
MEV
0
−0.5 Figure 9: RMSEE of FREPE, FREFE, FRE-3PE, and FRE-2FREPE
−1 estimators for different values of ES /N0 and relevant bounds,—solid
−1.5
lines: theory—marks: simulation. Doppler-rate algorithms: FREFE
versus FREPE versus FRE-3PE versus FRE-2FEPE, N = 44(45),
−2
M = 385(384), vT = 1.0 × 10−3 .
−2.5
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
αT 2 (Hz/s × s2 ) ×10−4
×10−5 ×10−5
1.5 1.5
1 1
0.5 0.5
MEV
MEV
0 0
−0.5 −0.5
−1 −1
−1.5 −1.5
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
αT 2 (Hz/s × s2 ) ×10−5 ×10−5
αT 2 (Hz/s × s2 )
Figure 10: MEV of FRE-3PE estimator for different values of Figure 11: MEV of FRE-3PE estimator for different values of the
ES /N0 —simulation only. 3 blocks of pilots DA ML phase estima- Doppler-shift vT—simulation only. 3 blocks of pilots, N = 45, M =
tion, N = 45, M = 384, vT = 1.0 × 10−3 . 384, ES /N0 = 10 dB.
they basically mimic those in Figures 10 and 11 for the [12]. After modulation removal, the generic sample within
FRE-3PE algorithm. The estimation error variance of (30) the preamble and the postamble is given by (5).
is found to be The Fisher information matrix (FIM) [18] can be written
as
σθ2
2
σFRE-2FEPE (α) = Fθθ Fθν
(2M/3 + N/2)2 (M/3 + N/4)2 π 2 T 2 F=
−1 Fνθ Fνν
2 · Es /N0 ⎡ ⎤
= . ∂2 ln p(r |
ν, θ) ∂2 ln p(r |
ν, θ)
π 2 T 4 N(2M/3 + N/2)2 (M/3 + N/4)2 ⎢−Er
−Er
⎥
⎥
⎢ ∂
θ 2 ∂
θ∂ ν
(31) =⎢
⎢
⎥
⎥,
⎣ ∂ ln p(r |
2
ν, θ) ∂ ln p(r |
2
ν, θ) ⎦
−Er −Er
Figure 9 shows also the curves of the RMSEE of FRE-2FEPE ∂ν∂θ ∂ν2
and its respective CRB. The FRE-2FEPE algorithm reaches its (A.1)
own VCRB4p (α) and thus, as demonstrated in Appendix B, it
gains 10 log10 (7.19) = 18.5 dB in terms of Es /N0 with respect is the probability density function of r =
where p(r | ν, θ)
to the performance of the previous algorithms with the 2P {r(k)}, k ∈ K, conditioned on ( and r(k) is a random
ν, θ),
format. Also, the FRE-2FEPE loses only 0.4 dB with respect Gaussian variable with variance equal to σ 2 = N0 /(2Es ) and
to the FRE-3PE algorithm and can thus be a valid alternative mean value equal to
to the 3P format.
As a final remark, we briefly address the issue of Doppler- s(k) = e j(θ+2π νkT) .
(A.2)
shift estimation in the 4P format. The best method is found
as
Therefore, we write p(r | ν, θ)
by applying the FEPE algorithm to the two extreme pilot
fields of the burst. Its variance is close to the VCRB4P (ν) cal-
=
p(r | ν, θ) p rk | ν, θ
culated setting x = 1 in (B.8) and (B.9), that is 2.4 dB worse k∈K
than the VCRB2P (ν) of the optimal 2P format.
1 1 r(k) −
2
= N exp − s(k) .
6. CONCLUSIONS 2πσ 2 2σ 2 k∈K
(A.3)
In this paper, we presented and analyzed some very-
low-complexity algorithms for carrier Doppler-shift and Taking the logarithm of (A.3), we obtain
Doppler-rate estimation in burst digital transmission. To
ln p(r | ν, θ)
achieve enhanced accuracy, the burst configurations that
1 1
r(k) 2 +
2
minimize the CRB for the estimation of Doppler-shift and = N ln − s(k)
Doppler-rate are derived. Our analysis showed that the 2P 2πσ 2 2σ 2 k∈K
format is optimum for Doppler-shift estimation and that the
s∗ (k)
− 2Re r(k)
3P format is optimum for Doppler-rate estimation. These
1
two configurations can be practically thought as repetition of =C + Re r(k)s∗ (k) ,
two/three consecutive conventional (preamble-only) bursts. σ 2 k∈K
Despite preventing from real-time processing of the data pay- (A.4)
load section, the 2P and 3P formats greatly outperform the
where C is a constant term that includes all the quantities
estimation based on conventional preamble-only pilot dis- After differentiating twice (A.4) with
tribution. Performance assessment has shown that all of the independent of ν and θ.
respect to ν and θ, calculating the expectation of the various
proposed algorithms are unbiased in practical operating con-
ditions, and that their accuracy in terms of estimation vari- terms with respect to r, we get
ance gets remarkably close to their respective CRBs down to
a b
very low Es /N0 values. F= , (A.5)
c d
APPENDICES where
1
∗
A. VCRB FOR JOINT CARRIER PHASE/DOPPLER-SHIFT a = (1)Er Re r(k)s (k) ,
ESTIMATION WITH 2P FORMAT σ2 k∈K
1
∗
In this appendix, we calculate the VCRB for the error vari- b = (2πTk)Er Re r(k)s (k) ,
σ2 k∈K
ance of any unbiased estimator of Doppler-shift in the case of
(A.6)
joint estimation of phase/Doppler-shift using the preamble- 1
∗
c = (2πTk)Er Re r(k)s (k) ,
postamble (2P) format. We explicitly mention that we have σ2 k∈K
chosen a set K of pilot locations that is symmetrical with
1
∗
respect to the middle of the burst, since a symmetrical K de- d = 4π 2 T 2 k2 Er Re r(k)s (k) .
σ2
couples phase from Doppler-shift estimation, as discussed in k∈K
10 EURASIP Journal on Wireless Communications and Networking
NP MD NP MD NP MD NP MD NP MD NP
Er Re r(k)s∗ (k) = 1, (A.7)
P P P P P P
we obtain 0
⎡ ⎤ (a)
⎢
(1) 2πT k ⎥
1 ⎢ k∈K k∈K ⎥ NP MD NP MD NP MD NP MD NP MD NP MD NP
F= ⎢ ⎥
⎢ ⎥, (A.8)
σ2 ⎣2πT k 4π 2 T 2 k2 ⎦ P P P P P P P
k∈K k∈K
0
L
where, considering the symmetry of the range K,
(b)
(1) = N, k = 0, (A.9)
k∈K k∈K Figure 12: Generic symmetric burst format.
2
N/2 N N
k2 = 8 −6 +1
k∈K
3 2 2
(A.10) pilot symbols, and (2xeven + 1) is the (odd) number of sub-
2 N groups of data symbols; in Figure 12(b), (2xodd + 1) is the
+ 3M + 3M 3 −1 .
2 (odd) number of subgroups of pilot symbols, and 2xodd is
the (even) number of subgroups of data symbols. In the se-
After calculation of F−1 , the VCRB for ν in case of joint quel we find the values of x that minimize the VCRBs of ν
phase/Doppler-shift estimation is found to be and α, for fixed values of L, N, and M.
In the case of joint phase/Doppler-shift/Doppler-rate es-
−1 1 1
Fνν = VCRB2P (ν) = timation, the fisher information matrix (FIM) of the generic
2π 2 T 2 k∈K k2 Es /N0 bursts of Figure 12 can be written as
−1 (A.11)
3 · Es /N0 ⎡ ⎤
=
. F F F
4π 2 T 2 (N/2) 4(N/2)2 + 3M 2 + 3MN − 1 ⎢ θθ θν θα ⎥
F=⎢ ⎣ Fνθ Fνν Fνα ⎦
⎥
respect to r, we have Note that, thanks to the symmetry of the burst, the el-
⎡ ⎤ ements Fθν , Fνθ , Fαν , Fνα are all zero, which means that
(1) 2πT k πT 2 k2 the joint phase/Doppler-shift and Doppler-shift/Doppler-
⎢ k∈K ⎥
⎢ k∈K k∈K ⎥ rate estimations are decoupled.
⎢ ⎥
⎢ ⎥ Calculating F−1 , we obtain the VCRBs for the estimation
1 ⎢
⎢ 2πT 2
k 4π T 2 2
k 2π T2 3 3⎥
k ⎥ of ν as follows:
F= 2⎢ ⎥ , (B.4)
σ ⎢⎢ k∈K k∈K k∈K ⎥
⎥
⎢ ⎥
⎢ ⎥
⎣πT 2 k2 2π 2 T 3 k3 π 2 T 4 k4 ⎦ −1 1 1
Fνν = VCRB(ν) = , (B.9)
k∈K k∈K k∈K 2π 2 T 2 k∈K k2 Es /N0
where, thanks to the symmetry of range K,
as the one found in (A.11) without any Doppler-rate. The
k = 0, k3 = 0. (B.5) optimal burst configuration that minimizes the VCRB for ν
k∈K k∈K is thus the 2P format found in [14] also in the presence of
We finally get the expression of the FIM matrix as Doppler-rate effects.
The VCRB for α is
⎡ ⎤
N 0 πT 2 k2
⎢ ⎥
⎢ k∈K ⎥ 2N 1
1 ⎢ 4π 2 T 2 2 ⎥ −1
Fαα = VCRB(α) = −
2 .
F= 2⎢ 0 k 0 ⎥. (B.6)
σ ⎢ ⎥ π2T 4 k∈K k
2 − N k∈K k 4 E s /N0
⎢ k∈K ⎥
⎣ ⎦ (B.10)
πT 2 k2 0 π2T 4 4
k
k∈K k∈K
100 78.6 tions on Signal Processing, vol. 39, no. 5, pp. 1203–1205, 1991.
71.9 [7] T. J. Abatzoglou, “Fast maximum likelihood joint estima-
80
tion of frequency and frequency rate,” IEEE Transactions on
60
Aerospace and Electronic Systems, vol. 22, no. 6, pp. 708–715,
40 1986.
20 [8] T. S. Rappaport, Wireless Communications: Principles & Prac-
1.09
0 tice, Prentice-Hall, Englewood Cliffs, NJ, USA, 1999.
−20 [9] J. A. Gansman, J. V. Krogmeier, and M. P. Fitz, “Single fre-
−40 quency estimation with non-uniform sampling,” in Proceed-
0 10 20 30 40 50 60 70 80 90 100 ings of the 30th Asilomar Conference on Signals, Systems and
N Computers, vol. 1, pp. 399–403, Pacific Grove, Calif, USA,
November 1996.
VCRB2P (α)/VCRB3P (α) [10] V. Lottici and M. Luise, “Embedding carrier phase recovery
VCRB2P (α)/VCRB4P (α)
into iterative decoding of turbo-coded linear modulations,”
VCRB4P (α)/VCRB3P (α)
IEEE Transactions on Communications, vol. 52, no. 4, pp. 661–
669, 2004.
Figure 13: VCRB2P (α)/VCRB3P (α), VCRB2P (α)/VCRB4P (α) and [11] S. Benedetto, R. Garello, G. Montorsi, et al., “MHOMS: high-
VCRB4P (α)/VCRB3P (α) ratios as function of the total number of pi- speed ACM modem for satellite applications,” IEEE Wireless
lots N. Communications, vol. 12, no. 2, pp. 66–77, 2005.
[12] F. Rice, “Carrier-phase and frequency-estimation bounds for
transmissions with embedded reference symbols,” IEEE Trans-
The various VCRBs can be easily calculated from (B.10) actions on Communications, vol. 54, no. 2, pp. 221–225, 2006.
using the appropriate x and (B.7) and (B.8). We report here [13] H. Minn and S. Xing, “An optimal training signal structure for
the final expressions frequency-offset estimation,” IEEE Transactions on Communi-
cations, vol. 53, no. 2, pp. 343–355, 2005.
VCRB2P (α) [14] A. Adriaensen, A. Van Doninck, and W. Steinert, “MF-TDMA
−1 burst demodulatir design with pilot-symbol assisted frequency
= Fαα xeven =1
−1 estimation,” in Proceedings of the 8th International Workshop
360 · Es /N0 on Signal Processing for Space Communications (SPSC ’03),
= 2 4 3 ,
Catania, Italy, September 2003.
π T N − 4N 4N 2 + 15M 2 + 15MN − 4
[15] L. Giugno and M. Luise, “Carrier frequency and frequency
rate-of-change estimators with preamble-postamble pilot
(B.12) symbol distribution,” in Proceedings of IEEE International Con-
ference on Communications (ICC ’05), vol. 4, pp. 2478–2482,
VCRB4P (α)
−1
Seoul, Korea, May 2005.
= Fαα xodd =2 [16] A. Zhuang and M. Renfors, “Combined pilot aided and de-
−1 2 4 cision directed channel estimation for the RAKE receiver,” in
=25920 · Es /N0 / π T N 288N 4 + 1305N 3 M Proceedings of the 52nd IEEE Vehicular Technology Conference
+ 240NM 8M 2 − 15 (VTC ’00), vol. 2, pp. 710–713, Boston, Mass, USA, September
2000.
+ 30N 2 77M 2 − 48
[17] U. Mengali and A. N. D’Andrea, Synchronization Techniques
+ 32 20M 4 − 75M 2 + 36 . for Digital Receivers, Plenum Press, New York, NY, USA, 1997.
(B.13) [18] F. Gini, R. Reggiannini, and U. Mengali, “The modified
Cramér-Rao bound in vector parameter estimation,” IEEE
Transactions on Communications, vol. 46, no. 1, pp. 52–60,
REFERENCES
1998.
[1] I. Ali, N. Al-Dhahir, and J. E. Hershey, “Doppler characteriza- [19] G. R. J. Povey and J. Talvitie, “Doppler compensation and code
tion for LEO satellites,” IEEE Transactions on Communications, acquisition techniques for LEO satellite mobile radio commu-
vol. 46, no. 3, pp. 309–313, 1998. nications,” in Proceedings of the 5th International Conference on
[2] “Special Issue on communications in the intelligence trans- Satellite Systems for Mobile Communications and Navigation,
portation system,” IEEE Communications Magazine, vol. 34, pp. 16–19, London, UK, May 1996.
1996. [20] M. Luise and P. Reggiannini, “Carrier frequency recovery
[3] M. Morelli and U. Mengali, “Feedforward frequency estima- in all-digital modems for burst-mode transmissions,” IEEE
tion for PSK: a tutorial review,” European Transactions on Transactions on Communications, vol. 43, no. 2–4, pp. 1169–
Telecommunications, vol. 9, no. 2, pp. 103–116, 1998. 1178, 1995.
[4] F. Giannetti, M. Luise, and R. Reggiannini, “Simple carrier fre- [21] D. C. Rife and R. R. Boorstyn, “Single-tone parameter estima-
quency rate-of-change estimators,” IEEE Transactions on Com- tion from discrete-time observations,” IEEE Transactions on
munications, vol. 47, no. 9, pp. 1310–1314, 1999. Information Theory, vol. 20, no. 5, pp. 591–598, 1974.
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 23923, 8 pages
doi:10.1155/2007/23923
Research Article
TCP-Call Admission Control Interaction in
Multiplatform Space Architectures
The implementation of efficient call admission control (CAC) algorithms is useful to prevent congestion and guarantee target qual-
ity of service (QoS). When TCP protocol is adopted, some inefficiencies can arise due to the peculiar evolution of the congestion
window. The development of cross-layer techniques can greatly help to improve efficiency and flexibility for wireless networks.
In this frame, the present paper addresses the introduction of TCP feedback into the CAC procedures in different nonterrestrial
wireless architectures. CAC performance improvement is shown for different space-based architectures, including both satellites
and high altitude platform (HAP) systems.
Copyright © 2007 Georgios Theodoridis et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
In this frame, the present paper investigates the possibil- users are met, and (ii) maximize revenue from the network’s
ity of introducing an interaction between TCP and CAC in perspective, that is, optimize the utilization of the available
several nonterrestrial wireless architectures, in order to im- resources [8]. However, achieving these objectives is rather
prove CAC efficiency by taking TCP dynamics into account. difficult, since CAC is inherently an “in advance” procedure
More specifically, the exploitation of TCP feedback as input and no traffic model can offer a priori a completely accu-
for the CAC algorithm at regular time intervals has proved rate prediction, in particular considering the heterogeneity
to be of primary importance for maximizing the utilization of multimedia telecommunication traffic sources. Therefore,
of the network resources [10]. However, as the TCP perfor- real-time measurements of each connection’s load and con-
mance is rather dependent on the characteristics of the com- ditions are considered essential for the CAC’s effectiveness
munication path, the implementation of the CAC-TCP inter- [9].
action on different system architectures will introduce mean- CAC functionality is based on the concept that the used
ingful improvements in all the architectures showing such bandwidth plus the bandwidth of the upcoming user should
characteristics, demonstrating a more general importance of be lower or equal to the total capacity. As a matter of fact the
the concept. Additionally, so far, limited work can be found following condition must be always respected:
in the literature on this topic.
The paper is organized as follows: Section 2 provides a
brief analysis of the TCP driven CAC concept, while Section 3
N
Bi + B f ≤ c. (1)
includes an overview of possible architectures. Section 4 i=1
presents a description of the reference architectures and the
simulation scenario, along with results comparing the effi-
ciency of the proposed algorithm in the several space archi- Since always B j TCP datarate ≤ B j nominal datarate, the
tectures. Finally, Section 5 summarizes the conclusions. exploitation of the TCP feedback leads to a decrease in the
system overall blocking probability. Moreover, the band-
width assigned to each connection is equal to the real data
2. THE CONCEPT OF A TCP-BASED CAC rate of the connection monitored via the TCP performance.
Therefore, having maximized the average number of the
TCP is a transport layer protocol based on sending data pack- users simultaneously active in the network and having min-
ets upon reception of acknowledgement of previously sent imized the over-assignment of resources, the throughput of
packets, thus guaranteeing high reliability. When the net- the network, defined as the percentage of the aggregate ca-
work is characterized by significant round trip time (RTT), pacity that is actually occupied by the set of active connec-
as in the case a satellite path is included, this process can sig- tions, is radically improved.
nificantly slow down data transfer. In this frame, the possibility to get feedback informa-
TCP can exploit congestion control either through an tion about TCP congestion window actual evolution would
ACK counting mechanism (the actions on the sliding win- be of primary importance in order to efficiently drive CAC
dow are just based on the number of received ACKs) [3] or scheme. In fact, since the CAC algorithm, by taking into ac-
through the byte-counting scheme (the actions on the sliding count the actual amount of capacity necessary to exploit all
window are based on the actual number of bytes acknowl- the TCP connections, could prevent the over provision of
edged) [11]. bandwidth to the aforementioned connections, a better uti-
When the communication path is not error-free (usual lization of the network resources would be achieved. In this
in wireless networks) TCP misinterprets the data loss due to way, the admission/rejection of the new user would be based
the harsh wireless reception conditions as congestion occur- on the actual occupancy of the channel by the active users at
rence. As a consequence, for every packet loss, TCP reduces the time instant of a new user arrival, computed according to
the actual transmission rate, limiting the bandwidth utiliza- the TCP congestion window state of the connections instead
tion of the connection far below its nominal value. of their nominal data rate. The above scheme is depicted in
This inefficiency is meaningful in wireless networks since the flow chart of Figure 1.
the radio resource is usually scarce and expensive. Particu-
larly in GEO satellite, the large footprints limit the imple-
mentation of frequency reuse, thus reducing system capac- 3. SUITABLE ARCHITECTURES FOR
ity. Therefore, achieving maximum utilization of the avail- CAC-TCP INTERACTION
able bandwidth must be the primary goal of every network
configuration. The potential improvement introduced by the implementa-
On the other hand, CAC is implemented by the network tion of the integrated CAC-TCP scheme is addressed in var-
manager as a preventive congestion control scheme. CAC al- ious nonterrestrial wireless architectures, where either the
gorithms decide upon the admittance/rejection of new con- high propagation delay and/or the occurrence of transmis-
nections based on the network conditions (traffic load, link sion errors negatively impact TCP performance by leading to
capacity, buffer size, etc.) as well as the traffic characteristics an unjustified decrease of the transmission data rate.
and the QoS objectives of both the candidate and the already In particular, four different architectures are introduced
active users. In this framework, the aim of CAC is twofold: (i) and described, focusing on the potential drawbacks concern-
to guarantee that the QoS requirements of all the admitted ing optimal TCP working.
Georgios Theodoridis et al. 3
Yes
N
BW occupied = B j datarate;
j =1
Yes
No
A is blocked A is admitted
Core network
Core network
be assumed. Moreover, the gateway-satellite link is typically RTT. Finally, the integrated GEO-HAP scenario combines
dimensioned to be error free. the characteristics of both of them, that is maximum RTT
and maximum PER. Moreover, beyond the fact that these
Stand-alone HAPS network scenarios present a wide range of RTT and PER val-
ues, they are also the most significant in terms of services
In comparison with the previous architecture, the GEO satel- and applications. Therefore, the analysis of these case stud-
lite has been replaced with a HAP, while the data flow main- ies can provide solid conclusions regarding the ability of the
tains the same characteristics. The proximity of the HAP to proposed TCP-CAC interaction to improve the network per-
the earth greatly decreases link latency and facilitates the con- formance, in terms of both blocking probability and average
nectivity of mobile users. In more detail, in the GEO satellite throughput, in a great variety of channel conditions.
scenario, a mobile terminal should be equipped with high
power transmission amplifier as well as sizeable antennas, so 4.2. Simulation scenario and parameters
as to compensate the high free-space attenuation imposed
by the long propagation path. These features lead to bulky All the users are classified into three QoS classes accord-
and power consuming (limited autonomy) terminals, com- ing to the nominal rate of their connections: 128, 256, and
pletely inappropriate for mobile use. On the contrary, pro- 512 kbps. The implementation of a weighted priority CAC
viding access via a HAP located at an altitude of 20 km al- scheme, as the one proposed in [15], guarantees the provi-
lows the use of small, cost-efficient, and user-friendly de- sion of equitable service of multiple parallel flows with dif-
vices. Consequently, the stand-alone HAPS scenario consid- ferent bandwidth requirements. According to this admission
ers mobile users, which are further divided into three cate- control algorithm, the aggregate capacity of the system is di-
gories based on their mobility characteristics: highway-users, vided into a number of segments equal to the number of QoS
suburban-users, urban-users. In particular, highway-users classes. The width of each segment (i.e., the capacity percent-
move in open areas with maximum LoS probability, while, as age assigned to each QoS class) is determined by manipulat-
the city centre (suburban and urban users) is approached, the ing the desired blocking probability ratio between the QoS
higher building in combination with the narrow streets hin- classes. Thus, a new flow belonging to the QoSi class is ad-
der the LOS path and the received signal is the result of suc- mitted to the network on the basis of the bandwidth commit-
cessive reflections (multipath). Moreover, according to the ted to the particular QoSi class. Instead, in the case of a CAC
channel model, even in the case of a highway user, the av- scheme without any prioritization based on QoS class, the
erage PER is much higher than in case of a fixed user that is users of the higher QoS classes would be practically excluded
served by a GEO system. from the network, as it would be difficult to satisfy their ex-
cessive bandwidth needs and they would be usually blocked
Integrated GEO satellite—HAP in favor of users with lower data rate requirements. There-
fore, a weighted priority CAC scheme as defined in [15] has
The rationale behind the integration of the two systems is been taken as reference in our analysis presented hereafter.
that one satellite can provide connectivity to multiple HAPs Furthermore, the TCP driven CAC scheme has been derived
both among each other and toward the core network, with- from exactly the same notion, with the only difference that,
out the deployment of extra infrastructures. In this case, as as it has been described in Section 2, the TCP-CAC algorithm
described in Section 3.4, the PER of the end-to-end link is de- takes into account the TCP feedback of the flows instead of
termined by the PER of the user-HAP segment (equal to the their nominal data rate in the process of computing the uti-
of stand-alone HAP system), while the long RTT is imposed lization of the channel and the availability of resources.
by the GEO-satellite segment (equal to the case of stand- Both the TCP driven and the reference CAC scheme, have
alone GEO system). been simulated through an offline combination of two sim-
As it becomes evident, the scenario involving a stand- ulation tools that run sequentially. In particular, the network
alone GEO system with fixed users presents the highest RTT simulator ns-2 [16] is used to configure the communica-
and the lowest PER, while the scenario involving a stand- tion scenario (nodes, link parameters, and communication
alone HAP system presents the highest PER and the lowest protocols) and to obtain TCP statistics. Additionally, a C++
6 EURASIP Journal on Wireless Communications and Networking
100
100
Decrease in blocking probability (%)
90
90
40 40
30 30
7000 8000 9000 10000 11000 12000 13000 0.1 1 10
Average traffic load Blocking probability (%)
GEO GEO, TCP-CAC GEO-HAPS, basic-CAC
GEO-HAPS GEO, basic-CAC HAPS, TCP-CAC
HAPS GEO-HAPS, TCP-CAC HAPS, basic-CAC
Figure 7: Blocking probability decrease versus average traffic load. Figure 8: Average throughput versus blocking probability.
This means that users of the GEO-HAPS network leave a that the occupancy of the network capacity is equal to the
great percentage of the system resources unutilized and thus, aggregate of the nominal rates of all the active users. Conse-
in comparison with other architectures, the number of users quently, the requests for new connections are rejected while
that can be simultaneously served by a channel of given ca- there is still spare bandwidth. The average throughput for a
pacity is much higher (lower blocking probability). given blocking probability relies now upon the amount of
Moreover, results shown in Figure 7 lead us to the con- TCP data rate degradation. Therefore, the stand-alone GEO
clusion that the exploitation of TCP-feedback is much more case presents the higher average throughput and the inte-
crucial in a stand-alone HAP system (high PER, low RTT) grated GEO-HAPS architecture the minimum one, as they
than in a stand-alone GEO (high RTT, low PER) configura- present, respectively, the minimum and the maximum de-
tion. Then, an error prone communication path, even with crease in the TCP data rate.
low RTT, can cause abrupt decrease in the connection trans- Finally, according to Figure 9 the lower the network
mission rate. blocking probability is, the higher the gain from the utiliza-
Blocking probability and average throughput are the two tion of the TCP feedback is. Moreover, the gain for the sce-
main metrics of the network performance, each dealing with narios with the worst reception conditions is higher, since the
the issue of the system efficiency from a different perspec- basic-CAC algorithm severely limits the system throughput.
tive. Blocking probability must be minimized to maximize
the QoS (minimum delay) guaranteed to the users, while av-
erage throughput must be maximized to maximize revenues 5. CONCLUSIONS AND FUTURE PERSPECTIVES
for the network administrator. Figure 8 shows that there is
always a tradeoff between these two factors: increased aver- New and innovative wireless telecommunication architec-
age throughput leads to increased blocking probability, while tures (including HAPs and satellite segments) are identified
limitation of the blocking probability results in a low band- to provide broadband services in a cost-efficient and ubiqui-
width utilization. In addition, from Figure 8 it is evident that, tous manner, ensuring seamless interoperation with the ex-
regardless of the network scenario, the implementation of isting infrastructure. To ensure network efficiency for such
the integrated TCP-CAC scheme results in the same average architectures it is worth optimizing the performance of pro-
throughput for any given value of blocking probability. This tocols originally designed for terrestrial networks and for
is due to the fact that the admission control algorithm bases classical architectures. Cross-layer techniques are becoming
the acceptance/rejection decision upon the knowledge of the fundamental to cope with the dynamic variations character-
real traffic load forwarded at that given time through the net- izing wireless environments. The present paper focuses on
work. optimal utilization of the precious wireless resources when
Therefore, a new connection is blocked only if there is flows running TCP share the channel. Referring to 5 differ-
no further available bandwidth. Thus, since the availability ent architectures based on HAP/satellite links, we have an-
of resources occurs on the basis of the new connection nom- alyzed the potential drawbacks leading to suboptimal end-
inal rate, for a given throughput, the blocking probability is to-end performance. A TCP driven CAC scheme has been
the same for all the possible scenarios (GEO, GEO-HAPS, proposed in order to guarantee QoS for multimedia ser-
HAPS). On the contrary, the basic-CAC algorithm assumes vices with different bandwidth requirements, guarantee an
8 EURASIP Journal on Wireless Communications and Networking
REFERENCES
[1] T. C. Tozer and D. Grace, “High-altitude platforms for wireless
communications,” Electronics and Communication Engineering
Journal, vol. 13, no. 3, pp. 127–137, 2001.
[2] S. Uskela, “Key concepts for evolution toward beyond 3G net-
works,” IEEE Wireless Communications, vol. 10, no. 1, pp. 43–
48, 2003.
[3] J. Postel, “Transmission Control Protocol,” IETF RFC 793,
September 1981.
[4] W. Stevens, TCP/IP Illustrated. Volume 1: The Protocols,
Addison-Wesley, Reading, Mass, USA, 1994.
[5] C. Partridge and T. J. Shepard, “TCP/IP performance over
satellite links,” IEEE Network, vol. 11, no. 5, pp. 44–49, 1997.
[6] P. Loreti, M. Luglio, R. Kapoor, et al., “Satellite systems per-
formance with TCP-IP applications,” in Proceedings of IEEE
Military Communications Conference on Communications for
Network-Centric Operations: Creating the Information Force
(MILCOM ’01), vol. 2, pp. 811–815, McLean, Va, USA, Oc-
tober 2001.
[7] W. Stevens, “TCP Slow Start, Congestion Avoidance, Fast re-
transmit, and Fast recovery Algorithms,” Internet RFC 2001,
1997.
[8] H. G. Perros and K. M. Elsayed, “Call admission control
schemes: a review,” IEEE Communications Magazine, vol. 34,
no. 11, pp. 82–91, 1996.
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 72626, 20 pages
doi:10.1155/2007/72626
Research Article
Efficient Delay Tracking Methods with Sidelobes
Cancellation for BOC-Modulated Signals
Institute of Communications Engineering, Tampere University of Technology, P.O. Box 553, 33101 Tampere, Finland
In positioning applications, where the line of sight (LOS) is needed with high accuracy, the accurate delay estimation is an im-
portant task. The new satellite-based positioning systems, such as Galileo and modernized GPS, will use a new modulation type,
that is, the binary offset carrier (BOC) modulation. This type of modulation creates multiple peaks (ambiguities) in the envelope
of the correlation function, and thus triggers new challenges in the delay-frequency acquisition and tracking stages. Moreover, the
properties of BOC-modulated signals are yet not well studied in the context of fading multipath channels. In this paper, sidelobe
cancellation techniques are applied with various tracking structures in order to remove or diminish the side peaks, while keep-
ing a sharp and narrow main lobe, thus allowing a better tracking. Five sidelobe cancellation methods (SCM) are proposed and
studied: SCM with interference cancellation (IC), SCM with narrow correlator, SCM with high-resolution correlator (HRC), SCM
with differential correlation (DC), and SCM with threshold. Compared to other delay tracking methods, the proposed SCM ap-
proaches have the advantage that they can be applied to any sine or cosine BOC-modulated signal. We analyze the performances of
various tracking techniques in the presence of fading multipath channels and we compare them with other methods existing in the
literature. The SCM approaches bring improvement also in scenarios with closely-spaced paths, which are the most problematic
from the accurate positioning point of view.
Copyright © 2007 Adina Burian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
into the range of the ±1 chips around the maximum peak This scheme employs only the prompt correlator and in pres-
of the correlation envelope. Since the receiver can lock on ence of multipath, it has an unbiased tracking error, unlike
a sidelobe peak, the tracking process has to cope with these the narrow or strobe correlators schemes, which have a bi-
false lock points. In conclusion, the acquisition and track- ased tracking error due to the nonsymmetric property of the
ing processes should counteract all these effects, and different correlation output. However, the performance measure was
methods have been proposed in literature, in order to allevi- solely based on the multipath error envelope curves, thus its
ate multipath propagation and/or side-peaks ambiguities. potential in more realistic multipath environments is still an
In order to minimize the influence of multipath errors, open issue. One algorithm proposed to diminish the effect
which are the dominating error sources for many GNSS ap- of multipath for GPS application is the multipath estimating
plications, several receiver-internal correlation approaches delay locked loop (MEDLL) [15]. This method is different in
have been proposed. During the 1990’s, a variety of receiver that it is not based on a discriminator function, but instead
architectures were introduced in order to mitigate the multi- forms estimates of delay and phase of direct LOS signal com-
path for GPS C/A code or GLONASS. The traditional GPS re- ponent and of the indirect multipath components. It uses
ceiver employs a delay-lock loop (DLL) with a spacing Δ be- a reference correlation function in order to determine the
tween the early and late correlators of one chip. However, due best combinations of LOS and NLOS components (i.e., am-
to presence of multipath, this wide DLL, which should track plitudes, delays, phases, and number of multipaths) which
the incoming signal within the receiver, is not able to align would have produced the measured correlation function.
perfectly the local code with the incoming signal, since the As mentioned above, in the case of BOC-modulated sig-
presence of multipath (within a delay of 1.5 chips) creates a nals, besides the multipath propagation problem, the side-
bias of the zero-crossing point of the S-curve function. A first lobes peaks ambiguities should be also taken into account. In
approach to reduce the influences of code multipath is the order to counteract this issue, different approaches have been
narrow correlator or narrow early minus-late (NEML) track- introduced. One method considered in [16] is the partial
ing loop introduced for GPS receivers by NovAtel [8]. Instead Sideband discriminator, which uses weighted combinations
of using a standard (wide) correlator, the chip spacing of a of the upper and lower sidebands of received signal, to obtain
narrow correlator is less than one chip (typically Δ = 0.1 modified upper and lower signals. A “bump-jumping” algo-
chips). The lower bound on the correlator spacing depends rithm is presented in [17]. The “bump-jumping” discrimi-
on the available bandwidth. Correlator spacings of Δ = 0.1 nator tracks the ambiguous offset that arises due to multi-
and Δ = 0.05 chips are commercially available for GPS. peaked Autocorrelation Function (ACF), making amplitude
Another family of tracking loops proposed for GPS are comparisons of the prompt peak with those of neighbor-
the so-called double-delta (ΔΔ) correlators, which are the ing peaks, but it does not resolve continuously the ambigu-
general name for special code discriminators which are ity issue. An alternative method of preventing incorrect code
formed by two correlator pairs instead of one [9]. Some tracking is proposed in [18]. This technique relies on sum-
well-known implementations of ΔΔ concept are the high- mation of two different discriminator S-curves (named here
resolution correlator (HRC) [10], the Ashtech’s Strobe Cor- restoring forces), derived from coherent, respectively non-
relator [11], or the NovAtel’s Pulse Aperture Correlator [12]. coherent combining of the sidebands. One drawback is that
Another similar tracking method with ΔΔ structure is the there is a noise penalty which increases as carrier-to-noise
Early1/Early2 tracking [13], where two correlators are lo- ratio (CNR) decreases, but it does not seem excessive [18]. A
cated on the early slope of the correlation function (with new approach which design a new replica code and produces
an arbitrary spacing); their amplitudes are compared with a continuously unambiguous BOC correlation is described
the amplitudes of an ideal reference correlation function and in [19].
based on the measured amplitudes and reference amplitudes, The methods proposed in [16–19] tend to destroy the
a delay correction factor is calculated. The Early1/Early2 sharp peak of the ACF, while removing its ambiguities. How-
tracker shows the worst multipath performance for short- ever, for accurate delay tracking, preserving a sharp peak of
and medium-delay multipath compared to the HRC or the the ACF is a prerequisite. An innovative unambiguous track-
Strobe Correlator [9]. ing technique, that keeps the sharp correlation of the main
The early late slope technique [9], also called Multipath peak, is proposed in [20]. This approach uses two correlation
Elimination Technology, is based on determining the slope channels, completely removing the side peaks from the corre-
at both sides of autocorrelation function’s central peak. Once lation function. However, this method is verified for the par-
both slopes are known, they can be used to perform a pseu- ticular case of SinBOC(n, n) modulated signals, and its ex-
dorange correction. Simulation results showed that in multi- tension to other sine or cosine BOC signals is not straightfor-
path environments, the early late slope technique is outper- ward. A similar method, with a better multipath resistance, is
formed by HRC and Strobe correlators [9]. Also, it should introduced in [21].
be mentioned that in cases of Narrow Correlator, ΔΔ, early- Another approach which produces a decrease of sidelobes
late slope, or Early1/Early2 methods the BOC(n, n) modu- from ACF is the differential correlation method, where the
lated signal outperforms the BPSK modulated signals, for correlation is performed between two consecutive outputs of
multipath delays greater than approximately 0.5 chips (long- coherent integration [22].
delay multipath) [9]. A scheme based on the slope differen- In this paper, we analyze in details and develop further a
tial of the correlation function has been proposed in [14]. novel class of tracking algorithms, introduced by authors in
Adina Burian et al. 3
BOC-modulated
[23]. These techniques are named the sidelobes cancellation
1
methods (SCM), because they are all based on the idea of
code
suppressing the undesired lobes of the BOC correlation en- 0
velope and they cope better with the false lock points (ambi-
−1
guities) which appear due to BOC modulation, while keeping 0 1 2 3 4 5
the sharp shape of the main peak. It can be applied in both Chips
acquisition and tracking stages, but due to narrow width of
PRN sequence (NBOC1 = 1)
the main peak, only the tracking stage is considered here.
In contrast with the approach from [20] (valid only for sine
BOC-modulated
BOC(n, n) cases), our methods have the advantage that they 1
code
can be generalized to any sine and cosine BOC(m, n) modu- 0
lation and that they have reduced complexity, since they are
based on an ideal reference correlation function, stored at re- −1
0 1 2 3 4 5
ceiver side. In order to deal with both sidelobes ambiguities Chips
and multipath problems, we used the sidelobes cancellation
idea in conjunction with different discriminators, based on NBOC1 = 2
the unambiguous shape of ACF (i.e., the narrow correlator,
BOC-modulated
the high resolution correlator), or after applying the differ- 1
code
ential correlation method. We also introduced here an SCM 0
method with multipath interference cancellation (SCM IC),
where the SCM is used in combination with a MEDLL unit, −1
0 1 2 3 4 5
and also an SCM algorithm based on threshold comparison.
Chips
This paper is organized as follows: Section 2 describes the
signal model in the presence of BOC modulation. Section 3 NBOC1 = 3
presents several representative delay tracking algorithms,
employed for comparison with the SCM methods. Section 4 Figure 1: Examples of time-domain waveforms for sine BOC-
introduces the SCM ideas and presents the SCM usage in modulated signals.
conjunction with other delay tracking algorithms or based
solely on threshold comparison. The performance evalua-
tion of the new methods with the existing delay estimators,
in terms of root mean square error (RMSE) and mean time SinBOC(15, 10) (odd BOC-modulation order NBOC1 = 3)
to lose lock (MTLL), is done in Section 5. The conclusions together with the original PRN sequence (NBOC1 = 1) are
are drawn in Section 6. shown in Figure 1. In order to consider the cosine BOC-
modulation case, a second BOC-modulation order NBOC2 =
2 has been defined in [25], in a way that the case of sine BOC-
2. SIGNAL MODEL IN PRESENCE OF modulation corresponds to NBOC2 = 1 and the case of cosine
BOC MODULATION BOC modulation corresponds to NBOC2 = 2 (see the expres-
sions of (1) to (4)). After spreading and BOC modulation,
At the transmitter, the data sequence is first spread and the the data sequence is oversampled with an oversampled factor
pseudorandom (PRN) sequence is further BOC-modulated. of Ns , and this oversampling determines the desired accuracy
The BOC modulation is a square subcarrier modulation, in the delay estimation process. Thus, the oversampling fac-
where the PRN signal is multiplied by a rectangular sub- tor Ns represents the number of samples per BOC interval,
carrier which has a frequency multiple of code frequency. A and one chip will consists of NBOC1 NBOC2 Ns samples (i.e, the
BOC-modulated signal (sine or cosine) creates a split spec- chip period is Tc = Ns NBOC1 NBOC2 Ts , where Ts is the sam-
trum with the two main lobes shifted symmetrically from the pling rate).
carrier frequency by a value of the subcarrier frequency fsc The BOC-modulated signal sn,BOC (t) can be written, in
[5]. its most general form, as a convolution between a PRN se-
The usual notation for BOC modulation is BOC( fsc , fc ), quence sPRN (t) and a BOC waveform sBOC (t) [25]:
where fc is the chip frequency. For Galileo signals, the
BOC(m, n) notation is also used [5], where the sine and co-
sine BOC modulations are defined via two parameters m and sn,BOC (t)
n, satisfying the relationships m = fsc / fref and n = fc / fref , +∞
SF
where fref = 1.023 MHz is the reference frequency [5, 24]. = bn (−1)nNBOC1 ck,n sBOC t − nT − kTc
From the point of view of equivalent baseband signal, BOC n=−∞ k=1
modulation can be defined via a single parameter, denoted +∞
SF
by the BOC-modulation order NBOC1 = 2m/n = 2 fsc / fc . The = sBOC (t) ⊗ bn ck,n (−1)nNBOC1 δ t − nT − kTc
factor NBOC1 is an integer number [25]. n=−∞ k=1
Examples of sine BOC-modulated waveforms for Sin- = sBOC (t) ⊗ sPRN (t),
BOC(1, 1) (even BOC-modulation order NBOC1 = 2) and (1)
4 EURASIP Journal on Wireless Communications and Networking
where bn is the nth complex data symbol, T is the symbol Examples of PSD for different BOC-modulated signals
0
period (or code epoch length) (T = SF Tc ), ck,n is the kth
chip corresponding to the nth symbol, Tc = 1/ fc is the chip
−20
period, SF is the spreading factor (i.e., for GPS C/A signal
and Galileo OS signal, SF = 1023), δ(t) is the Dirac pulse,
−40
⊗ is the convolution operator and sPRN (t) is the pseudo-
PSD (dB/Hz)
random (PRN) code sequence (including data modulation)
−60
of satellite of interest, and sBOC (·) is the BOC-modulated
signal (sine or cosine) whose expression is given in (2) to
−80
(4). We remark that the term (−1)nNBOC1 is included to take
into account also odd BOC-modulation orders, similar with
[26]. The interference of other satellites is modeled as addi- −100
⎪
⎪ NBOC1 πt
⎪
⎨sign sin for SinBOC, Figure 2: Examples of baseband PSD for BOC-modulated signals.
ssin / CosBOC (t) = ⎪ Tc
⎪
⎪
NBOC1 πt
⎩sign cos for CosBOC,
Tc
(2) SinBOC(15, 10)), the interference around the DC frequency
is not completely suppressed.
respectively, that is, for SinBOC-modulation [25], The baseband model of the received signal r(t) via a fad-
ing channel can be written as [25]
NBOC1 −1
Tc n=
+∞
L
sSinBOC (t) = (−1)i pTB1 t − i , (3) r(t) = Eb e+ j2π fD t bn αn,l (t)
i=0
NBOC
n=−∞ l=1 (6)
and for CosBOC-modulation [25], × sn,sin / CosBOC t − τl + η(t),
NBOC1 −1 NBOC2 −1
where Eb is the bit or symbol energy of signal (one symbol is
sCosBOC (t) = (−1) i+k equivalent with a code epoch and typically has a duration of
i=0 k=0 (4) T = 1 ms), fD is the Doppler shift introduced by channel, L
is the number of channel paths, αn,l is the time-varying com-
Tc Tc
× pTB t − i −k . plex fading coefficient of the lth path during the nth code
NBOC1 NBOC1 NBOC2
epoch, τl is the corresponding path delay (assuming to be
In (3) and (4), pTB1 (·) is a rectangular pulse of sup- constant or slowly varying during the observation interval)
port Tc /NBOC1 and pTB (·) is a rectangular pulse of support and η(·) is the additive noise component which incorporates
Tc /NBOC1 NBOC2 . For example, the additive white noise from the channel and the interfer-
⎧ ence due to other satellites.
⎪
⎨1 Tc At the receiver, the code-Doppler acquisition and track-
if 0 ≤ t < ,
pTB (t) = ⎪ NBOC1 NBOC2 (5) ing of the received signal (i.e., estimating the Doppler shift fD
⎩0 otherwise. and the channel delay τl ) are based on the correlation with a
reference signal sref (t − τ, fD , n1 ), including the PRN code and
We remark that the bandlimiting case can also be taken into the BOC modulation (here, n1 is the considered symbol in-
account, by setting pTB (·) to be equal to the pulse shaping dex):
filter.
Some examples of the normalized power spectral den- sref t − τ, fD , n1
sity (PSD), computed as in [25], for several sine and cosine
SF NBOC1 −1 NBOC2 −1
BOC-modulated signals, are shown in Figure 2. It can be ob- = e− j2π fD t ck,n1 (−1)i+ j pTB
served that for even-modulation orders such as SinBOC(1, 1) k=−1 i=0 j =0
or CosBOC(10, 5) (currently selected or proposed by Galileo
Tc Tc
Signal Task Force), the spectrum is symmetrically split into t − n1 T − kTc − i −j − τ .
NBOC1 NBOC1 NBOC2
two parts, thus moving the signal energy away from DC fre- (7)
quency and thus allowing for less interference with the exist-
ing GPS bands (i.e., the BPSK case). Also, it should be men- Some examples of the absolute value of the ideal ACF for
tioned that in case of an odd BOC-modulation order (i.e., several BOC-modulated PRN sequences, together with the
Adina Burian et al. 5
BPSK case, are illustrated in Figure 3. As it can be observed, Ideal ACF for BOC-modulated signals
1
for any BOC-modulated signal, there are ambiguities within
the ±1 chips interval around the maximum peak. 0.9
After correlation, the signal is coherently averaged over 0.8
Nc ms, with the maximum coherence integration length dic-
0.7
tated by the coherence time of the channel, by possible resid-
Normalized ACFs
ual Doppler shift errors and by the stability of oscillators. If 0.6
the coherent integration time is higher than the coherence 0.5
time of the channel, the spectrum of the received signal will
be severely distorted. The Doppler shift due to satellite move- 0.4
ment is estimated and removed before performing the coher- 0.3
ent integration. For further noise reduction, the signal can be
0.2
noncoherently averaged over Nnc blocks; however there are
some squaring losses in the signal power due to noncoher- 0.1
ent averaging. The delay estimation is performed on a code- 0
Doppler search space, whose values are averaged correlation −1.5 −1 −0.5 0 0.5 1 1.5
functions with different time and frequency lags, with max- Chips
ima occurring at f = fD and τ = τl . BPSK SinBOC (15, 10)
SinBOC (1, 1) CosBOC (10, 5)
3. EXISTING DELAY ESTIMATION ALGORITHMS IN
Figure 3: Examples of absolute value of the ACF for BOC-
MULTIPATH CHANNELS modulated signals.
The presence of multipath is an important source of error
for GPS and Galileo applications. As mentioned before, tra-
ditionally, the multipath delay estimation block is imple- branches are combined noncoherently, and the S-curve is ob-
mented via a feedback loop. These tracking loop methods are tained as in (9),
based on the assumption that a coarse delay estimate is avail- 2 2
able at receiver, as result of the acquisition stage. The tracking SNEML (τ) = RLate (τ) − |REarly (τ) . (9)
loop is refining this estimate by keeping the track of the pre- The error signal given by the S-curve is fed back into
vious estimate. a loop filter and then into a numeric controlled oscilla-
tor (NCO) which advances or delays the timing of the ref-
3.1. Narrow early minus late (NEML) correlator erence signal generator. Figure 4 illustrates the S-curve in
single path channel, for BPSK, SinBOC(1, 1), respectively,
One of the first approaches to reduce the influences of code SinBOC(10, 5) modulated signals. The zerocrossing shows
multipath is the narrow early minus late correlation method, the presence of channel path, that is, the zero delay er-
first proposed in 1992 for GPS receivers [8]. Instead of us- ror corresponds to zero feedback error. However, for BOC-
ing a standard correlator with an early late spacing Δ of 1 modulated signals, due to sidelobes ambiguities, the early late
chip, a smaller spacing (typically Δ = 0.1 chips) is used. spacing should be less than the width of the main lobe of
Two correlations are performed between the incoming sig- the ACF envelope, in order to avoid the false locks. Typically,
nal r(t) and a late (resp., early) version of the reference code for BOC(m, n) modulation, this translates to approximately
srefEarly,Late (t − τ ± Δ/2), where srefEarly,Late (·) is the advanced or Δ ≤ n/4m.
delayed BOC-modulated PRN code and τ is the tentative
delay estimate. The early (resp., late) branch correlations 3.2. High-resolution correlator (HRC)
Rearly,Late (·) can be written as
Ideal S-curve (no multipath) for Ideal S-curve (no multipath) for two BOC-modulated signals
BOC-modulated and BPSK signals 1
1 0.8
0.8
0.6
0.6
0.4
Normalized S-curve
0.4
Normalized S-curve
0.2
0.2
0
0
−0.2
−0.2
−0.4
−0.4
−0.6
−0.6
−0.8
−0.8
−1
−1 −1.5 −1 −0.5 0 0.5 1 1.5
−1.5 −1 −0.5 0 0.5 1 1.5 Delay error (chips)
Delay error (chips)
SinBOC (1, 1)
SinBOC (1, 1) SinBOC (10, 5)
SinBOC (10, 5)
BPSK Figure 6: Ideal S-curves for noncoherent HRC with a = −1/2, for
two BOC-modulated signals and Δ = 0.1 chips.
Figure 4: Ideal S-curves for BOC-modulated and BPSK signals
(NEML, Δ = 0.1 chips).
Ideal ACFs (no multipath) for SinBOC (1, 1)-modulated signal and thus this method has a potential in reducing the side
1
peaks ambiguities.
0.9
0.8 3.5. Nonambiguous BOC(n, n) signal tracking
0.7
(Julien&al. method)
Normalized ACF
1 Tc 1 Tc
Non-coherent integration Rideal
BOC (τ) = ΛTc /2 (τ) − ΛTc /2 τ − − ΛTc /2 τ + ,
2 2 2 2
Differential correlation (12)
Figure 7: Envelope correlation function of traditional noncoher-
ent integration and differential correlation for a SinBOC(1, 1)- where ΛTc /2 (τ − α) is the value in τ of a triangular function1
modulated signal. centered in α, with a width of 1-chip, Tc is the chip period,
and τ is the code delay in chips.
The cross correlation of a SinBOC(1, 1) signal with the
the moment when introducing a new delay does not spreading pseudorandom code, for an ideal case (no multi-
improve the performance in the sense of root mean paths and ideal PRN code), can be expressed as [20]
square error between the original correlation function
1 T T
and the estimated correlation function. Rideal
BOC,PRN (τ) = ΛTc /2 τ + c + ΛTc /2 τ − c .
2 2 2
(13)
3.4. Differential correlation (DC)
Two types of DLL discriminators have been considered
Originally proposed for CDMA-based wireless communi- in [20], namely, the early-minus- late- power (EMLP) dis-
cation systems, the differential correlation method has also criminator and the dot-product (DP) discriminator. These
been investigated in context of GPS navigation system [22]. It examples of possible discriminators result from the use of
has been observed that with low and medium coherent times the combination of BOC-autocorrelation function and of
of the fading channel and in absence of any frequency error, the BOC/PRN-correlation function [20]. Based on (12) and
this approach provides better resistance to noise than the tra- (13), the ideal EMLP discriminator is constructed, as in (14),
ditional noncoherent integration methods. In DC method, where τ is the code tracking error [20]:
the correlation is performed between two consecutive out-
2 Δ 2 Δ
puts of coherent integration. These correlation variables are Sideal ideal
EMLP (τ) = RBOC τ + − Rideal
BOC τ −
then integrated, in order to obtain a differential variable. The 2 2
differential detection variable z is given as ideal2 Δ ideal2 Δ
− RBOC,PRN τ + − RBOC,PRN τ − .
2 2
M −1
1 ∗ (14)
zDC = y yk+1 2 , (11)
k
M−1 k=1 The alternative DP discriminator variant [20] does not
where yk , k = 1, . . . , M are the outputs of the coherent in- have a linear variation as a function of code tracking error:
tegration and M is the differential integration length. For a
Sideal
DP (τ)
fair comparison between the differential noncoherent and
traditional noncoherent methods, here it is assumed that 2 Δ 2 Δ 2
= Rideal
BOC τ + − Rideal
BOC τ − Rideal
BOC (τ)
M = Nnc , where Nnc is the noncoherent integration length.
2
2
Since the differential coherent correlation method was no- 2 Δ 2 Δ 2
− Rideal
BOC,PRN τ + − Rideal
BOC,PRN τ − Rideal
BOC (τ).
ticed to be more sensitive to residual Doppler errors, only 2 2
the differential noncoherent correlation is considered here. (15)
The analysis done in [22] is limited to BPSK modulation.
From Figure 7, it can be noticed that applying the DC to a
BOC-modulated signal, instead of the conventional nonco- 1 Our notation is equivalent with the notation triα (x/ y) used in [20], via
herent integration, the sidelobes envelope can be decreased, triα (τ/ y) = ΛTc /2 (τ − αTc / y).
8 EURASIP Journal on Wireless Communications and Networking
TB
are very close to zero, which means a potentially strong resis- (−1)i× j+k+l ΛTB τ + (i − j)TB + (k − l) ,
NBOC2
tance to long-delay multipath. (16)
In practice, the discriminators SEMLP (τ) or SDP (τ), as
given in [20], are formed via continuous computation, at re-
ceiver side, of correlation functions RBOC (·) and RBOC,PRN (·) where TB = Tc /NBOC1 NBOC2 is the BOC interval, ΛTB (·)
values, not on the ideal ones. In practice, RBOC (·) is the is the triangular function centered at 0 and with a width
correlation between the incoming signal (in the presence of of 2TB -chips, NBOC1 is the sine BOC-modulation order
multipaths) and the reference BOC-modulated code, and (e.g., NBOC1 = 2 for SinBOC(1, 1), or NBOC1 = 4
RBOC,PRN (·) is the correlation between the incoming signal for SinBOC(10, 5)) [25], and NBOC2 is the second BOC-
and the pseudorandom code (without BOC modulation). modulation factor which covers sine and cosine cases, as ex-
This method has been applied only to SinBOC(n, n) signals. plained in [25] (i.e., if sine BOC modulation is employed,
Moreover, instead of making use of the ideal reference func- NBOC2 = 1 and, if cosine BOC modulation is employed,
tion Rideal NBOC2 = 2).
BOC,PRN (·) (which can be computed only once and
stored at the receiver side), the correlation RBOC,PRN (·) needs As an example, the simplest case of SinBOC(1, 1)-
to be computed for each code epoch in [20]. Of course, in or- modulation (i.e., the main choice for Open Services in
der to make use of the Rideal Galileo), (16) becomes
BOC,PRN (·) shape, we also need some
information about channel multipath profile. This will be ex-
plained in the next section. Rideal
sub,SinBOC(1,1) (τ) = ΛTB τ − TB + ΛTB τ + TB , (17)
Adina Burian et al. 9
which is similar with Julien& al. expression of (13) with the CosBOC (10, 5) modulation, ACFs of
BOC-modulated and subtracted signals
exception of a 1/2 factor (here, TB = Tc /2). 1
The Sin- and CosBOC(m, n)-based ideal autocorrelation Continue line:
BOC-modulated signal
function can be written as [25]
Dashed line:
0.5 subtracted signal
NBOC1 −1 NBOC1 −1 NBOC2 −1 NBOC2 −1
Rideal
BOC (τ) =
i=0 j =0 k=0 l=0 0
−1.5 −1 −0.5 0 0.5 1 1.5
T
(−1)i+ j+k+l ΛTB τ + (i − j)TB + (k − l) B . Delay (chips)
NBOC2
(18) CosBOC (10, 5) modulation, ACF of unambiguous signal
1
Again, for SinBOC(1, 1) case, the expression of (18) reduces Unambiguous signal
to 0.5
Rideal
SinBOC(1,1) (τ) 0
= 2ΛTB (τ) − ΛTB τ − TBOC − ΛTB τ + TBOC ,
−1.5 −1 −0.5 0 0.5 1 1.5
(19)
Delay (chips)
which is, again, similar to Julien& al. expression of (12) with
Figure 9: CosBOC(10, 5)-modulated signal: examples of the am-
the exception of a 1/2 factor (for SinBOC(1, 1), TBOC = Tc /2, biguous correlation function and subtracted pulse (upper plot)
NBOC1 = 2 and NBOC2 = 1). and obtained unambiguous correlation function (lower plot), in a
We remark that the difference between (16) and (18) single-path channel.
stays in the power of −1 factor, that is, (16) stands for an ap-
proximation of the sidelobe effects (no main lobe included),
while (18) is the overall ACF (including both the main lobe
and the side lobes). The next step consists in canceling the ef- normalization of reference function (i.e., to find the weight
fect of sidelobes (16) from the overall correlation (18), after factors w), the peaks magnitudes of RBOC (·) function are first
normalizing them properly. found out and sorted in increased order. Then the weighting
Thus, in order to obtain an unambiguous ACF shape, the factor w is computed as the ratio between the last-but-one
squared function (Rideal
sin (·)) , (Rcos (·)) , respectively, has to
2 ideal 2 peak and the highest peak. We remark that the above algo-
be subtracted from the ambiguous squared correlation func- rithm does not require the computation of the BOC/PRN
tion as shown in correlation anymore, it only requires the computation of
2 2
RBOC (τ) = Rn (τ) correlation. The pulses to be subtracted are
Rideal ideal
unamb (τ) = RBOC (τ) − w Rideal
sin / cos (τ) , (20) always based on the ideal functions Ridealsin / cos (τ), and therefore,
they can be computed only once (via (16)) and stored at the
where w < 1 is a weight factor used to normalize the reference receiver (in order to decrease the complexity of the tracking
function (to achieve a magnitude of 1). unit).
For example, for SinBOC(1, 1) and w = 1, we get from By comparison with Julien&al. method, here the num-
(17), (19), and (20), after straightforward computations, that ber of correlations at the receiver is reduced by half (i.e.,
RBOC,PRN (·) computation is not needed anymore). Thus the
Rideal 2
unamb (τ) = 4 ΛTB (τ) − ΛTB (τ)ΛTB τ − TBOC SCM technique offers less computational burden (only one
(21)
− ΛTB (τ)ΛTB τ + TBOC , correlation channel in contrast to Julien&al. method, which
uses two correlation channels).
and if we plot Rideal
unamb (τ) (e.g., see the lower plot of Figure 8),
Figures 8 and 9 show the shapes of the ideal ambigu-
we get a main narrow correlation peak, without sidelobes. ous correlation functions and of the subtracted pulses, to-
All the derivations so far were based on ideal assumptions gether with the correlation functions, obtained after subtrac-
(ideal correlation codes, single path static channels, etc.). tion (SCM method). Figure 8 exemplifies a SinBOC(1, 1)-
However, in practice, we have to cope with the real signals, modulated signal, while Figure 9 illustrates the shapes for a
so the ideal autocorrelation function RidealBOC (τ) should be re-
CosBOC(10, 5)-modulation case. As it can be observed, for
placed with the computed correlation RBOC (τ) between the both SinBOC and CosBOC modulations, the subtractions
received signal and the reference BOC-modulated pseudo- removes the sidelobes closest to the main peak, which are
random code. Thus, (20) becomes the main threats in the tracking process. Also, it should be
mentioned that the Figure 8, for a SinBOC(1, 1) modulated
2 2
Runamb (τ) = RBOC (τ) − w Rideal
sin / cos (τ) . (22) signal, is also illustrative for the Julien&al. method, since the
shapes of correlation functions are similar with those pre-
Here comes into equation the weighting factor, since vari- sented in [20].
ous channel effects (such as noise and multipath) can mod- Equation (20) is valid for single path channels. How-
ify the levels of RBOC (τ) function. In order to perform the ever, in multipath presence, delay errors due to multipaths
10 EURASIP Journal on Wireless Communications and Networking
are likely to appear. When (22) is applied in this situation, Exemplification of SCM IC method (steps 1 to 4)
one important issue is to align the subtraction pulse to the 1
LOS peak (otherwise, the subtraction of (22) will not can-
cel the correct sidelobes). This can be done only if some ini- 0.8
tial estimate of LOS delay is obtained. For this purpose, we
employ and compare several feedback loops or feedforward 0.6
algorithms, as it will be explained next.
0.4
Rideal
unmab (τ) is the unambiguous reference function
given by (20). The shape of residual function is and the maximum global peak is re-estimated from
exemplified in Figure 10, lower plot (drawn with R(2)
n,unamb (τ) = (Rn,unamb (τ)) − (a
2 1,n Rideal
unamb (τ)(τ −
1,n ideal 2,n 2
continuous line). τ1,n )e j θ + a2,n Runamb (τ)(τ − τ2,n )e ) .
j θ
(6) The new maximum peak of the residual function (7) The steps (3) to (6) are repeated until all desired peaks
R(1) are estimated and until the residual function is below
n,unamb is found out (e.g., at 44th-sample delay,
Figure 10, lower plot), with its corresponding de- a threshold value. In the example of Figure 10, after 6
steps both path delays are estimated correctly.
lay τ2,n , amplitude a2,n and phase θ2,n . The con-
tributions of both peaks 1 and 2 are subtracted These steps of SCM IC method are illustrated in
from unambiguous correlation function Rn,unamb (τ) Figure 10, for 2-path fading channel.
Adina Burian et al. 11
Ideal S-curve (no multipath), SCM NEML method SinBOC (1, 1), Δ = 0.1 chips
1
envelope (meters)
Multipath error
10
0.8
0.6 0
0.4 −10
Normalized S-curve
−0.6
SinBOC (10, 5), Δ = 0.1 chips
envelope (meters)
Multipath error
−0.8 10
−1 0
−1.5 −1 −0.5 0 0.5 1 1.5
Delay error (chips) −10
Ideal S-curve (no multipath), SCM HRC method Ideal ACF (no multipath) for SinBOC (10, 5) modulated signal
1
0.8
0.8
0.6
0.4
Normalized S-curve
0.6
Normalized ACF
0.2
0 0.4
−0.2
0.2
−0.4
−0.6 0
−0.8
−0.2
−1
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
Delay error (chips) Delay error (chips)
Figure 13: SCM HRC method: ideal S-curves (no multipath), for Figure 15: Envelopes of correlation functions obtained with am-
two BOCmodulation cases, with a = −1/2 and Δ = 0.1 chips. biguous correlation, DC method, SCM approach, and SCM DC
method, for a SinBOC(10, 5)-modulated signal.
SinBOC (1, 1), Δ = 0.1 chips SinBOC(10, 5) cases. As it can be noticed, there is a slight im-
provement brought by the SCM HRC method over the HRC
10
envelope (meters)
Multipath error
correlator.
5
0
4.5. SCM using differential correlation (DC) in
−5
conjunction with feedback and feedforward
−10
tracking algorithms
0 0.2 0.4 0.6 0.8 1
Multipath spacing (chips) It has been observed that the DC method has potential to de-
crease the sidelobes amplitudes, thus lowering the possibility
HRC method
SCM HRC method
to detect a wrong side peak. To enhance the performance of
the DC method, we use it in conjunction with different track-
SinBOC (10, 5), Δ = 0.1 chips ing algorithms, such as NEML or HRC methods, or with IC
method. These algorithms are applied in similar ways as ex-
envelope (meters)
10
Multipath error
the out-of-peak values, similar to [4]. Using this estimated SinBOC (1, 1), AWGN single-path channel
100
noise variance, a linear threshold γ is computed, based on the
second peak γ2 of the ideal unambiguous correlation func-
tion Rideal 10−1
unamb (τ) (i.e., for SinBOC(1, 1) γ2 = 0.5, as seen in
Figure 3), together with the estimate of the noise variance σn2 :
10−2
RMSE (chips)
γ = γ2 + σn2 . (25)
10−3
Then the LOS delay is estimated, based on the unambigu-
ous correlation function Rn,unamb (τ), using this threshold. If
the peak of the estimated first path is too low (i.e., ten times 10−4
lower than the global peak), then this path is discarded and
the next estimate is considered. 10−5
NEML DC NEML
We first test the performance of the proposed algorithms in
Julien & al. EMLP DC SCM NEML
the ideal AWGN channel (single path), in order to check SCM NEML
whether SCM algorithm introduces a deterioration with re-
SinBOC (1, 1), AWGN single-path channel
spect to the standard narrow and high-resolution correla-
tors (it is known that NEML is able to attain the Cramer-
Rao bound in AWGN channels [8]). We will show that no
deterioration is incurred when SCM is applied. The perfor-
mance criteria are root mean square error (RMSE) and mean 102.4
time to lose lock (MTLL). The simulations were carried out
in Matlab. The MTLL is computed as the average value for
MTLL (s)
SinBOC (1, 1), AWGN single-path channel to v = 3 km/h, while for outdoor profiles, the mobile speeds
100
of 25, 45, or 75 km/h have been selected). Two main chan-
nel profiles have been considered: either with fixed Rayleigh
10−1 distribution of all paths and with average path power of −1,
−2, 0 and −3 dB, or a 2-paths decaying power delay profile
(PDP) channel, with Rician distributions for the first path
10−2
RMSE (chips)
and Rayleigh distribution for the next path. Similar with the
AWGN case in Section5.1, during simulations, the first path
delay of the channel is assumed to be linearly increasing, with
10−3
a slope of 0.05 chips per block of Nc Nnc millisecond, thus the
tracking algorithms should capture this linear delay increase.
10−4 The successive channel path delays have a random spacing
with respect to the precedent delay, uniformly distributed be-
tween 1/(Ns NBOC1 NBOC2 ) and xmax , where xmax (in chips) is
10−5 the maximum separation between successive paths (i.e., for
20 25 30 35 40
closed-spaced paths scenario, xmax = 0.1 chips). In order to
CNR (dB-Hz)
have independent and reliable results for each method, the
HRC DC HRC search interval is different for each algorithm. which means
Julien & al. DP SCM DC HRC that once the lock is lost for one method, this will not affect
SCM HRC the other algorithms. The search window has few chips (typ-
ically between 4 and 12 chips), depending on the number
SinBOC (1, 1), AWGN single-path channel of paths, the distance between them and on the used BOC-
modulation orders. The search window is sliding around the
350 previous delay estimate and if we have erroneous estimates,
the lock is lost at some point. For the feedback algorithms
(i.e., NEML, HRC, or Julien&al. methods), the search for
zero crossing is conditioned by the previous delay estimates.
Similar with AWGN case, he coherent integration length is
200 set to Nc = 20 milliseconds, the noncoherent integration is
MTLLS (s)
SinBOC (1, 1), Rayleigh fading channel SinBOC (1, 1), Rayleigh channel,
with 2 paths, xmax = 1 chip speed mobile = 75 km/h, xmax = 0.1 chips
0.6
RMSE (chips)
10−0.4
ACFs
0.4
0.2
10−0.5
0
−0.2 10−0.6
40 60 80 100 120 20 25 30 35 40
Delay error (samples) CNR (dB-Hz)
0.4
0.2
−0.2
20 25 30 35 40
180 200 220 240 260 280 300 320
CNR (dB-Hz)
Delay error (samples)
NEML DC NEML
Ambiguous ACF Julien & al. EMLP SCM DC NEML
Subtracted pulse SCM NEML
Unambiguous ACF
Figure 19: Comparison of feedback delays estimation algorithms
Figure 18: Exemplification of SCM method for a 2-paths Rayleigh employing the NEML discriminator and of the Julien&al. method,
fading channel. Upper plot: SinBOC(1, 1)-modulated signal and as a function of CNR; SinBOC(1, 1) modulation, Rayleigh channel
xmax = 1 chip. Lower plot: SinBOC(10, 5)-modulated signal and with an average pathspower delay profile of −1, −2, 0, and −3 dB,
xmax = 0.5 chips. v = 75 km/h, closely spaced paths with xmax = 0.1 chips.
SinBOC (1, 1), 2-paths Rician channel, CosBOC (10, 5), Rayleigh channel,
xmax = 1 chip, mobile speed = 25 km/h xmax = 0.1 chips, mobile speed = 45 km/h
10−0.3
10−0.3
10−0.4
RMSE (chips)
RMSE (chips)
10−0.4 10−0.5
10−0.6
10−0.5
10−0.7
20 25 30 35 40 20 25 30 35 40
CNR (dB-Hz) CNR (dB-Hz)
SinBOC (1, 1), 2-paths Rician channel, CosBOC (10, 5), 4-paths Rayleigh channel,
xmax = 1 chip, mobile speed = 25 km/h xmax = 0.1 chips
MTLL (s)
MTLL (s)
101 101
20 25 30 35 40
20 25 30 35 40
CNR (dB-Hz)
CNR (dB-Hz)
HRC DC HRC
HRC DC HRC
Julien & al. DP SCM DC HRC
Julien & al. DP SCM DC HRC
SCM HRC
SCM HRC
Figure 21: Comparison of feedback delays estimation algorithms
Figure 20: Comparison of feedback delays estimation algorithms employing the HRC discriminator and of the Julien&al. method, as
employing the HRC discriminator and of the Julien&al. method, as a function of CNR; CosBOC(10, 5) modulation, 4-paths Rayleigh
a function of CNR; SinBOC(1, 1) modulation, 2-paths Rician chan- channel, with paths PDP of −1, −2, 0, and −3 dB, v = 45 km/h,
nel with decaying PDP of 0 and −2 dB, v = 25 km/h, maximum closely spaced paths xmax = 0.1 chips.
separation between paths xmax = 1 chip.
signal, for a 4-paths Rayleigh channel, with closely spaced method exceeds those of HRC and SCM HRC algorithms,
paths xmax = 0.1 chips and v = 45 km/h. which both give similar results. On the other hand, for the
From all plots of Figures 20 and 21, it can be ob- CosBOC(10, 5) modulation, the Julien& al. DP method ap-
served that, in both RMSE and MTLL terms, there is a proaches the results provided by the HRC and SCM HRC
small improvement brought by the DC HRC and SCM DC algorithms, which still offer a deterioration in performance
HRC methods, which have similar performance. For the of about 1 dB, comparing to DC HRC and SCM DC HRC
SinBOC(1, 1) case, the performance of the Julien& al. DP methods.
Adina Burian et al. 17
SinBOC (1, 1), Rayleigh channel, SinBOC (1, 1), 2-paths Rician channel,
speed mobile = 3 km/h, xmax = 0.1 chips xmax = 0.5 chips, mobile speed = 45 km/h
100
10−0.3
RMSE (chips)
RMSE (chips)
10−0.5
10−0.7
10−1
10−0.9
20 25 30 35 40 20 25 30 35 40
CNR (dB-Hz) CNR (dB-Hz)
MEDLL DC IC MEDLL DC IC
SCM IC SCM DC IC SCM IC SCM DC IC
SCM thr. SCM thr.
SinBOC (1, 1), Rayleigh channel, SinBOC (1, 1), 2-paths Rician channel,
4 paths, xmax = 0.1 chips xmax = 0.5 chips, mobile speed = 45 km/h
102
101
MTLL (s)
MTLL (s)
101
100 100
20 25 30 35 40 20 25 30 35 40
CNR (dB-Hz) CNR (dB-Hz)
MEDLL DC IC MEDLL DC IC
SCM IC SCM DC IC SCM IC SCM DC IC
SCM thr. SCM thr.
Figure 22: Comparison of feedforward delays estimation algo- Figure 23: Comparison of feedforward delays estimation algo-
rithms employing the MEDLL and IC methods and of the SCM rithms employing the MEDLL and IC methods and of the SCM with
with threshold approach, as a function of CNR; SinBOC(1, 1) mod- threshold approach, as a function of CNR; SinBOC(1, 1) modula-
ulation, 4-paths indoor Rayleigh channel, with PDP of −1, −2, 0, tion, 2-paths decaying PDP Rician channel, v = 45 km/h, xmax = 0.5
and −3 dB, v = 3 km/h, closely spaced paths with xmax = 0.1 chips. chips.
CosBOC (10, 5), Rayleigh channel, CosBOC (10, 5), Rician channel, 2 paths, xmax = 0.5 chips
speed mobile = 3 km/h, xmax = 0.1 chips 102
10−0.3
10−0.4
101
MTLL (s)
10−0.5
RMSE (chips)
10−0.6
100
10−0.7
10−0.8 10−1
20 25 30 35 40
20 25 30 35 40 CNR (dB-Hz)
CNR (dB-Hz)
MEDLL DC IC
MEDLL DC IC SCM IC SCM DC IC
SCM IC SCM DC IC SCM thr.
SCM thr.
Figure 25: Comparison of feedforward delays estimation algo-
rithms employing the MEDLL and IC methods and of the SCM with
CosBOC (10, 5), 4-paths Rayleigh channel, threshold approach, as a function of CNR; CosBOC(10, 5) modula-
xmax = 0.1 chips tion, 2-paths decaying PDP Rician channel, v = 45 km/h, xmax = 0.5
102 chips.
SinBOC (1, 1), 4-paths Rayleigh channel, SCM does not bring substantial improvement, since HRC
CNR = 35 dBHz, v = 4 km/h has already rather good performance in multipath channels.
10−0.3 Also, the higher BOC-modulation order, the more advanta-
geous is to apply SCM technique in order to cope better with
the false lock points.
10−0.4
ACKNOWLEDGMENTS
RMSE (chips)
10−0.5 This work was carried out in the project “Advanced Tech-
niques for Personal Navigation (ATENA)” funded by the
Finnish Funding Agency for Technology and Innovation
10−0.6 (Tekes). This work has also been supported by the Academy
of Finland. The authors would like to thank the anonymous
reviewers for their valuable comments to improve this paper.
10−0.7
the Institute of Navigation (ION-GPS ’97), vol. 1, pp. 559–568, R. Lucas, “Status of Galileo frequency and signal design,”
Kansas City, Mo, USA, September 1997. in CDROM Proceedings of the International Technical Meet-
[12] J. Jones, P. Fenton, and B. Smith, “Theory and perfor- ing of the Satellite Division of the Institute of Navigation
mance of the pulse aperture correlator,” Tech. Rep., NovA- (ION-GPS ’02), Portland, Ore, USA, September 2002.
tel, Calgary, Alberta, Canada, September 2004, http://www [25] E. S. Lohan, A. Lakhzouri, and M. Renfors, “Binary-offset-
.novatel.com/Documents/Papers/PAC.pdf. carrier modulation techniques with applications in satel-
[13] A. V. Dierendonck and M. Braasch, “Evaluation of GNSS re- lite navigation systems,” Wireless Communications and Mobile
ceiver correlation processing techniques for multipath and Computing, vol. 7, no. 6, pp. 767–779, 2006.
noise mitigation,” in Proceedings of the National Technical [26] E. Rebeyrol, C. Macabiau, L. Lestarquit, et al., “BOC power
Meeting of the Institute of Navigation (ION-NTM ’97), pp. spectrum densities,” in CDROM Proceedings of the National
207–215, Santa Monica, Calif, USA, January 1997. Technical Meeting of Institute of Navigation (ION-NTM ’05),
[14] C. Lee, S. Yoo, S. Yoon, and S. Y. Kim, “A novel multipath San Diego, Calif, USA, January 2005.
mitigation scheme based on slope differential of correlator
output for Galileo systems,” in Proceedings of the 8th Inter-
national Conference on Advanced Communication Technology
(ICACT ’06), vol. 2, pp. 1360–1363, Phoenix Park, Korea,
February 2006.
[15] R. van Nee, J. Siereveld, P. Fenton, and B. Townsend, “The
multipath estimating delay lock loop: approaching theoreti-
cal accuracy limits,” in Proceedings of IEEE Position Location
and Navigation Symposium, pp. 246–251, Las Vegas, Nev, USA,
April 1994.
[16] P. A. Bello and R. L. Fante, “Code tracking performance for
novel unambiguous M-code time discriminators,” in Proceed-
ings of the National Technical Meeting of the Institute of Nav-
igation (ION-NTM ’05), pp. 293–298, San Diego, Calif, USA,
January 2005.
[17] P. Fine and W. Wilson, “Tracking algorithms for GPS offset
carrier signals,” in Proceedings of the National Technical Meet-
ing of the Institute of Navigation (ION-NTM ’99), San Diego,
Calif, USA, January 1999.
[18] V. Lin, P. Dafesh, A. Wu, and C. Cahn, “Study of the impact
of false lock points in subcarrier modulated ranging signals
and recommended mitigation approaches,” in Proceedings of
the 59th ION Annual Meeting & CIGTF Guidance Test Sympo-
sium, pp. 156–165, Albuquerque, NM, USA, June 2003.
[19] P. Ward, “A design technique to remove the correlation ambi-
guity in binary offset carrier (BOC) spread spectrum signals,”
in Proceedings of the National Technical Meeting of the Institute
of Navigation (ION-NTM ’04), pp. 886–896, San Diego, Calif,
USA, January 2004.
[20] O. Julien, C. Macabiau, M. Cannon, and G. Lachapelle, “BOC
signal acquisition and tracking method and apparatus,” US
Patent Application Publication 2005/0270997 A1, December
2005.
[21] V. Heiries, J.-A. Avila-Rodriguez, M. Irsigler, G. Hein, E. Re-
beyrol, and D. Roviras, “Acquisition performance analysis of
composite signals for the L1 OS optimized signal,” in Proceed-
ings of the 18th International Technical Meeting of the Satel-
lite Division of the Institue of Navigation (ION-GNSS ’05), pp.
877–889, Long Beach, Calif, USA, September 2005.
[22] A. Schmid and A. Neubauer, “Differential correlation
for Galileo/GPS receivers,” in Proceedings IEEE Interna-
tional Conference on Acoustics, Speech, and Signal Processing
(ICASSP ’05), vol. 3, pp. 953 –956, Philadelphia, Pa, USA,
March 2005.
[23] A. Burian, E. S. Lohan, and M. Renfors, “Sidelobe cancellation
method for unambiguous tracking of binary-offset-carrier
modulated signals,” in CDROM Proceedings of the 3rd ESA
Workshop on Satellite Navigation User Equipment Technolo-
gies (NAVITEC ’06), Noordwijk, The Netherlands, December
2006.
[24] G. Hein, J. Godet, J.-L. Issler, J. C. Martin, T. Pratt, and
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2007, Article ID 25178, 11 pages
doi:10.1155/2007/25178
Research Article
Analysis of Filter-Bank-Based Methods for Fast Serial
Acquisition of BOC-Modulated Signals
Institute of Communications Engineering, Tampere University of Technology, P.O. Box 553, 33101 Tampere, Finland
Binary-offset-carrier (BOC) signals, selected for Galileo and modernized GPS systems, pose significant challenges for the code ac-
quisition, due to the ambiguities (deep fades) which are present in the envelope of the correlation function (CF). This is different
from the BPSK-modulated CDMA signals, where the main correlation lobe spans over 2-chip interval, without any ambiguities or
deep fades. To deal with the ambiguities due to BOC modulation, one solution is to use lower steps of scanning the code phases
(i.e., lower than the traditional step of 0.5 chips used for BPSK-modulated CDMA signals). Lowering the time-bin steps entails
an increase in the number of timing hypotheses, and, thus, in the acquisition times. An alternative solution is to transform the
ambiguous CF into an “unambiguous” CF, via adequate filtering of the signal. A generalized class of frequency-based unambigu-
ous acquisition methods is proposed here, namely the filter-bank-based (FBB) approaches. The detailed theoretical analysis of
FBB methods is given for serial-search single-dwell acquisition in single path static channels and a comparison is made with other
ambiguous and unambiguous BOC acquisition methods existing in the literature.
Copyright © 2007 Elena Simona Lohan. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
1. INTRODUCTION however, they are also related to the deep fades within this
interval. The terminology used here refers to the deep fades
The modulation selected for modernized GPS and Galileo of CF envelope.
signals is BOC modulation, often denoted as BOC(m, n), The number of fades or ambiguities within 2-chip inter-
with m = fsc / fref , n = fc / fref . Here, fc is the chip rate, fsc val depends on the NBOC order (e.g., for SinBOC, we have
is the subcarrier rate, and fref = 1.023 MHz is the reference 2NBOC − 2 ambiguities around the maximum peak, while for
chip frequency (that of the C/A GPS signal) [1]. Alterna- CosBOC, we have 2NBOC ambiguities [4]). The distance be-
tively, a BOC-modulated signal can also be defined via its tween successive ambiguities in the CF envelope sets an up-
BOC modulation order NBOC 2 fsc / fc [2–4]. Both sine and per bound on the step of searching the time-bin hypotheses
cosine BOC variants are possible (for a detailed description (Δt)bin , in the sense that if the time-bin step becomes too
of sine and cosine BOC properties, see [3, 4]). The acqui- high, the main lobe of the CF envelope might be lost during
sition of BOC-modulated signals is challenged by the pres- the acquisition. Typically, a step of one-half the distance be-
ence of several ambiguities in CF envelope (here, CF refers to tween the correlation peak and its first zero value, or, equiva-
the correlation between the received signal and the reference lently, one quarter of the main lobe width is generally consid-
BOC-modulated code). That is, if the so-called ambiguous- ered [9]. For example, acquisition time-bin steps of 0.5 chips
BOC (aBOC) approach [5–7] is used (meaning that there are used for BPSK modulation (such as for C/A code of GPS),
is no bandlimiting filtering at the receiver or that this filter where the width of the main lobe is 2 chips, and steps of 0.1–
has a bandwidth sufficiently high to capture most energy of 0.2 chips are used for SinBOC(1,1) modulation, where the
the incoming signal), the resultant CF envelope will exhibit width of the main lobe is about 0.7 chips (such as for Galileo
some deep fades within ±1 chip interval around the correct Open Service) [5, 10, 11].
peak [5, 8], as it will be illustrated in Section 4. We remark In order to be able to increase the time-bin step (and,
that sometimes the term “ambiguities” refers to the multi- thus, the speed of the acquisition process), several Filter-
ple peaks within ±1 chip interval around the correct peak; Bank-Based (FBB) methods are proposed here, which exploit
2 EURASIP Journal on Wireless Communications and Networking
Time uncertainty Δtmax The time-frequency bin defines the final time-frequency
error after the acquisition process and it is characterized by
. .
··· parallel (if one decision variable is formed for the whole un-
certainty space) [13]. This paper focuses on the serial search
Time-bin step (Δt)bin approach.
One of the main features of Galileo system is the intro-
Figure 1: Illustration of the time/frequency search space. duction of longer codes than those used for most GPS sig-
nals. Also, the presence of BOC modulation creates some ad-
ditional peaks in the envelope of the correlation function, as
well as additional deep fades within ±1 chip from the main
the property that by reducing the signal bandwidth before
peak. For this reason, a time-bin step of 0.5 chips is typically
correlation, we are able to increase the width of the CF
not sufficient and smaller steps need to be used [5, 10, 11].
main lobe. A thorough theoretical model is given for the
On the other hand, decreasing the time-bin step will increase
characterization of the decision variable in single-path static
the mean acquisition time and the complexity of the receiver
channels and the theoretical model is validated via sim-
[9].
ulations. The proposed FBB methods are compared with
In the serial search code acquisition process, one decision
two other existing methods in the literature: the classical
variable is formed per each time-frequency bin (based on the
ambiguous-BOC processing (above-mentioned) and a more
correlation between the received signal and a reference code),
recent, unambiguous-BOC technique, introduced by Fish-
and this decision variable is compared with a threshold in
man and Betz [9] (denoted here via B&F method, but also
order to decide whether the signal is present or absent. The
known as sideband correlation method or BPSK-like tech-
ambiguous-BOC (aBOC) processing means that, when form-
nique) and further analyzed and developed in [2, 6, 7, 10, 11].
ing the decision variable, the received signal is directly corre-
It is mentioned that FBB methods have also been studied by
lated with the reference BOC-modulated PRN sequence (all
the author in the context of hybrid-search acquisition [12].
the spectrum is used for both the received signal and refer-
However, the theoretical analysis of FBB methods is newly
ence code).
introduced here.
SinBOC(1, 1) spectrum
1
Normalized PSD
0.8
0.6
0.4
0.2
0
−4 −3 −2 −1 0 1 2 3 4
Frequecy (MHz)
Figure 2: Block diagram of B&F method, single-sideband processing (here, upper sideband).
similar manner with the received signal, hence the autocor- since the noise power depends on the filter bandwidth (i.e.,
relation function is no longer the CF of a BOC-modulated the noise power is constant from one band to another for
signal, but it will resemble the CF of a BPSK-modulated sig- the FBBefw case, and it is variable for the FBBep case). How-
nal. However, the exact shape of the resulting CF is not iden- ever, the incoming (filtered) signal is correlated with the ref-
tical with the CF of a BPSK-modulated signal, since some in- erence BOC-modulated code. Thus, the noise, which may
formation is lost when filtering out the sidelobes adjacent to be assumed white before the correlation, becomes coloured
the main lobe (this is exemplified in Section 4). This filtering noise after the correlation with BOC signal, and its spectrum
is needed in order to reduce the noise power. When the B&F (after the correlation) takes the shape of the BOC power
dual-sideband method is used, we add together the USB and spectral density. Therefore, after the correlation stage at the
LSB outputs and form the dual-sideband statistic. receiver (e.g., immediately before the coherent integration
block), both signal power density and noise power density
4. FILTER-BANK-BASED (FBB) METHODS are shaped by the BOC spectrum. Thus, the denominations
used here (FBBep and FBBefw ) are suited for both signal and
The underlying principle of the proposed FBB methods is noise parts, as long as the focus is on the processing after the
illustrated in Figure 3 and the block diagram is shown in correlation stage (as it is the case in the acquisition).
Figure 4. The number of filters in the filter bank is denoted As seen in Figure 4, the same filter bank is applied to
by Nfb and it is related to the number of frequency pieces per both the signal and the reference BOC-modulated pseudo-
sideband Npieces via: Nfb = 2Npieces if dual sideband (SB) is random code. Then, filtered pieces of the signal are corre-
used, or Nfb = Npieces if single SB is used. In Figure 3, the lated with filtered pieces of the code (as shown in Figure 4)
upper plot shows the spectrum of a SinBOC(1,1)-modulated and an example of the resultant CF is plotted in the lower
signal, together with several filters (here Nfb = 4) which cover part of Figure 3. For reference purpose, also aBOC and B&F
the useful part of the signal spectrum (the useful part is con- cases are shown. It is noticed that, when Npieces = 1, the pro-
sidered here to be everything between the main spectral lobes posed FBB methods (both FBBep and FBBefw ) become identi-
of the signal, including these main lobes). Alternatively, we cal with B&F method, and the higher the Npieces is, the wider
may select only the upper (or lower) SB of the signal (i.e., the main lobe of the CF envelope becomes, at the expense of
single-SB processing). a higher decrease in the signal power.
The filters may have equal or unequal frequency widths. The block diagram in Figure 4 applies not only to FBB
Two methods may be employed and they have been denoted methods, but also to other GPS/Galileo acquisition meth-
here via equal-power FBB (FBBep ), where each filter lets the ods, such as single/dual SB, and ambiguous-/unambiguous-
same signal’s spectral energy to be passed, thus they have un- BOC acquisition methods (i.e., aBOC corresponds to the
equal frequency widths (see upper plot of Figure 3), or equal- case when no filtering stage is applied to the received and
frequency-width FBB (FBBefw ), where all the filters in the fil- reference signals, while B&F corresponds to the case when
ter bank have the same bandwidth (but the signal power is Npieces = 1). The complex outputs yi (·), i = 1, . . . , Nfb of the
different from one band to another). An observation ought coherent integration block of Figure 4 can be written as
to be made here with respect to these denominations: indeed,
before the correlation takes place and after filtering the in-
1 nT+Tcoh
coming signal (via the filter bank), the noise power density yi τ, fD , n = ri (t)ci (t − τ)e j2π fD t dt, (1)
is exactly in reverse situation compared to the signal power, Tcoh nT
4 EURASIP Journal on Wireless Communications and Networking
0.7
Dual sideband processing, equal-power pieces nel is available (such as it is the case of Galileo L1 band), thus
the received signal r(t) (before filtering) has the form
0.6
r(t) = Eb c(t − τ)e− j2π fD t + ηwb (t), (2)
0.5
where τ and fD are the delay and Doppler shift introduced
Spectrum
0.4 by the channel, ηwb (t) is the additive white Gaussian noise at
wideband level, and Eb is the bit energy.
0.3 The coherent integration outputs yi (·) are Gaussian pro-
cesses (since a filtered Gaussian processes is still a Gaussian
0.2
processes). Their mean is either 0 (if we are in an incorrect
time-frequency bin) or it is proportional to a time-Doppler
0.1
deterioration factor Eb F (Δτ, Δ fD ) [11], with a proportion-
0
ality constant dependent on the number of filters and of the
−3 −2 −1 0 1 2 3 acquisition algorithm, as it will be shown in Section 5. Here,
Frequency (MHz) F (·) is the amplitude deterioration in the correct bin due
Filter 3
to a residual time error Δτ and a residual Doppler error Δ fD
BOC PSD
Filter 1 Filter 4 [11]
Filter 2
sin πΔ fD Tcoh
(a)
F Δτ, Δ fD =
R Δτ
. (3)
0.5
where CNR = Eb BW /N0 is the Carrier-to-Noise Ratio, ex-
pressed in dB-Hz [5, 7, 11], BW is the signal bandwidth after
0 despreading (e.g., BW = 1 kHz for GPS and Galileo signals),
−3 −2 −1 0 1 2 3
Delay error (chips) and N0 is the double-sided noise spectral power density in
the narrowband domain (after despreading or correlation on
BOC FBBep , dual SB 1 millisecond in GPS/Galileo). The proportionality constants
B&F, dual SB FBBefw , dual SB are presented in Section 5. The decision statistic Z of Figure 4
(b) is the output of noncoherent combining of Nnc Nfb complex
Gaussian variables, where Nnc is the noncoherent integration
Figure 3: Illustration of the FBB acquisition methods, SinBOC(1,1) time (expressed in blocks of Nc ms):
case. Upper plot: division into frequency pieces, via Nfb = 4 filters
(FBBep method). Lower plot: squared CF shapes for 2 FBB meth-
1 1 nc fb
N N
ods, compared with ambiguous BOC (aBOC) and unambiguous
Z=
yi τ, fD , n
2 . (5)
Betz&Fishman (B&F) methods. Nnc Nfb n=1 i=1
Optional stage y1
Coherent Nnc
r1 (t) ||2
FB integr. . . . Nfb Z
Rx sign. .. . . .
. . .
. rNfb (t) ∗ Coherent Nnc
r(t) Nfb filters integr. yNfb ||2
∗
c1 (t)
Ref code FB .
.
c(t) . cNfb (t)
Nfb filters
5. THEORETICAL MODEL FOR FBB density (PSD) function. Pml can be easy computed analyti-
ACQUISITION METHODS cally, using, for example, the formulas for PSD given in [3, 4]
and some illustrative examples are shown in Figure 5; the
5.1. Test statistic distribution normalization is done with respect to the total signal power,
thus Pml < 0.5.; Pml factor is normalized with respect to the
As explained above, the test statistic Z for aBOC, B&F, and total signal power, thus Pml < 0.5 (e.g., Pml = 0.428 for Sin-
proposed FBBep approaches1 is either a central or a noncen- BOC(1,1)). The decrease in the signal and noise power after
tral χ 2 -distributed variable with Ndeg degrees of freedom, ac- the correlation in B&F method (and thus, the decrease in ξ λH1
cording to whether we have an incorrect (bin H0 ) or a and ξ σ 2bin parameters) is due to the fact that both the signal
correct (bin H1 ) time-frequency bin, respectively. Its non- and the reference code are filtered and the filter bandwidth is
centrality parameter λZ and its variance σ 2Z are thus given by adjusted to the width of the PSD main lobe. Also, in dual-
SB approaches, the signal power is twice the signal power
for single SB, therefore, the noncentrality parameter (which
λZ = ξ λbin
F Δτ, Δ fD
, is a measure √ of the amplitude, not of the signal power) in-
σ2 (6) creases by 2. Furthermore, in dual-SB approaches, we add
σ 2Z = ξ σ 2bin , a double number of noncoherent variables, thus the num-
Nnc
ber of degrees of freedom is doubled compared to single-SB
approaches.
where F (·) is given in (3), σ 2 is given in (4), and ξ σ 2bin and The derivation of χ 2 parameters for FBBep is also straight-
ξ λbin are two algorithm-dependent factors shown in Table 1 forward by keeping in mind that the variance of the vari-
(they also depend on whether we are in a correct bin or in an ables yi is constant for each frequency piece (the filters were
incorrect bin). We remark that the noncentrality parameter designed in such a way to let equal power to be passed
used here is the square-root of the noncentrality parameter through them). Thus, the noise power decrease factor is
defined in [17], such that it corresponds to amplitude lev- ξ σ 2bin = Pml /Npieces , bin = H0 , H1 , and the signal power de-
els (and not to power levels). The relationship between the
creases to Npieces (P 2ml /Npieces
2
), thus xλbin = Pml / Npieces for
distribution functions and their noncentrality parameter and √
variance will be given in (8). single SB (and xλbin = 2Pml / Npieces for dual SB).
All the parameters in Table 1 have been derived by in- For FBBefw , the reasoning is not so straightforward (be-
tuitive reasoning (explained below), followed by a thorough cause the sum of squares of Gaussian variables of different
verification of the theoretical formulas via simulations. For variances is no longer χ 2 distributed) and the bounds given
clarity reasons, we assumed that the bit energy is normalized in Table 1 were obtained via simulations. It was noticed (via
to Eb = 1 and all the signal and noise statistics are present simulations) that the noise variance in the correct and in-
with respect to this normalization. correct bins is no longer the same. It was also noticed that
Clearly, for aBOC algorithm, ξ σ 2bin = 1 and the noncen- the distribution of FBBefw test statistic is bounded by two χ 2
trality factor ξ λbin is either 1 (in a correct bin) or 0 (in an in- distributions. Moreover, Pmaxpp is the maximum power per
correct bin) [5, 7, 19]. Also, Ndeg = 2Nnc for aBOC, because piece (in the positive or in the negative frequency band). For
we add together the absolute-squared valued of Nnc complex example, if Npieces = 2 and FBBefw approach is used for Sin-
variables (or the squares 2Nnc real variables, coming from BOC(1,1) case, the powers per piece of the positive-sideband
real and imaginary parts of the correlator outputs). For B&F, lobe are 0.10 and 0.34, respectively (hence, Pmaxpp = 0.34).
the noncentrality deterioration factor and the variance dete- Again, these powers can be derived straightforwardly, via the
rioration factor depend on the normalized power per main formulas shown in [1, 3, 4, 20].
lobe (positive or negative) Pml of the BOC power spectral Figure 6 compares the simulation-based complementary
CDF (i.e., 1-CDF) with theoretical complementary CDFs
for FBBep case (similar plots were obtained for aBOC,
1 The case of FBBefw is discussed separately, later in this section. B&F, and FBBefw but they are not included here due to
6 EURASIP Journal on Wireless Communications and Networking
Table 1: χ 2 parameters for the distribution of the decision variable Z, various acquisition methods.
Matching to χ 2 complementary CDF for SSB, FBB sition process if a false alarm state is reached), and the total
1
number of bins in the search space [21]:
0.9
0.8 2 + 2 − Pd (q − 1) 1 + Kpenalty Pfa
T acq = τd, (11)
2Pd
0.7
0.6 where τ d = Nnc Tcoh is the dwell time, q is the total num-
1-CDF
0.5
ber of bins in the search space, and Pd and Pfa are given by
(7) to (10). An example of the theoretical average detection
0.4 probability Pd compared with the simulation results is shown
0.3 in Figure 7, where a very good match is observed. The small
mismatch at high (Δt)bin for the dual B&F method can be ex-
0.2
plained by the number of points used in the statistics: about
0.1 5000 random points have been used to build such statistics,
0 which seemed enough for most of (Δt)bin ranges. However, at
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 very low detection probabilities, this number is still too small
Test statistic levels for a perfect match. However, the gap is not significant, and
Sim, non-central Sim, central
low Pd regions are not the most interesting from the analysis
Th, non-central Th, central point of view.
An example of performance (in terms of average and
Figure 6: Matching with χ 2 distributions, (complementary CDF: worst detection probabilities) of the proposed FBB methods
1-CDF), theory (th) versus simulations (sim), FBBep , Nfb = is given in Figure 8. The gap between proposed FBB methods
Npieces = 4. and aBOC method is even higher from the point of view of
the worst Pd . Here, SinBOC(1,1)-modulated signal was used,
and Nc = 20 ms, Nnc = 2. The other parameters are specified
with σ 2Z , Ndeg , and λZ given in (6) and in Table 1, and in the figures captions. The small edge in aBOC average per-
QNdeg /2 (·) being the generalized Marcum-Q function [17]. formance at around 0.7 chips is explained by the fact that a
Due to the fact that the time-bin step may be smaller than time-bin step equal to the width of the main lobe of CF en-
the 2-chip interval of the CF main lobe, we might have velope (i.e., about 0.7 chips) would give worse performance
several correct bins. The number of correct bins is: Nt = than a slightly higher or smaller steps, due to ambiguities in
2Tc /(Δt)bin , where Tc is the chip interval. Thus, the global the CF envelope. Also, the relatively constant slope in the re-
detection probability Pd is the sum of probabilities of detect- gion of 0.7–1 chips can be explained by the combination of
ing the signal in the ith bin, provided that all the previous high time-bin steps and the presence of the deep fades in the
tested hypotheses for the prior correct bins gave a misdetec- CF: since the spacing between those deep fades is around 0.7
tion [11]: chips for SinBOC(1,1), then a time-bin step of 0.7 chips is the
worst possible choice in the interval up to 1 chip. However,
t −1
N there is no significant difference in average Pd for time-bin
Pd Δτ 0 = Pdbin Δτ 0 + k(Δt)bin , Δ fD steps between 0.7 and 1 chip, since two counter-effects are
k=0 superposed (and they seem to cancel each other in the region
k−1 (9)
of 0.7 till 1 chip from the point of view of average Pd ): on
1 − Pdbin Δτ 0 + i(Δt)bin , Δ fD . one hand, increasing the time-bin step is deteriorating the
i=0
performance; on the other hand, avoiding (as much as possi-
ble) the deep fades of CF is beneficial. This fact is even more
In (9), Δτ 0 is the delay error associated with the first sam- visible from the lower plot of Figure 8, where worst-case Pd
pling point in the two-chip interval, where we have Nt cor- are shown. Clearly, having a time-bin step of about 0.7 chips
rect bins. Equation (9) is valid only for fixed sampling points. would mean that, in the worst case, we are always in a deep
However, due to the random nature of the channels, the sam- fade and lose completely the peak of the main lobe. This ex-
pling point (with respect to the channel delay) is randomly plains the minimum Pd achieved at such a step. Also, for steps
fluctuating, hence, the global Pd is computed as the expecta- higher than 1.5 chips, there is always a sampling sequence
tion E(·) over all possible initial delay errors (under uniform that will miss completely the main lobe of the envelope of CF
distribution, we simply take the temporal mean): (thus, the worst Pd will be zero).
It is noticed that FBB methods can work with time-bin
Pd = EΔτ 0 Pd Δτ 0 , (10) steps higher than 1 chip, due to the increase in the main lobe
of the CF envelope. Moreover, the proposed FBB methods
and the worst detection probability is obtained for the worst (both single and dual SB) outperform the B&F and aBOC
sequence of sampling points: Pd,worst = minΔτ 0 (P d (Δτ 0 )). method if the step (Δt)bin is sufficiently high. Indeed, the
The mean acquisition time T acq for the serial search is higher the time-bin step, the higher is the improvement of
computed according to the global Pd , the false alarm Pfa , the FBB methods over aBOC and B&F methods. We remark that
penalty time Kpenalty (i.e., the time lost to restart the acqui- even at (Δt)bin = 1 chip, we have a significantly high Pd ,
8 EURASIP Journal on Wireless Communications and Networking
Pd at Pfa = 0.001, dual B&F, CNR = 27 dB-Hz Average Pd , Npieces = 2, CNR = 30 dB-Hz
100 1
0.9
0.8
10−1
0.7
Pd
Pd
0.6
0.5
10−2
0.4
0.3
10−3 0.2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.5 1 1.5 2 2.5 3
(Δt)bin (chips) Time-bin step (Δt)bin
Pd
0.5
0.4
0.3
0.2
0.1
10−2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0
(Δt)bin (chips) 0 0.5 1 1.5 2 2.5 3
Time-bin step (Δt)bin
Sim, average Sim, worst
Th, average Th, worst aBOC Single FBB
Single B&F Dual FBB
(b)
Dual B&F
Figure 7: Comparison between theory and simulations for Sin- (b)
BOC(1,1). Left: dual-sideband B&F method. Right: Dual-sideband
FBBep method, Npieces = 2. Nc = 10 milliseconds, Nnc = 5, CNR = Figure 8: Average (upper) and worst (lower) detection probabili-
27 dB-Hz, Ns = 5. ties versus (Δt)bin ambiguous and unambiguous BOC acquisition
methods (FBBep was used here).
due to the widening of the CF main lobe. The constant Pd search space (see (11) is directly proportional to (Δt)bin . For
at higher time-bin steps is explained by the fact that, if the example, if the code epoch length is 1023 chips and only
step increases with respect to the correlation function width, one frequency bin is searched (assisted acquisition), q =
only noise is captured in the acquisition block. Thus, increas- 1023/(Δt)bin . Moreover, the computational load required
ing the step above a certain threshold would not change the for implementing a correlator acquisition receiver per unit of
serial detection probability, since the decision variable will time uncertainty is inversely proportional to (Δt)2bin [9], thus,
only contains noise samples. when (Δt)bin increases, the computational load decreases.
On the other hand, by increasing the time-bin step in An example regarding the needed time-bin step in or-
the acquisition process, we may decrease substantially the der to achieve a certain detection probability, at fixed CNR
mean acquisition time, because the number of bins in the and false alarm probability, is shown in what follows. We
Elena Simona Lohan 9
Step needed to achieve a target Pd = 0.9, (average case) Step needed to achieve a target Pd = 0.9, (average case)
2 1.5
(Δt)bin (chips)
(Δt)bin (chips)
1.5
1
1
0.5
0.5
0 0
25 26 27 28 29 30 31 25 26 27 28 29 30 31
CNR (dB-Hz) CNR (dB-Hz)
103
MAT
MAT
102
102
101 101
25 26 27 28 29 30 31 25 26 27 28 29 30 31
CNR (dB-Hz) CNR (dB-Hz)
Figure 9: Step needed to achieve a target average Pd = 0.9, at false alarm Pfa = 10−3 and corresponding mean acquisition time, SinBOC(1,1)
signal. Code length 4092 chips, penalty factor Kpenalty = 1, single frequency-bin. Npieces = 2 for FBBep . Left: dual sideband. Right: single
sideband.
assume a SinBOC(1,1)-modulated signal, a CNR = 30 dB- lection is in general related to the quality of the following
Hz, and a target average detection probability of Pd = 0.9 at code tracking circuit. There is a wide range of values that
Pfa = 10−3 . For these values, we need a step of (Δt)bin = 1.2 Kpenalty may take and no general rule about the choice of
chips for the dual-sideband B&F method (which will cor- Kpenalty has been given so far, to the author’s knowledge. For
respond to a mean acquisition time T acq = 86.24 s for sin- example, in [22] a penalty factor Kpenalty = 1 was consid-
gle frequency serial search and 4092-chip length code) and a ered; in [23] simulations were carried out for Kpenalty = 2, in
step of (Δt)bin = 1.7 chips for dual-sideband FBBep method [24] a penalty factor of Kpenalty = 103 was used, while in [25]
with Npieces = 2 (i.e., T acq = 58.14 s). Thus, the step can be we have Kpenalty = 106 . Penalty factors with respect to dwell
about 50% higher for dual-sideband FBB case than for dual- times were also used in the literature, for example: Kpenalty =
sideband B&F case, and we may gain about 48% in the MAT 105 /(Nc Nnc ) [26, 27], or Kpenalty = 107 /(Nc Nnc ) [27] (in our
(i.e., MAT is 48% less in dual-SB FBB case than in dual-SB simulations, Nc Nnc = 40 ms). Therefore, Kpenalty may spread
B&F case). For single-sideband approaches, the differences over an interval of [1, 106 ], therefore, in our simulations we
between FBB and B&F methods are smaller. An illustrative considered the 2 extreme cases: Kpenalty = 1 (Figure 9) and
plots is shown in Figure 9, where the needed steps and the Kpenalty = 106 (Figure 10). Figure 10 uses exactly the same
achievable mean acquisition times are given with respect to parameters as Figure 9, with the exception of the penalty
CNR. We notice that FBB methods outperform B&F meth- factor, which is now Kpenalty = 106 . For Kpenalty = 106 of
ods at high CNRs. Below a certain CNR limit (which, of Figure 10, MAT for the dual-sideband B&F method becomes
course, depends on the (Nc , Nnc ) pair), B&F method may T acq = 8.62 ∗ 104 , which is still higher than MAT for the
be better than FBB method. dual-sideband FBBep (T acq = 5.8 ∗ 104 s). Similar improve-
The optimal number of pieces or filters to be used in the ments in MAT times via FBB processing (as for Kpenalty = 1)
filter bank depends on the CNR, on the method (single or are observed if we increase the penalty time.
dual SB), and on the BOC modulation orders. From simu- The plots with respect to the receiver operating charac-
lation results (not included here due to lack of space), best teristics (ROC) are shown in Figure 11 for a CNR of 30 dB-
values between 2 and 6 have been observed. This is due to Hz. ROC curves are obtained by plotting the misdetection
the fact that a too high Npieces parameter would deteriorate probability 1 − Pd versus false alarm probability Pfa [28]. The
the signal power too much. lower the area below the ROC curves is, the better the per-
We remark that the choice of the penalty factor has not formance of the algorithm is. As seen in Figure 11, the dual
been documented well in the literature. The penalty time se- sideband unambiguous methods have the best performance.
10 EURASIP Journal on Wireless Communications and Networking
Achieved MAT [s] at considered step Achieved MAT [s] at considered step
106 107
106
MAT
MAT
105
105
104 104
25 26 27 28 29 30 31 25 26 27 28 29 30 31
CNR (dB-Hz) CNR (dB-Hz)
Figure 10: Mean acquisition time corresponding to the step needed to achieve a target average Pd = 0.9, at false alarm Pfa = 10−3 , Sin-
BOC(1,1) signal. Code length 4092 chips, penalty factor Kpenalty = 106 , single frequency-bin. Npieces = 2 for FBBep . Left: dual sideband. Right:
single sideband.
ROC, (Δt)bin = 0.5 chips, CNR = 30 dB-Hz ROC, (Δt)bin = 1.5 chips, CNR = 30 dB-Hz
1 1
0.9 0.9
0.8 0.8
Mis-detection probability 1-Pd
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
10−10 10−8 10−6 10−4 10−2 10−10 10−8 10−6 10−4 10−2
False alarm probability Pfa False alarm probability Pfa
Figure 11: Receiver operating characteristic for CNR = 30 dB-Hz, SinBOC(1,1) signal, Nc = 20, Nnc = 2. Left: (Δt)bin = 0.5 chips; right
(Δt)bin = 1.5 chips.
At low time-bin steps (e.g., (Δt)bin = 0.5 chips), the FBB and pends on the CNR, on the integration times, and on the time-
B&F methods behave similarly, as it has been seen before also bin step and it is typically quite low (below 10−5 ).
in Figure 8. The main advantage of FBB methods is observed
for time-bin steps higher than one chip, as shown in the left 6. CONCLUSIONS
plot of Figure 11. For both time-bin steps considered here,
the single sideband unambiguous methods have a threshold This paper introduces a new class of code acquisition meth-
false alarm, below which their performance becomes worse ods for BOC-modulated CDMA signals, based on filter bank
than that of ambiguous BOC approach. This threshold de- processing. The detailed theoretical characterization of this
Elena Simona Lohan 11
new method has been given and theoretical curves were val- [11] E. S. Lohan, “Statistical analysis of BPSK-like techniques for
idated via simulations. The performance comparison with the acquisition of Galileo signals,” in Proceedings of the 23rd
other methods (i.e., ambiguous BOC and Betz&Fishman AIAA International Communication Systems Conference (IC-
sideband correlator) showed that FBB techniques can be suc- SSC ’05), Rome, Italy, September 2005, CDROM.
cessfully employed if the target is to increase the time-bin [12] E. S. Lohan, “Filter-bank based technique for fast acquisition
step of the acquisition process and to minimize the mean ac- of Galileo and GPS signals,” in Proceedings of the 17th IEEE
International Symposium on Personal, Indoor and Mobile Ra-
quisition times and the computational load of the correlator.
dio Communications (PIMRC ’06), pp. 1–5, Helsinki, Finland,
September 2006.
ACKNOWLEDGMENTS [13] E. D. Kaplan, Understanding GPS: Principles and Applications,
Artech House, London, UK, 1996.
This work was carried out in the project “Advanced Tech- [14] P. W. Ward, “GPS receiver search techniques,” in Proceedings
niques for Personal Navigation (ATENA)” funded by the of the IEEE Position Location and Navigation Symposium, pp.
Finnish Funding Agency for Technology and Innovation 604–611, Atlanta, Ga, USA, April 1996.
(Tekes). This work has also been supported by the Academy [15] M. Katz, Code acquisition in advanced CDMA networks, Ph.D.
of Finland. thesis, University of Oulu, Oulu, Finland, 2002.
[16] J. Betz and P. Capozza, “System for direct acquisition of re-
REFERENCES ceived signals,” US patent no. 2004/0071200 A1, April 2004.
[17] J. Proakis, Digital Communications, McGraw-Hill, New York,
[1] J. W. Betz, “The offset carrier modulation for GPS moderniza- NY, USA, 4th edition, 2001.
tion,” in Proceedings of the International Technical Meeting of [18] R. R. Rick and L. B. Milstein, “Optimal decision strate-
the Institute of Navigation (ION-NTM ’99), pp. 639–648, San gies for acquisition of spread-spectrum signals in frequency-
Diego, Calif, USA, January 1999. selective fading channels,” IEEE Transactions on Communica-
[2] A. Burian, E. S. Lohan, and M. Renfors, “BPSK-like methods tions, vol. 46, no. 5, pp. 686–694, 1998.
for hybrid-search acquisition of Galileo signals,” in Proceedings [19] F. Bastide, O. Julien, C. Macabiau, and B. Roturier, “Analysis
of the IEEE International Conference on Communications (ICC of L5/E5 acquisition, tracking and data demodulation thresh-
’06), vol. 11, pp. 5211–5216, Istanbul, Turkey, June 2006. olds,” in Proceedings of the International Technical Meeting of
[3] E. S. Lohan, A. Lakhzouri, and M. Renfors, “Binary-offset- the Satellite Division of the Institute of Navigation (ION-GPS
carrier modulation techniques with applications in satel- ’02), pp. 2196–2207, Portland, Ore, USA, September 2002.
lite navigation systems,” Wireless Communications and Mobile [20] S. H. Raghavan and J. K. Holmes, “Modeling and simulation of
Computing, vol. 7, no. 6, pp. 767–779, 2006. mixed modulation formats for improved CDMA bandwidth
[4] E. S. Lohan, A. Lakhzouri, and M. Renfors, “Spectral shap- efficiency,” in Proceedings of the 60th IEEE Vehicular Technol-
ing of Galileo signals in the presence of frequency offsets and ogy Conference (VTC ’04), vol. 6, pp. 4290–4295, Los Angeles,
multipath channels,” in Proceedings of 14th IST Mobile & Wire- Calif, USA, September 2004.
less Communications Summit, Dresden, Germany, June 2005, [21] J. Holmes and C. Chen, “Acquisition time performance of PN
CDROM. spread-spectrum systems,” IEEE Transactions on Communica-
[5] S. Fischer, A. Guérin, and S. Berberich, “Acquisition concepts tions, vol. 25, no. 8, pp. 778–784, 1977.
for Galileo BOC(2,2) signals in consideration of hardware lim- [22] G. J. R. Povey, “Spread spectrum PN code acquisition using
itations,” in Proceedings of the 59th IEEE Vehicular Technology hybrid correlator architectures,” Wireless Personal Communi-
Conference (VTC ’04), vol. 5, pp. 2852–2856, Milan, Italy, May cations, vol. 8, no. 2, pp. 151–164, 1998.
2004.
[23] W. Zhuang, “Noncoherent hybrid parallel PN code acquisition
[6] N. Martin, V. Leblond, G. Guillotel, and V. Heiries, “BOC(x,y)
for CDMA mobile communications,” IEEE Transactions on Ve-
signal acquisition techniques and performances,” in Proceed-
hicular Technology, vol. 45, no. 4, pp. 643–656, 1996.
ings of the 16th International Technical Meeting of the Satellite
[24] E. A. Homier and R. A. Scholtz, “Hybrid fixed-dwell-time
Division of the Institute of Navigation (ION GPS/GNSS ’03),
search techniques for rapid acquisition of ultra-wideband sig-
pp. 188–198, Portland, Ore, USA, September 2003.
[7] B. Bandemer, H. Denks, A. Hornbostel, A. Konovaltsev, and nals,” in Proceedings of the International Workshop on Ultra-
P. R. Coutinho, “Performance of acquisition methods for Wideband Systems, Oulu, Finland, June 2003.
Galileo SW receivers,” in Proceedings of the European Navi- [25] B.-J. Kang and I.-K. Lee, “A performance comparison of code
gation Conference (ENC-GNSS ’05), Munich, Germany, July acquisition techniques in DS-CDMA system,” Wireless Per-
2005, CDROM. sonal Communications, vol. 25, no. 2, pp. 163–176, 2003.
[8] B. C. Barker, J. W. Betz, J. E. Clark, et al., “Overview of the [26] O.-S. Shin and K. B. Lee, “Utilization of multipaths for spread-
GPS M code signal,” in Proceedings of the International Tech- spectrum code acquisition in frequency-selective Rayleigh fad-
nical Meeting of the Institute of Navigation (ION-NTM ’00), ing channels,” IEEE Transactions on Communications, vol. 49,
Anaheim, Calif, USA, January 2000, CDROM. no. 4, pp. 734–743, 2001.
[9] P. Fishman and J. W. Betz, “Predicting performance of direct [27] D. DiCarlo and C. Weber, “Multiple dwell serial search: perfor-
acquisition for the M-code signal,” in Proceedings of the Inter- mance and application to direct sequence code acquisition,”
national Technical Meeting of the Institute of Navigation (ION- IEEE Transactions on Communications, vol. 31, no. 5, pp. 650–
NTM ’00), pp. 574–582, Anaheim, Calif, USA, January 2000. 659, 1983.
[10] V. Heiries, D. Roviras, L. Ries, and V. Calmettes, “Analysis of [28] S. Soliman, S. Glazko, and P. Agashe, “GPS receiver sensitiv-
non ambiguous BOC signal acquisition performance,” in Pro- ity enhancement in wireless applications,” in Proceedings of
ceedings of the 18th International Technical Meeting of the Satel- the IEEE MTT-S International Tipical Symposium on Technolo-
lite Division of the Institute of Navigation (ION-GNSS ’05), gies for Wireless Applications, pp. 181–186, Vancouver, Canada,
Long Beach, Calif, USA, September 2005, CDROM. February 1999.