Research Article Modeling and Performance Analyses of Hybrid Cellular and Broadcasting Networks

Transcription

1 Internatonal Journal of Dgtal Multmeda Broadcastng Volume 29, Artcle ID 32973, 9 pages do:.55/29/32973 Research Artcle Modelng and Performance Analyses of Hybrd Cellular and Broadcastng Networks Peter Unger and Thomas Kürner Insttute for Communcatons Technology, Braunschweg Techncal Unversty, Schlentzstrasse 22, 386 Braunschweg, Germany Correspondence should be addressed to Thomas Kürner, t.kuerner@tu-bs.de Receved March 29; Revsed 28 October 29; Accepted 3 November 29 Recommended by Sandro Scalse Moble communcaton servces are gettng more and more mportant and, n partcular, multmeda servces have attracted the nterest of the users. Moble TV s one of the most demanded canddates. Powerful and effcent communcaton systems are needed, whch provde hgh capactes, especally at the downlnk. Furthermore, nteractvty s essental for supportng the user needs and to extend the servce offerng. As one possble soluton to meet the mentoned requrements, we consder the combnaton of the cellular network UMTS and the moble broadcast network DVB-H, whch form a hybrd network. We nvestgate the performance of hybrd networks and develop a system model, whch descrbes the hybrd network and the load swtchng between both networks. One of the contrbutons s the defnton of the swtchng bound concept, whch represents an effcent tool to assess the necessty and the feasblty of hybrd networks and the amount of load swtchng. The performance ndcators cell load and grade of servce are analyzed by usng theoretcal and realstc scenaros. Copyrght 29 P. Unger and T. Kürner. Ths s an open access artcle dstrbuted under the Creatve Commons Attrbuton Lcense, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted.. Introducton Wth the ntroducton of the 3rd Generaton cellular network Unversal Moble Telecommuncatons System (UMTS), hgher data rates have been made possble and the varety of servces has been ncreased, compared to the prevous cellular networks such as General Packet Rado Servce (GPRS). In addton to the tradtonal servces, such as voce telephony and messagng, the multmeda servces and hgh data rates have been made possble and accepted by the users. The new offered servces, for example, vdeo telephony, Internetbased data communcatons, and streamng servces, make hgh downlnk capactes and effcent networks necessary. The currently avalable termnals are equpped wth several recevers, whch allow for the recepton of data from dfferent networks, such as UMTS and DVB-H. Hybrd networks are possble for these types of termnals and the broadcast downlnk channel provded by DVB-H can be used very effcently to serve many users at the same tme wth hgh data rate servces. The DVB-H technology s based on the terrestral dgtal TV system DVB-T, whch has been successfully ntroduced n many countres [, 2]. The sgnal has been made more robust n order to enable moble recepton wth hgh veloctes and at moble envronments, for example, to cope wth multpath recepton. An addtonal error protecton scheme has been ncluded at the lnk layer, called Multprotocol Encapsulaton Forward Error Correcton (MPE-FEC) [3]. Furthermore, the tme slcng concept has been ntroduced for power savng at the termnal sde, whle servces are transmtted n tme slots and the recever front end can be swtched off n the meantme. Furthermore, the tme slcng approach enables soft handover and optmzed handover algorthms [4], and local content areas n sngle frequency networks [5]. Dependng on the modulaton and codng scheme, and the used channel bandwdth, a total downlnk data rate of about 6 Mbps can be acheved. The ndvdual user data rate depends on the tme slcng setup and s typcally set to 4 kbps. The currently deployed cellular system UMTS s standardzed by the 3rd Generaton Partnershp Project (3GPP) and makes use of the Wdeband Code Dvson Multple Access (W-CDMA) technology. All users are served at a common frequency wth a 5 MHz bandwdth and are

2 2 Internatonal Journal of Dgtal Multmeda Broadcastng separated by orthogonal codes. The network capacty and coverage suffer from nterference and have to be planned carefully. The network s deployed n a cellular structure, whch s dense n urban envronments and larger cells occur n less populated areas. Moble termnals are assgned to one or more cells, whch are establshed by the Node-Bs. Dfferent bearer servces can be used n a crcut or packet swtched mode and thus dfferent types of user servces are avalable, such as voce telephony, web browsng, streamng multmeda, and fle download. The combnaton and the cooperaton of both network types has been enabled by the IP Datacast standard, whch has been specfed wthn the Internatonal DVB Project [6]. It defnes the hgher layers and form an Internet Protocol- (IP-) based end-to-end system, together wth the DVB-H specfcaton as the physcal layer [7]. In addton to the broadcast downstream, the IP Datacast reference archtecture defnes an optonal lnk wth an nteractve cellular network, whch s used, for example, for servce requests. In terms of hybrd networks, the connecton to a cellular network s defned as a mandatory feature. The combnaton of both network types has several advantages. On the one hand, popular content, whch s requested by several users at the same tme, can be swtched from UMTS to a broadcast transmsson for a more effcent delvery. The UMTS network can be unloaded n order to avod a cell overload or to reduce the overall cell power and the ntercell nterference. On the other hand, the broadcast servces are enhanced by an uplnk channel and new features are possble, such as on-demand servces. Hybrd networks do not exst nowadays. But at ths stage, the deployment and operaton are possble, snce the fully standardzed and currently deployed networks UMTS and DVB-H, the IP Datacast standard, and load balancng algorthms as, for nstance, proposed n [8] exst. Nevertheless, effcent models are stll needed for hybrd network plannng and network performance analyss. We propose a performance model, performance crtera, and constrants for load swtchng n hybrd network, whch s called the swtchng bound concept. The performance s shown by applyng dfferent extents of load swtchng for a theoretcal and a realstc scenaro. Ths paper s structured as follows. In Secton 2 the servces and the scenaro are defned, whch have been consdered for our analyses. Secton 3 descrbes models for the ndvdual networks UMTS and DVB-H. In Secton 4, the model for hybrd networks s derved, the network performance s evaluated, and the swtchng bounds are defned. In Secton 5, the smulaton results are shown by means of a realstc network scenaro. 2. Defnton of Servces and Scenaro As mentoned above, several types of servces are enabled by UMTS, and eght dfferent types have been summarzed n [9]. The defned servce set S ncludes voce and vdeo telephony, web browsng and emal, locaton-based servces, short/multmeda massagng servce, fle download, and streamng multmeda. It s assumed, that some of these Fgure : UMTS network structure and user ntensty for the streamng multmeda servce n a 7.5 km 7.5 km scenaro wth a resoluton of 5 m. The network s based on the publc reference scenaro of Berln avalable at []. servces, namely, fle download and streamng multmeda, are sutable to be transmtted by both network types, UMTS and DVB-H. For ths subset of servces, denoted as S (H),load swtchng can be appled by transferrng the transmsson from UMTS to the broadcast network. In ths paper, we focus on the streamng multmeda servce wth an average stream data rate of 28 kbps. It has been turned out that ths servce generates a much hgher contrbuton to the cell load, compared to the fle download servce. The other sx servces are called cellular servces and have always to be delvered by the UMTS network. In addton to the defnton of the servce characterstcs and the requrements, several scenaros have been proposed n [], whch contan reference network archtectures wth reference stes, antenna confguratons, and predctons of the sgnal propagaton loss. Furthermore, realstc traffc demands have been provded n the form of user ntensty maps, whch descrbe the spatal dstrbuton of the servce request rate. We select the publc reference scenaro of Berln for our nvestgaton, whch s publcly avalable at []. Fgure shows the selected UMTS stes and the user ntensty of the streamng multmeda servce for the entre scenaro area. A subset of 48 stes has been selected from the reference scenaro, n whch the angles of azmuth and elevaton have not been modfed. However, new sgnal predcton maps have been calculated wth a more sophstcated propagaton model, based on [2]. 3. Modelng the UMTS and the DVB-H Network The UMTS network s a bdrectonal network, whch has to be planned for both uplnk and downlnk. In general, the downlnk part s the more crtcal one, due to a hgher amount of traffc demand, caused by asymmetrc servces. The hybrd network can be benefcal for the downlnk part, User ntensty (users/m 2 )

3 Internatonal Journal of Dgtal Multmeda Broadcastng 3 f the addtonal capacty and the combng capablty of the broadcast network are used effcently. It s assumed that all users are connected to the UMTS network at the uplnk, regardless of the way of downlnk delvery. Thus, load swtchng does not effect the uplnk and tradtonal models can be appled. Therefore, we focus on the modelng of the downlnk part. 3.. Modelng the UMTS Downlnk. An effcent model for UMTS network evaluaton has been proposed n [3] and descrbed n detal n [4]. We revew ths model n ths secton, snce t s the bass for developng the model for hybrd networks. The UMTS model descrbes the servce requrements, the generated user load, the ntracell nterference and the nterference couplng between cells, whch s caused by ntercell nterference. It provdes an effcent estmaton of the necessary transmt power for the moble termnals and the Node-Bs by solvng a lnear equaton system, whch s called nterference couplng system. The model can be used n conjuncton wth snapshot analyses usng Monte-Carlo smulatons. Several detals can be mplemented wth ths approach and thus, t generates accurate results for a hgh number of snapshots. But t s computatonal expensve. A more effcent approach has been proposed n [5] and s descrbed more n detal n [4], whch uses expected values for the user demand, nstead of ndvdual users. Thus, t replaces the computatonal expensve snapshot approach wth an average vew on the current network load. The drawback s that the accuracy of the results s somewhat decreased and some detals are not modeled, for example, dfferent veloctes. In the context of ths work, both methods have been consdered. Frst, the downlnk of hybrd networks s modeled and the prncpal behavor of load swtchng s shown by means of the expected approach. For network smulatons performed n Secton 5, a realstc network and the snapshot approach are appled. For determnng the cell load and the couplng between the cells, moble termnals are assgned to the Node-Bs by the best server approach, whereas that cell s selected, whch provdes the strongest plot (CPICH) sgnal. The best server area A of cell s determned by usng the plot channel power p (CPICH) and the end-to-end channel gan γ (x) from antenna to locaton x, whch bascally ncludes the propagaton loss and antenna gans. Each servce requres a specfc carrer-to-nterferenceand-nose rato (CINR) target, denoted as μ s, whch ncludes the processng gan of the selected bearer servce and the requred E b /N target [3]. The actvty of the bearer usage s denoted by α s and the orthogonalty loss factor of a locaton x A s denoted by ω(x). The orthogonalty of the sgnals s partly lost due to multpath propagaton effects and thus, ntracell nterference s caused. The caused load of servce s at locaton x s calculated by l s (x) := α s μ s +ω(x)α s μ s. () The nterference couplng approach defnes the couplng factors for cell (2) as expected values, whch form the nterference-couplng downlnk matrx C = (c j ),j {,...,Ncell}. The parameter T s (x) represents the user ntensty of servce s at locaton x: c = ω(x) l s (x)t s (x)dx, x A s S γ j (x) c j = l s (x)t s (x)dx. x A γ (x) s S The couplng system for all cells can be wrtten as p = c p + c j p j + p (η) j (2) + p (CC) (3) wth p (η) = x A (η(x)/γ (x)) s S l s (x)t s (x)dx beng the average requred power to overcome the nose n a non nterfered cell. The constant power values of all common channels ncludng the plot channel are summed up n p (CC). The parameter p denotes the average transmt power of cell n order to serve all users and to overcome nose and nterference. The resultng power values p must not exceed the maxmum feasble power p max. In order to avod cell overloadng caused by hgh traffc demand and hgh nterference, the scalng factor λ [, ] has been proposed n [4]. It s used to scale the couplng matrx wth dag(λ)c n order to reduce the traffc demand per cell and the couplng to the other cells. Ths approach assumes perfect load control wth applyng an exact downgrade of cell load. The scalng factor λ represents the Grade of Servce (GoS); that s, the fracton of served traffc compared to the offered traffc n cell, and( λ ) defnes the blockng probablty of cell. In[4] an teratve process has been defned to estmate the parameters λ and p for each cell. The nterference couplng concept assumes perfect power control, whch enables to precsely assgn the necessary power to each user. Another measure of the qualty n network plannng s the average other-to-own nterference ι, whch descrbes the total ntercell nterference to cell (4). Note that ths defnton has to be dvded by the average orthogonalty loss (e.g., ω =.673 for urban envronments) n order to be comparable to tradtonal defntons, for example, n [3]: ι = j c j p j c p. (4) In [4], the so-called pole equatons have been defned to calculate the necessary average transmt power p (5) and the grade of servce λ (C) (6) per cell wth a gven other-toown nterference. We have used the notaton (C) to denote

4 4 Internatonal Journal of Dgtal Multmeda Broadcastng the cellular network, snce we have also defned the GoS crtera for the broadcast (B) and the hybrd (H) network: p p (CC) + p (η) f (+ι )c < p max p (CC) p (η), = (+ι )c p max p max, otherwse, λ (C) f (+ι )c < p max p (CC) p (η), p max = p max p (CC) p max (+ι )c + p (η), otherwse. (6) The transmt power p and the GoS λ C form a complementary system. The transmt power ncreases untl the offered traffc load can be served or the maxmum power s reached. Addtonal traffc load s blocked, whch reduces the GoS of the cell. In ths work, alternatve concepts of solvng a network overload as descrbed n [6] are not consdered. The GoS of the entre UMTS network λ (C) s estmated by a weghtng sum, consderng the offered traffc loadτ = x A s S l s (x)t s (x)dx: λ (C) = (5) τ λ (C) τ. (7) The utlzaton of the transmt power n each cell s measured wth the downlnk cell load. It s defned by the rato of the necessary downlnk transmt power compared to the total avalable maxmum power: L (C) = p /p max Modelng the Broadcast Network. In contrast to the UMTS network, the requred system capacty of broadcast network s ndependent on the actual number of users. The necessary amount of capacty depends on the number of dfferent content tems, that are requested by the users [7]. The requests for dfferent content have dfferent probabltes; that s, some tems are more popular compared to others. A common descrpton of the popularty s the Zpf-dstrbuton defned n (8). Detaled analyses and measurements have been performed and descrbed n [8]. The probablty that a user request corresponds to tem s defned as P(X = ) = N tem Ω κ wth Ω = κ. (8) The parameter N tem represents the number of tems offered by the content provder and κ s the shape parameter of the popularty dstrbuton. For a value of κ =, the requests = are unformly dstrbuted to all tems. For an ncreasng value of κ, the popularty concentrates more and more on specfc tems. As mentoned above, we consder the streamng multmeda servce for hybrd delvery, whereas several requested streams are swtched to the broadcast network. Snce streamng s a real-tme servce, the broadcast capacty s shared among all swtched streams n a parallel transmsson. Each stream requres a constant data rate d n order to guarantee suffcent Qualty of Servce (QoS). We defne the broadcast cell load L (B) n (9), whch depends on the number of delvered streams and the capacty D (B), reserved for swtchable servces. If L (B) exceed one, the broadcastcellsoverloaded,somestreamshavetobeblocked and the correspondng users cannot be served: n L (B) d =. D (B) (9) = The maxmum number of streams, whch can be delvered n parallel, s calculated by n max = D (B) /d wth d beng the average requred data rate of all streamng servces. 4. Modelng Hybrd Networks In a hybrd system, the delvery of servces can be swtched between the downlnks of both networks, UMTS and DVB- H. The overall network performance can be optmzed wth properly applyng load swtchng. The load swtchng n hybrd networks s modeled by specfyng the amount of traffc, whch s swtched from the UMTS network to the broadcast network. In contrast to GoS scalng, the swtched traffc s not lost, but served by the broadcast system. Thus, the UMTS network s unloaded, whle mprovng the overall grade of servce. Frst of all, the offered traffc has to be classfed nto the cellular servce demand and swtchable servces (see Secton 2). Snce the couplng matrx C s generated by summng up the user demand for each servce, ths matrx can be splt nto the cellular C (C) and the hybrd part C (H). All elements n the matrces can be calculated by usng (2) and the approprate set of servces S/S (H) and S (H). We defne the load share parameter β, whch specfes the fracton of cellular (nonswtchable) traffc load as β = c (C) /(c (C) + c (H) ). For β =, t s possble to swtch the entre traffc load to the broadcast network and for β =, no swtchable traffc load exsts at all. The rato parameter β s dependent on the locaton. An average value of about 43% occurs for evaluatng the area of the Berln reference scenaro by consderng the streamng multmeda servce for load swtchng. That means, about 57% of the traffc can be used for unloadng the UMTS network. 4.. Defnton of Load Swtchng. The fracton of traffc load s estmated, whch s actually needed to be swtched to the broadcast network for a successful unloadng of the UMTS network. For example, a typcal target crtera for network plannng s to acheve a cell blockng rate of 2% or less, whch corresponds to λ.98.

5 Internatonal Journal of Dgtal Multmeda Broadcastng 5 Number of streams Swtchng rato δ κ = κ =.5 κ = κ =.5 κ = 2 κ = 3 Fgure 2: Comparng the swtchng parameter δ wth the number of streams n, whch are offered to the broadcast system, dependng on the popularty dstrbuton. We defne the swtchng parameter δ [, ], whch s used to control the amount of traffc load, swtched between both networks. A value of δ = ndcates that all requests for the swtchable servce are swtched to the broadcast network. For example, all streamng servces are served by the DVB-H network. Therewth, we defne c j as the new couplng factor for the UMTS network wth c j = c (C) j + δ c (H) j. () It has been shown n [8] that those servce tems are benefcal for optmzng hybrd networks, whch causes a hgh load n the UMTS network. The product of arrval rate and requred resources (.e., data rate) has been taken as a measure to defne the swtchng order. In ths work, we assume equal data rates for all streams and therefore, the tem request rate s drectly proportonal to ts popularty. Thus, we assume that the most popular tems are swtched frst. As mentoned before, both networks can be descrbed by dfferent schemes, the user and content dependence. In order to estmate the broadcast cell load, we have to transform the appled load swtchng wth the swtchng parameter δ nto the correspondng number of streams n δ, whch have to be delvered by the broadcast network. The fracton ( δ) corresponds to the n most popular tems. Thus, we look for the mnmum n, whch fulfls δ P(X n). In other words, the (N tem n) low popular tems have to be served by the UMTS network. Thus, the relaton between n and δ can be defned as n δ = mn n {n P(X >n) δ}. () On the other hand, the estmaton of δ for a gven number of streams n, whch have to be swtched, can be descrbed by δ n = P(X >n). In Fgure 2, the number of streams s shown, whch s offered to the broadcast system, dependng on δ and the popularty shape parameter κ. The total number of tems s set to N tem =. It can be seen that, wth κ =, all tems have the same popularty and thus, n ncreases lnearly for more broadcast traffc. For an ncreasng κ, a hgher fracton of traffc corresponds to the more popular tems. The broadcast technology allows for combnng user requests to the correspondng hgh popular tems. Thus, for the same δ, the broadcast network s less loaded for hgher values of κ Defnton of Swtchng Bounds. Dependng on the system parameters and the traffc demand, the possble range of swtchng may be lmted, f cell load values of L (B) and L (C) have to be guaranteed. We ntroduce the swtchng boundconcept,whchdefnesaboundforδ for each network separately. In order to apply reasonable load swtchng n a hybrd network, the broadcast network should defne the lower and the UMTS network the upper bound of δ. The swtchng bounds ndcate the necessty and feasblty of applyng a hybrd network. It s obvous, f no bound has been set by the UMTS network, no unloadng s requred Broadcast Bound. The bound δ (B) [, ] s defned by the parameters of the broadcast network and the streamng servce. Snce the broadcast capacty s lmted, the number of parallel transmttable streams s lmted too. If the broadcast cell load defned n (9) equals one, the maxmum number of streams n max s reached. The broadcast bound δ (B) s calculated by ( ) D δ (B) (B) = P(X >n max ) = P X>. d (2) Therefore, δ (B) depends on the broadcast capacty D (B), the servce data rate d, and the popularty dstrbuton, that s, N tem and κ. Fgure 3 shows δ (B) dependng on the popularty shape parameter κ and the broadcast cell capacty D (B). The number of tems s set to N tem = and the average streamng data rate s d = 28 kbps. It s obvous that for hgher broadcast data rates, more traffc canbeswtched to the broadcast network and δ (B) s decreased and less restrctve. For a more even popularty dstrbuton (smaller values of κ), the bound ncreases due to a hgher varety of the requested streams and thus, a hgher broadcast capacty demand. For a very large values of κ, the popularty corresponds to a very small number of tems, n extremum to one sngle tem. Thus, all requests can be delvered by the broadcast network. For D (B) = Nd = 2.8 Mbps, t s possble to serve all tems by the broadcast network, thus δ (B) = and no restrctons exst from the broadcast network sde. For a too small capacty, that no stream can be transmtted, the broadcast bound equals one for any value of κ Cellular Bound. The cellular bound δ (C) [, ] descrbes the mnmum necessary unloadng of cell by swtchng streamng servces to the broadcast network, untl

6 6 Internatonal Journal of Dgtal Multmeda Broadcastng Broadcast bound δ (B) UMTS bound δ (U) D B =.Mbps D B =.5Mbps D B = Mbps D B = 2Mbps Popularty shape κ D B = 4Mbps D B = 7Mbps D B = Mbps D B = 2.8Mbps Fgure 3: Broadcast bound δ (B) dependng on the popularty shape κ and the total broadcast capacty Fracton of UMTS traffc β c =.45 c =.5 c =.6 c =.8 c =.2 c = 2 Fgure 4: Cellular bound δ (C) for cell dependng on the load share parameter β andthetotaloffered traffcload. the GoS target λ s reached. In ths work, we use λ =.98. Once the cell s able to handle the amount of offered traffc, a further unloadng yelds to a reducton of the requred transmt power p.inordertoestmateδ (C), we use (6) wth c = c (C) t to δ : + δ c (H) and p (η) = p (η,c) + δ p (η,h), and transform δ (C) p max p (CC) λp max (+ι )c (C) p (η,c) = λp max (+ι )c (H) + p (η,h), λ (C) < λ,, λ λ (C). (3) It s obvous that a soluton s only possble, f swtchable traffc exsts n cell, so that c (H) >.Ifallservcesare swtchable (β = ) and the GoS target s set to one, the bound becomes equal to the GoS λ (C). δ (C) In the followng, we nvestgate a theoretcal scenaro of a sngle cell, n whch the nfluence from the surroundng cells, that s, the average other-to-own nterference ι, s assumed to be constant. The curves n Fgure 4 depct the cellular bound δ (C). The bound s shown for dfferent cell loads c and dependng on load share parameter β. For β =, no hybrd traffc exsts and no load swtchng can be appled. For β =,allservcescanbe transmtted by both network types. Ths case shows the less restrctve bound values, snce no cellular background traffc exsts. In ths example, the followng parameters have been assumed: p max = 4 W, p (CC) = 4W, ι =.6, and p (η) =.96 mw. If λ (C) λ, no traffc needs to be swtched, snce the QoS target s already acheved wthout applyng unloadng. The threshold n our case s c =.45. Ths threshold can be calculated by usng (6) wthλ (C) = λ. Thus, δ (C) s set to one for all values of c, whch are lower or equal as ths threshold. In general, t can be seen that δ (C) decreases for hgher cell load, snce the cell needs to be unloaded more and more. For hgh traffc loads, the proporton of swtchable traffc needs to be hgh, that s, low values of β,nordertoapplysuffcent load swtchng for successfully unload the cells Dscusson on Swtchng Bounds. We have defned two bounds for the load swtchng, whch depend on the offered traffc, the load share, and on the parameters of the broadcast and UMTS network, respectvely. The broadcast bound s vald for the entre broadcast cell. Each UMTS cell defnes an ndvdual bound, dependng on the current loadng, the grade of servce, and the cell parameters. In order to successfully apply unloadng n a hybrd network the broadcast bound needs to be lower than the cellular bound, δ (B) δ (C). All UMTS cells, whch are covered by a broadcast cell and whch are consdered for a hybrd mode, need to be evaluated. Therefore the mnmum of all δ (C) has to be consdered, f the qualty target has to be reached. Nevertheless, strateges for network plannng and estmatng the approprate cellular bound are beyond the scope of the paper. If no broadcast bound s defned, the necessary broadcast data rate can be determned. The estmated swtchng parameter δ s based on the evaluaton of the UMTS cells. It defnes the necessary capacty of the broadcast network, whch can be estmated by the number of streams to be transmtted (). By summng up the requred data rates of all tems swtched to broadcast the necessary broadcast data rate can be calculated.

7 Internatonal Journal of Dgtal Multmeda Broadcastng Grade of Servce n Hybrd Networks. The grade of servce of a broadcast network can be defned n two ways, a contentor a user-based descrpton. The user-based defnton has been selected, due to compatblty reasons to the cellular network and for a common GoS estmaton of hybrd networks. It s estmated by evaluatng the probablty of the blocked streams. If δ<δ (B), the capacty of the broadcast network s not suffcent for transmttng the swtched content and users have to be blocked. The probablty of not served streams s ( P(X n δ ) P(X n max ) = ( δ) ) δ (B) (4) = δ (B) δ. Thus the grade of servce of the broadcast network s calculated by λ (B) = mn{, (δ (B) δ)}. It s obvous that for δ δ (B), no blockng occurs and λ (B) =. In order to estmate the grade of servce λ (H) for the hybrd network, the GoS parameters λ (C) and λ (B) have to be combned. Equaton (5) descrbes the rato of the traffc served by the hybrd network compared to the total offered traffc. Ths rato s transformed nto a weghted term; whereas the weghts descrbe the fracton of cellular and swtchable traffc: ( ) c (C) + λ (B) λ (H),δ = λ (C),δ + δc (H) c δ ( δ)c (H) = [ δ ( ) β + β ] λ (C),δ + ( δ)( ) β λ (B) δ. (5) For δ =, the total traffc has to be transmtted by the UMTS network, thus, λ (H) = λ (C). In the case that δ =, the UMTS network s unloaded by all swtchable traffc, but ths traffc s partly lost due to an overloaded broadcast network. The GoS of the broadcast network results n λ (B) = δ (B). The GoS of the UMTS network can be calculated by (6) usng c = c (C) and p (η) = p (U,η) Performance Analyss. A sngle cell s consdered wth fxed nterference from the surroundng cells. Ths smplfed scenaro s used n order to show the basc behavor of the network performance ndcators cell load (L (C) and L (B) )and grade of servce (λ (C), λ (B),andλ (H) ). Fgure 5 shows these performance ndcators. In ths example, the parameters of Secton 4.2 and the followng parameters have been used: D (B) = 2Mbps, d = 28 kbps, N tem =, κ =, c =.5, and β =.6. The used parameter values result n the bound values of δ (C) =.78 and δ (B) =.36. The cell load of the broadcast cell ncreases wth a decreasng δ untl the broadcast bound s reached. The nonunform steps are caused by the dfferent content popularty values. The UMTS cell load decreases whle unloadng beyond the cellular bound, caused by a lower necessary transmt power p. The grade of servce s shown for both network types and the hybrd combnaton. For hgh values of δ the curves of λ (C) and λ (H) are smlar due to a small contrbuton of the broadcast part wth a small amount of traffc swtched to the Cell load Grade of servce L (B) L (C).2 δ (B).4.6 δ (C) λ (C) λ (H) λ (B) (a).6.2 δ (B).4.6 δ (C).8 Swtchng rato δ (b) Fgure 5: Performance ndcators cell load and grade of servce n a hybrd network for a sngle cell, dependng on the swtchng rato δ. broadcast network. A sgnfcant gan of the GoS exsts, f the ntal case (λ (C) δ= ) s compared to compared to the hybrd network wth appled unloadng (λ (H) δ=δ ). (B) In the depcted case, a wde operatng range s shown for achevng suffcent GoS. From a network plannng pont of vew, applyng an δ<δ (B) s not preferable, snce the overall GoS λ (H) decreases very fast, due to the hgh nfluence of the overloaded broadcast network (see (5)). 5. Smulaton of a Realstc Scenaro In ths secton, the behavor of the performance ndcators GoS and the cell load s shown by means of a realstc scenaro. In the shown example, t s assumed that the broadcast cell covers the entre scenaro area. Ths can be assured by, for example, reusng a DVB-T transmtter for DVB-H, a real transmtter s avalable at the Berln scenaro. As mentoned before, the snapshot approach has been used n order to acheve accurate results of the performance ndcators. The snapshot method has been mplement, consderng ndoor and outdoor users, and sgnal shadowng wth spatal correlaton. We have performed a suffcently large number of snapshots and tested tconfdencenterval [9, Ch. 4] n order to acheve average values of the transmt power and the GoS, that are wthn a range of ±% around the expected value wth a 99% confdence. 5.. Smulaton Results. For the network scenaro, the parameters have been used as descrbed n Secton 4.4. The cells of the UMTS network are unloaded, f content s swtched from the UMTS network to the broadcast network. The grade of servce λ s ncreased, and the transmt power p s decreased sgnfcantly.

8 8 Internatonal Journal of Dgtal Multmeda Broadcastng Cumulatve probablty δ = δ = δ (B) δ = Grade of servce Fgure 6: Emprcal CDF for grade of servce dependng on the swtchng parameter δ. Dotted curves represent the cases wth the broadcast network n overload. Grade of servce λ (B) λ (C) λ (H).2 δ (B) Swtchng rato δ Fgure 8: Grade of servce of the entre network dependng on the amount of load swtchng..4 Fgure 7: Gan of grade of servce for unloadng the UMTS cells by δ = δ (B). Fgure 6 shows the emprcal cumulatve dstrbuton of the GoS of all cells n the network. Each curve represents a swtchng value for δ from zero to one n. steps and the broadcast bound δ (B) =.36. It can be seen that the GoS ncreases for lower values of δ. For the full-loaded network (δ = ) about 3% of the cells are above the GoS target λ =.98. By unloadng the cells wth δ = δ (B) =.36 about 55% and for δ = about 89% of the cells meet the target threshold. The dotted lnes represent the cases wth δ<δ (B), n whch the broadcast network s overloaded. In order to estmate, whch UMTS cells beneft from load swtchng, the GoS gan s defned by the rato of the unloaded cells (λ (C),δ=δ ) to the fully loaded case (λ (C) (B),δ= ). Fgure 7 shows the gan values, based on the best server cell structure. A hgh gan occurs for those cells, n whch a hgh traffc demand exsts (compare to Fgure ). For the cells wth a gan equal to one, unloadng has not been necessary due to alowtraffc load. A maxmum gan of about 75% ncrease n Gan of grade of servce per cell Fgure 9: Power gan per cell resultng from cell unloadng wth δ = δ (B). GoS can be seen n some selected cells, whch beneft most of usng a hybrd mode. We use (7) n order to calculate the grade of servce of the entre network. For network plannng the grade of servce should be λ = 98% or hgher. In Fgure 8 the performance of the hybrd network s shown. The GoS of the UMTS network λ (C) ncreases by unloadng but does not meet λ for an unloadng tll δ (B) (λ (C) δ=δ = 92.6%). The UMTS bound (B) δ (C) s not depcted snce for δ = agosofλ (C) = 97.9% < λ s acheved. Snce the δ (C) <δ (B) the maxmum total GoS occurs for δ = δ (B) wth λ (H) = 95.5%. If cells are unloaded and the GoS target s reached, a further unloadng leads to a reducton of the necessary transmt cell power. In Fgure 9, the power reducton gan s shown per cell. A maxmum reducton gan of.45 s acheved. It can be seen that cells wth hgh GoS gan do not show a power gan, snce the GoS target has not been reached and thus the power s stll at ts maxmum p max.the complementarty to the GoS can also be seen n ths case. Powerreductonganpercell

9 Internatonal Journal of Dgtal Multmeda Broadcastng 9 6. Concluson We have proposed a system and performance model for hybrd networks, whch are composed of the cellular network UMTS and the moble broadcast network DVB-H. The developed model s based on the exstng nterference couplng model of the UMTS network, whch has been enhanced by the load swtchng concept. Furthermore, a broadcast system model has been developed. The two network performance ndcators cell load and grade of servce (GoS) have been defned; whereas the GoS parameter s used as the prmary performance crtera for hybrd networks. Based on the developed models and the GoS defnton, the swtchng bound concept has been been proposed, whch defne bounds for the load swtchng for each network separately. They determne thresholds for the necessary unloadng of the UMTS network and for the maxmum loadng of the broadcast network n order to acheve the GoS target. These bounds are used to evaluate the condtons for reasonable load swtchng and defne the necessty and feasblty of hybrd networks. The mpact of load swtchng for the streamng multmeda servce has been analyzed by a theoretcal and a realstc scenaro. We have shown the behavor of the cell load, the grade of servce, and the approprate gan values on a cell bass and for the entre network. The results have shown that a hybrd network can be appled to unload the UMTS network and to serve the multmeda servces very effcently. References [] B. Rakocz, E. R. Fledderus, B. Hedeck, P. Lourenço, and T. Kürner, Reference scenaros, Tech. Rep. IST MOMENTUM, D5.2, IST, 23, paper/momentum-d52.pdf. [] Momentum Project, IST , Momentum publc UMTS plannng scenaros, 23, [2] T. Kürner, Propagaton models for macro-cells, n Dgtal Moble Rado Towards Future Generaton Systems (COST 23 Fnal Report), COST Telecom Secretarat, pp , CEC, Brussels, Belgum, 999, [3] M. Nawrock, H. Aghvam, and M. Dohler, Eds., Understandng UMTS Rado Network Modellng, Plannng and Automated Optmsaton: Theory and Practce, John Wley & Sons, New York, NY, USA, 26. [4] H.-F. Geerdes, UMTS rado network plannng: masterng cell couplng for capacty optmzaton, Ph.D. dssertaton, Veweg and Teubner, Wesbaden, Germany, 28. [5] A. Esenblätter, H.-F. Geerdes, T. Koch, A. Martn, and R. Wessäly, UMTS rado network evaluaton and optmzaton beyond snapshots, Mathematcal Methods of Operatons Research, vol. 63, no., pp. 29, 26. [6] J. Laho, A. Wacker, and T. Novosad, Rado Network Plannng and Optmsaton for UMTS, John Wley & Sons, New York, NY, USA, st edton, 22. [7] C. Heuck and P. Unger, Load balancng and network plannng for DVBH/UMTS hybrd networks, ITS-Europe 25, Hanover, Germany, June 25. [8] J. Aaltonen, Content dstrbuton usng wreless broadcast and multcast communcaton networks, Ph.D. thess, Tampere Unversty of Technology, Tampere, Fnland, 23. [9] A. M. Law and D. W. Kelton, Smulaton Modellng and Analyss, McGraw-Hll, New York, NY, USA, 2. [] U. Remers, DVB The Famly of Internatonal Standards for Dgtal Vdeo Broadcastng, Sprnger, Berln, Germany, 2nd edton, 25. [2] ETSI, Dgtal Vdeo Broadcastng (DVB); transmsson system for handheld termnals (DVB-H), Tech. Rep. ETSI Std. EN V.., ETSI, Sopha Antpols, France, 24. [3] G. Fara, J. A. Henrksson, E. Stare, and P. Talmola, DVB-H: dgtal broadcast servces to handheld devces, Proceedngs of the IEEE, vol. 94, no., pp , 26. [4] G. May, Loss-free handover for IP datacast over DVB-H networks, n Proceedngs of the Internatonal Symposum on Consumer Electroncs (ISCE 5), pp , June 25. [5] G. May and P. Unger, A new approach for transmttng localzed content wthn dgtal sngle frequency broadcast networks, IEEE Transactons on Broadcastng, vol. 4, no. 53, pp , 27. [6] ETSI, Dgtal Vdeo Broadcastng (DVB); IP datacast over DVB-H: archtecture, Tech. Rep. TR V.., ETSI, Sopha Antpols, France, May 26. [7] M. Kornfeld and G. May, DVB-H and IP datacast broadcast to handheld devces, IEEE Transactons on Broadcastng, vol. 53, no., pp. 6 7, 27. [8] C. Heuck, Optmerung hybrder (Rundfunk-/Moblfunk-) Netze durch Steuerung der Lastvertelung, Ph.D. dssertaton, Insttut für Nachrchtentechnk, TU Braunschweg, Braunschweg, Germany, 29. [9] L. Ferrera, L. M. Correa, D. Xaver, I. Vasconcelos, and E. R. Fledderus, Fnal report on traffc estmaton and servces charactersaton, Tech. Rep. IST MOMENTUM, D.4, IST, 23, paper/momentum-d4.pdf.

10 Internatonal Journal of Rotatng Machnery Engneerng Journal of Volume 24 The Scentfc World Journal Volume 24 Internatonal Journal of Dstrbuted Sensor Networks Journal of Sensors Volume 24 Volume 24 Volume 24 Journal of Control Scence and Engneerng Advances n Cvl Engneerng Volume 24 Volume 24 Submt your manuscrpts at Journal of Journal of Electrcal and Computer Engneerng Robotcs Volume 24 Volume 24 VLSI Desgn Advances n OptoElectroncs Internatonal Journal of Navgaton and Observaton Volume Chemcal Engneerng Volume 24 Volume 24 Actve and Passve Electronc Components Antennas and Propagaton Aerospace Engneerng Volume 24 Volume 2 Volume 24 Internatonal Journal of Internatonal Journal of Internatonal Journal of Modellng & Smulaton n Engneerng Volume 24 Volume 24 Shock and Vbraton Volume 24 Advances n Acoustcs and Vbraton Volume 24