Matching Practice and Theory Communications in ITU-T Work -- …kkleung/papers/grillo.pdf · 2002. 5. 15. · “directed retry”, “cell load sharing”, [2], and “queuing”

D. Grillo, R. A. Skoog, S. Chia, and K. K. Leung, "Teletraffic Engineering for Mobile Personal Communications in ITU-TWork -- The Need for Matching Practice and Theory”, to appear in IEEE Personal Communications

��

��

Guidance on traffic engineering for fixed networks and telephone service has been successfully developed by ITU-T(formerly CCITT). This guidance has been based on a communication paradigm which has included regulated operation

environment, operation domains matched to national boundaries, centralized control and predictable service quality. Theseassumptions are being more and more challenged with the introduction of new communication modes, the increasing

popularity of mobile and personal communications services, and deregulated operation. As a result, new paradigms areemerging for the 21st century telecommunications. In such a framework, continued user satisfaction and operator revenue

growth with personal communications services require that suitable traffic engineering methods be devised. These methodsshould help in reconciling the expected service quality with cost-effective dimensioning and operation of networks and

infrastructure for supporting a range of services as well as provide means for capitalizing on investments intraditional/existing telecommunication infrastructure. The paper notes that in order to arrive at sensible ITU-T

Recommendations on traffic engineering for personal communications networks, practice and theory should be mutuallysupportive.

Teletraffic Engineering for Mobile PersonalCommunications in ITU-T Work -- The Need for

Matching Practice and Theory-----------

DAVIDE GRILLO, FONDAZIONE UGO BORDONI (+39.6.5480 3430, [email protected])RONALD A. SKOOG, AT&T LABS (+1.732.949 7915, [email protected])

STANLEY CHIA, AIRTOUCH COMMUNICATIONS (+1.925.210 3470,[email protected])

KIN K. LEUNG, AT&T LABS (+1.732.345 3153, [email protected])

raffic engineering has as its ultimateobjective the cost-effectivedimensioning of network resources tohandle the user's demand for

telecommunications services - and hence the induceduser information and signaling traffic streams. At theinternational level, traffic engineering is being guidedthrough the activity of ITU-T1 (formerly CCITT2)where the related procedures are cast inRecommendations covering such issues as trafficcharacterization, service quality targets, dimensioningmethods, and measurements. As opposed toRecommendations dealing with the procedural aspectsof interconnection/interworking in a multi-vendor andmulti-operator environment, Recommendations ontraffic engineering provide advice on “good practice”for network operation and are not binding. Howevertheir recognized value is based on input fromAdministrations, ROA's (Recognized OperatingAgencies) and scientific bodies who collectivelyprovide a broad range of experiences and knowledge,and who have developed an understanding of therelationships (by a given technological scenario and

1 International Telecommunication Union TelecommunicationStandardization Sector.2 International Telegraph and Telephone ConsultativeCommittee.

communications paradigm) between such interrelatedaspects as traffic demand, cost of communicationsinfrastructure, sustainable service quality, andacceptable tariffs. The bulk of ITU-T work on trafficengineering assumes the existence of operationdomains comprised within national boundaries forsupporting telecommunications services between userslocated anywhere in the world. This work has beeninstrumental in enabling (national) operators to conductbusiness on a global scale, and has performed well inensuring user satisfaction and continued revenuegrowth. In particular, traffic engineering for fixedcommunications, and PSTN-based services, has a longtradition in ITU-T. The underlying communicationsparadigm has been characterized, in addition toregulated operation and operators’ domains matched togeographical boundaries, by centralized networkcontrol, fixed bandwidth allocation, and predictableservice quality - among others. This paradigm whichhas held throughout the 20th century, is now beingchallenged, [1], since: i) its cornerstones are either nolonger valid (spread of deregulated operation,ownership of network segments and Points-of-Presencebeyond national boundaries) or, ii) new services andother operation modes are flanking the traditional ones(e.g. multi-party communications, highly variable

T


��bandwidth requirements, on-demand bandwidthallocation, terminal and personal mobility support 3).

An intensive study activity is being undertaken tounderstand and model the key aspects of the emergingtelecommunications systems associated with thischanging scenario so as to arrive at a sensible trafficengineering methodology for them and continue tosuccessfully operate telecommunications servicesunder the developing paradigms. The progress of thesestudies in ITU-T shows a variegated picture, with areaswhere investigation is approaching saturationcomplemented with substantial traffic engineeringstandardization activity (ISDN, ATM), areas wheretraffic engineering is progressing (IN and signalingsystems), and areas where traffic engineering has beenstarted but progress is not yet adequate to the advancesin system deployment and service penetration (mobilenetworks).

Personal communications4, plays a vital role inshaping the 21st century communications paradigms.Indeed, the associated dimensions of space/timedependence of traffic demand, “hostile” operationenvironment, unpredictable quality, and changing user'snetwork attachment point represent major deviationswith respect to the traditional communicationsparadigm. Although mobile services have beencommercially available since the late seventies, trafficengineering for personal communications has beenbased - and to a great extent still is - on “currentpractice” of mobile operators, and has been dominateduntil recently more by radio transmission and coverageconsiderations rather than by classical traffic loadingand service quality arguments. The justification for thishas been that an (elite) customer base has traded-offimpairments of service quality against ubiquitousservice that is perceived as a major value. In addition,there are many factors which have been hiding theinefficiency of network planning for operators. As amatter of fact, to circumvent poor network quality,users have been inclined to develop a pattern of 3 Personal mobility refers to the ability of users to have flexibleaccess to telecommunication services from any terminal, fixedor mobile, to meet the users requirements. These requirementsmay then be relocated from terminal to terminal. Personalmobility involves the network capability to locate the user onthe basis of a unique personal telecommunication identity (i.e.,"personal" number) for the purposes of addressing, routing andcharging of the users calls.Terminal mobility involves the ability of the user to be incontinuous motion whilst accessing and usingtelecommunication services and the capability of the networkto keep track of the user's terminal. This requires thetelecommunication services to be available as the terminalmoves within the radio coverage and ideally at all times (ITU-TRecommendation I.114, [3]).

4 The term personal communications will be used to refer to thecollection of terminal (PCS, Personal CommunicationsServices) and personal (UPT, Universal PersonalTelecommunication) mobility services.

complex avoidance behavior in order to work aroundthe perceived network congestion and impairment.Furthermore, in order for an operator to meet both wide-area and in-building coverage requirements of networkroll-out, a large spare capacity has been implicitly builtinto the network from an early day of operation. Thisprocess is further exaggerated by the large “build-ahead” margin adopted by most operators which isnecessary to meet the predicted intensive subscribergrowth and to alleviate long delay in site acquisitionfor building new base sites. Finally, infrastructurevendors have introduced many temporary capacityrelief features such as “directed retry”, “cell loadsharing”, [2], and “queuing” to dynamically exploit the(temporary) unused resources due to short-term trafficfluctuations5. Naturally, all these processes andtechniques have adverse implications to the cost andefficiency of a network.

In the last decade mobile services have beenexperiencing an accelerated penetration which hasculminated in the explosive growth to mass marketdimensions over the last few years, with outlooks forcontinued growth. Accordingly, mobile-related traffic isforecast to be comparable in volume with that relatedto fixed networks in a not too distant future. Even withthe prediction of such a strong growth in the mobilemarket, increasing competition in service provision hasmeant that mobile operators have to be more prudent intheir cost-management and vigilant in maximizing theirincome. While over-dimensioning of a network isequivalent to poor capital investment, congestion atbusy hours could mean lost calls and lost revenues.These factors, combined with the impact that mobile-related traffic may have on the fixed infrastructure, andthe convergence of mobile and fixed services, drivetowards a rationalization of the resource allocation andmanagement procedures (both inter- and intra-operators'domains) and make it urgent to address trafficengineering for personal communications at theinternational standardization level. Furthermore, withthe introduction of mobile data services such asGeneral Packet Radio Services (GPRS) for GSM andIS-707 for cdmaOne (also known as IS-95, [4]), theintegration of voice and data traffic is becoming areality. This adds yet more dimensions to thecomplexity of traffic engineering. Table 1 tries toillustrate the challenges in the planning and operationof personal communications networks and the expectedbenefits from a traffic engineering activity.

To accommodate the need for standardization ontraffic engineering for personal communications, ITU-T

5 "Directed retry" means a terminal is redirected by thenetwork to setup a call with a base station which is not the bestserver due to network congestion reasons. "Cell load sharing"is to dynamically move the boundary of the cells so that calls inprogress can continue in cells which are less congested."Queuing" is used to temporarily buffer calls e. g. on adedicated control channel when all the traffic channels arecongested.


��

has started a dedicated series of Recommendations, theE.750 series, [5]. Having first addressed frameworksetting aspects for a traffic engineering activity - suchas reference configurations and GOS6 parameters, andtarget values - the E.750 series is now giving priority tostudies on traffic demand modeling and dimensioningmethods.

In parallel with the increasing penetration ofpersonal communications, a range of related studies onarchitecture, service offerings and performance aspectshave been proliferating in the open literature, alsofueled by the international efforts aimed at designingadvanced, "third generation" mobile systems such asIMT-2000, [7], [8], developed in ITU and UMTS, [9],[10], studied in ETSI. Although these studies do not

6 Grade of Service (GOS): "A number of traffic engineeringvariables used to provide a measure of adequacy of a group ofresources under specified conditions; these grade of servicevariables may be probability of loss, dial tone delay, etc.”,ITU-T Recommendation E.600, [6].

necessarily exhibit the combination of simplicity andcoverage of key aspects which matter for trafficengineering, and frequently favor the analysis ofspecific technical solutions rather than relatefundamental parameters, they have a great potential fortraffic engineering work. The current operators’ practiceand this wealth of literature are, in a sense, twoextremes that traffic engineering (and related ITU-Tactivity) has to reconcile in order to provide a sensibleguidance for a cost-effective use of network resourceswhile meeting users expectation on service quality.

Although personal communications implies bothterminal and personal mobility, this note concentrateson terminal mobility for which Fig. 1 gives twofundamentally different supporting architectures. Asshown in the figure, the key difference lies in theorganization (signaling and database arrangement) ofthe mobility management functions resulting in “stand-alone” (or separated mobile and fixed network) and in“integrated” architectures. The interfaces andfunctionality shown in the figure, in addition to having

TERMINAL MOBILITY

Key Aspects & ProblemsRequirements & SolutionsAspects of Traffic Engineering &Challenges

Limited amount of radio spectrumSpectrum re-use (cellular lay-outarchitecture) to meet capacityrequirements.Interference management, dynamicchannel allocation (DCA), use of smartantennas

Spectrum partitioning between celllayers, considering traffic overflow to“umbrella cells” (or mutual overflows innon-hierarchical cell lay-outs).DCA performance models.

Hostile transmission environment forwireless communications

Channel quality monitoring andrecovery to combat adversetransmission conditions.Smart antennas, advanced equalization,handover and admission control, highfrequency re-use.

Handover (combining) handling andpriority service; Call Admission Controlfor trading off service quality againstcarried traffic; modeling of co-channelinterference as dependent on traffic load.

Mobility behavior of the customer baseLocation, registration and authenticationof the customer base to track users andprovide a seamless communicationsspace shielding from mobilityimplications and roaming technicalities.

Mobility models. Space/time trafficdemand dependence modeling, mappinguser mobility into signaling traffic.Signaling network dimensioning.

Long term subscriber growth, intensivein-building coverage and wide-areacoverage

Minimum infrastructure build and just-in-time delivery of capacity.

Description of the traffic characteristicsand modeling of the traffic processes.Generic dimensioning methodscomplemented with network trafficdimensioning for specific mobiletechnology classes.

Traffic demand fluctuationCapability of moving spare resourcesaround in the network to accommodateperiodic and instantaneous trafficvariations.

Traffic modeling for dynamic trunking ofshared resources.

Integrated voice and data servicesProviding additional resources for dataservices on a dedicated basis or sharedbasis.

Service quality requirements for dataservices. Traffic dimensioning forpacket data.

PERSONAL MOBILITY

Key Aspects & ProblemsRequirements & SolutionsAspects of Traffic Engineering &Challenges

Registration on and interaction fromcurrent user equipment

Authentication of user data, negotiationcapabilities to cope with actualequipment characteristics/performance,impact of actual service providerofferings and visited networkconstraints.

Determination of traffic mix resultingfrom user-network negotiation actions.Signaling network dimensioning.

Interaction with and personalization ofthe user profile

Manipulation capability of user data toprovide a familiar (VHE, Virtual HomeEnvironment) and user-friendlycommunications space.

Modeling IN services; mapping usermobility into signaling traffic.Signaling network dimensioning.

Table 1. Key aspects and problems in personal communications and related traffic engineering implications.


��

logical significance, also indicate the scope forteletraffic engineering.

The thread followed in the paper is as follows.Initially, the specifics of mobile related traffic demandare briefly introduced. Obviously, these specifics arekey to the whole traffic engineering process. Since

mobile services have beensupported before asystematic trafficengineering activity hadbeen initiated in ITU-T, thecurrent practice on whichoperation of mobilenetworks is based is thendescribed. This descriptionhighlights the relationshipbetween the dimensionsassociated with radiotransmission and networkplanning, and considerssome typical actions to betaken for estimating andaccommodating the trafficdemand assuming that theradio coverage problem issolved. Subsequently, theorganization of the E.750series is briefly reviewedand it is indicated that theunderlying traffic modelingand dimensioning studiestry to provide tools to beused for rationalizing keyphases comprising theoperator's practice. In theface of the scope ofteletraffic engineering asenvisaged in the E.750series, a selection ofrepresentative theoreticalcontributions addressingboth mobility and trafficdemand modeling as wellas typical trafficengineering issues issurveyed. The survey isintended to provide anoverview on the kind oftheoretical support currentlyavailable for thestandardization activity.The paper finishes by listingand commenting on somecommonly acceptedassumptions underlyingtheoretical work, andstating the ITU-T needs forprogressing a usefulstandardization activity.

�� !�"

Mobility Support Signaling & Database

Infrastructure

#$%&'()*+),-$. /01#*2314*5

1616*-67 1(8&%99(&8*+),-$.

:!"�;�


��H��

Specifics of Mobile RelatedTraffic

n a sense, user mobility has always beenconsidered also for fixed networks andtelephone traffic. As far as the originationand destination of calls is concerned, a

fixed network may be viewed as a collection ofcommunication devices rigidly associated with networkattachment points. “Fixed” users, i.e. users of fixeddevices, may change their location over a specificarea, for example when moving from home to theworking place and vice versa, and then initiate (placeor receive) calls from the fixed device location theyhave reached.

For traffic engineering purposes, the traffic loadduring “busy hours” for a geographic region is oftenreferred to. The location of the busy-hour during the daydepends on, for example, whether the region is used asa business or residential area. Although users movefrom one place to another, the peak-hour traffic load

implicitly considers such mobility. This, combined withthe condition of sufficiently large (“infinite”)population has led to models of the fixed trafficdemand based on characterizing call arrival statistics(e.g., exponential law between call arrivals) thatcapture the aggregate behavior of the users in ageographical area.

To dimension transmission, switching andprocessing resources bound to a geographical area, theother two needed elements are the distribution of thecall length and the arrival rate of calls during thereference (“busy hour”) period. Again, under the aboveconditions, the call length distribution - at least in thecase of telephony - has been found to be onlydependent on aggregate models of user “behavior”which has resulted in a robust, parametric and universalmodel with very little dependence on user class,country, etc. In conclusion, although the basicassumptions underlying traffic engineering of fixednetworks are continuously challenged and tested (forexample see [11], [12] and [13]), for any practicalpurpose the fixed traffic demand can be expressed in

I

Calldensity

x-dimension

y-dimension

Call density

x-dimension

y-dimension

1315

17S1

S2

13

57

911

S3

S4

S5

S6

S7

S8S9

0

100

200

300

400

300-400200-300100-2000-100

1113

1517

S1

S2

13

57

9S3

S4

S5

S6

S7

S8S9

0

100

200

300

400

300-400200-300100-2000-100

y-dimension

Userdensity

x-dimension

y-dimension

Userdensity

x-dimension

1315

17S1

S2

13

57

911

S3

S4

S5

S6

S7

S8S9

0

100

200

300

400

300-400200-300100-2000-100

1315

17S1

13

57

911S2

S3

S4

S5

S6

S7

S8S9

0

100

200

300

400

300-400200-300100-2000-100

User density at time t

User density at time t + ∆tCall density at time t + ∆t

Call density at time t Mapping user

densityinto call density

Figure 2. Illustrative example of distribution of user density at different times and related terminal mobility traffic demandfor a generic service.


��terms of the distribution of call inter-arrival times, thecall arrival rate and the distribution of call duration,with virtually no dependence on space per se and adependence on time dictated by the general level ofdaily and/or seasonal activity.

By contrast, as concerns the initiation of calls, amobile network may be viewed as a collection ofcommunication devices in no rigid association withnetwork attachment points (radio ports), and normallytraveling together with the (mobile) users. Toappreciate the specifics of mobile related traffic onehas to note that:

•The user mobility is no longer confined to possiblytraveling within a few places with long intervals ofstationariness (as in the case of fixed users), but isgenerally characterized by wider range usermovements and more frequent location change;

•The association between calling device and networkattachment point is dynamic (in-between callattach/detach) and may also change during thesame call (in-call association re-arrangement due tohandover or combining);

•The resources to be dimensioned (e.g., radiochannels carrying user and signaling traffic, and thefixed infrastructure for supporting mobility) continueto be bound to a geographical area.

As a consequence, the underlying user behavior hasa higher traffic impact (both in space and in time)than in the fixed network case. For example, thedistribution of population density may be stationary, butcalls are initiated while users are “on the move” thusgiving rise to mobility related traffic loads and “trafficvolatility”, i.e.: i) the call initiation sites are scatteredand dynamically changing over a geographical area; ii)bandwidth associated with a connection may have tobe provided to different sites throughout the call, viz.,changing radio cells during a call.

For terminal mobility traffic, Fig. 2 illustrates anexample of the variation of the user density in spaceand time over a geographical area. The same figurealso suggests that the mapping of the user density intotraffic demand may not be simply a matter of scalingbut may require more elaborate considerations sincethe call initiation rate may not be linearly dependenton the user density. For example, a high user density asa result of a traffic jam may also trigger users to placecalls at an extraordinarily high rate. Moreover, in thecase of cellular systems this mapping may be impactedby deep fading and/or insufficient radio coveragecausing tides in the distribution of the traffic demandnot present in the user density. In fact, the trafficdemand in such systems may be conditioned (to avarying degree) by the provisions made by the networkoperators and service providers (e.g. deployment ofcommunication facilities, roaming agreements, tariffstructure, marketed service features, etc.). Thisconditioning, however, shall not impact the logicalframework for carrying out the traffic engineeringactivities.

Finally, the in-call behavior of mobile users doesnot necessarily align with that related to the fixednetwork, for example as concerns the average lengthand length distribution of the calls. This area is theobject of intensive study and investigation, [14].

Current Practices for theOperation of Cellular

Systems

s engineering procedures for mobilesystems have only recently started to bestandardized, the operation of cellularsystems is frequently based on simple

rules for traffic demand estimation and resourceallocation, complemented with monitoring and tuningthe system performance in the field as the networkevolves, [15]. To illustrate this process, it is instructiveto assume a “green field” situation, although in manyinstances similar issues could be faced by operatorswho are still evolving their infrastructure. Key aspectsare summarized in Fig. 3. In the figure, a simplifiedview of the complex relationship between the radioplanning and the capacity dimensioning process is alsoshown. In addition, a numerical example of thedimensioning process is shown in Table 2.

The dimensioning process starts with an estimationof the user population by using the density of theinhabitants in a specific area together with ananticipated service penetration rate. Radio and networkplanners continue with the identification of sites wherethe cellular infrastructure has to be laid down(typically, base transceiver stations, base stationcontrollers and mobile switches), and the mapping ofuser density into traffic demand. The process thenfinishes with the allocation of the traffic (radio)channels making judicious use of the availablespectrum7.

The accomplishment of the dimensioning cyclerequires that numerous optimization problems be solved.These range from minimizing the number of the basesites while guaranteeing sufficient coverage andacceptable service quality, to planning the re-use ofspectrum so as to accommodate the traffic demandwhile ensuring stable system operation and usersatisfaction - to name just a few.

7 It should be noted that the ultimate traffic capacity of a basesite is highly dependent on the exact locations of the base sites.While every effort is usually made during the planning stage toensure that a base site can be well positioned, the actual siteacquisition process is subject to many factors including thephysical location, real estate cost, the height of the location,the availability of equipment and antenna space, etc.

A


��I��

The success in operating a system is assessed,among others, by the degree of control exercised oversuch phenomena as dropped calls, repeated callattempts and handover cases. All these phenomena willpenalize user expectations about good service qualityand, more often than not, many of these shortcomingsare due to poor balancing of operation parameters.The following sections are intended to illustrate thecurrent approach to the operation and management ofcellular systems (together with some deficiencies).They provide a reference list of key issues to becovered in future standardization work on trafficengineering in ITU-T.

Figuring out the Traffic Demand

For traffic engineering of cellular systems,information on geographical population distribution is ofvital importance to an operator. For new entrants to amarket, this information can only be estimated usingpublished census information. This may vary inresolution, with the better ones being rather detailed,and may resolve down to municipal or district level.From the census database and the size of thegeographical area, it is possible to estimate thepopulation density for the location. Together with theyear-on-year user penetration forecast and the averagetraffic intensity per subscriber, the traffic demand canbe obtained. (Occasionally, road traffic information mayalso be available. This information is either too detailedor very specific, which renders its use difficult. Thereason is that this information is collected usually in the

JKLMNOPMQMRSPTOTU VNMWXMTYZ[\OTTSTP

]OUSK^M_`KNaQSbMTRSKTSTP

cd

cd

efghifjdkflmnf dopfjqhkfr

sftfjqomrfrqmqhdi tdjmqhdir

ulfvhjqjdkflmnfw ofrqrflkfllfnhdirxlfqyfjdkflmnf lfz{hlf|fiqr

rmqhrghfv}xrrhniqlmgghj ~fhnyqhinrmjjdlvhinqd

nfdnlmyhjmtgfmq{lfrxvvjmmjhq omrfrhqfr

xrrhniglfz{fij nld{mivydrfqqdfmjyjftt

mtj{tmqfhiqflgflfijf tfkftgld||mhihiqflgflflr

uflgdl|rhnimttfkft |fmr{lf|fiqr~hqyqfrqqlmir|hqqflr

ijdldlmqfvhnhqmt qfllmhiw|dlydtdnvmqm

xlfhiqflgflfijf lfz{hlf|fiqrrmqhrghfv}

mtholmqflmvhd tmiihin|dvfthiqyffinhifflfvqddt

fr

fr

i{qfmldifml j{rqd|flomrfnld~qy

efqfl|hif nfdnlmyhjmtvhrqlho{qhdidghiymohqmiqr

rqh|mqfqlmgghj jmmjhqlfz{hlf|fiqr

xttdjmqfjmllhflrqd hivhkhv{mtjfttr

xlfqyfjmmjhq lfz{hlf|fiqr|fq}

xvvmimvvhqhdimt jmllhflqdqyfjftt

rqyfi{|ofldg jmllhflrfjffvhinqyfrfjql{|th|hqr}

i{qfmldifml qlmgghjhiqfirhq fl{rfl

efghiflmvfdg sflhkjfqmlnfqrwvh|firhdihinl{tfr

cd

cd

fr

fr

Figure 3. A summary of the radio planning and capacity dimensioning processes.


��form of number of vehicles passing a specific junctionper 24 hour period rather than the number of vehiclesusing the road. As urban roads could have many turn-offs, extrapolating the information to give traffic volumefor a coverage area is by no means a simple task).

In parallel with the traffic engineering process, radiocoverage planning is also performed to enable networkinfrastructure roll-out. Based on the terrain database andthe morphology database together with the desiredsignal level necessary to provide suitable in-buildingand outdoor services, base site locations are identified.Frequently, contiguous coverage is required and, hence,the coverage area of base sites are packed closelytogether in order to eliminate coverage gaps as far as

possible. In reality, the terrain is rather undulated. Inorder to eliminate the majority of the coverage gaps andto provide adequate in-building and in-vehicle services,one needs to significantly overlap the coverage betweenbase stations. Thus the dividing lines defined by theequal signal level from two or more base sites will formthe boundary of the “best server” region for individualbase sites. In other words, when a mobile is within thebest server region of a specific base site, it will receivethe strongest signal from that base site even though thesignal from other base sites may still be adequate forcommunications. By associating mobile stations withthe base site of a best server region, the highestdownlink carrier-to-interference ratio can be obtained. In

Year 0through

10

Year 0Year 5Year 10

Traffic demand estimation

Population1,000,000

Penetration10%20%50%

Number of subscribers100,000200,000500,000

Traffic per subscriberE0.020.0120.012

Traffic demandE2,0002,4006,000

Coverage design

Size of service areakm2

1,000

Traffic distributionUniform

Nominal cell radiuskm2

Nominal cell area (assuming circular)km2

12.6

Number of base sites for coverage80

Number of cells / site (3-sector sites)3

Frequency planning and network dimensioning

Spectrum allocationMHz7.4

Radio carrier spacing (as an example)MHz0.2

Number of radio carriers37

Nominal frequency reuse12

Average number of carriers / cell3

Average number of voice channels / cell (8 channels percarrier, excluding control channels)

22

Call carrying capacity per cell (2% blocking, Erlang B)E14.9

Capacity dimensioning

Total call carrying capacityE3,575

Average base site traffic efficiency due to terrain)60%

Actual call carrying capacityE2,145

Spare capacity (negative means insufficient capacity)E145-255-3,855

Additional capacity

Additional capacityE02553,855

Limiting factorCoverageCapacityCapacity

Number of additional base sites0687

Adjusting for base site traffic efficiency010145

Total number of base sties8090225

Infrastructure increase-12.50%181.25%

Table 2. A simple numerical example of the radio planning and capacity dimensioning process.


��addition, when power control is used in the uplink, aminimum transmit power is required which could in turnprolong battery life of the handsets.

As the best server regions are rarely regular in shapedue to terrain undulation and geographical features, thetraffic capture ability of each base site could be quitedifferent. By mapping the best server region into thepopulation density map, a first order estimation of thetraffic demand per sector can be realized. However, thismay lead to substantial inaccuracies since users aregenerally congregated along roads and buildings andrarely located in open spaces. A method to obtain amore accurate estimation of the traffic demand is topolarize the population into areas where it is mostlikely to be located.

This can be achieved by assigning weightings todifferent geographical features. Based on these

weightings, the traffic for each cell can be moreaccurately estimated. Evidently, for a cell whichcontains lots of open spaces, the amount of traffic isexpected to be very low. By contrast, for a cell whichcontains buildings and shopping areas, the trafficdensity is expected to be high. Fig. 4 tries to illustratethis principle. Specifically, Fig. 4a shows thegeographical features and the best server regions of anarea under study. Fig. 4b shows a 100 x 100 m

2 grid

overlay on the area. This represents the resolution of thedigital terrain database which will eventually determinethe resolution of the best server map as well as thetraffic distribution map. It should be noted that both thebest server map and the traffic maps now become anapproximation of the real life situation. Assuming thatcensus information indicates that this 1.1 x 1.6 km

2 area

has a population of 17,600 people, this corresponds to a

0.060.170.170.110.060.060.060.060.060.060.06

0.060.170.170.110.170.170.170.170.060.060.060.060.170.170.110.170.170.170.170.060.060.060.060.170.170.110.060.170.170.170.060.060.060.060.170.170.110.170.170.170.170.060.060.060.060.170.170.110.110.110.170.170.170.110.110.060.060.170.110.060.060.110.110.110.110.110.060.060.170.110.060.000.000.060.170.060.06

0.060.060.060.170.110.060.060.170.170.170.060.060.060.060.170.110.110.170.170.170.060.06

0.060.060.060.170.110.110.060.170.170.060.060.060.060.060.170.170.110.110.060.060.060.060.060.060.170.170.060.170.110.110.110.110.06

0.060.060.170.170.060.060.060.060.110.110.110.060.060.170.060.060.060.060.060.060.060.060.060.060.170.060.060.060.060.060.060.060.06

Cell 1Cell 1

1

1

1

1

1

1

11

1

1

1

11

11

1

1

3

3

3

3

3

311

1

1

11

11

1

3

3

3

3

3

3

333

1

1

13

33

1

1

2

2

2

2

2

222

3

3

33

31

3

1

1

3

3

1

3

211

2

2

31

11

2

1

1

3

3

3

3

210

1

2

23

11

2

1

1

3

3

3

3

320

1

3

22

11

1

1

1

3

3

3

3

321

3

3

12

11

3

1

1

1

1

1

1

323

3

3

12

11

3

1

1

1

1

1

1

221

3

1

12

21

1

1

1

1

1

1

1

221

1

1

11

21

1

a)b)

d)e)f)

Road

Building

Base station

Best serverboundary

Pond

Cell 3

Cell 2

Cell 4

Cell 6Cell 5

100m

100m

Traffic estimationfor each cell

Computerpredicted bestserver regionsbased onterrain map

Determining theresolution forfiguring out thetraffic demand

Figuring out thetraffic weighting

Generatetrafficmap

c)

Mapping trafficto best serverregion

0.170.170.170.170.170.170.170.17

0.110.060.170.170.17

0.170.110.170.170.170.170.110.110.11

0.110.060.060.110.060.00

0.110.060.060.110.17

4.15Erlangs

Cell 1

Cell 2

Cell 3

KEY

a) Geographical area with a population of 17600 and a predicted traffic of 17.6 Erlangsb) Overlaying a grid of 100x100 m

2 on the geographical area for traffic weighting

c) Signal level computer prediction of the best server regiond) Traffic weighting map for the geographical areae) Traffic distribution map based on the traffic weightingf) Predicted traffic for individual cells in the geographical area

Figure 4. Estimating traffic demand from population distribution, city lay-out and radio coverage arrangements.


�� population density of 1 person per 100 m

2. With a

service penetration rate of 5%, there will be 880subscribers in the area. Given that each user willgenerate 20 mErlangs of traffic, the total traffic in thearea amounts to 17.6 Erlangs. This corresponds to anaverage traffic density of 1 mErlangs per 100 m

2. As the

resolution of the digital terrain database is accurate to100 x 100 m

2, computer prediction of the signal level

for the best server region will be quantized into bins ofthe same size.

FeatureWeightRoad2Open1Water0

Road and building3Open and road2

Open and building3Open and water1

Table 3. Weighting factors for the offered traffic inrelationship to the geographical features.

This is shown in Fig 4c. Finally, also assume someweighting factors for the traffic load in relationship tothe geographical features as represented in Table 3.

By mapping the best server region to the traffic bins,the traffic demand for each cell can be calculated.Applying the weighting to the geographic area as shownin Fig 4b, a weighting map as shown in Fig 4d can beobtained. Summing the total weights in the area andknowing the total traffic, the traffic demand for eachindividual bin can be apportioned as shown in Fig 4e.Finally, mapping the best server region to the trafficmap, the traffic prediction for each cell can beobtained, Fig. 4f.

To show the importance of using the weightingfactor, consider the demanded traffic of Cell 1. Withoutthe weighting factor, a traffic load of 3.3 Erlangs ispredicted. However, with the weighting, a traffic load of4.2 Erlangs (30% higher) can be anticipated. It shouldbe noted that the weighting factors shown in thisexample are indicative and for a real application morecalibrations are necessary to ensure good accuracy. Theboundary effects between geographical areas will alsohave to be accounted for but this is beyond the scope ofthis discourse.

Evidently, the example shown here is related to agreen field deployment where an operator has no “apriori” knowledge of the actual traffic and mobilitypattern of the users. As the network evolves, with trafficstatistics collected through the mobile switches overtime, a much more accurate picture of the trafficdistribution down to the resolution of a cell can be built.With this information, the above technique can beapplied again to refine the network optimization.Furthermore, as the radio network is alwaysdimensioned for the peak traffic together with a safetymargin, variations of the daily traffic due to themobility pattern of the users are usually adequatelytaken into account as well as extraordinary events such

as major incidents or scheduled gatherings. In the latterinstances, operators have to employ special measures tocope with the surge in traffic demand. An example ofthese temporary measures is the “Cell Site on Wheels”deployed by an operator following the Los Angelesearthquake in 1994.

Sizing the Channel Capacity of Cells

Based on the knowledge of the traffic for each cell,the number of traffic channels - and hence oftransceivers - can be determined. Using the GSMsystem as an example, the relationship between offeredtraffic and the number of transceivers is indicated inTable 4. Specifically, for GSM one transceiver supportsone carrier which in turn supports eight time slots. Thetime slots can be assigned as traffic channels or controlchannels depending on the specific configuration of anetwork. (The results in Table 4 consider therequirement of control channel assignment and assumea 2% probability of blocking for fresh calls using theErlang B model).

Offered trafficper cell

[Erlangs]

Number of trafficchannels (or time

slots)

Number oftransceivers

2.3618.214214.922318.4264

Table 4. Examples of allocation of transceivers to a cellas a function of offered traffic.

As a network is normally dimensioned for growthand traffic fluctuations, it is not uncommon that a threeto six months build ahead is incorporated in the trafficdimensioning plan in order to accommodate safetymargins, [16]. These margins have proven to beadequate in most cases.

In addition to normal calls, handover requests alsorequire radio resources especially for a “make-before-break” scheme as in some implementations of the GSMsystem, [17]. In a real network, the number of handoversper call is dependent on the length of the calls as wellas the mobility pattern of the users. The contribution ofhandovers to the total traffic loading is difficult topredict and is normally assumed as an acceptableoverhead to a system. When an operator detects anexceptionally high volume of handovers in a cell,measures will have to be taken to bring it under control.Typical techniques at the disposal of an operator are to:Increase the hysteresis margin; change the handoverthresholds; and reduce the overlap between adjacentcells by using narrower beam antennas. As an example,60

o antennas are frequently used among three sector

cell sites in dense urban areas for minimizing thenumber of repeated handovers. Otherwise, the small cellsize and large coverage overlap between neighboringsectors coupled with highly variable shadow fading inan urban area might trigger an excessive number ofhandover requests.


��

The ability to prioritize handover handling withrespect to fresh calls is infrastructure equipmentspecific. In dense urban environments where cells areheavily overlapped with each other in outdoor areas dueto their short inter-site distances and the requirement forindoor coverage, it is generally not necessary toprioritize handover handling as calls can be maintainedeven if the mobile moves away from the best serverregion. This large overlap can be visualized byunderstanding that the indoor penetration margin forurban areas is typically required to be 15 dB or more. Ahandover failure will not normally lead to a droppedcall as calls, which are not successfully handed-over,can be reverted back to the originating base station.Evidently this will degrade the service quality due tothe increased uplink interference as calls are draggedoutside the best server region.

Similar to handover prioritization, the handoveralgorithm and parameters are typically infrastructureequipment specific and are usually omitted fromstandardization. For instance, for GSM, it is possible toemploy received signal level, received signal quality,timing advance as well as traffic reasons for initiatinghandovers, [18]. For a properly engineered network, themajority of the handovers should be initiated because ofinsufficient signal level, i.e. the system operationshould be power-limited. A high volume of handoversdue to poor received signal quality normally indicatesthat there are interference problems in the network(interference-limited operation). In this situation, thedropped call rate is expected to be high as well. Anoperator will have to minimize this by re-tuning thefrequency plan and optimizing the base sites in thevicinity of the affected region.

As an example, Table 5 shows a break-down ofhandover cases according to initiating reason for a GSMsystem. In the table, “signal criterion” signifies that thehandover is triggered by insufficient signal level,whereas “UL quality” and “DL quality” stand for

handover due to excessiveerror rate on the uplink anddownlink, respectively. InGSM, signal quality isdirectly related to the biterror rate measured prior tochannel decoding. There areeight quality levels definedand the threshold for poorquality is between 5 and 6.Evidently, all handoverthresholds are operationparameters. As shown in thetable, not all unsuccessfulhandovers lead to droppedcalls, many are reverted backto the original channel wherea repeated handover requestmay be initiated.

As it may be observedfrom the table, in an evolvingcellular network, on theaverage about 25% of thehandover cases could be due

to interference impairments though the scatter aroundthis value is very significant. This is similar to whatobserved in [19]. It should be noted that the example isfor a typical European capital city and only the caseswhich are in excess of 300 handovers per observationperiod are cited.

More recently, with the proliferation of the use offrequency hopping and power control in the GSMcommunity, it was found that quality based intra-cellhandover can help to improve the robustness of theradio link significantly. Up to now, operators havenormally used this more as a safety feature rather thanas an active approach to optimize the frequency reuse.The dimensioning rules and the relationship withinterference are still not fully understood.

As for code division multiple access (CDMA)systems such as cdmaOne, [4], the dimensioning rulesare, in principle, similar in many respects to othercellular systems. For instance, handovers from basestation to base station will be largely based on thepower level measurements from adjacent base stations.However, a CDMA system has the flexibility of softquality degradation which allows more room for facingtraffic fluctuations. As CDMA is inherently aninterference-limited system, soft handover (bycombining or selection) is required to controlinterference. To this end, additional radio hardware isrequired at the base station. This amounts to a 40% to70% increase in channels and this is included as part ofthe hardware dimensioning rules. Thus, there is a subtletradeoff between the system loading, mobility andcapacity.

In order to maintain QOS8 objectives, CDMAadmission control can be either enforced by physically 8 Quality of Service (QOS): “The collective effect of serviceperformances which determine the degree of satisfaction of auser of the service. The quality of service is characterized by

Handover Triggering ReasonOutcome Of Handover Handling

HO’spercell

Signalcriterion

Bad ULquality

Bad DLquality

SuccessFailure

SuccessFailure:Total

Failure:Returned

to oldchannel

Failure:Dropped

calls

[%][%][%][%][%][%][%]37777.1916.186.6346.9553.0539,5212.5340766.3415.2318.4386.4913.5111,32.2148170.4810.8118.7184.4113.5912,473.11166190.015.724.2752.4447.5646,421.4464893.212.014.7854.9445.0644,60.4686556.6522.3121.0486.0113.999,5954.3961854.538.4137.0683.3316.6713,273.3931437.903.5058.6086.9413.069,5543.5058383.537.209.2689.0210.983,0877.8930276.1618.874.9750.3349.6747,022.64

Table 5. Examples of classification of handover cases for a set of ten cells withexceptionally high volume of handover (courtesy of AirTouch International).


��limiting the number of user codes which can be used,or it can achieved by allowing the interference level ornoise rise in the system to determine the capacitynaturally. However, as most of the CDMA systems arestill in their early phase of deployment, theeffectiveness and the adequacy of these techniquesneed to be better understood. At present, where capacityis expected to be a problem, usually more carriers areadded. Thus, the relationship between admission controland QOS control is still very much an undeterminedissue from a practical network operation point of view.

Adjusting the System Dimensioning

As a network evolves, the number of subscribers willgrow in time while the average traffic intensity persubscriber will gradually decline over time. This isbecause at network start up, the initial customers aregenerally business customers who generate a highamount of “minutes of use”. However, as timeprogresses, more (non-business) subscribers are addedand the minutes of use by them are much lower than thebusiness users. This will serve to dilute the averagetraffic intensity per subscriber across the network. Forinstance, at network startup, the traffic per subscriber istypically around 18-20 mErlangs per subscriber and thisdeclines to 12-13 mErlangs as the network matures. Asfor the net effect, although the subscriber growth mayinduce a decline in minutes of use, the growth in thevolume of traffic demand is still quite significant. Anoperator has to closely monitor the traffic statistics ofthe busy hour traffic channel utilization for all the cells.This information is obtained from the mobile switchstatistics on an ongoing basis. When the level ofblocking reaches a pre-determined threshold, action hasto be taken to increase the number of transceivers percell. Of course this is subject to the constraint of thespectrum allocation and availability. Once themaximum number of transceivers for a specific reusepattern is reached, the frequency reuse has to betightened in order to enable an operator to increase thenumber of carriers per cell. This normally takes time toplan the frequency re-tune and the installation of newequipment at the appropriate base sites. The overalloptimization cycle for network capacity is summarizedin Fig. 5.

When spectrum and frequency reuse become thelimiting factors, for the longer term capacity relief, itwill be necessary to increase the number of base sitesin order to cell split9, or to introduce a hierarchical cell

the combined aspects of service support performance, serviceoperability performance, service integrity and other factorsspecific to each service.", ITU-T Recommendation M.60, [20].

9 This is different from tightening the frequency reuse as cellsplitting is to increase the spatial reuse of the frequency setrather than tightening the frequency reuse pattern. Bytightening the frequency reuse pattern, the number of carriersin the frequency group per cluster is reduced. For instance, anominal frequency reuse pattern for GSM is a 12 carriers

structure (HCS) with microcells overlaying themacrocell network. The traffic dimensioning rules forestimating the number of traffic channels requiredbecome more complex for HCS. For example, directedretry, [21], i.e. attempting to serve a call in the secondbest server cell when the best cell has no channelsavailable, could improve the trunking efficiency of themicrocells but not the macrocells 10.

A recent approach, [22], may enable efficient reuseof frequencies between microcell and macrocells,although the new technique is yet to be tested in thefield. In addition, the effectiveness of speed sensitivehandover algorithms, [23], [24], [25], etc., could alsoimpact the traffic capturing ability of the microcells. Atthe time of writing, operators and infrastructuresuppliers are still actively testing the viability of thehandover algorithms and the appropriate channelallocation strategies.

As mentioned before, for shorter term capacityrelief, it is possible to temporarily borrow the sparecapacity in the neighboring cells. In practice, thetemporal distribution of the traffic among the base sitesin a dense urban area could be rather uneven and not allthe cells neighboring to a congested cell aresimultaneously congested. Evidently, cells have to besignificantly overlapped in order to be able to sharecapacity with each other.

In general, there are many optimization issuessurrounding a cellular radio system. Most of these havemultiple variables and constraints. Precisemathematical formulations are often difficult if notintractable. For instance, the traffic loading for aspecific cell is dependent on the traffic distribution, themobility pattern, spectrum allocation, handoveralgorithm, switch parameters setting, and so on.However, automated planning tools are beginning toemerge to assist engineers to plan and optimize theirnetwork. An example is the proliferation of automaticfrequency assignment tools which put theory, [26], [27],[28], into practice. Yet there is still a long way for toolsto evolve to become fully effective for the many otheroptimization problems of more direct relevance to radionetwork dimensioning.

reuse. However, operators with a generous spectrum allocationmay relax the reuse pattern to 15 carriers or more. Although, toincrease the network capacity, a tighter reuse pattern of 9carriers or lower may be required after the first transceivers.10 Assuming a split band arrangement for the microcell and themacrocell, the coverage of the microcells is only a subset of themacrocell. Splitting the spectrum allocation would imply thatthe number of carriers for the macrocell will be reduced. Inareas where there is microcell implementation, it is possible tohave better trunking efficiency as the mobiles in the microcellwill be able to access both the microcell and the macrocell.However, in areas beyond the coverage of the microcell, themacrocell will have less carriers and can only offer a lowervolume of traffic.


��

ITU-T Framework for TrafficEngineering of Personal

Communications

s already mentioned, trafficengineering for networks supportingmobile and UPT services is addressed inthe ITU-T E.750-series of

recommendations. This series covers traffic engineeringfor the user plane, while traffic engineering for thecontrol plane is handled under separate ITU-Trecommendations, typically those related to commonchannel signaling systems and IN (IntelligentNetwork)11. The characterization of personalcommunications user traffic demand is expected todeliver input to the demand processes for trafficengineering of the control plane.

11 A separation between the definition of the user plane and thecontrol plane is that the former is associated with user datahandling, whereas the latter relates to signaling trafficnormally consisting of short messages and packet data.

The E.750-Series ofRecommendations

The goal of the E.750-series is to recommendprocedures for cost-effectively dimensioning networkresources for terminal and personal mobility support.The scope of the E.750-series covers land, maritimeand aeronautical services, as well as terrestrial andsatellite based networks. To achieve the objectives ofthe E.750-series the following study areas need to beaddressed:

•Mapping of user density and mobility for typicaloperating scenarios into user traffic demand (offeredtraffic);

•Definition of GOS parameters with user perceptionsignificance and related target values to setobjectives for traffic engineering;

•Development of methods for meeting the GOStargets for specified demand patterns;

•Specification of measurement procedures formonitoring the GOS target attainment in the serviceoperations environment.

Correspondingly, the series is organized into five majorgroupings (general aspects, traffic modeling, Grade ofService, dimensioning methods, and trafficmeasurements). Orthogonal to this grouping is theorganization of the series into recommendations

¡¢£¤¥¦§¨©¡ ª«¬£¤¥®£¯

°£±ª²³¢´µ¶¶¥¶·

°£±ª²³ ¸£¢´ª§¹£¶

º»¼½¼¾¿À¾Á¾Â¼½Ã Ä¼ÅÆ»Ç¼È»¼»É

ÊÆ½ËÈÌÍÀ¾Á¾Â¼½Ã ÎÁ½¼Å¼Ï¾½¼È»ÀÃÂ¿Æ

Ðª¶·£²¹¢²ª«´£¹¯

Ñª´´£¤¯±¥¤Ò¯µ¥¯¥¤¯Ó¦¥£´¸£¯

²£¯Ô´¯Õ¤Ô¯ª¹£²¤ª¹¢´µ¥¶¯

Öª¶¥ª²´ªµ¸µ¶¸¢£²¦ª²¹µ¶¤£ °£±ª²³¥¶ ª

¢£²µ¥ª¶

×¥µ·¶ª¯£¢²ª«´£¹¯

Ø¸£¶¥¦§¤µ¢µ¤¥§¯ª´Ô¥ª¶¯

Ø¸£¶¥¦§µ¶¸²µ¶³¢²ª«´£¹¯

Ø¹¢´£¹£¶µ ¢£²¹µ¶£¶¯ª´Ô¥ª¶

Ø¹¢´£¹£¶µ £¹¢ª²µ²§¯ª´Ô¥ª¶¡Òª²£²¹

¢²ª«´£¹¯

Figure 5. The network capacity optimization cycle.

A


ÙÚÛÜÝÞßàáÚâÛãÜäßåæâçèé

addressing terminal and personal mobility12. Theorganization of the E.750-series is summarized in Fig.6.

The development of the series has been mainlyfocused until recently on reference connections andGOS aspects for both land and maritime/aeronauticalservices, with particular stress on terminal mobilityissues. The E.750-series consists currently of twelverecommendations whose status ranges fromapproved/revised to draft recommendations to befinalized.

The Traffic Engineering Cycle

The approach taken in ITU-T is that for both fixedand mobile services the traffic demand should becharacterized based on those aspects which are underno or little control of network operators and serviceproviders. In the case of terminal mobility, theseaspects relate essentially to the environmentalpropagation conditioning (e.g., indoor/outdoor,metropolitan/urban), mobile terminal speed, and userbase characteristics (e.g., mobility behavior, fresh andrepeated call arrival processes). As an example, Table6 lists some parameters which characterize theoperating scenarios envisaged for IMT-2000 .

12 Recommendations on terminal mobility apply to bothexisting, - e. g. GSM (Global System for Mobilecommunications, Europe), NADC (North American DigitalCellular, North America), and PDC (Personal Digital Cellular,Japan) - and developing mobile systems, e. g. IMT-2000 andUMTS (Universal Mobile Telecommunication System, ETSI).

ApplicationDelivery Mode

PhysicalAttributes/

Propagation

Level ofMobility

Public cellularTerrestrialIndoor and/oroutdoor

Stationary(0 km/h)

Privatebusiness

SatelliteOutdoor inurban,suburban, rural,hilly or coastalareas

Pedestrian(up to 10 km/h)

Residentialcordless

Terrestrial orsatelliteoperation

Typicalvehicular(up to 100km/h)

Fixedsubscriber loopreplacement

Land, maritime,or aeronauticaloperation

High speedvehicular (up to500 km/h)

Residentialneighborhood

Aeronautical(up to 1,500km/h)

Mobile basestation

Satellite(up to 27,000km/h)

Paging

Table 6. Operating scenarios for IMT-2000, [29].

A key task in characterizing the traffic demand ishow to capture the space-time relationship with dueconsideration given to the scope of ITU-T trafficengineering. Once the traffic demand is defined, trafficengineering for mobile systems should proceed byexercising the dimensioning methods with specificGOS/QOS objectives and cost constraints. Possibly, thedimensioning methods have to be repeatedly exercisedto meet in the field GOS/QOS objectives and to copewith the necessary adjustments of the many operationparameters which, by necessity, cannot always beexplicitly accommodated in the dimensioningprocedures (e.g. power control, interleaving depth,frequency hopping patterns, source/channel coding,etc.). With a focus on radio resources for systemssupporting terminal mobility, Fig. 7 schematicallyshows the envisaged traffic engineering study areas tobe covered in the E.750-series. Note that these areas areclosely related to the stages followed in the practice ofradio planning and capacity dimensioning as depictedin Fig. 3. As a matter of fact, existing and envisagedrecommendations in the E.76x (TrafficCharacterization), E.78x (Engineering Methods), andE.79x (Measurements and Performance Monitoring)decades, relate, respectively, to the actions describedunder “figuring out the traffic demand”, “sizing thechannel capacity of cells”, and “adjusting the systemdimensioning” described in the section on the currentpractices for the operation of cellular systems. Asindicated in the figure, the tasks associated with trafficengineering of cellular systems are part of a complexcycle which includes radio coverage design andfrequency planning as key components.

êëìí îïðïñòó

êëìôêëìõ

êëìì êëìö

÷ñòøøùú ûüòñòúýïñùþòýùÿð �ïò��ñï�ïðý� ��ïñøÿñ�òðúï �ÿðùýÿñùð�

îñò�ïÿø�ïñùúï ð�ùðïïñùð� �ïýüÿ��

TerminalMobility

PersonalMobility

Figure 6. Organization of the E.750-series ofRecommendations.


ÙÚÛÜÝÞßàáÚâÛãÜäßå�âçèé

Scope of Traffic Engineering forNetworks Supporting Terminal

Mobility

The scope of traffic engineering for networkssupporting terminal mobility can be classified accordingto two inter-related dimensions, i.e. thegeographical/operation-domain dimension and thefunctional dimension. As for the former dimension, as ageneral rule, traffic engineering in ITU-T has beenrelated to the international segments in the

communication path. However, with the trend towardsderegulation, competition and proliferation of roles inthe provision of telecommunication services and theincurred dependency of service performance oninterworking of an increasing number of subsystems inthe communication path, the scope of trafficengineering has widened. Correspondingly, the span oftraffic engineering for networks supporting terminalmobility ranges from metropolitan to internationalareas, as indicated in Fig. 8. As for the functionaldimension, three major network segments can beidentified (see also Fig. 1): i) the radio interface, ii) thefixed infrastructure of the mobile network; and, iii) thefixed core network. The radio interface comprises the

transport and signaling functionality between the mobileterminal and the Base Transceiver Station (forsimplicity in Fig. 1, BSS, Base Station System). Thefixed infrastructure of the mobile network, spannedbetween the Base Transceiver Station and MobileSwitching Center (MSC), comprises the functionalityfor exercising/activating control on radio channelquality and availability as impacted by user populationactivity and mobility. Finally, the fixed core network,extending beyond the MSC, includes the functionalityfor user location, tracking, location updating, and callrouting. For the case of separated fixed and mobilenetworks, [30], the figure also shows the allocation of

mobility management functions(MMF) within the mobile network, asis typically the case with secondgeneration mobile systems (e.g.GSM). Depending on the actualimplementations and trafficrequirements, the MSC (MobileSwitching Center) can be connectedwith the fixed network at the LE(Local Exchange) or TE (TandemExchange) level. This is succinctlyindicated in Fig. 8 through thecombination LE/TE. As a matter offact, the allocation of MMF has arange of possible options includingthe arrangements resulting fromintegrated mobile and fixed network,[31]. The figure indicates two obviousteletraffic interfaces at which trafficdemand has to be characterized fortraffic engineering purposes. Onetraffic demand relates to the radiointerface and has collected most ofthe contributions in the literature. Theother is associated with thecharacterization of mobile relatedtraffic which requires fixed networkresources.

The traffic engineering tasks fornetworks supporting terminal mobilityrelate to all three above functionalsegments. For the traffic engineeringof the radio interface, the key GOSparameters are “probability of call

blocking” and “probability of handover failure”, [32],[33]. The traffic engineering problems that must beaddressed include (see also [34]):•A tradeoff of radio spectrum reuse with call

blocking and handover failure;•Estimating the signaling load and allocating

adequate bandwidth to handle mobility relatedsignaling functions (e.g., paging and locationupdates);

•Radio resource allocation policies;•Admission control strategies.

For the fixed part of the network, the key GOSparameters related to mobility are:

��

��

E.76x

E.79x

E.77x &E.78x

�� !"�# ��$��

%&'

%&'

()*+,-&./01)0-&./ 2+',13&/4.'& 5/16,7

8&,&/39)& 5&15/.*792.:;9',/904+,91)1


ÙÚÛÜÝÞßàáÚâÛãÜäßåéâçèé•Post selection delay (consisting of authentication

delay, paging/alerting delay, time to obtain therouting number, and the fixed network ISDN/PSTNdelays);

•Call blocking and lost signaling transactions;•Profile lookup and update response times.

Fig. 8 shows the two parts of the fixed network: Thefixed infrastructure of the mobile network (the middleband in Fig. 8 consisting of the BS, MSC, MMF andassociated trunking and signaling) and the core network(the upper part of Fig. 8).

The mobility related traffic engineering problemsthat must be addressed for the fixed network are:

•Dimensioning the mobile infrastructure real-timeresources in BS, MSC, and MMF systems to handlethe mobility processing needs;

•Developing database architectures anddimensioning database real-time capacity in boththe mobile infrastructure and core network to handlemobility functions;

•Dimensioning signaling and trunking capacity inboth the mobile infrastructure and core network.

To characterize the traffic demand for theseproblems, models of user behavior must be developed.These models must characterize user density andmobility, user calling behavior (e.g., arrival, destinationand holding time statistics), and user re-attemptbehavior. These user behavior models are then coupledwith system characteristics and operations to derive

traffic load parameters such as rate of handovers, rate oflocation updates, channel occupancy time, etc. Thesetraffic loads are then used to dimension the system andmeet specified GOS targets.

Probing the Literature

Much research efforts have been spent tocharacterize user density and mobility, calling behaviorand their performance impacts on wireless networks.Given the large volume of results in the literature, it isimpossible to give a comprehensive review of themhere. Rather, the purpose here is to present a briefoverview of some of the models and results that, in ourview, are representative. Readers can find other workon the subject referenced by the papers cited here.When appropriate, areas that require further study willalso be pointed out.

The basic purpose of the mobility and teletrafficmodels is to capture the movement and callingbehavior of subscribers as a means to predict orevaluate capacity and performance of wirelessnetworks. Specifically, user movement in terms ofdirection and speed affects the time duration in whichusers stay in a cell or location area. In turn, a shorttime duration results in a frequent call handover whenusers are making calls, or a frequent update to locationdatabases for call delivery even when users are notmaking calls. Highly mobile users may also require

MetropolitanArea

GHGH

IHJ

GHGH

IHJ

KLMNL

GHGH

IHJ

GHGH

IHJ

KLMNL

OHJ

GHGH

IHJ

GHGH

IHJ

KLMNL

GHGH

IHJ

GHGH

IHJ

KLMNL

OHJ

Wide/NationalArea

InternationalArea

ININ

LandMobileNetwork

NationalFixedNetwork

InternationalFixedNetwork

PPQPPQ

PPQPPQ

PPQPPQ

Teletraffic Interface

Teletraffic Interface

otherLE/TE’s

otherMSC’s

KEYMTMobile TerminalBSBase StationMSCMobile Switching CenterLELocal ExchangeTETandem ExchangeISCInternational Switching CenterMMFMobility Management Function

Figure 8. Scope of traffic demand characterization for cellular networks(separated mobile and fixed network, mobile-to-mobile communication).


ÙÚÛÜÝÞßàáÚâÛãÜäßåRâçèémore network resources for paging and other signalingfunctions for call delivery than is required for slowermoving users. Furthermore, as discussed earlier, userdensity and service penetration play an important rolein network planning and engineering. An area with ahigh user density is likely to yield high traffic demands,for which sufficient equipment has to be deployed tohandle the projected traffic load. Last but not least,these models should capture the calling behavior interms of call arrival rate, re-attempt, and call holdingtime distribution for proper system dimensioning. Astraffic load of new and handover calls, locationupdates, call delivery and other signaling load havesignificant impacts on service quality (e.g., callblocking and dropping probability), there is always astrong demand for mobility and teletraffic models foranalyzing and engineering cellular networks.

Radio Interface

Mobility and teletraffic models related to the radiointerface typically address traffic load characterizationand handover performance. By assuming uniform userdensity, and randomly chosen fixed movementdirection and speed, the model proposed by [35]expresses the cell-crossing (i.e., handover) rate as afunction of mobile density, mean velocity, and cellperimeter. The model is often referred to as the fluid-flow model. Although it neglects many aspects ofpractical situations, it appears to be the simplestmobility model with closed-form formulas. In fact, theassumption of uniform user density and movementdirection or its variant are commonly used to studynetwork performance; for example, see [36], [37], [38],and [39]. As pointed out below, [40] shows that thefluid model can closely approximate certain practicalscenarios, but fails to do so in others.

In [38], the authors prove that the uniformassumption leads to a biased sampling condition. Thatis, the speeds of cell-crossing terminals are statisticallydifferent from those that remain within a cell. Theprobability distribution function for the speed of cell-crossing terminals are derived for use in performanceanalysis and simulation to obtain consistentcomparisons among different design alternatives. Usingthe same assumption, [39] shows that (a) the handoverrate (i.e., the mean number of handovers per call)increases as the square root of the increase in the cellsper unit area; and (b) the handover rate is given by theratio of mean call duration to the mean cell sojourntime. These results offer an understanding of designtrade-offs and sizing issues in evolving wirelessnetworks.

Although [41] also assumes uniform user density,the author proposes that the speed and direction of amobile terminal are regenerated randomly andindependently after an exponentially distributed randomtime (i.e., a random traveling distance). The mobilitymodel is motivated by the need of reducing thecomplexity of simulation models for studying channeloccupancy time, but [41] also presents arguments tojustify the mobility assumption in certain settings. It is

found that the channel occupancy time can be closelyapproximated by an exponential distribution inpractical situations.

Recently, [42] develops a mathematical formulationfor tracking movement of mobile terminals, whichmove randomly with degrees of freedom matching themobility conditions in practical networks. The modelcan be used to characterize cell residence time,channel holding time, and the average number ofhandovers per call. It is found that cell residence timecan be described by generalized gamma distributions,while channel holding time can be approximated byexponential distributions. The latter is consistent withthe findings in [41], although the formulation in [42]provides additional modeling flexibility and capability.

Instead of tracking user movement, [43] studiesvarious channel assignment strategies for handovercalls by assuming that the residence times of a mobileterminal staying in different cells are independent andidentically distributed. Such an assumption is made fortractability reasons. If it can be validated by actualfield measurements, the mobility model will be usefulin system design and engineering.

It is evident that the assumption of random usermovement is not appropriate for cases such as usersplacing phone calls while driving. For this reason,researchers have observed the need of combiningteletraffic theory and vehicular traffic theory [44] toestimate call and signaling traffic load. For example,[40] proposes a simulation method, which keeps trackof terminal locations in cellular networks by using avehicular traffic model based on realistic relationshipsamong vehicle speed, density and volume. For uniformstreet layouts such as the Manhattan street patterns, itis found that the fluid-flow model [35] yields accuratecell/area boundary crossing rate when compared withthe detailed simulation. However, the fluid-flow modelcan over or under-estimate the crossing rate in non-uniform street patterns. Furthermore, [45] proposes asimulation and analytic model to consider teletrafficload and vehicle movement. The authors also use themodels to study the impacts on call blockingprobability in case of sudden change of vehiculardensity in a ring-shaped service area.

Another mobility model used to capture time andspace dynamics in cellular networks along a highway ispresented in [46]. Assuming that each vehiclealternates between calling and non-calling staterandomly, the model uses differential equations todescribe the movement of both vehicle types. As aresult, call traffic load for each cell along the highwayand handover rates at cell boundaries can be obtained.With vehicles arriving to the system according to time-dependent Poisson processes, new call load andhandover also form Poisson processes. The model canbe viewed as a traffic demand and useful in studyingthe time and spatial dynamics in mobile networks. Forinstance, their analysis shows that a significantincrease of offered load results if users initiate callswith a rate inversely proportional to their speed (e.g., ina traffic jam). In fact, this coupling effect betweenmovement and calling behavior has not been fully


ÙÚÛÜÝÞßàáÚâÛãÜäßåSâçèéunderstood. Additional study will be worthwhile on thesubject as well as on re-attempt behavior, and two-dimensional traveling space. These aspects areaddressed in other studies, although not in theframework of one comprehensive model. As examplesof recent work: The impact of the user re-attemptbehavior on protecting handovers against fresh calls isaddressed in [47] (although disregarding the spatialdynamics); a two-dimensional space is addressed in[48] by reducing the problem to a combination of one-dimensional geometries (but assuming, among others,that the distribution of users on a segment is uniform);“spatial point patterns” modeling (two-dimensional)space, time-frozen distributions of mobile users whileaccommodating in a flexible and computation-friendlyway a range of possible geometrical constraints areconsidered in [49] as a multi-dimensional extension ofthe Markovian Arrival Process (MAP), [50], based onarrival rates dependent on the “environment state” ofthe system.

It is worth noting the potential importance of theconcept of different realms of traffic and mobilitymodels with different levels of details for street, region,metropolitan and national areas, as reported in [51] and[52]. This is so because different amounts of detailssimplify the models, while adequately capturing themajor essence of mobility and traffic situations inquestion. As a result, the simple models may beproven to be sufficient and useful in planning andengineering practical networks. Towards this goal,additional research work will be highly desirable.Recently, [52] introduces a set of mobility models withscope ranging from city to street level. Three modelsthat cover city, zone and street, respectively, areintended to capture mobility at different scales as ameans to estimate mobility and traffic parameters forengineering procedures. (As discussed later, [51] alsomakes a similar observation that mobility models withdifferent levels of details are needed.)

Before leaving the discussion on the radio interface,it must be pointed out that mobility and traffic modelsfor overlaid, micro/macrocell architecture [53], [54] areeven more complicated than those for the single layerarchitecture. Due to the overlapping coverage of microand macrocell, a call can be served either by micro ormacrocell. This brings about new approaches ofdynamic channel assignment according to the mobilityof users; see e.g., [19], [20], [55], [56] and [21]. Further,efficient reuse of radio spectrum in the overlaidnetworks also becomes an issue [57], [18]. In terms ofteletraffic issues, when a call is blocked by amicrocell, it can be “overflowed” to the associatedmacrocell to see if the latter has spare radio channels.Existing methods such as [58], [59] and [60] may beuseful for analyzing the call overflow, but additionalfactors such as mobility need also be considered to

balance performance for new and handover calls [61].A survey of some of these issues for micro/macrocelloverlays can be found in [62].

It is clear that additional research will be needed tofully address the mobility and teletraffic issues for theoverlaid networks.

Fixed Infrastructure

In the study of teletraffic issues for the signalingfunctions and the fixed infrastructure, terminal orpersonal mobility need not, in general, be consideredas detailed as in most of the models discussed above.Rather, models with large scales will be adequate.

Adding new base stations and reducing cell size area common approach to meet increased traffic demand.Small cell and registration areas however tend toincrease signaling traffic. Using the fluid-flow model[35], [63] reveals a potentially significant increase oftraffic load on the signaling links due to a combinationof high terminal density and mobility, and smalllocation area in PCS networks. Using the number oflocation updates between two calls, [64] proposes aframework for estimating the signaling load. Similarly,[65] and [66] predict a large increase of workload forthe network databases to support mobility whencompared with IN network services.

To avoid the performance impacts of signaling anddatabase load due to mobility, many new mobilitymanagement or location tracking algorithms have beendevised and analyzed. The common goal of these newmethods is to reduce the network signaling (includingpaging over the radio channels [67], [68], [69]) anddatabase load, thus improving call-setup delay, networkcapacity and perceived service quality, whileefficiently delivering calls to mobile users. Since thedetails of the algorithms lie beyond the main scope ofthis paper, instead of discussing them here, readers arereferred to the papers and their cited work on thesubject that are published recently in two special issuesof IEEE JSAC [70] and [71].

Last but not least, it is worth noting that [51]proposes a realistic teletraffic modeling framework,which consists of topology, call and mobility model.The call model is characterized by actual call data inan existing telephone network. The mobility modelconsiders user movement at three different scales,resulting in metropolitan, national, and internationalsubmodel. The mobility parameters in the firstsubmodel is estimated from personal transportationsurveys, while those for the latter two are approximatedfrom the airplane passenger traffic data. Using thisframework, workload for the location database can bestudied by simulation. Potentially, it can also be usefulfor evaluating various mobility management algorithmsand network topology design.


ÙÚÛÜÝÞßàáÚâÛãÜäßåTâçèé

A summary of these mobility and traffic models,and related teletraffic issues for mobile networks isgiven in Table 7. The purpose there is to show thepotential of the models in studying networkperformance and design issues.

Where Do We Stand?

espite a large volume of (at times)quite sophisticated mobility, trafficdemand and dimensioning models,theorists, for the sake of tractability,

often make simplifying assumptions about user density,and assume certain statistical properties of channelholding time, cell residence time and other mobilityrelated parameters, when modeling mobilecommunication networks. As the mobile and UPTservices will undoubtedly provide users with richfeatures and multimedia capability, mobility andteletraffic issues for the future networks will becomemore complicated than those in the current secondgeneration networks. It will be a challenge to theteletraffic community to provide engineering tools fordifferent system generations meeting the robustness andsimplicity requirements demanded for a smooth systemoperation.

Some Popular Assumptions: TrafficEngineering "Myths"?

In the area of terminal mobility and cellularsystems, there has been increasing consensus in theopen literature on a series of working assumptionswhich have led to mathematically tractable problems.Given the complexity of mobile system operation andthe need for traffic engineering procedures with ITU-Tsignificance, it is important to revive considerations ofhow well the models being used represent “real world”systems. This is by no means meant to undermine thevalue of traffic models, but rather to stress the need ofvalidating with field data the indications from themodels and determine that the models are accurateenough to justify their adoption in a sound teletraffic

engineering practice. A point to note is that trafficstatistics, and in general information on networkoperation, for a specific cellular radio network arehighly sensitive proprietary information for an operatordue to the competitive nature of the industry. Thisinformation is rarely published in the public domain norshared outside the companies. (Infrastructure suppliersmay occasionally have access to a limited amount ofthis information, but this could mostly be restricted tothe start-up phase of a network). Since access to thewealth of information on real life network operation iseffectively very restricted, contributions on trafficengineering to the open literature usually have to workfrom abstractions whose rationale and impact may notbe backed in all cases by deep knowledge about realsystem operation and needs. These factors contribute tosome of the disconnects between theory of trafficengineering and the real world.

Some of the most popular assumptions related totraffic and mobility modeling have included:

•Radio cells have regular (hexagonal) shape.Cells are defined in terms of radio coverage asprovided by the power emitted by the basetransceiver station/radio port antenna around whicha cell is constituted. As such, the cell “boundary” isassociated with the limiting distance from theantenna site beyond which communication with amobile terminal becomes troublesome. Due toterrain characteristics and existence of obstacles ofvarious nature interfering with propagation of radiowaves, the boundary of a cell is normally fuzzy anda cell coverage may even be jeopardized over anarea.

•Handovers (in FDMA/TDMA systems) areaccomplished as soon as a user crosses theboundary between adjacent cells).Since cells are usually not regularly shaped, theirradio coverage must overlap to some extent. Thisoverlap provides a window during w

Documents

Matching Practice and Theory Communications in ITU-T Work -- …kkleung/papers/grillo.pdf · 2002. 5. 15. · “directed retry”, “cell load sharing”, [2], and “queuing”