Tài liệu Sổ tay của các mạng không dây và điện toán di động P22 pdf

CHAPTER 22Topological Design, Routing, and Handover in Satellite Networks AFONSO FERREIRA, JÉRÔME GALTIER, and PAOLO PENNA Mascotte Team, INRIA/CNRS/UNSA, Sophia Antipolis, France 22.1 I

CHAPTER 22

Topological Design, Routing, and

Handover in Satellite Networks

AFONSO FERREIRA, JÉRÔME GALTIER, and PAOLO PENNA

Mascotte Team, INRIA/CNRS/UNSA, Sophia Antipolis, France

22.1 INTRODUCTION

A low earth orbit (LEO) satellite constellation consists of a set of satellites orbiting theEarth with high constant speed at a relatively low altitude (a few thousand kilometers) [1].Each satellite is equipped with a fixed number of antennas that allow it to communicatewith ground transmitters/receivers and with other satellites One of the major advantages

of LEO satellites (as opposed to geostationary—GEO—satellites) is that they are closer tothe Earth’s surface This reduces the communication delay and the energy required to di-rectly connect a user with a satellite On the other hand, two major issues arise due to theirlow altitude First, because a single satellite can only cover a small geographical area(called footprint) at the Earth’s surface, many satellites are required to provide global cov-erage Second, the footprint of each satellite moves continuously, implying a high mobility

of the whole network, in contrast with other cellular systems

In the following, we will see how the topology of LEO constellations is limited byphysical constraints Then we will review how these factors have been taken into account

in the design of routing and handover policies

22.2 TOPOLOGIES

During the systems design phase, several parameters come into play, such as satellites’ titude, number of satellites, number of orbits and satellites per orbit, how to deploy the or-bits, and how to interconnect the satellites All such factors determine the topology of thenetwork, as shown in this section

al-22.2.1 Orbits

A closer look at the feasible types of orbits shows that unless the orbits have the samealtitude and inclination, their relative positions change so often that intersatellite links

473

Handbook of Wireless Networks and Mobile Computing, Edited by Ivan Stojmenovic´

Copyright © 2002 John Wiley & Sons, Inc ISBNs: 0-471-41902-8 (Paper); 0-471-22456-1 (Electronic)

(ISLs) can hardly connect them for a sufficient amount of time (for more details on bit mechanics with respect to telecommunication services see [1, 39]) Under such con-straints, different kinds of constellations can be obtained according to how the orbits aredeployed.

or-The so-called ␲-constellations form the structure of the Iridium system [20, 22] andwere the basis of the original plans for the Teledesic system [21, 30] The basic structure

of a ␲-constellation consists of a set of orbits that are deployed along a semicircle whenviewed from a pole, as shown in Figure 22.1(a) The satellites are placed along the orbits

so as to obtain maximum coverage of the Earth’s surface In Figure 22.1(c) the ment of satellites along with their footprints is shown We can see that in a ␲-constellationthere are two extreme orbits that are adjacent, but whose satellites move in opposite direc-tions As a result, a seam appears that divides the network into two parts: those satellitesmoving from south to north and those moving from north to south [see Figure22.1(a)–(b)]

deploy-From a communication network viewpoint, the seam is the main drawback of ␲constellations, as will be seen later Also, ␲-constellations suffer from excessive polarcoverage Finally, their unique coverage in many areas and, therefore, sensitivity to manyobstacles, like trees and buildings, does not always ensure sufficient radio signal quality

-In order to avoid these kinds of problems, 2␲-constellations have been proposed A 2␲constellation is constructed by spacing the orbits along a complete circle as shown in Fig-ure 22.2 The 2␲-constellation is used in the Globalstar constellation [9], and has alsobeen planned for the future Skybridge project and the now abandoned Celestri

-Another important aspect concerns the use of “inclined” orbits, that is, orbits whose clination is between the equatorial inclination (0 degrees) and the polar one (90 degrees).Usually, ␲-constellations use polar orbits (informally, orbits that “roughly” cross the polaraxis) for coverage reasons (see Section 22.2.5), and therefore are called “polar” constella-tions On the other hand, inclined orbits allow a better optimization of 2␲-constellations,

in-Figure 22.1 The structure of ␲-constellations (a) View from the north pole (b) View from theequatorial plane (c) The position of satellites on adjacent orbits and the resulting coverage

hence the name “inclined” constellations The use of inclined orbits allows for an increase

in the number of simultaneously visible satellites on more populated or wealthier areas

It is worth observing that there is no technical reason to forbid the use of polar orbits

on 2␲-constellations, and vice versa Moreover, the use of inclined orbits does not affectthe network topology [for instance, ␲-constellations that use inclined orbits still result inthe mesh-like topology shown in Figure 22.4(b)]

Fig-gy, we have to take into account the seam and the relative position of satellites crossing thepoles, as follows

For ␲-constellations, one has to consider the problem of connecting two satellitesmoving in opposite directions, which is too expensive or even infeasible with the exist-ing technology (see Section 22.2.5) Hence, it is commonly assumed that two such satel-lites cannot be directly connected over the seam, even though they are “physically” closeone to each other Therefore, long user-to-user delay can occur even when the two par-ties are geographically close to each other but the covering satellites are separated by theseam Also notice that two adjacent satellites swap their relative position whenevercrossing the poles [see Figure 22.4(a)] Hence, the network topology can be represented

22.2 TOPOLOGIES 475

Figure 22.2 2␲-constellations (a) View from the north pole (b) View from the equatorial plane

as a two-dimensional mesh in which columns are wrapped around, but rows are not [seeFigure 22.4(b)]

In [15] the impact of the ISLs architecture (for instance, the use of antennas that port higher angular velocity) has been studied, and further patterns to connect the satel-lites of a ␲-constellation have been proposed Such patterns use interorbital links that con-nect satellites in nonadjacent orbits, typically the neighboring orbit of the neighboring

sup-Figure 22.3 Some intersatellite link patterns

Figure 22.4 The relative position of two adjacent satellites crossing the pole and the resultingtopology of ␲-constellations

south

north

south

orbit This reduces the user-to-user delay when the communication takes place betweentwo positions that are quite far apart (or when the communications have to go across theseam) Assuming use of ISLs that support high angular velocity, the delay effects of ISLsthat cross the seam have also been investigated in [15]

With respect to intersatellite links for 2␲-constellations, neither Globalstar nor bridge have implemented ISLs in their design, although it seems that they had been con-sidered in the early phases of these projects, as was the case in Celestri At that point intime, many designers thought those projects were too innovative accelerate the introduc-tion of this additional new feature Nevertheless, there is a strong belief that future designs

Sky-of 2␲-constellations will introduce such links

From the topology point of view, it is worth observing that the regular torus turns into askewed torus if an inclined ISL pattern such as the one of Figure 22.3(b) is adopted [17].Notice that 2␲-constellations do not present any seam Thus, their coverage has smootherproperties On the other hand, a unique position may be covered by two satellites quite farone from another in the network topology (e.g., two satellites that move in opposite direc-tions), especially when the user is close to the equator

22.2.3 ISLs versus Terrestrial Gateways

The use of ISLs is intended to implement communications that do not use any terrestrialinfrastructure However, the use of terrestrial gateways still present some advantages such

as a reduced number of computing devices on board the satellites For instance, gatewayscan be used to compute the routing tables that are used by the satellites

A more extensive use of the gateways has been adopted in the Globalstar system, inwhich the satellites operate in a “bent pipe” mode Their main function is to redirect usersignals to ground gateways, and vice versa As a result, the operator has to build manygateways, one for each area in which the service is opened Additionally, part of the radiospectrum is used to support the communications between the satellites and the gateways.Unfortunately, radio resources are becoming a scarce resource Currently, several systemsshare the same spectrum of frequencies (Globalstar, ICO, and probably Ellipso), which isthe source of several interference problems

We note that the use of ISLs presents significant advantages, like reducing the nications between the satellites and the gateways, reducing the number of gateways, bal-ancing the load between the gateways, and preventing gateway faults

commu-22.2.4 Multiple Coverage

Another important issue for satellite constellations with ISLs is multicoverage goals.From the radio and signal propagation points of view, a single satellite may not suffice toensure the real-time connection, especially if some obstacles exist between the user andthe satellite Systems like Globalstar [9] answer this problem using multipath techniques.Instead of being received by one satellite, the signal is received by two to four satellitesand merged to recover a clear signal When a new satellite is visible to the user, its signalcontribution is introduced progressively into the global merging of the signals

We remark that routing with multipath techniques in a satellite constellation is very

22.2 TOPOLOGIES 477

challenging A single user may be directly connected to two (or more) satellites that arevery far one from another in the network topology, mainly in inclined constellations Fromthe algorithmic point of view, this characteristic essentially turns a basic network routingproblem into a multicasting problem

22.2.5 Physical and Technological Constraints

In this section we discuss some of the main physical and technological factors that impact

on many of the above design choices

The main technological constraints to take into account in ISL design are the relative lar velocity of the endpoints and their visibility [17] This is because antennas cannot toler-ate excessive angular speed and the atmosphere is also a source of fading of the signal

As a satellite moves along its orbit, the set of satellites visible from it changes

continuous-ly This happens for those satellites that are not in adjacent orbits and, in polar tions, whenever the satellite approaches the poles This is due to the small distance betweenadjacent satellites approaching the pole, which results in a higher angular velocity [1, 15].Additionally, ISLs between adjacent orbits must be turned off when crossing the poles be-cause of the satellites’ relative position switching (see Figure 22.4) As observed in [15],ISLs that support higher angular velocity allow maintainence of intraorbital links at higherlatitudes An unexpected side effect of the angular velocity is that the tracking system mayaffect the stability of the satellite within its orbit and therefore result in an additional con-sumption of fuel, which in turn impacts the satellite’s weight and time in service

It is worth observing that the distance between two adjacent polar orbits decreases as theyget closer to the poles Hence, for ␲-constellations using the “W” ISLs pattern, for in-stance, the minimum delay path is the one that uses a minimal number of ISLs and in-terorbital links whose latitude is the maximum latitude between the two satellites to beconnected

Notice that routing algorithms on mesh-like topologies may return suboptimal time/delaypaths, since such models do not consider that the orbit distance varies with the latitude In[10] a model that takes this issue into account has been investigated

22.3 NETWORK MOBILITY AND TRAFFIC MODELING

There are two main factors that should be taken into account when designing routing rithms for LEO satellite constellations:

algo-1 Users’ distribution: the fact that the position of the users and the duration of the

communications are not known in advance

2 Network mobility: the fact that satellites move, constantly changing the network

topology

Although the first aspect has been extensively studied for classical cellular networks,such networks use wired connections in order to connect two base stations Hence, themain issue in these “terrestrial networks” is to provide enough resources for the user’sconnection to last There is a lot of flexibility in the size of the cells, but the users maymove from one to another, and at different speeds Conversely, LEO cells are big enough

to consider the users immobile However, routing problems occur, since on-board sources, in particular the maximum number of connections using ISLs—are scarce.The second aspect, namely the network mobility, is a distinguishing factor of LEO con-stellations Indeed, even if we assume a static set of communications (i.e., pairs of usersthat want to communicate one with each other), the problem of maintaining active connec-tions over time is not a trivial task—the satellite’s movement triggers both handovers andconnections updates (rerouting) when a topology change occurs

re-In both cases, mobility is the main cause of call blocking, call dropping, and

unbound-ed delay in communications However, there is a fundamental difference between theusers’ mobility and the network’s mobility: The users’ behavior is not deterministic,whereas changes to the network topology are predictable Hence, two different approachesare generally adopted:

1 The network’s behavior is deterministic and can be “predicted” quite accurately (seeSection 22.3.1)

2 The users’ behavior is usually modeled by means of a probability distribution (seeSection 22.3.2)

It is worth observing that if we consider the relative movement between a user and thesatellites, then the major part of such movement is due to the satellites’ speed Hence, theprobability distributions used to model users’ mobility mainly focus on the issue of man-aging requests whose position and duration are not known prior to their arrival

22.3.1 Satellite Mobility

One of the main differences between “classical” cellular networks and LEO constellations

is the high mobility of the system Complicating factors such as the satellite movementand the Earth’s self-rotation make the problem of connecting “immobile” users nontrivial

In the following, we describe the interplay between these two factors and the previouslymentioned aspects, and also how network mobility can be modeled

Satellite movement is the main cause of handovers Two types of handover may occur:

1 A satellite handover is the transfer of a user from one satellite to another during acommunication

22.3 NETWORK MOBILITY AND TRAFFIC MODELING 479

2 A cell handover is the transfer of a user from one spot beam to another within thesame satellite A satellite antenna directed to terminals is composed of a series ofbeams Such a decomposition of the satellite footprint allows reuse of the radio fre-quencies several times in its coverage area These handovers have no impact on in-tersatellite routing, but seriously impact on-board computations

If a user is just on the border of the coverage area of a satellite, his/her connection time

to an individual satellite can be extremely small Hence, in general, constellations aredesigned in such a way that the footprints overlap and extremely small connection times

to an individual satellite never happen Nevertheless, the maximum connection time isstill limited A user’s trajectory, viewed from the satellite, will resemble a straight linecrossing the center of the coverage area The apparent (or relative) speed of the user isthen the speed of the satellite This causes the following undesirable phenomena: visi-bility changes, varying topologies (ISLs changes), footprint handover, and need forrerouting

The Earth’s self-rotation introduces some more complication in the system In Figure

22.5, we plot the maximum time between two satellite handovers against the altitude h and

the elevation angle ␧of a constellation, in two cases:

1 The Earth’s self-rotation is not taken into consideration and the satellite’s inclinationcan be arbitrary

2 The Earth’s self-rotation is taken into account and the orbit of the satellite is torial

Notice that the maximum handover time, shown in Figure 22.5, can vary from some utes up to several hours Also, inclined orbits can be used to exploit the Earth’s self-rota-tion to increase the visibility period Hence, the mobility of the network can also vary alot Roughly speaking, one can distinguish between low and high mobility, depending onthe maximum handover time

min-Low Mobility (periodic) In [5], the mobility of a satellite constellation is described in

terms of finite state automation (FSA) by a series of states described along the time period

in round robin fashion The main advantage of this model is that we have to consider only

a finite set of configurations of the satellite constellation (in which the satellites are sumed to be immobile), and provide efficient routing solutions for each of them, inspired

as-by classical telecommunication problems

Low Mobility (aperiodic) It is worth observing that the “periodicity” assumption of the

FSA model may be, in some cases, too strong This is essentially due to the combination

of “physical” factors, such as the Earth’s self-rotation, the satellite’s speed, and the use ofinclined orbits They make the system aperiodic for all practical purposes, i.e., a satellitewill find again the same position only after such a long time that too many intermediatestates would be necessary In this case, a possible approach consists in taking a series of

snapshots or fixed constellation topologies, a method sometimes referred to as tion [11, 37, 38] Then the routing problem is solved with respect to that fixed “constella-tion.”

discretiza-High Mobility The above two models are interesting when the mobility of the satellite

network is negligible with respect to the mobility of the users’ requests, e.g., if most of therequests have very low duration, let us say a few minutes, while the handover time would

be one hour or more In that case, before the network configuration changes ly) several (many) requests will have been satisfied

(significant-Moreover, these models do not take into account the dependence between consecutivestates of the network Thus, between two states, the complete routing scheme of the con-stellation should be changed Clearly, in the case of highly dynamic constellations and/orlong call durations, almost all requests may pass through several states and thus may bererouted several times

22.3.2 User Distribution: Common Traffic Assumptions

Depending on the application, three major scenarios can be identified for satellite kets The first and most natural one states that satellites will serve countries where thetelecommunication infrastructure is insufficient or nonexistent The second one, whichappears to be more and more probable, is that the satellites will provide additional capaci-

mar-22.3 NETWORK MOBILITY AND TRAFFIC MODELING 481

Minimum elevation angles

10 degrees, without Earth’s self-rotation

20 degrees, without Earth’s self-rotation

30 degrees, without Earth’s self-rotation

40 degrees, without Earth’s self-rotation

10 degrees, equatorial orbit

20 degrees, equatorial orbit

30 degrees, equatorial orbit

40 degrees, equatorial orbit

Figure 22.5 Maximum time between two satellite handovers

ties to countries that already have good telecommunication infrastructures, but which fer from overload of their resources A third market concerns people who require a seam-less connection in their international activities Of course, depending on the scenario, thetraffic may have different characteristics, as summarized in Table 22.1

suf-Little is known about the two first classes of applications The last one has been tigated in [35], where an analysis of the international activities led to a map of differentzones, worldwide In this model, the planisphere is divided into 288 cells, with 24 bandsalong the longitude and 12 along the latitude The intensity levels from 0 to 8 shown inFigure 22.6 correspond to traffic expectations for the year 2005 of 0, 1.6, 6.4, 16, 32, 95,

inves-191, 239, and 318 millions of addressable minutes/year In [15] the traffic requirementmatrix is obtained from trading statistics, namely the imports/exports between any two re-gions Further market studies on satellites can be found in [23]

In the following, we describe how the users mobility can be modeled by means of sometraffic assumptions In particular, we group traffic assumptions into three categories:

TABLE 22.1 Characteristics of foreseeable usages of satellite constellations

Location Poor countries/oceans Rich countries International

5 1

2 5 5 5 2

7

7 6

8

5 7

2

4

6 3