Dimensioning of Hieachical B3G etwoks with Multiple Classes of Taffic M. Rubaiyat ibia and Abbas Jamalipou chool of Electical and Infomation Engineeing The Univesity of ydney, W 2006, Austalia {kibia, abbas}@ee.usyd.edu.au Abstact Dimensioning of mobile netwoks (e.g. 2G, 3G, WLA etc.) is moe o less esticted to the sevice aea pojection and wieless capacity fomulation (calculation of numbe of base stations/access points (Bs/APs) equied) in suppoting the anticipated taffic load. As the Beyond Thid Geneation (B3G) netwok pomises to povide inte-connectivity among all existing technologies with minimal softwae and hadwae upgade, any significant change to coveage aea and capacity of existing netwoks will violate this key pomise. To adhee to the B3G definition, in this pape we have poposed a dimensioning famewok that is confined to the wied pat of the netwok and will suppot wide aea coveage (typically fo high speed subscibes) in the pesence of multiple classes of taffic. The famewok deives an initial estimate bound of wied netwok elements (i.e. numbe of Bs pe, numbe of s pe MAP and numbe of MAPs pe GW) in a multi-tie hieachical system based on use capacity limits (soft-capacity), futue taffic pojection and mean taffic load (o aveage data ate) fo each class of taffic. imulation esults demonstate the wied netwok bound unde diffeent taffic distibutions and fo bounded aveage cell capacities (soft-capacity).. Intoduction Access netwoks ae deployed and optimized fo diffeent sevices that have thei own set of chaacteistics (aveage bit ate, bandwidth etc.) and gade of sevice equiements (E b / 0, delay, R etc.). Fo example 2G netwok has been optimized fo voice taffic, while 3G (backwad compatible to 2G) and WLA offe high speed/high bandwidth data sevices. The dimensioning and subsequent deployment of each of these netwoks in pinciple follows the methodology adopted in 2G netwoks which is lagely based on a tedious tial and eo pocess []. This involves calculating the equied numbe of base stations (Bs) upon the pojection of sevice aea and maximum allowable cell loading, and This wok is patly suppoted by the Intenational cience Linkages pogam established unde the Austalian Govenment s innovation statement Backing Austalia s Ability. thei successful deployment. In the deployment phase, an expet adio planne manually selects sites to host a B, specifies popagation model and sevice equiements, and analyzes the measuement data taken aound the B ove vaious instances of time. If the esults do not meet the Qo equiements (e.g. blocked calls, cell loading etc.) of the sevice then eithe changes ae made to the B paametes (tansmit powe, R, antenna tilt and azimuth etc.) o else the B is moved to a new site and the entie pocess is epeated. ubscibe gowth is addessed by adding new netwok elements (e.g. new cell sites o educed cell sizes) to the initial dimensioning estimate in ode to satisfy the inceased taffic. ince the Beyond Thid Geneation (B3G) netwoks epitomize on evolution athe than evolution by inteconnecting all existing and emeging access technologies though a common IPv6 coe netwok, the dimensioning pocess involved will be notably diffeent fom conventional netwoks. This is because any significant change to the adio netwok (coveage aea and capacity) of existing technologies will violate the B3G definition of inte-connectivity. As a consequence the dimensioning pocess will lagely be confined to oganizing the wied netwok elements in a stuctued way guaanteeing minimum Qo. Unfotunately extensive liteatue eview has evealed a death of dimensioning wok in B3G netwoks, in paticula fo the wied segment. Most of the eseach wok so fa has focused on the wieless dimensioning aspect of existing netwoks i.e. the cell coveage (in km) and equied capacity (e.g. numbe of Bs) once the system aea (in km 2 ) and maximum allowable cell loading (eithe in Mbps o Elangs) is detemined [2-4]. Few of the woks that have addessed coe netwok dimensioning include [5] which descibes a GUI based foecasting tool albeit in a fixed channel envionment. Futhemoe the legacy netwoks mainly conside single class of taffic (e.g. poisson distibuted voice call aival) duing the dimensioning pocess. Even when multiple classes ae consideed as in 3G WCDMA netwok, calculations ae ovesimplified by adopting a single class taffic appoach [6]. The cell loading and subsequent planning which esults fom such single taffic appoach theefoe do not eflect the statistical chaacteistics of othe taffic
distibutions. Fo example, multimedia can be modeled as an autoegessive pocess while data taffic can be modeled as self-simila taffic with long ange dependence exhibiting bustiness ove diffeent time scales. Unfotunately analytical expession fo cell loading based on diffeent taffic distibutions is difficult to deive, leading to the adoption of aveage chaacteistics of multiple taffic pofile to simplify the calculations. The netwok designes need to conside these issues in the dimensioning pocess and the ensuing deployment stage of the B3G netwok. In this pape we have theefoe poposed a famewok that pedicts a bound on the wied pat of the hieachical B3G netwok achitectue [7] taking into consideation multiple classes of taffic pofile. The hieachical achitectue, constituting of ties of base stations/access points (Bs/APs), access outes (s), mobility ancho points (MAP) and gateways (GWs), offes: localized handoff enabling wide oaming facility to domant teminals without initiating egistation, faste connectivity and data tansfe between adjacent domains etc. The poposed algoithm offes an appoximate solution to the analytical expession poblem by adopting the aveage taffic pofile of multiple sevices and povides an initial estimate of netwok elements (including numbe of Bs pe, numbe of s pe MAP and numbe of MAPs pe GW) based on the futue taffic pojection and the mean taffic load (in Mbps) pe taffic class. As cell loading elies on the use capacity in the undelying netwok, the softcapacity in CDMA based netwoks (detemined by the bandwidth and powe limits of the B in the fowad link and mobile teminals in the evese link [8]) fa outweighs the fixed capacity of 2G netwoks. The poposed dimensioning algoithm in this pape eflects this vey point by consideing soft-capacity in an intefeence limited (evese link) 3G cell (altenatively in a WCDMA based cell) as the deciding facto along with futue taffic pojection and mean taffic load (aveage data ate) fo each class of taffic in deiving the netwok bounds of a B3G netwok. The pape is oganized as follows. ection II povides a bief oveview of the hieachical system stuctue. ection III explains the poposed dimensioning algoithm while ection IV shows the simulation esults. Conclusions ae dawn at the end of the pape. 2. Hieachical ystem Fig. shows the poposed hieachical and modula B3G netwok with distibuted system achitectue [7] and diffeent link capacities. Within a domain, the lowest tie is fomed by ovelapping Bs/APs fom diffeent undelying access technologies and is contolled by an. All subscibes within a domain access sevices though the MAP. The gateway (GW) connects multiple MAPs to offe faste connectivity and data tansfe duing inte-domain handoves. As shown in Fig., the MAP is connected to multiple GWs and the coe netwok via seveal wied links. ince significant amount of data tansfe takes place between adjacent domains, multiple links assist in distibuting this taffic among seveal GWs theeby alleviating the netwok esponse time. GWs not only enhance inte-domain communication, but duing heavy load condition when the diect link fom the MAP to the coe netwok is nea satuation, they povide emote data tansfe facility as well. A distibuted Bandwidth Boke (BB) based achitectue is adopted to cay out esouce management and esouce negotiation within the domain and along the communication path between pee end teminals [7]. Although each at the edge of the hieachy is capable of becoming a BB, MAP offes the best choice to take on the ole of a BB. Management of each domain is caied out by an administative body and sevice ageement is assumed to exist among diffeent netwok povides. B L G GW L AM Wieless Inteface IPv6 Coe etwok L MG MAP AP GW Figue. Multi-tie hieachical system achitectue ince B3G netwok will be expected to offe ubiquitous mobility, to suppot high speed subscibes (in vehicles) in addition to stationay and pedestian ones, the sevice aea will have to incopoate the macocell appoach which offes wide aea coveage. The poposed algoithm theefoe consides macocells in its dimensioning of wied netwok elements. Once the initial estimate is deived, smalle cell sizes (micocells, picocells etc.) can be consideed to compliment the wide aea coveage. The following section descibes the poposed algoithm in detail.
3. Poposed Dimensioning Algoithm Cell loading is closely associated with the aveage cell esidence ate and call aival pocess. With diffeent pobability distibutions modeling diffeent classes of taffic aival (e.g. voice, data, multimedia etc.), analytical appoach to fomulate aggegate cell loading in B3G netwok soon poves to be intactable necessitating the adoption of an altenative appoach (consides aveage taffic pofile). uppose that thee ae k classes of taffic in the netwok: k =, 2, 3,. Assuming unifom distibution of aveage cell esidence ate, the ate fo k classes of taffic in cell i is given as, i =(,i, 2,i,.,,i ). If the mean taffic load/data ate fo k classes of taffic is, R =(R, R 2, R 3,., R ) then the aveage taffic capacity in cell i can be witten as, Ti = k, i. Rk, i=,2,3.m cells. Aveage Cell Load Calculation: As mentioned ealie, in the poposed algoithm a maco-cell based appoach (wide aea coveage) is utilized whee the cell adius is deived fom [9]. Refeence [9] dimensions a B3G netwok in the pesence of voice taffic only to ensue that the infastuctue has sufficient numbe of elements to handle at least voice sevice wheneve thee is an outage of othe sevices. It can be seen fom [9] that in a B3G netwok the maximum cell adius (macocell) achievable with a Walfisch-Ikegami model (WIM) fo a cetain Qo (E b / 0 =5 db) is 2.5 km. With such a cell adius, the signal powe fom the mobile teminal will diminish significantly as it eaches the seving B in the pesence of path loss and log nomal shadowing. In ode to maintain the desied R at the B, closed o open loop powe contol is used to egulate the teminal tansmit powe so that signals fom all teminals within the cell coveage aea aive at the B with equal powe. ince CDMA cell capacity is coveage limited (usually in the evese diection) as well as being intefeence limited, we calculate the minimum allowable eceived powe to achieve the taget E b / 0. Teminals nea the edge of the cell will be the most affected by the popagation channel impaiments. Theefoe in ou calculation of minimum eceived signal powe we conside the tansmit powe fom this distant teminal as, = 0 ξ /0 d 4 t () whee is the eceived signal powe, t is the maximum allowable tansmit powe (teminal), d is the cell adius and ξ is a zeo mean Gaussian andom vaiable with standad deviation of aound 8 db. ince cell loading in CDMA based netwoks ae govened by intefeence in the evese link, total intefeence at the B needs to be estimated befoe capacity can be detemined. With multiple classes of taffic, individual taffic pofiles ae equied to calculate the aggegate cell loading. The eceived signal powe fom () and a pioi knowledge of expected E b / 0 fo each class of taffic enables the calculation of maximum intefeence level fo the k-th taffic and is given by, w a k. R I k k = (2) E b o k whee I k is the total intefeence, w is the chip ate, R k is the data ate, a k is the activity facto and (E b / 0 ) k is the bit enegy to noise density atio of the k-th taffic. The total intefeence petaining to each class of taffic defines the uppe and lowe theshold of intefeence which ae given by I th,low = min(i k ) and I th,up =max(i k ): k =, 2, 3,,. These intefeence thesholds then enable the calculation of the use capacity pe cell as follows, ( i ) + η ( i ) + η = I th, x (3) whee the fist tem is inta-cell intefeence, the second tem is inte-cell intefeence govened by the loading facto η [9] (η is a facto between 0% and 00% epesenting pecentage incease of intefeence fom neighboing cells above those intoduced by inta-cell uses only) and the thid tem is backgound noise, while I th,x denotes the intefeence theshold level. Hence fom (3) the lowe cell capacity can be witten as, I th, low η L = + (4) (+ η ) and highe cell capacity is given as, I th, up η U = + (5) (+ η ) The aveage cell esidence ate (unifomly distibuted in the entie netwok) is theefoe bounded by ( L, U ). etwok Dimensioning: It is expected that by 200, 80% of the total taffic will be data (eal-time as well as noneal time packet taffic) while only 20% will be voice taffic. Assuming unifom taffic distibution thoughout the netwok, each cell taffic will have the following distibution, p = (p, p 2, p 3,, p ): p k =. Based on the aveage cell esidence bound ( L, U ) fom (4) and (5), taffic distibution fom above and mean taffic load (o data ate) fo each class of taffic, the soft-capacity of a B in cell i can be calculated as, C B, ix = p Rk. Ith, x = Rk. Tth, x. (6) whee T th,x denotes the aveage taffic bound in cell i. Hee we conside that all the B capacities within the netwok ae bounded by this soft-capacity limit. Assuming fibe optics to eplace existing wied links so that moe connections can be accommodated via fo
example wavelength division multiplexing (WDM) theeby offeing high data tansmission ate in the wied backbone, the following links, diffeentiated by thei capacity, ae consideed in the poposed algoithm (shown in Fig. ). L AM =link between and MAP =622.08 Mbps (optical caie (OC)-2) L MG =link between MAP and GW=.244 Gbps (OC-24) L G =link between GW and the coe netwo2.488 Gbps (OC-48) With technology impovement in CMO design, highe speed links may become available that will be able to eplace the abovementioned ones. Using these OC links the numbe of Bs pe can be deduced as, _ link _ capacity L n AM B = = (7) B _ capacity C B, ix ow each MAP has a total of l links to GWs and the coe netwok whee each of the links is denoted by L MG. Based on the total capacity of a single MAP, the numbe of s pe MAP can be calculated as follows, MAP _ link _ capacity L MG. l n = = (8) _ taffic n B. C B, ix Assuming that β faction of the total taffic fom a MAP goes to multiple GWs while (-β) faction goes diectly to the coe netwok, numbe of MAPs pe GW can be calculated as, GW _ link _ capacity L = = G LG n MAP = Faction_ MAP_ taffic n. n B. C B, ix LMG. l β. β. l l (9) L. whee. MG l β epesents the aveage taffic sent to a l single GW fom each MAP. 4. imulation Results Table summaizes the paametes used in the poposed dimensioning algoithm. The focus hee is on calculating an initial estimate bound of wied netwok elements offeing wide aea coveage. This is eflected on the choice of data ate fo the thee classes of taffic consideed, especially data and multimedia taffic with espect to mobile subscibes (possibly in vehicles). imulations ae caied out ove the uppe and lowe limit of aveage cell esidence ate (in this case 6 and 54 uses pe cell) fo a cell adius of 2.5 km. Fig. 2 and 3 show the wied netwok element bounds fo diffeent taffic distibutions and link choices. Hee taffic distibution signifies vaiation of voice taffic in compaison with data and multimedia taffic. The voice taffic is vaied fom 00% to 20% as pe the futue taffic pojection. It can be seen fom Fig. 2 that as the data and Table. imulation paametes Taffic class, k Voice, data, multimedia Cell adius, d 2.5 km [9] Mobile Tansmit powe, t Chip ate, w watt 3.84 Mchip/s Bit enegy to noise E density, b 5 db,.5 db, 5 db [] o k Mean taffic load (data 2.2 bps, 44 bps, 64 R bps [6, 0] ate), ( b ) k omal andom vaiable, ζ tandad deviation: 0 db Activity facto, a k 0.5,, [6, ] Backgound noise, η -03 dbm [2] Loading facto, η 30% Faction of MAP taffic, β 50% Aveage cell esidence ate Lowe limit: 6 ( L ) Uppe limit: 54 ( U ) multimedia taffic stats to dominate voice, the numbe of Bs pe (l =3) deceases, taking into consideation the highe data ate and E b / 0 equiements. The choice of the equied numbe of Bs pe will ely entiely on the netwok povide, bounded by the extemities shown in Fig. 2. The issue of monetay cost and povide s own taffic pojection will subsequently dictate this choice. Fig. 3 on the othe hand shows the selection of numbe of s pe MAP and numbe of MAPs pe GW as l is vaied with 20% voice taffic pojection. It is evident fom the figue that l has a pofound effect on the numbe of s pe MAP with lowe esidence ate. This is because with finite link capacity (L MG ) lowe esidence ate allows moe s to be included unde the same MAP as l is vaied while with a highe esidence ate the cell loading quickly conveges to the maximum limit. We can also see fom Fig. 3 that as l inceases so does the numbe of MAPs pe GW with lowe esidence ate because the MAP can distibute the adjacent-domain taffic moe equitably among inceased numbe of GWs thus allowing highe numbe of MAPs to be included unde the same GW. At highe esidence ate the taffic unde each MAP conveges to the maximum limit and hence the choice of numbe of MAPs pe GW is affected entiely by the links l wheeas the choice conveges to a value of 3 at lowe esidence ate. When deploying the B3G netwok the onus will theefoe be with the netwok povides to choose the
numbe of wied netwok elements in addessing the pojected taffic load within the afoementioned bounds. Pope selection of the wied links will be the key to meeting this taffic. 5. Conclusions B3G netwok is envisaged to glue togethe all existing and emeging access technologies though a common IPv6 coe netwok. ince it will be an evolutionay step fom cuent netwoks athe than a evolutionay new technology and would involve minimal softwae and hadwae upgades, significant changes to the existing cell coveage and capacity will violate the key pomise of inte-connectivity. Theefoe the dimensioning of B3G netwok will lagely involve estuctuing the hieachical system though pope selection of wied links. In this pape we poposed a dimensioning famewok that deives the uppe and lowe bounds of wied netwok elements i.e. the numbe of Bs pe, the numbe of s pe MAP and the numbe of MAPs pe GW, in the pesence of multiple classes of taffic. Futue taffic pojection (data taffic including multimedia dominating othe taffic classes) and mean taffic load (o data ates) ae utilized in deiving the aveage B capacity (cell loading) bound. The CDMA based soft-capacity that govens the cell loading in the poposed algoithm is calculated fom an intefeence theshold based mechanism. The simulation esults pesented at the end of the pape offe the netwok povides with an estimate of the wied netwok elements that would be equied duing the deployment phase of a B3G netwok. [5] C.. onstantinopoulou,. A. outsopoulos, G. L. Lybeopoulos and M. E. Theologou, Coe netwok planning, optimizing and foecasting in GM/GPR netwoks, IEEE ymposium on Communications and Vehicula Technology (CVT 00), Octobe 2000, pp. 55-6. [6] T. Dahlbeg,. R. ubamanian and B. Cao, oft capacity modeling fo thid geneation adio esouce management, IEEE Intenational Mobility and Wieless Access Wokshop (MobiWac 02), Octobe 2002, pp. 33-39. [7] M. R. ibia, V. Michandani, and A. Jamalipou, A Consolidated achitectue fo 4G/B3G netwoks, IEEE Wieless Communications and etwoking Confeence (WCC 05), vol. 4, Mach 2005, pp. 2406-24. [8]. Dimitiou, etwok planning & esouce management issues fo mobile multimedia CDMA systems, IEEE IEEE 59 th Vehicula Technology Confeence (VTC 04), vol. 4, May 2004, pp. 234-2345. [9] A. Jamalipou, V. Michandani and M. R. ibia, Dimensioning of an enhanced 4G/B3G infastuctue fo voice taffic, IEEE Pesonal and Indoo Mobile Radio Communications (PIMRC 2005), Belin, Gemany, eptembe 2005. [0]. C. Tang, CDMA RF ystem Engineeing, Atech House, Inc., London, 998. [] H.-P. Lin, R.-T. Juang, D.-B. Lin, C.-Y. e and Y. Wang, Cell planning scheme fo WCDMA systems using genetic algoithm and measued backgound noise floo, IEE Poceedings on Communications, vol. 5, no. 6, Decembe 2004, pp. 595-600. [2] E. Dinan, A. uochkin and. ettani, UMT adio inteface system planning and optimization, Bechtel Telecommunications Technical Jounal, vol., no., Decembe 2002, pp. -0. 6. Refeences [] J. Zhang, L. Guo and J. Y. Wu, An integated appoach fo UTRA planning and optimization, IEEE 59 th Vehicula Technology Confeence (VTC 04), vol. 4, May 2004, pp. 2360-2364. [2]. Ions, C. Johnson, A. ing and D. McFalane, uppoting the successful deployment of thid geneation public cellula technologies-system dimensioning and netwok planning, IEE Fist Intenational Confeence on 3G Mobile Communication Technologies, Mach 2000, pp. 56 60. [3] W. Moh, R. Lude and -H. Mohmann, Data ate estimates, ange calculations and spectum demand fo new elements of systems beyond IMT-2000, IEEE 5 th Intenational ymposium on Wieless Pesonal Multimedia Communications, vol., 2002, pp. 237-24. [4] A.Heas-Bandin, P. Batolome-Pascual, D. Gomez-Mateo and J. Izquiedo-Ace, A multisevice dimensioning pocedue fo 3G CDMA, IEE Fist Intenational Confeence on 3G Mobile Communication Technologies, Mach 2000, pp. 406-40.
3500 Dimensioning of Bs fo multiple classes of taffic (l=3) numbe of Bs (uppe limit) numbe of Bs (lowe limit) 3000 2500 numbe of netwok elements 2000 500 000 500 0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 % of voice taffic Figue 2. election of Bs pe fo diffeent taffic pojections 4 2 umbe of s and MAPs fo diffeent l numbe of s pe MAP (cell esidence=6) numbe of MAPs pe GW (cell esidence=6) numbe of s pe MAP (cell esidence=54) numbe of MAPs pe GW (cell esidence=54) 0 numbe of netwok elements 8 6 4 2 0 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 numbe of MAP uplinks, l Figue 3. election of s pe MAP and MAPs pe GW fo vaying numbes of l at 80% data taffic pojection