Admission Control and Handover Management for High-Speed Trains in Vehicular Geostationary Satellite...

INTERNATIONAL JOURNAL OF SATELLITE COMMUNICATIONS

Int. J. Commun. Syst. Network 2010; 28:1–27Published online 30 July 2009 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/sat.940

Admission control and handover management for high-speedtrains in vehicular geostationary satellite networks with

terrestrial gap-filling

Fabio Lattanzi1,�,y, Giovanni Giambene2, Guray Acar1 and Barry Evans1

1Centre for Communications Systems Research, Faculty of Engineering and Physical Sciences, University of Surrey,Guildford, Surrey GU2 7XH, U.K.

2Information Engineering Department, University of Siena, Siena, Italy

SUMMARY

Following a recent upgrade, the Digital Video Broadcasting—Return Channel Satellite (DVB-RCS)standard sets up to support terminal mobility. In this scenario, integration with terrestrial systems becomesa primary concern to ensure network connectivity in urban areas. This article proposes an integratedsatellite–terrestrial architecture for the provision of broadband services onboard high-speed trains, inwhich terrestrial cellular networks are seen as viable gap-fillers for discontinuous satellite coverage. Wederive an analytical model of the hybrid DVB-RCS-cellular system by exploiting analogies between themobility pattern predictability of LEO constellations and that of high-speed trains. Terminals whose QoScannot be guaranteed by the satellite segment are proposed to temporarily divert the connections towardsthe terrestrial infrastructure, where available. Using an iterative approach based on the Erlang fixed-pointapproximation, we show performance improvements with respect to stand-alone satellite systems in termsof handover failure probability and overall resource utilization. The analytical model is also validated viaour ns2-based DVB-RCS packet-level simulator. Detailed modelling of synchronization and signallingmechanisms confirms the accuracy of the analytical results, and shows that topology and mobilityinformation can contribute to refine radio resource utilization optimality when used jointly. Copyright r2009 John Wiley & Sons, Ltd.

KEY WORDS: mobile DVB-RCS; admission control; vertical handover; terrestrial gap-filler

1. INTRODUCTION AND RELATED WORK

In the last two decades, wireless communications have undergone a fast growth and the currenttrends are to evolve into integrated networks [1]. Mobile users expect demanding communica-tion services such as voice or real-time video to be available at anytime, regardless of their actual

*Correspondence to: Fabio Lattanzi, Centre for Communications Systems Research, Faculty of Engineering and PhysicalSciences, University of Surrey, Guildford, Surrey GU2 7XH, U.K.yE-mail: [email protected]

Contract/grant sponsor: European SatNEx II Network of Excellence; contract/grant number: IST-027393

Copyright r 2009 John Wiley & Sons, Ltd.

position. Multi-mode terminals are already available to communicate without service disruptionin multi-tier coverage regions via several access networks. Interaction is therefore requiredamong several domains to coordinate the execution of inter-segment handover and refine theradio resource assignment.

This article sets up to investigate improvements in the connection-level performance of next-generation Digital Video Broadcasting—Return Channel Satellite (DVB-RCS) systems bymeans of interaction with terrestrial systems. In particular, we will demonstrate that low forcedtermination probability and high utilization can be achieved via inter-segment resourcemanagement without sacrificing much bandwidth in aggressive resource reservation strategieswhen the network load increases. Two solutions are currently available to provide multi-modeReturn Channel Satellite Terminals (RCSTs) with a temporary point of attachment to theterrestrial infrastructure [2]. The first, known as the integral gap-filler, is suggested when themobile terminals lose the line-of-sight (LOS) and ultimately the access to the serving satellite.This might be the case of a train entering a tunnel. In order to preserve ongoingcommunications, the RCST must necessarily connect with a gap-filler, which, in the simplestcase, acts as a transparent repeater and an access point. The Network Control Centre (NCC) isunaware of the gap-filler taking over the communication; hence, no specific resourcemanagement issues are observed. The second solution considers the utilization of theterrestrial gap-fillers. This represents a more sophisticated approach entailing inter-segmenthandover and integrated resource management as well as potential adaptations due to differentair interfaces and/or packet formats.

In the following, we envisage a network scenario in which terrestrial gap-fillers located in theproximity of the overlapping area between adjacent spot-beams provide backup connectivity forthe mobile collective communities. It is foreseen that the DVB-RCS system resorts to thediversion of active connections towards available terrestrial networks only when intra-domainmobility cannot be handled due to high congestion. A mathematical model of this hybridsatellite–terrestrial network is developed with the objective of defining an accurate means for theperformance analysis of Connection Admission Control (CAC) and handover policies aiming toincrease both the user satisfaction and the resource utilization. From the perspective of ageostationary coverage, cellular networks have a patchy nature. These are usually located inurban centres, where the LOS to the serving satellite may be obstructed or seriously affected bymultipath effects. Terrestrial gap-fillers therefore represent an attractive solution for the designof new advanced radio resource management (RRM) algorithms improving the QoS perceivedfrom passengers onboard high-speed trains.

The admission control and handover policies presented in this article are based on thecustomary assumption that preserving the QoS of ongoing connections has a higher prioritythan accepting new connections. In order to minimize the forced termination probability, werely on information about the joint satellite–terrestrial network topology. In particular,terrestrial gap-fillers are proposed as temporary access providers whenever intra-satellitehandover cannot be immediately executed. Because the resource management is defined by thecapacity requirements of terminals, we do not pose any limitation in terms of terrestrialinfrastructure. Systems such as UMTS and WiMAX may be used indifferently as accessnetworks; thus, we will refer to a generic terrestrial wireless system in the rest of the article. Inaddition, due to the patchy nature of the terrestrial coverage, it is reasonable to expect that onlya portion of active terminals will reside in a multi-tier region at handover time. As shown in thefollowing, our analysis accounts for this aspect via a specific probabilistic factor. Many issues

F. LATTANZI ET AL.2

Copyright r 2009 John Wiley & Sons, Ltd. Int. J. Commun. Syst. Network 2010; 28:1–27

DOI: 10.1002/sat

must be considered when analysing the effectiveness of inter-segment resource managementsolutions. Among these, we cannot neglect the fact that terrestrial gap-fillers entail trafficre-routing between the gateway and the terrestrial base station, hence arising layer 2 and layer 3mobility issues. Because IP mobility mechanisms are usually demanding in terms of processingand bandwidth due to tunnelling, a reasonable approach is to consider terrestrial diversion asthe last option to maintain the satellite connections active when the terminal is approaching thebeam boundary and resources are not immediately available in the target spot-beam. Moreover,the diverted traffic should be handed over back to the satellite network as soon as possible tomaximize the resource utilization.

Despite the different propagation delays, we have identified some analogies between themobility pattern predictability of LEO constellations and that of vehicular GEO networks. Inparticular, as the user velocity is traditionally neglected with respect to the mobility of LEOsatellites (from 18000 to 25 000km/h), GEO footprints are considered fixed whereas velocity androutes of mobile collective terminals (e.g. on trains) are assumed to be predictable [3]. Manyarticles have already addressed admission control and handover management in LEO networks.The subject has been extensively investigated and several solutions have been proposedmaximizing different cost metrics. In guaranteed handover [4], a new connection is admitted onlyif sufficient bandwidth can be reserved in the first two spot-beams that the terminal is expected tovisit. In predictive reservation [5], an amount of resources, which is proportional to theprobability that the connection will be still active at handover time, is reserved in each spot-beamalong the route followed by the terminal. As a result, CAC takes into account not only thecurrent traffic but also an estimation of the future loading. In order to increase the channelutilization, Ming-Hsing and Bassiouni [6] proposed to delay the reservation until the distancebetween the terminal and the spot-beam boundary is below a given threshold. Theaforementioned policies prioritize the handover traffic by rejecting new admission requestswhenever there is a lack of resources either in the current spot-beam or in one of the successivespot-beams. In addition, handover requests can be temporarily queued to decrease the forcedtermination probability and increase the resource utilization. In this regard, Del Re et al. [7]provide a comparison between the traditional first-in-first-out (FIFO) discipline and a handoverprioritization mechanism based on the residual time spent in the overlap area (handover area)between two adjacent cells. Handover strategies have been investigated extensively also interrestrial networks [8, 9]. Traditionally, the decision about transferring a connection from onecell to another is made by comparing the link quality of adjacent base stations, while severalparameters are taken into account to drive this procedure such as the received signal strength(RSS), estimations of power degradation and bit error rate (BER) [10]. However, inheterogeneous communication environments, mobile terminals may fall into the coverage ofmultiple networks with different access technologies. Positioning information may thereforeprove particularly effective at removing handover uncertainties due to temporary degradations ofchannel quality [11, 12]. Stemm first introduced the concept of vertical handover in multi-tiernetworks [13]. In his model, the technology at the lowest tier delivers the highest data rate and istherefore always preferred. A different approach was proposed by Fang and McNair [14]. Theirwork is focused on the minimization of a multi-dimensional cost function, which identifies thetarget network by jointly considering network condition and system performance as well as userpreferences and terminal conditions. Finally, Liu et al. [15] presented a scheme that dynamicallyswitches connections among different networks according to a profitability function. Their strategyaims to maximize the user data rate and to minimize the connection blocking and dropping rates.

ADMISSION CONTROL AND HANDOVER MANAGEMENT 3


DOI: 10.1002/sat

The rest of this article is organized as follows. Section 2 introduces the main issues associatedwith the DVB-RCS mobility extension as well as the integration with terrestrial gap-fillers. InSection 3, a mathematical characterization of the hybrid satellite–terrestrial network is proposedand analysed. In Section 4, several admission control and handover strategies are proposed andtheir performance analysed using both the mathematical model and packet-level simulations.Our conclusions are drawn in Section 5.

2. MOBILE DVB-RCS NETWORKS AND TERRESTRIAL GAP-FILLERS

Since its publication, ETSI DVB-RCS specification [16] has been successfully implemented by alarge number of equipment manufacturers around the world. The standard and the guidelinesfor implementation define a framework for satellite physical- and MAC-layer solutions, whichhave enabled mass production of low-cost equipments. Yet, the details of most RRMalgorithms are left open to the manufacturers.

Recently, both EU and ESA have sponsored R&D programmes to introduce mobilitycapabilities to DVB-RCS in order to provide broadband satellite services to collective mobilecommunities such as passengers of airlines, maritime vessels and high-speed trains [17–19]. Inparticular, high-speed trains are expected to be one of the major communities that will exploitbroadband satellite services, even if they have posed special challenges due to the peculiarcharacteristics of the railroad satellite channel [20].

A simplified mobile satellite interactive network (Figure 1) consists of a space and a groundsegment [21]. The space segment comprises a geostationary bent-pipe satellite working at Ku orKa band. The terrestrial segment consists of RCSTs installed onboard trains (note that trainsare expected to cross the spot-beams in Figure 1 in both directions), one Ground Earth Station

Figure 1. Reference network architecture.

F. LATTANZI ET AL.4


DOI: 10.1002/sat

(GES) and the NCC. GES acts as a feeder for the forward link and as a traffic gateway for thereturn link. The NCC is responsible for monitor and control operations as well as resourceallocation and mobility management in the satellite network. In Figure 1, TGF 1 and TGF 2 aretwo terrestrial gap-fillers in two adjacent cells of the same GEO satellite.

In the eventuality that interaction with the terrestrial infrastructure is needed, a furtherelement may become necessary to coordinate resource negotiation during a handover, theNetwork Operations Centre (NOC).

It is assumed that the forward link of the satellite interactive network is compliant with theDVB-S2 standard. The multiplexing scheme is therefore based on Time-division Multiplexing(TDM), while the return RCST access scheme is Multi-frequency—Time-division MultipleAccess (MF-TDMA). Resource management in DVB-RCS specification is defined as acombination of static and dynamic (on-demand) allocations. As such, five categories of capacityare supported [16]. Continuous rate assignment (CRA) is suited for traffic that requires a fixedguaranteed rate with minimum delay and jitter. Once allocated, this typology of capacity ismade available to the terminals for the whole duration of the communication. Rate-baseddynamic capacity (RBDC) can be used in combination with CRA to increase the transmissionrate above a given minimum. Despite being guaranteed, RBDC capacity must be explicitlyrequested by the terminal; otherwise, it is made available to other users. Volume-based dynamiccapacity (VBDC) is only allocated for delay jitter-tolerant applications like Web browsing ore-mailing. Finally, free capacity assignment (FCA) distributes unused resources (if any) amongterminals. No signalling is involved in this process, thus terminals have no control over it. In ourstudy, we aim to improve the connection-level performance of real-time applications such asVoIP that are very demanding in terms of QoS. As a result, only CRA traffic is considered in therest of this article. We assume that the standard forward signalling service of the DVB-S2standard is used by RCSTs for link acquisition. As a result, data and signalling information areencapsulated into MPEG transport streams via padded segmentation and reassembly (asdictated by AAL5). The Terminal Burst Time Plan (TBTP) periodically specifies thefrequency–time slots each terminal is allocated in the upcoming super-frame. Yet, RCSTsmust first synchronize with forward and return links in order to be able to transmit data. Thisrequires the acquisition of timing information and frame organization. Synchronizationrepresents a key aspect of DVB-RCS and it must be maintained throughout the period ofactivity to avoid packet collisions. SYNC bursts and Correction Message Tables (CMT) are themechanism used by the NCC to tune terminals transmitting parameters and provide them withthe appropriate timing, frequency and amplitude corrections.

Previous work from the authors focused on the adaptations required by the DVB-RCSstandard to support broadband access onboard mobile collective vehicles [3, 22, 23]. The impactof mobility on DVB-RCS synchronization mechanisms was evaluated in [22]. We then proposeda simple position-based strategy for spot-beam handover management and a more sophisticatedmechanism considering velocity and mobility pattern information as a means for prioritizingdifferent handover requests [3]. A similar approach was followed in [23], where a new CACalgorithm for vehicular GEO networks was developed. This scheme relies on traffic estimationsto predict future network congestion and decide whether new requests can be accepted withoutdegrading the performance of ongoing connections. Knowledge of terminal actual locationswithin the satellite coverage represents a fundamental assumption in our study. This is due tothe intrinsic necessity of mobile RCSTs for positioning information to be able to minimize theguard time among slots and access network services while roaming around the coverage [24].



DOI: 10.1002/sat

Present study aims to represent a further step towards the definition of next-generation DVB-RCS networks. It is expected that they will natively support terminal mobility both in LOS andnon-LOS scenarios. As a result, full integration with terrestrial infrastructure is requiredwherever the satellite channel is obstructed (e.g. in a tunnel) or disturbed (e.g. in urban areas).We consider such eventuality as an opportunity to develop more sophisticated resourcemanagement algorithms that may increase the user-perceived QoS. In the following, asimulation scenario is envisaged that considers intra-system mobility (i.e. spot-beam handover)as well as inter-system mobility (i.e. inter-segment handover between a satellite network andterrestrial gap-fillers). It should be noted that each RCST is assumed to be connected only to asingle domain at a time. As a result, the execution of a spot-beam (i.e. horizontal) handoverautomatically excludes any concurrent interaction with the terrestrial domain. Vice versa, whenan inter-segment (i.e. inter-domain, vertical) handover is completed all satellite resources arereleased. The scope of our handover mechanism is limited to DVB-RCS, while RCSTs areassumed supporting mobile IP to ensure seamless IP mobility [25].

In accordance with the mechanisms presented in [3, 22], we suggest that the jointsatellite–terrestrial system implements a mobile-assisted handover scheme. This means that adistributed handover detection scheme is implemented. However, the decision whether thisshould be completed still depends on a centralized entity. In stand-alone DVB-RCS networks, theNCC is responsible for resource management and mobility management. Nevertheless, if inter-segment handover is required, the correspondent entity in the target system will be involved in theprocess. In mobile DVB-RCS, each mobile RCST necessitates a positioning system (i.e. globalpositioning system (GPS)) and a map of the satellite interactive network for synchronizationpurposes [24]. It results that terminals such as high-speed trains that follow pre-determined routescan derive accurate estimations about their mobility pattern [3]. We advocate a position-basedmechanism for intra-satellite handover detection completed by radio channel quality estimationswhen necessity for inter-segment operations arises. A handover recommendation is generated bythe RCST and destined for the NCC via the standard DVB-RCS signalling messages. The NCC(directly or via the NOC) is then responsible for contacting its peer entity within the targetwireless network and negotiating the connection transfer. This approach requires additionalcomplexity on the network side as broad knowledge of the hybrid network topology is required,but it is our belief that this may represent a viable solution for smooth handover and prompttraffic redirection. Note that in previous studies [22], the authors showed that less than 2% ofRCSTs exceeded 2-s handover delay in a typical spot-beam handover scenario. This is the timefrom the reception of the switching command to the transmission of the first synchronization(SYNC) burst in the target uplink. In particular, the handover time-off depends on theDVB-RCS synchronization mechanism, which requires the ordered reception of a set ofsignalling tables. Their repetition rates may therefore have a strong impact on handoverduration. For simplicity, in this study we assume link-layer handover delay as the dominantfactor, whereas layer 3 mobility does not entail any additional significant delay.

3. ANALYTICAL FORMULATION

As any mathematical abstraction describing the behaviour of a real system, our analysis relieson a certain number of assumptions. In particular, the recursive approach employed in ouranalytical model may be adopted to describe the behaviour of a belt of satellite spot-beams.

F. LATTANZI ET AL.6


DOI: 10.1002/sat

However, due to the topology-related computational complexity of packet-level simulations,our reference scenario comprises a joint satellite–terrestrial coverage consisting of two adjacentGEO spot-beams (beams i and j) and two terrestrial cells (TCi and TCj) located in proximity ofurban areas (Figure 2). As a consequence of this simplified topology and bi-directional mobilityscenario (all users move at the same speed along parallel linear trajectories), there is only onespot-beam contributing to the handover traffic towards the adjacent satellite spot-beam.

In our study, RCSTs are not permitted to access terrestrial networks (where available) unlessthey are specifically instructed to do so during handover operations. The NCC is therefore theonly entity responsible for authorizing connection transfers from the satellite network toterrestrial infrastructure. On the other side, periodic updates about the satellite resource statusshould be exchanged between the NCC and its peer entity in the terrestrial domain to allow theterminals to move back to the satellite segment whenever resources are made available.

The new connection arrival process in each spot-beam is Poisson with rate lnew. For the sakeof simplicity, we approximate the channel holding time for new and handover connections asexponentially distributed with an average 1/m. As a consequence, the bandwidth occupancy ismodelled as a birth–death process. Details about how to derive 1/m are provided in the next sub-section.

With reference to Figure 2, we define Pj!i as the intra-satellite handover probability fromspot-beam j to spot-beam i and PTCi!i as the inter-segment handover probability between theterrestrial wireless network TCi and the spot-beam i, respectively. As previously mentioned, weassume mobile terminals following straight trajectories and moving at almost-constant velocitywithin the current serving segment (bi-directional linear mobility is considered). Yet, the velocitymay differ if the terminal is connected to the satellite or the terrestrial infrastructure. Theseassumptions, despite their simplicity, well describe the mobility pattern of high-speed trains,usually characterized by large radii of curvature [26]. In particular, PTCi!i is evaluated assumingthat both the diameter of the terrestrial cell and the velocity of the terminal should be reducedwith respect to the values in the satellite network. This is consistent with the assumption thatterminals move at a lower velocity in urban areas. Table I resumes main mobility variables that

Beam j

2R

Beam i

Terrestrial coverage

Terrestrial coverage

Satellite coverage

Satellite coverage

j iP→iTC iP →

iTC

jTC

Figure 2. Network topology.



DOI: 10.1002/sat

have been considered in our analytical model and simulations. In order to characterize the userrelative mobility in the joint satellite/terrestrial network, we also refer to the dimensionlessparameter a [7], defined as the ratio between the cell diameter (2R) and the product betweenaverage service duration and terminal velocity. Generically referred as vsat;ter in Table I, theterminal velocity may assume different values when the RCST is connected to the GEO satellitesegment (i.e. vsat) or the wireless terrestrial segment (i.e. vter) in accordance with the assumptionthat the trains move at a lower velocity in the urban centres rather than in the rural areas.Instead, Ts denotes the average service duration. The value of a represents a measure ofhandover necessity. In particular, the smaller this parameter, the higher the likelihood that theconnection will require a handover during its lifetime. According to values in Table I, we cannote that the degree of mobility in the satellite segment is higher than that in the terrestrial case.

Our admission control strategy is based on the fractional guard channel (GC) mechanismfirstly proposed by Ramjee et al. in [27]. This enforces a probabilistic reservation policy aimed toattain low blocking probability at the expense of increased handover failures. In particular,handover traffic is prioritized by randomly rejecting some new connections during congestedintervals. Let ni be the number of channels occupied at a certain time in a spot-beam, C the totalcapacity in each spot-beam and Tload a given threshold identifying the set of channels that couldbe used as GCs in a probabilistic way. In traffic balanced conditions, the new connectionadmission probability for a generic spot-beam i when there are ni channels occupied, Padm

ni, is

then calculated as

Padmni¼

1; 0pnipC� Tload

1�niCPi!j; C� Tload þ 1pnioC

0; ni ¼ C

8>><>>: ð1Þ

where Pi!j is the intra-satellite handover probability between spot-beams i and j. Our schemeprioritizes handover requests over new calls via the random rejection of some new connectionswhen there is a certain degree of satellite cell congestion. In particular, when the handover trafficincreases, the admission probability decreases so that ongoing connections are preserved. Asproposed by Ramjee, upon the arrival of a new connection request in the satellite segment, thesystem generates a random number P that is uniformly distributed between 0 and 1. If this valueis smaller than a reference probability Padm

ni, the connection is admitted; otherwise, it is rejected.

On the other hand, intra-satellite handover requests are accepted as long as there are freeresources in the target spot-beam. During congestion, terminals within the dual-coverage areamay benefit from connection diversion towards the terrestrial wireless network. When theRCSTs approach the end of the terrestrial coverage, connections that cannot be accommodatedin the target spot-beam are forced to terminate.

New connection (NBP) and handover blocking probabilities are obtained recursively via thereduced load approximation [28], also known as Erlang fixed-point approximation. This

Table I. Mobility parameters.

Cell radius (R)(km)

Terminal velocity(km/h)

Mean serviceduration (Ts) (h)

a ¼2R

vsat;terTs

Satellite network 800 vsat 5 343 2 2.33Terrestrial network 200 vter 5 60 2 3.33

F. LATTANZI ET AL.8


DOI: 10.1002/sat

approach requires that the total traffic offered to each spot-beam is specified at each step. Forsimplicity, the two spot-beams’ topology shown in Figure 2 is considered and formulas belowrefer to spot-beam i. However, a symmetric approach can be followed to evaluate the trafficoffered to spot-beam j (we assume equilibrium in the mobility from i to j and vice versa).Assuming that each connection consumes exactly one frequency–time slot (i.e. 1 CRA-TRF) persuper-frame (note that in our study one super-frame is formed by just one frame), the trafficoffered to spot-beam i is

li ¼ linew þHj!iHO þ Vj!i

HO ð2Þ

where Hj!iHO and Vj!i

HO represent the handover arrival rate from the adjacent spot-beam j and theterrestrial network TCi, respectively. The diagram in Figure 3 shows the handover processacross the seam of the satellite/terrestrial network. In particular,

Hj!iHO ¼ Hsingle

direct þHdualdirect þHdual

rejected ð3Þ

Equation (3) is the sum of three arrival rate terms. The first term, Hsingledirect, represents the

amount of traffic originating in spot-beam j, which is directly offered to spot-beam i:

Hsingledirect ¼ ljnewð1�NBPjÞPj!ið1� RTCi

Þ ð4Þ

where NBPj is the new connection blocking probability in spot-beam j and RTCiis a factor

describing the probability of being in the overlay coverage served by spot-beam i and terrestrialnetwork TCi at handover time. The second term, Hdual

direct, is the amount of traffic that yet can besupported by the satellite system despite being located in the dual-coverage region. To this end,we assume that the satellite network has the capacity of estimating the current intra-satellitehandover failure rate (HFP) over long time observation windows [29]. As a result, we define a

Handover arrival rate contributions towards spot-beam i

New Connection Attempt

Blocking of the new connection attempt

1 jNBP−

new

jNBP

Connection terminated in the source spot-beam

j iP→

1 j iP →−

Handover request to the adjacent spot-beam

Dual Coverage Area

iTCR1iTCR−

singledirectH Need for terrestrial diversion

needP1 needP−

dualdirectH Terrestrial network availability

1 iTCrejP−

iTCrejP

dualrejectedH

Connection terminated inthe terrestrial domain

1iTC iP →−

iTC iP →

j iHOV →

Figure 3. Diagram of the handover arrival rates for our integrated system in Figure 2. Details of theimplementation are provided in the paragraph devoted to the packet-level simulator description.



DOI: 10.1002/sat

new parameter Pneed ¼ minf1;b �HFPg where b40 is an integer value that is used to amplifythe current HFP in order to increase the amount of traffic that is diverted towards the wirelessterrestrial network to alleviate the congestion in the satellite segment. It results that Hdual

direct isevaluated as

Hdualdirect ¼ ljnewð1�NBPjÞPj!iRTCi

ð1� PneedÞ ð5Þ

Finally, the last component, Hdualrejected, represents those connections that would require

diversion, but are rejected by the terrestrial network:

Hdualrejected ¼ ljnewð1�NBPjÞPj!iRTCi

PneedPTCi

rej ð6Þ

where PTCi

rej is the probability that the connection transfer from the satellite to the terrestrialdomain is rejected. Of course, this value depends on the current network status at handover time(e.g. the traffic load in the terrestrial segment), but it is not the only parameter that needs to betaken into account for an exhaustive analysis. Several physical characteristics, such as the powerlevel that regulates the access to the terrestrial network, should also be considered. However,due to the complexity of modelling the interaction between these variables, we have decided tocollectively describe the rejection event as a single parameter. Similarly, Vj!i

HO is derived as

Vj!iHO ¼ ljnewð1�NBPjÞPj!iRTCi

Pneedð1� PTCirej ÞPTCi!i ð7Þ

where, due to the memoryless nature of the exponential distribution, Pj!iPTCi!i describes theoverall probability that the connection will be offered back to the satellite network after crossingthe terrestrial network TCi. Similar equations can be obtained with reference to spot-beam j.

Let Pini

be the probability that spot-beam i has ni channels occupied. Then, the balanceequation for this spot-beam is

Pini�1

linewPadmni

liþ

Hj!iHO þ Vj!i

HO

li

!li ¼ Pi

ninim; 0pnipC ð8Þ

From the standard solution for birth–death processes we obtain the probability that the ithspot-beam has ni TRF slots occupied:

Pini¼

li

m

� �ni Qnij¼1

linewPadmj

liþ

Hj!iHO þ Vj!i

HO

li

!

ni!Pi0; 1pnipC ð9Þ

Pi0 ¼ 1þ

XCni¼1

li

m

� �ni Qnij¼1

linewPadmj

liþ

Hj!iHO þ Vj!i

HO

li

!

ni!

266664

377775

�1

ð10Þ

New and handover blocking probabilities (NBPi and HFPi, respectively) can be easilyobtained from the equations above:

NBPi ¼XC�1ni¼0

Pinið1� Padm

niÞ þ Pi

C ð11Þ

HFPi ¼ PiC ð12Þ

F. LATTANZI ET AL.10


DOI: 10.1002/sat

The iterative procedure starts with setting NBPi and HFPi equal to 0. Successively, lt isevaluated using Equations (2)–(7). At this stage, new values for NBP and handover blockingprobability are calculated using (11) and (12). The algorithm terminates when the sum ofdifferences between NBP and handover blocking probability calculated at two consecutive steps[k�1, k] is below a given threshold. This results in satisfying the condition defined in

jNBPik �NBPi

k�1j þ jHFPik �HFPi

k�1jpe ð13Þ

where e is an arbitrary positive value (i.e. 10�6). Finally, we assume that NBPi is equal to NBPj

and that HFPi is equal to HFPj.The admission strategy specified above refers to the general case of limited fractional GC

policy [27], where Tload40. However, our analytical model can be easily adapted to carry outperformance evaluations for several different rejection strategies. This is achieved byintroducing minor modifications to the admission probability vector defined in Equation (1).In particular, if Tload ¼ 0 we obtain a pure fractional GC policy [27], whereas a fixed GCstrategy [30] can be derived by specifying Padm

n ¼ 1 for 0pnpC� Tload and 0 otherwise. Inaddition, a performance analysis of stand-alone satellite systems without gap-fillers can bederived if we let Pneed ¼ 0.

3.1. Channel holding time estimation

It is well known that the channel holding time for a new/handover connection in a cell isthe minimum between the cell residence time and the residual time of the new/handoverconnection [30]. In our model, we approximate both satellite spot-beams and terrestrialcoverage regions as rectangular areas (Figure 2). In addition, for ease of tractability, we assumethat the routes followed by terminals are parallel to the equator.

As shown in [31], the following formula holds:

E½Tholding� ¼ Tsð1� PexcessÞ ð14Þ

where E½Tholding� is the average channel holding time in a generic spot-beam, Ts is the averageservice duration and Pexcess is the excess lifetime probability. This is the generic probability thatthe terminal will require a handover. Assuming that the terminals are initially located within theinterior rectangular area of each spot-beam at a distance from the beam edge that is uniformlydistributed between 0 and vsatTmax, Equation (14) becomes

E½Tnew�¼Ts 1�1

Tmax

Z Tmax

0

e�t=Ts dt

� �¼Ts 1�

1� e�Tmax=Ts

Tmax

Ts

0BB@

1CCA¼Ts 1�

1� e�a

a

� �ð15Þ

E½THHO� ¼ Tsð1� e�Tmax=TsÞ ¼ Tsð1� e�aÞ ð16Þ

E½TVHO� ¼ Tsð1� e�jTmax�Tt j=TsÞ ¼ Tsð1� e�jaTs�Tt j=Ts Þ ð17Þ

Equation (15) applies to the holding time in the spot-beam where the connection originateswith the assumption that the connection duration is exponentially distributed with average Ts.Note that Tmax represents the time necessary for terminals to cover a distance that is equal to thediameter of the spot-beam (i.e. distance 2R). Instead, Equations (16) and (17) apply to transitspot-beams after a horizontal or a vertical handover, respectively. In particular, Tt represents



DOI: 10.1002/sat

the time needed to cross the entire terrestrial coverage. We estimate the connection averagechannel holding time in a generic spot-beam [32] as

1

m¼

lnewlnew þHHO þ VHO

E½Tnew� þHHO

lnew þHHO þ VHOE½THHO

� þVHO

lnew þHHO þ VHOE½TVHO

�

ð18Þ

where HHO and VHO are the average arrival rates of horizontal and vertical handover traffic,which are related to lnew through Equations (4)–(7).

4. PERFORMANCE EVALUATION

4.1. Analytical results

The objective of the first part of this study is to analyse the sensitivity of our mathematicalmodel to important system parameters such as the extension and the availability of theterrestrial network. As shown in Equations (6) and (7), the satellite network tries to divert excesshandover traffic towards the terrestrial infrastructure whenever the success of a spot-beamhandover cannot be guaranteed. By monitoring the current status of the network, the systemdecides whether a connection should be handed over to the wireless terrestrial domain toalleviate traffic congestion in the satellite segment. In particular, the aggressiveness of suchbehaviour is regulated by the penalty parameter b. The effects on the network performance areevaluated by continuously measuring the NBP and the HFP. In our calculations, we assumethat the total shared capacity C in each spot-beam is equal to 19 frequency–time slots and thatthe average service duration Ts is equal to 2 h. The spot-beam radius and the train speed areassumed to be 800 km (at equator) and 345 km/h (conservative case) [33], respectively. As aresult, Tmax duration is set at 4.6 h. NBP and handover blocking probability are depicted inFigure 4 with respect to the new connection arrival rate. To isolate the effects of the penaltyvalue, we keep fixed the terrestrial network extension RTC at 0.5 and the probability thathandover traffic is rejected by the terrestrial network ðPTC

rej Þ at 0.1 (conservative choice). Theneed for inter-segment handover is obtained by multiplying the current estimation of HFP by avalue of b that spans in the range [0, 25, 50, 75]; such a range has been used to account for manycases, including extreme cases ðPneed � 0 and Pneed ! 1Þ. As a reference, the case in which Pneed

is always 1 independently of the network load and the measured performance is also plottedwith a dashed line. As expected, both NBP and HFP increase with increasing the connectionarrival rate, but while the performance of stand-alone satellite systems sensibly decreases, oursolution attains lower blocking probabilities even at medium and high loads. In particular, thehigher the inter-segment handover rate (that depends on b), the lower result the blockingprobabilities, as clearly shown in Figure 4. This would suggest always using high b values;however, the operator should carefully tune this parameter in order to achieve a trade-offbetween maintaining a high satellite resource utilization and minimizing the roaming coststowards the terrestrial segment.

If the system realizes that an ongoing connection cannot be supported by the target spot-beam, an attempt is made to hand it over to the terrestrial network, if available. Connections arere-distributed according to the two criteria that are discussed in the following sub-sections.



DOI: 10.1002/sat

4.1.1. Degree of overlap between terrestrial and satellite coverage. The first criterion depends onthe degree of overlap, that is, the probability that the RCST resides in a two-tier region (i.e. RTC).In general, this value also depends on the mobility pattern that is followed by the train. Inparticular, in urban routes RTC equal to 1 may be assumed, but in rural regions this value is likelyto be very low. In order to isolate the effects of the terrestrial coverage extension, we maintainfixed the probability that connections are rejected by the terrestrial network ðPTC

rej Þ at 0.1 and b at50 as in the previous experiment. The eventuality that a terminal crosses the spot-beam edge in atwo-tier zone (i.e. RTC) varies from 0.25 to 1. As a reference, the case of a stand-alone satellitenetwork is also plotted as a dashed line. As one may expect, HFP increases with increasing theconnection arrival rate because so does the channel occupancy. However, we can see that a lowernumber of connections with respect to stand-alone satellite systems are forced to terminate wheninter-segment handover is allowed to take place. The performance also improves by increasing theterrestrial coverage extension (we implicitly assume that RCSTs always detect the terrestrialcarriers when available). An important effect of this temporary traffic re-distribution is thathandover from source to destination spot-beam is actually delayed by the interval terminals spendin the terrestrial domain (virtual queuing effect). This increases the chances that somefrequency–time slots are released in the target spot-beam and are successfully re-assigned. Inaddition, the satellite resources that are temporarily made free through this procedure allow thesystem to accept more logon requests, as is clearly shown in Figure 5.

4.1.2. Percentage of calls rejected by the terrestrial segment. The second criterion affecting trafficre-distribution is the availability of resources in the terrestrial domain. In our study, we assume

2 4 6 80

0.005

0.01

0.015

0.02

0.025

0.03

New Connection Arrival Rate [cnx per hour]

New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

Never Diverted Beta=25 Beta=50 Beta=75 Always Diverted

2 4 6 80

0.002

0.004

0.006

0.008

0.01

0.012

New Connection Arrival Rate [cnx per hour]H

ando

ver

Failu

re P

roba

bilit

y

Figure 4. NBP and HFP for different b values.



DOI: 10.1002/sat

that not all inter-segment handover requests are accepted. The percentage of connections thatcannot be accommodated in the terrestrial network is modelled as PTC

rej . Despite being in needfor diversion, some traffic is therefore offered back to the satellite system where it is likely to bedropped during handover. This is consistent with the hypothesis that the terrestrial networkmay also be overloaded or that the capacity reserved for the roaming of satellite connectionsmay not be sufficient to match the current demand. Figure 6 shows the sensitivity of NBP andHFP to the terrestrial rejection probability ðPTC

rej Þ with respect to the new connection arrival rate.Same assumptions hold as in the previous case (i.e. b5 50), with the exception that the overlayregion extension RTC is now fixed at 0.5. This means that only 50% of RCSTs reside in a dual-coverage area at handover time; the remaining population can only go through a pure spot-beam handover. As we can see, both NBP and HFP increase with increasing the offered traffic,but always remain below the dashed line representing stand-alone satellite systems. Values alsospan in a narrower range with respect to the previous cases depicted in Figure 5. This isimportant as it gives the design of hybrid networks some flexibility. In an effort to increase thepopulation of terminals that may benefit from segment diversion and minimize the NBP, theDVB-RCS operator may decide to focus only on wireless terrestrial networks along busy routes.In addition, the low sensitivity to the PTC

rej value suggests that roaming agreements withterrestrial operators should be driven by cost-effectiveness rather than performance issues.

4.2. Admission control strategies

In this section, we provide the results obtained when analysing the fractional GC [27] and thestatic GC [30] strategies through our mathematical model. Behaviours of NBP and HFP areplotted with respect to the new connection arrival rate. As a reference, same policies are

2 4 6 80

0.005

0.01

0.015

0.02

0.025

0.03


New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

RTC

=0.25 S+T RTC

=0.5 S+T RTC

=0.75 S+T RTC

=1 S+T RTC

=0 S

2 4 6 80

0.002

0.004

0.006

0.008

0.01

0.012


ando

ver

Failu

re P

roba

bilit

y

Figure 5. NBP and HFP for different RTC values (S, satellite only; S1T, satellite1terrestrial coverage).



DOI: 10.1002/sat

enforced in stand-alone satellite systems and relative results are shown with dashed lines. Usualassumptions as in the previous cases hold, with the exception that dual-coverage (RTC) andterrestrial rejection ðPTC

rej Þ probabilities are fixed at 0.5 and 0.1, respectively. Again, a b value of50 has been considered.

Figure 7 presents the results obtained using three different threshold Tload values for thenumber of virtual GCs. As shown in Equation (1), the threshold defines the limit above whichnew requests for initial access to the satellite resources are randomly rejected in order toprioritize the handover traffic. As expected, both NBP and HFP increase with increasing thenew connection arrival rate. In addition, NBP decreases when decreasing the threshold valuebecause it directly affects the admission probability. This results in fewer resources for handoverpurposes that therefore see an increase in the rate of failures. One should note the improvementsachieved in terms of both NBP and HFP with respect to stand-alone satellite systems. From ourperspective, this confirms the capacity of terrestrial gap-fillers to increase resource utilizationwithout affecting the QoS of ongoing connections. The effects of this temporary diversion resultmore explicit for low threshold values. This is related to the amount of resources that arevirtually reserved for handover traffic. In fact, a high threshold means that new connectionsstart getting rejected at lower load levels. As a consequence, more resources can be utilized forhandling handover requests without recurring to terrestrial diversion, but channel utilizationinevitably decreases.

A similar trend is shown in Figure 8, which refers to the static GC strategy. This admissioncontrol scheme statically reserves a number of TRF slots for the exclusive use of handovertraffic. As a consequence, new connections are admitted into the system only if the congestionlevel is below a given threshold; otherwise, they are rejected. In this case, the impact of

2 4 6 80

0.005

0.01

0.015

0.02

0.025

0.03


New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

Rej=0 S+T Rej=0.25 S+T Rej=0.50 S+T Rej=0.75 S+T Sat Only

2 4 6 80

0.002

0.004

0.006

0.008

0.01

0.012


ando

ver

Failu

re P

roba

bilit

y

Figure 6. NBP and HFP for different PTCrej values (Rej, rejection probability; S1T, satellite1terrestrial

coverage).



DOI: 10.1002/sat

inter-segment handover on the network performance is less visible due to the discretization ofresources in terms of frequency–time slots. In fact, with respect to what proposed by Ramjeeet al. [27], this strategy puts aside a given number of channels for the exclusive use of handovertraffic. In our network configuration, GC values in the range [1, 2, 3] correspond to reserve 5, 10and 15% of the total capacity, respectively. This scheme is therefore far less flexible and theimpact of terrestrial diversion is less visible for high reservation policies. As shown in Figure 8,improvements due to terrestrial diversion are more evident when a single TRF slot (i.e. a singleGC) is used for handover purposes. This is the consequence of higher resource utilization withrespect to the cases in which the number of GCs is 2 or 3. In fact, the hybrid satellite–terrestrialapproach is more effective when there is shortage of resources for handover traffic. Obviously, ahigh level of resource reservation for intra-system handover traffic such as the cases in whichGC5 2 or 3 (i.e. 2 or 3 TRF slots, channels) sensibly decreases the need for inter-systemhandover. Nevertheless, when the proportion between reserved and shared resources increasesin favour of the latter (i.e. GC5 1 TRF slot), improvements in terms of both HFP and NBPbecome more noticeable with respect to stand-alone satellite systems due to the availability ofbackup capacity in the terrestrial domain.

4.3. Mobile DVB-RCS network simulator

This section of the article introduces the simulation platform that has been employed to verifythe results obtained via our mathematical model. A detailed description of the handoverprocedures and packet processing at RCST side and NCC side is also provided. In an effortto attain a means of performance evaluation as accurate as possible, we have developeda packet-level ns-based simulator [34] of a multi-satellite, multi-beam DVB-RCS network,

2 4 6 80

0.01

0.02

0.03

0.04

0.05

0.06

0.07


New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

Thr=1 S+T Thr=1 S Thr=2 S+T Thr=2 S Thr=3 S+T Thr=3 S

2 4 6 80

0.002

0.004

0.006

0.008

0.01

0.012


ando

ver

Failu

re P

roba

bilit

y

Figure 7. NBP and HFP for fractional guard channel strategy (Thr, threshold; S, satellite only; S1T,satellite1terrestrial coverage).



DOI: 10.1002/sat

which incorporates multi-tier technology and node mobility scenarios, Super-frame Composi-tion Table (SCT)/Frame Composition Table (FCT)/Timeslot Composition Table (TCT)/TBTP-based dynamic capacity allocation mechanisms, full Network Information Table (NIT)/RCSMap Table (RMT)/Programme Association Table (PAT)/Programme Map Table (PMT)-basedlogon sequence and synchronization, SYNC-CMT time corrections, Programme ClockReference (PCR)/Network Clock Reference (NCR) synchronization, random tuning delaysand MPE/MPEG and ATM segmentation and reassembly. The simulator supports the DVB-S2standard mechanisms for signalling services as well as MPE/MPEG segmentation andreassembly on the forward link. However, FEC, high-order modulations and ACM have notbeen implemented, unlike the DVB-RCS standard descriptors.

In our DVB-RCS simulator, we have removed the simplifications made in previoussections: both terrestrial cells and satellite spot-beams are assumed as being circular withthe terrestrial cells providing backup capacity for terminals within the overlapping area.Coverage radii are specified in Table I. RCSTs are initially created in locations thatare uniformly distributed within the spot-beam coverage. However, latitudes above theintersection points between adjacent spot-beams are excluded in order to ensure coveragecontinuity.

The admission control phase is triggered every time the NCC successfully receives a commonsignalling channel (CSC) burst. The NCCAgent class is responsible for evaluating the networkstatus by estimating the current load and the amount of handover traffic that is expectedfrom the neighbouring beams. The policy specified in Equation (1) is then enforced and auniformly distributed number is extracted and compared with the admission probability value.If the connection is granted access to the network, a Logon TIM message is destined for the

2 4 6 80

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05


New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

GC=1 S+T GC=1 S GC=2 S+T GC=2 S GC=3 S+T GC=3 S

2 4 6 80

0.5

1

1.5

2

2.5

3

3.5

4

4.5x 10


ando

ver

Failu

re P

roba

bilit

y

Figure 8. NBP and HFP for static guard channel strategy (GC, number of guard channels; S, satelliteonly; S1T, satellite1terrestrial coverage).



DOI: 10.1002/sat

awaiting RCST; otherwise, a Logon Denied TIM is generated and the terminal abandons thesystem.

As mentioned in Section 2, terminal positioning information is a necessity forsynchronization acquisition. In addition, GPS equipment is commonly available in mostvehicular networks. Therefore, it is reasonable to further exploit the GPS-derived informationfor the purpose of devising DVB-RCS mobility management mechanisms, as explained later. Inmost DVB-RCS implementations, the NCC periodically assigns SYNC slots to each RCST fortiming/frequency/power correction purposes. Defever in [11] firstly proposed using such burstsalso for HO-related signalling. Unfortunately, only a portion [0x0003–0x7FFF] of the 16-bitM_and_C sub-field of the SAC field in SYNC bursts is currently reserved for future use. This isnot sufficient to carry the GPS-based location information. However, the DVB TechnicalModule has recently introduced in the new release of the DVB-RCS1M standardz a newoptional sub-field in the SAC that is specifically designed for mobility management. As a result,the aforementioned reserved space may be used to convey channel quality measurements or ameasure of urgency associated with the HO need.

Our HO mechanism can be classified as mobile-assisted and relying on positionmeasurements and channel quality assessment. RCSTs periodically monitor their ownposition (GPS) and scan different frequencies to detect other networks. It is assumed thateach RCST is equipped with the information regarding the centre positions and the radii of allspot-beams. This information is not expected to change often and thus can be loaded at the timeof installation and updated periodically. For the sake of simplicity, all spot-beams are equallydimensioned with two-tier regions dispersed along the overlapping area between adjacent spot-beams.

Details of the proposed HO strategy are explained within the context of our ns-basedsimulator (Figure 9).

A HOManager class entity is embedded into each RCST, which is responsible for periodically(every 2 s) deriving its own position within the hybrid network. In particular, whenever a spot-beam centre is identified, which is closer than the current serving spot-beam, the need for ahandover is detected. Owing to the predictability of high-speed train routes, the HOManagerhas the ability of foreseeing when the terminal is expected to cross the beam edge and derivingthe urgency indicator accordingly. RCSTs also scan a given range of frequencies to detectterrestrial wireless networks. The MAC class entity in our RCST nodes relays a SYNC burst tothe HOManager object before transmitting to the NCC (Figure 9(a)). A list containingthe target spot-beam and the detected terrestrial networks is specified together with theurgency indicator and an estimation of the received power level from the terrestrial network(Figure 9(b)). The SYNC burst is then returned to the MAC entity and finally transmitted overthe wireless medium. The NCC is responsible for HO decision and execution. In accordancewith the standard, each SYNC burst is acknowledged by a CMT table. However, if the SYNCalso carries HO information, the HO decision phase is triggered. It is instructive to walk throughthe NCC-side HO process in the context of our ns-based DVB-RCS simulator (Figure 10).

The NCCAgent class maintains separate RCST registers for each spot-beam, which containall terminal-related information. Upon reception of a handover recommendation, the NCCretrieves the appropriate details from the source uplink and tries to assign the RCST a new

zAt the time of writing this article, a draft version of the extended standard for mobility support was undergoing editingrevision at ETSI before being publicly released.



DOI: 10.1002/sat

Logon ID and a periodical SYNC slot for use in the target spot-beam.y If both operations returnsuccessfully, the NCCAgent verifies that the target beam can support the QoS of the incomingconnection. If also this task can be accomplished, the NCCAgent transfers the RCST detailsfrom source to the target uplink. This is followed by the generation of a unicast TIM tablecontaining all necessary descriptors to enable the terminal to receive and transmit over the newchannels. The process terminates by modifying the packet classifiers in the GW node so that IPdatagrams destined for the RCST are switched to the target downlink. This event is scheduledwith a delay that is equal to half of the system round-trip time as this is roughly the timenecessary to complete physical synchronization to the new forward link signalling service(FLSS) transport stream.

Unfortunately, physical and logical resources are not always sufficient to serve immediatelyall HO requests. SYNC slots and Logon IDs may not be available or the target spot-beam maynot be able to guarantee the request QoS in terms of CRA slots. Different strategies can beimplemented to minimize the effects of such events. For example, the NCCAgent may delay the

Figure 9. HOManager process at RCST.

yEach terminal is allocated a dedicated SYNC slot per super-frame in a round-robin manner. The total number ofterminals that can be simultaneously supported is therefore limited by the number of SYNC slots allocated per super-frame. In our simulations, the SYNC repetition period is set at 40 super-frames.



DOI: 10.1002/sat

HO decision according to traffic load or interference levels in the target spot-beam [3].Handover requests, whose urgency indicator is low enough, may also be discarded [22].

In this article, we analyse the case when the target spot-beam cannot accommodate anincoming connection, but a suitable terrestrial network is detected by the RCST. If this eventoccurs, the NCC is responsible for contacting (directly or via the NOC) its peer entity in thetarget network and negotiating the connection transfer. The NCCAgent may use the channelquality estimations provided by the RCST to decide if and when the terminal should proceedwith inter-segment handover. Resource availability must be also verified in the terrestrialwireless network. However, for the sake of simplicity, the inter-segment handover traffic isrejected according to a given probability, as discussed in previous paragraphs.

Upon the successful reception of the TIM message, the RCST node releases all satelliteresources and starts physical-layer synchronization to the new forward link. This may be eithera satellite or a terrestrial link. As a result, access discipline to the return link may differ,according to the technology adopted, but it is expected not to impact the simulation results. Theterrestrial network is therefore modelled as an independent domain that has the capacity ofhosting satellite connections as long as either resources are made available in the target

Figure 10. HO operations at NCCAgent.



DOI: 10.1002/sat

spot-beam or the terminals fall off the service coverage, whichever occurs first. When one ofthese events takes place, an attempt to transfer the connection back to the satellite domain ismade, which may or may not be successful.

4.4. Simulation scenario and results

A simple two spot-beam topology is considered in the simulations according to the scenario inFigure 2. Spot-beam centres are located on 5 and 151 longitudes and 01 latitude, whereas theradii are equal to 800 km. The GEO satellite nadir point is on 101 longitude. Table II(a) gives thesuper-frame channel organization [16] in the evaluation scenario. One should note that super-frames and frames have the same duration, which is equal to 0.02651 s. The return pathcomprises three carriers. CSC slots are used for network random access according to the slotted-Aloha protocol. SYNC slots are periodically assigned (every 1.0604 s) to terminals forsynchronization maintenance. Finally, 19 TRF slots carry user data.

The forward link contains three transport streams in each spot-beam (Table II(b)). The start-up transport stream carries the NIT table only. The RMT transport stream carries PAT, PMTand RMT tables. Finally, the FLSS transport stream carries PAT, PMT, SCT, FCT, TCT,TBTP tables as well as data traffic. Table III presents our DVB-RCS settings.

Terrestrial gap-fillers are located in proximity of the overlapping areas between neighbouringspot-beams. Exact position and extension of the terrestrial coverage is a simulation parameter

Table II. RTN-FWD link channel organizations.

(a)Frequency 1 CSC CSCFrequency 2 SYNC TRF TRF TRF TRF TRF TRF TRF TRF TRFFrequency 3 TRF TRF TRF TRF TRF TRF TRF TRF TRF TRF

(b)Start-up transport stream NITRMT transport stream PAT PMT RMTForward link signalling service PAT PMT SCT FCT TCT TBTP Data traffic

Table III. DVB-RCS simulator settings.

Broadcast periods (s) Timeout values (s)

NCR 0.2651 NCR 3NIT 5 TBTP 10RMT 5 CSC response 0.636PAT 1 SYNC response 0.636PMT 0.5 CSC channelSCT 0.02651 CSC max. losses Three timesFCT 0.02651 CSC max. retries Five timesTCT 0.02651 CSC max. backoff 1 sTBTP 0.02651 SYNC slotsCMT 1 SYNC period 1.0604 s

Guard times (ms) SYNC max. tries 6CSC 15 Achieve threshold 0.001 sTRF/SYNC 1 SYNC max. losses 3



DOI: 10.1002/sat

that is modelled by the RTC value. When switched on, terminals start moving towards theopposite spot-beam along straight trajectories that are parallel to the equator at 345 km/h [33].As a result, the handover arrival process is not under our control, but is the consequence ofseveral parameters such as initial location, terminal mobility and traffic conditions. Theconnection duration is modelled as an exponentially distributed random variable whose averageis 2 h. In order to increase the statistical reliability of the simulations, our results are the averageof 10 simulation runs (i.e. 86 400 s).

Figure 11 shows a comparison between the values derived from our DVB-RCS simulator andthe analytical model described in Section 3 when implementing the fractional GC admission(threshold, Tload 5 3) control strategy and the fixed GC strategy (GC5 1 TRF) in the inter-segment handover scenario. In order to prove the accuracy of our mathematical model undervarious conditions, the two CAC strategies have been tested using very different settings (i.e.different number of GCs). RCSTs are considered as having 50% probability of crossing thespot-beam edge in a two-tier region. New connection arrivals form a Poisson process whoseaverage spans from 2 to 7.5 connections per hour. In addition, the probability that the terrestrialnetwork rejects an inter-segment HO request is PTC

rej ¼ 0:1.Statistics about NBP and HFP derived from simulations (plotted in dashed lines) show a

small deviation with respect to theoretical values at high loadings. It is reasonable to believe thatthis difference is due to the simplifications regarding the handover arrival process and thechannel holding time distribution. In particular, the assumption regarding the Poissonian nature

2 4 6 80

0.01

0.02

0.03

0.04

0.05

0.06


New

Con

nect

ion

Blo

ckin

g P

roba

bilit

y

2 4 6 80

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5x 10


Han

dove

r Fa

ilure

Pro

babi

lity

Thr=3 (Theory) Thr=3 (Sim) GC=1 (Theory) GC=1 (Sim)

Figure 11. NBP and HFP simulation results (Thr, threshold value for the number of virtual guardchannels; GC, number of static guard channels).



DOI: 10.1002/sat

of the new connection arrival rate does not apply to handover traffic, which has a smoothedbehaviour [35].

The last part of our study is dedicated to deriving an overall performance figureassociated with each admission control strategy. To this end, we define a ‘relative costfunction’ as follows [36]:

COST ¼ wNBP

NBP0þ g

HFP

HFP0þ e

UTIL0

UTILð19Þ

where NBP, HFP and UTIL are the call blocking probability, the handover failure probabilityand the average resource utilization, respectively. NBP0, HFP0 and UTIL0 refer to the satellitestand-alone non-reservation scheme (in which new connections and handover traffic are treatedequally) and provide a normalization factor for all the schemes under analysis. Instead, w, g ande represent the weights associated with each parameter. In our analysis, we prioritize thehandover blocking probability by assigning g a higher weight in the cost function with respect toNBP and resource utilization: g ¼ 1

2, w ¼ e ¼ 14. Figure 12 shows the behaviour of the cost

function for the fractional GC (threshold, Tload 5 1 TRF) and the static GC (GC5 1 TRF)strategies with respect to the offered load. In both cases we set the dual-coverage and theterrestrial rejection probabilities at 0.5 and 0.1, respectively. In addition, the penalty value b isset at 50. One should note that the cost sensibly decreases when increasing the offered load forboth strategies since so does the resource utilization. In particular, when the connection arrivalrate increases, it becomes evident that the integrated network solution permits a lower cost thanthe stand-alone scenario. This is an important proof of the effectiveness of the gap-filler

2 3 4 5 6 7 8 9

0.8

1

1.2

1.4

1.6

1.8

2

2.2


Tota

l Cos

t

Thr=1, S+T

Thr=1, S

GC=1, S+T

GC=1, S

Figure 12. Cost function (Thr, threshold value for the number of virtual guard channels; GC, number ofstatic guard channels; S, satellite only; S1T, satellite1terrestrial coverage).



DOI: 10.1002/sat

approach. In particular, when the fractional GC strategy is implemented, the combined effect ofprobabilistic connection blocking and terrestrial diversion increases the resource utilizationand minimizes the HFP, thus permitting to achieve a sizeable improvement with respect tostand-alone solutions. With regard to the comparison between the two admission controlstrategies, we register a lower overall cost for the static GC. This is due to its strict reservationpolicy that minimizes the HFP. In fact, static approaches better adapt to scenarios characterizedby high handover traffic intensity like our railway mobility. In particular, despite providinga lower connection blocking probability in intermediate states above the loading threshold,the fractional GC policy does not perform as well as the GC strategy in terms of HFP. Fromour analysis, the fractional GC proves to be better suited for scenarios characterized by lowmobility where its high new connection acceptance rate can be better exploited. Nevertheless,the gap between conservative and probabilistic admission control strategies reduces whenincreasing the offered traffic and becomes negligible above eight connections per hour, wherethe fractional GC policy supporting the inter-segment handover outperforms stand-alonesystems.

5. CONCLUSIONS

In the last two decades, wireless communications have undergone a fast growth, which isgoing towards the integration of several domains and technologies. As a result, futuremulti-mode mobile terminals will be able to switch from one network to another withoutlosing connectivity. This transition will be technology-independent as mobile users expect to beable to access their favourite services at anytime, regardless of their mobility pattern andposition within the coverage. DVB-RCS standard is also evolving and the next release willsupport terminal mobility. To this end, we proposed a network architecture in which terrestrialgap-fillers provide backup connectivity to the satellite domain by hosting connections wheneverspot-beam handover cannot be completed due to temporary lack of resources in the satellitenetwork.

In the first part of this article, we presented an analytical formulation that models thebehaviour of this hybrid network. We analysed the sensitivity of the model and showed that theextension of the gap-filler coverage and the terrestrial connection rejection probability can beused as a means to achieve specific QoS targets in terms of NBP and handover blockingprobability. We also derived performance measurements for the fractional GC and the staticGC admission control strategies. The proposed analytical model permits to appreciate theconditions under which the use of inter-satellite handovers improves the QoS perceived bymobile users. In the second part of the article, we described our ns-based DVB-RCS simulator.Details regarding RCSTs and NCC implementations were also provided. We comparedanalytical and simulation results and demonstrated the accuracy of our mathematical model fortwo different network configurations. Finally, we specified a collective cost function, whichprovides the satellite operator with an effective means for describing the overall networkperformance. It is concluded that DVB-RCS networks completed with terrestrial gap-fillershave a lower cost with respect to stand-alone systems when the average offered load exceedsfour connections per hour. In addition, the gap between the two admission control schemesanalysed decreases when the average offered load increases, with both policies outperformingstand-alone solutions above 7.5 connections per hour.



DOI: 10.1002/sat

ACKNOWLEDGEMENTS

This work was supported by the European SatNEx II (contract No. IST-027393) Network of Excellence,http://www.satnex.org, within joint activity JA2330.

REFERENCES

1. Falowo OE, Anthony Chan H. Optimal joint radio resource management to improve connection-level QoS in nextgeneration wireless networks. Radio and Wireless Symposium, 2008. IEEE: New York, 2008.

2. Ryu J et al. The gap filler technology for mobile satellite system. Advanced Satellite Mobile Systems, 2008, 4th ASMS’08, Bologna, Italy, 2008.

3. Lattanzi F, Acar G, Evans BG. DVB-RCS spotbeam handover using residence time estimations in vehicular satellitenetworks. Advanced Satellite Mobile Systems, 2008, 4th ASMS ’08, Bologna, Italy, 2008.

4. Maral G et al. Performance analysis for a guaranteed handover service in an LEO constellation with a ‘satellite-fixedcell’ system. IEEE Transactions on Vehicular Technology 1998; 47(4):1200–1214.

5. El-Kadi M, Olariu S, Todorova P. Predictive resource allocation in multimedia satellite networks. GlobalTelecommunications Conference, 2001, GLOBECOM ’01, IEEE: New York, 2001.

6. Ming-Hsing C, Bassiouni MA. Predictive schemes for handoff prioritization in cellular networks based on mobilepositioning. IEEE Journal on Selected Areas in Communications 2000; 18(3):510–522.

7. Del Re E, Fantacci R, Giambene G. Different queuing policies for handover requests in low Earth orbit mobilesatellite systems. IEEE Transactions on Vehicular Technology 1999; 48(2):448–458.

8. Pollini GP. Trends in handover design. IEEE Communications Magazine 1996; 34(3):82–90.9. Tripathi ND, Reed JH, VanLandinoham HF. Handoff in cellular systems. IEEE Personal Communications 1998;

5(6):26–37.10. Verdone R, Zanella A. Performance of received power and traffic-driven handover algorithms in urban cellular

networks. IEEE Wireless Communications 2002; 9(1):60–71.11. Defever S. DVB-RCS for mobile applications: a way to reduce the costs through extension of the DVB-RCS

standard. AIAA, ICSSC, and 11th Ka and Broadband Communications Conference, Rome, Italy, September 2005.12. Zhao W, Tafazolli R, Evans BG. Internet work handover performance analysis in a GSM-satellite integrated mobile

communication system. IEEE Journal on Selected Areas in Communications 1997; 15(8):1657–1671.13. Stemm M, Randy HK. Vertical Handoffs in Wireless Overlay Networks. ACM: New York, 1998; 335–350.14. Fang Z, McNair J. Optimizations for vertical handoff decision algorithms.Wireless Communications and Networking

Conference, 2004, WCNC 2004, IEEE: New York, 2004.15. Liu X, Li VOK, Zhang P. NXG04-4: joint radio resource management through vertical handoffs in 4G networks.

Global Telecommunications Conference, 2006, GLOBECOM ’06. IEEE: New York, 2006.16. ETSI. EN 301 790 (V1.4.1), Digital Video Broadcasting (DVB): Interaction Channel for Satellite Distribution Systems,

ETSI, March 2006.17. Kasparis C, Acar G, Skoutaridis P, Evans B, Klaeyle A, Mateus J, Vincent P. Mobile wideband global link system

(MOWGLY)—aeronautical, train and maritime global high-speed satellite services. Joint 23rd AIAA, ICSSC, and11th Ka and Broadband Communications Conference, Rome, Italy, September 2005.

18. Morlet C, Ginesi A. Introduction of mobility aspects for DVB-S2/RCS broadband systems. 2006 InternationalWorkshop on Satellite and Space Communications, Madrid, Spain, 2006.

19. Conforto P, Losquadro G. Fast internet for fast train hosts: the FIFTH project. The 8th Ka-band UtilizationConference, Baveno, Italy, 2002.

20. Sciascia G et al. Statistical characterization of the railroad satellite channel at Ku band. International Workshop ofCOST Actions 272 and 280, ESA-ESTEC, Noordwijk, The Netherlands, May 2003.

21. Ray ES et al. Space/Terrestrial Mobile Networks: Internet Access and QoS Support. Wiley: New York, 2004.22. Acar G, Lattanzi F, Evans BG. DVB-RCS mobility management in vehicular satellite networks. The 25th AIAA/

ICSSC Conference, Seoul, Korea, April 2007.23. Lattanzi F, Acar G, Evans BG. Predictive reservation based connection admission control for mobile DVB-RCS

networks. The 26th AIAA/ICSSC Conference, San Diego, CA, U.S.A., June 2008.24. ETSI. TR 101 790, Digital Video Broadcasting (DVB): Interaction Channel for Satellite Distribution Systems:

Guidelines for the Use of EN 301 790, ETSI, March 2005.25. Perkins C. RFC No. 2002: IPv4 Mobility Support, 1996.26. Hernandez E, Helal A. Examining mobile-IP performance in rapidly mobile environments: the case of a commuter

train. The 26th Annual IEEE Conference on Local Computer Networks, 2001, Proceedings of LCN 2001, Tampa, FL,U.S.A., 2001.



DOI: 10.1002/sat

27. Ramjee R, Nagarajan R, Towsley D. On optimal call admission control in cellular networks. INFOCOM ’96, The15th Annual Joint Conference of the IEEE Computer Societies, Networking the Next Generation, Proceedings of theIEEE, San Francisco, CA, U.S.A., 1996.

28. Ross KW. Multiservice Loss Models for Broadband Telecommunication Networks. Springer: Berlin, 1995.29. Wee-Seng S, Kim HS. Dynamic bandwidth reservation in cellular networks using road topology based mobility

predictions. INFOCOM 2004, The 23rd Annual Joint Conference of the IEEE Computer and CommunicationsSocieties, Hong Kong, 2004.

30. Daehyoung H, Rappaport SS. Traffic model and performance analysis for cellular mobile radio telephone systemswith prioritized and nonprioritized handoff procedures. IEEE Transactions on Vehicular Technology 1986; 35(3):77–92.

31. Yi-Bing L, Li-Fang C, Noerpel A. Modeling hierarchical microcell/macrocell PCS architecture. IEEE InternationalConference on Communications, 1995, ICC ’95, Gateway to Globalization, 1995, Seattle, WA, U.S.A., 1995.

32. Fang Y, Zhang Y. Call admission control schemes and performance analysis in wireless mobile networks. IEEETransactions on Vehicular Technology 2002; 51(2):371–382.

33. Alstom_Transport. AGV Technical Sheet: Challenge, Concept and Key Figures, 2008. Available from: http://www.transport.alstom.com/home/elibrary/technical/products/_files/file_32033_13411.ppt.

34. NS-2. The Network Simulator: Online Documentation. Available from: http://www.isi.edu/nsnam/ns/doc/ns_doc.pdf,2009.

35. Delbrouck L. A unified approximate evaluation of congestion functions for smooth and peaky traffics. IEEETransactions on Communications (legacy, pre-1988] 1981; 29(2):85–91.

36. Su W, Gerla M. Bandwidth allocation strategies for wireless ATM networks using predictive reservation.Global Telecommunications Conference, 1998, GLOBECOM 98, The Bridge to Global Integration. IEEE:New York, 1998.

AUTHORS’ BIOGRAPHIES

Fabio Lattanzi received his BSc and MSc in Telecommunications Engineering at theUniversity of Rome Tor Vergata in 2002 and 2005, respectively. During his studies,he was employed as a visiting research scholar at the Ericsson Lab Italy and theEuropean Space Agency, The Netherlands. He worked as a junior consultant for thetelecom sector in Italy and is currently a PhD candidate at the University of Surrey.His research interests focus on the design, modelling and performance assessment ofmobile vehicular satellite systems. He is currently involved in the SatNEx II projectworking on the networking solutions for mobile DVB-RCS networks.

Giovanni Giambene was born in Florence, Italy, in 1966. He received the DrIng degreein Electronics in 1993 and the PhD degree in Telecommunications and Informatics in1997, both from the University of Florence, Italy. From 1994 to 1997, he was with theElectronic Engineering Department of the University of Florence, Italy. He was thetechnical external secretary of the European Community COST 227 Action. From1997 to 1998, he was with OTE of the Marconi Group, Florence, Italy. In 1999, hejoined the Information Engineering Department of the University of Siena, Italy, firstas a research associate and then as an assistant professor. He teaches the advancedcourse of telecommunication networks at the University of Siena. He was vice-chairof the COST 290 Action (www.cost290.org) for the whole of its duration 2004–2008,entitled ‘Traffic and QoS Management in Wireless Multimedia Networks’ (Wi-QoST). At present, he is involved in the SatNEx network of excellence of the FP6

programme in the satellite field, as work package leader of two groups on radio access techniques andcross-layer air interface design (www.satnex.org). He also participates in the FP7 Coordination ActionRADICAL as work package leader (http://www.radicalhealth.eu/).



DOI: 10.1002/sat

Guray Acar is an EPSRC RCUK academic fellow in satellite communications atCCSR. He completed his PhD studies in Broadband Satellite Networking and MScstudies in Communications and Signal Processing at the Imperial College, London, in2001 and 1997, respectively. He received his BSc degree in Electronic and ElectricalEngineering at the Middle East Technical University, Turkey, in 1996. During thesecond half of his PhD studies, he was employed as a visiting research scholar at thePurdue University, U.S.A. His research interests cover radio resource management,MAC protocols, reliable multicast transport protocols, mobility management andsystem- and packet-level simulation analyses in mobile and satellite networks.

Barry Evans is the pro-vice-chancellor for Research and Enterprise at Surrey and isresponsible for the strategy and implementation of the whole university researchportfolio and the knowledge transfer from research to business. He is also thedirector of the Centre for Communication Systems Research, which is a 160 strongpostgraduate research centre with around £12m research turnover covering mobile,wireless and satellite communications networking and multimedia research. Thecentre was the largest U.K. recipient of EU IST FP6 funds and now is involved innine FP7 projects. The centre is also a partner in EPSRC’s largest portfolio grant inintegrated electronics bringing together nanotechnology with signal processing andcommunications. He has a background in satellite communications and is the editorof International Journal of Satellite Communications and BNSC board member. He isa fellow of the Royal Academy of Engineering.



DOI: 10.1002/sat

Date post:	09-Jan-2023
Category:	Documents
Upload:	unisi
View:	0 times
Download:	0 times

Admission Control and Handover Management for High-Speed Trains in Vehicular Geostationary Satellite...

Documents