Analytical Pefomance Evaluation of Mixed Sevices with Vaiable Data Rates fo the Uplink of UMTS Laissa Popova and Wolfgang Koch Lehstuhl fü Mobilkommunikation, Univesität Elangen-Nünbeg, Gemany Email: {popova, koch}@lnt.de Abstact The contibution of this pape is twofold. Fist, we intoduce a new efficient analytical method fo uplink pefomance estimation in multi-sevice UMTS netwoks by taking the dynamic behaviou of the use tansmission ate into account. Ou pefomance measues ae the aveage use infomation ate and the blocking pobability of new calls. Next, a taffic management technique is applied to UMTS and incopoated into the analytical appoach to specify how esouces have to be adequately engineeed to meet quantitative pefomance objectives. Fo this we investigate the joint behaviou of the admission contol chaacteistics, scheduling, theshold and esevation scheme fo UMTS Teestial Radio Access Netwok (UTRAN) in tems of thei ageement with Gade of Sevice (GoS) equiements. The pefomance of the poposed algoithms is evaluated. I. INTRODUCTION As the capacities and flexibility of communication netwoks incease, and the amount of sevices, they povide, expand, the demand fo adio esouces becomes less homogeneous. In ode to povide sevice in this heteogeneous envionment moe efficiently, it is desiable to be able to estimate the blocking pobability fo all the specific classes of demands. Taffic theoy cuently plays a quite mino ole in the design of thid geneation cellula mobile netwoks. Netwok dimensioning of UMTS systems is geneally based on ough estimates athe than on ealistic analytical models fo the expected taffic behaviou. Consideable effot is spent on the design of a vaiety of quality of sevice (QoS) mechanisms. The ole of the latte is to ensue an appopiate level of sevice of diffeent use classes in UMTS netwoks. Nevetheless, the necessity of applying the appopiate taffic-pefomance elation, if the objective is to ensue that QoS meets its tagets fo a given population of specified kind of uses, is still emaining. The pefomance evaluation of multi-sevice wieless netwoks is a complex poblem, because the non-poison taffic chaacteistics, dynamic behaviou of use tansmission ate and the effect of the adio channel have to be consideed. Futhemoe, multi-band WCDMA allows many uses to tansmit simultaneously within the same fequency band. Fom a single-use pespective, all othe use s signals, located in the ai-inteface, ae noise. Evey time a new call is accepted, the intefeence level inceases, consequently, deceases signal-tonoise atios fo all othe uses. In ode to peseve the quality of aleady accepted calls, a new call should not be accepted by the system if this call inceases the noise fo any othe existing call above the toleable level. If the capacity is limited by the amount of intefeence and noise in the adio inteface, it is by definition a soft capacity, since thee is no single fixed value fo the maximal netwok capacity. Thus, the UMTS system has no had limits on call capacity, which makes the estimation of netwok capacity much moe complex. Since the actual load of the existing UMTS netwoks is still too low to anticipate the eally busy hou taffic values, thee ae two altenative ways to estimate the pefomance of 3G mobile netwoks: the analytical modelling and the simulation of the functionality of UTRAN. In the following, a bief oveview of the-state-of-the-at of taffic engineeing of UMTS netwoks is pesented. In 1981 Kaufman and Robets published ecusive equations which can be used fo computing the blocking pobabilities fo an Elang Multiate Loss Model [3]. In [5] and late [2] state dependent blocking pobabilities fo calculation of the pefomance of cellula systems with WCDMA adio inteface was intoduced. In [5], [2] the othe-cell intefeence is modelled as a lognomal andom vaiable and the blocking pobabilities ae computed using an appopiately modified Kaufman-Robets ecusion. In [6] an uplink UTRAN pefomance was estimated. To detemine the blocking pobability of a new call appeaing in the cell k the load geneated by the call in the neighbouing cells was taken into consideation. Howeve, in above mentioned studies, the specifics of such impotant featues of UMTS as vaiable use tansmission ate have not been included into the models so fa. Methodologies to analyze pefomance of UTRAN by the use of suitable simulation tools have been intoduced in [1]. Although simulation tools poduce good esults they need many input paametes and a huge amount of computation time. This pape pesents a new appoximation method fo pefomance estimation of multi-sevice UMTS netwoks, taking into account key featues of UTRAN, such as the intefeence 1-4244-0398-7/06/$20.00 2006 IEEE
limited soft capacity of a cell and vaiable use s tansmission ate. Ou algoithm is new fo its analytical consideation of two quite diffeent aspects, which ae associated with QoS in tems of acceptable intefeence level and aveage use infomation ate at the one hand, and blocking pobability of new calls at the othe hand. Examples fo evaluation of ealistic cases ae given. In contast to the widespead view, that analytical appoaches fo modelling a adio inteface ae too simplified and yield inaccuate esults, the analytical method poposed in this pape demonstates the flexibility and the accuacy of this appoach. Of couse, some degee of simplification is necessay. Fo this the dominant aspects of the eal quantities wee identified and epesented in the selected model and unimpotant aspects wee neglected. The pape is oganized as follows. In Section II some basic elations egading admission contol in UTRAN ae given and the system model is intoduced. In Section III the poposed method fo blocking pobability calculation fo UTRAN is pesented. Examples of incopoation of taffic management stategy into the model ae pesented fo ealistic UMTS scenaio in Section IV. The effects of this stategy on the call blocking pobability ae evaluated. Finally, Section V concludes the pape. II. SYSTEM MODEL A. Some basic elations In a UMTS system the numbe of effective channels is obtained in tems of the numbe of available codes, the pocessing gain, and the equied E b /N 0 (E b : aveage eceived bit enegy, N 0 : powe spectal density of white Gaussian noise). Accuate signal eception of diffeent taffic steam types is possible only when the elation of enegy pe bit to noise spectal density is appopiate. The E b /N 0 fo the jth use is calculated as : (E b /N 0 ) j = W P j (1) R j ν j I total P j whee P j is eceived signal powe fom the jth, W -chipate of speading signal, v j - activity facto of the jth use, R j - bit ate of the jth use, I total - total eceived wideband powe including themal noise powe. Befoe admitting a new connection in the system, connection admission contol ensues that a new call will not sacifice the quality of the existing connections. The admission of a new connection of a specific sevice class k in the consideed cell is a andom vaiable depending on the cuent state of the cell, the noise-ise of the new connection and the othe-cell intefeence. B. Model chaacteistics and assumptions The system unde consideation is a cellula UMTS netwok. We assume that the oveall system is homogeneous in statistical equilibium. Fo such a system any cell is statistically identical to any othe cell. The total capacity of each cell is N channels. The definition of a channel used in ou model is given in [2]. The following paametes and assumptions ae used: The netwok seves n independent classes of Poisson distibuted taffic steams with aival ates: λ 1, λ 2...,λ n. The sevice time fo the calls of paticula sevices classes has exponential distibution with paametes: µ 1 1, µ 1 2,...,µ 1 n, espectively. Each call is assigned one o multiple channels depending upon the souce taffic. Hee, a channel is defined by a data ate of 12.2 kbps. Any sevice class k is chaacteized by a data ate of d k = m k 12.2 kbps (m k is positive intege). The mean taffic offeed to the system by sevice class k is a k = λ k /µ k. Calls of diffeent sevice classes have diffeent esouce equiements (use tansmission ate and sevice time), which depend on the total numbe of aleady occupied channels, i.e. a new call of the sevice k can be admitted in the system with othe that initial tansmission ate. The sevice pocesses in the cells ae dependent, and an actual taffic load in one cell influence the sevice pobability in othe cells. The objective is to detemine the call blocking pobability and aveage use tansmission ate fo each sevice class. III. PROPOSED ALGORITHM In ode to detemine the call blocking pobability and aveage use tansmission ate fo each sevice class, we evisit the analysis fom [2], whee the following ecuent fomula is poposed fo estimation of the uplink pefomance of UTRAN: G() = 1 n a k m k G( m k )(1 b mk,k), (2) k=1 whee G() is the pobability of busy channels in the system and b is the state dependent blocking pobability (soft blocking), which chaacteizes the cuent othe-cell intefeence situation. We appopiately extend this fomula to study a ealistic UTRAN envionment. Accodingly, let us conside a pat of the two-dimensional Makov chain diagam constucted fo a system suppoted two taffic classes (Fig. 1). This should be undestood as follows. A new call of sevice class k with esouce equiement d k is only accepted if all channels equied to seve this call ae available. Additionally, we assume, that in UTRAN multiple channels equied to establish a new connection ae occupied simultaneously and theefoe the othe-cell intefeence emains the same duing this allocation pocess fo all channels. Then, with the above mentioned assumption, the sevice pobability fo a new call with esouce equiement d k is defined as: 1 b,k =P{ <N <N ( +1) <N ( +2)... <N ( + m k 1)} (3) =P{ <N ( + m k 1)}
(1 b (i+j+1)+mk 1)λ 1 i, j +1 i + m k,j+1 i+mk mk µ 1 (1 b i+j )λ 2 (j +1)µ 2 (1 b i+mk+j)λ 2 (j +1)µ 2 (1 b (i+j)+mk 1)λ 1 i, j i + m k,j i+mk mk µ 1 Fig. 1. A fagment of state-tansition diagam of Makov chain in a multisevice UMTS system with a soft blocking. whee and is the intefeence fom othe cell expessed as channel equivalent. Futhemoe, to incopoate a tansmission ate eduction policy (TRRP) into ou model, we employ a modified analytical appoach fo estimating the blocking pobability in Asynchonous Tansfe Mode (ATM) netwoks poposed in [4]. In this so-called theshold model fo blocking avoidance new calls of cetain sevice classes can get a connection with othe than the initial esouce and sevice time equests, which ae state dependent. Afte some modifications to the fist sum and the incopoation into the expession of the second sum we get: G() = 1 n a k m k δ k ()G( m k )+ k=1 1 n k=1 l=1 s a kl m kl δ kl ()G( m kl ) (4) fo =1,...,N,wheeδ k () =1when1 N o, when J k1 + m k,othewiseδ kl =1; l is a numbe of a possible vaiation of tansmission ate, J kl is the adjusted theshold(s). Fo calculation of the total blocking pobability fo a sevice while taking into account TRRP we modify popely the limits of the summations in the following expession: B k = N m kl =0 G 1 G()b,k, (5) whee G = N G(). This is the pobability of a call of sevice =0 class k to be blocked with its lowest possible tansmission ate, with taking into account the othe-cell intefeence. IV. PERFORMANCE EVALUATION ANALYSIS Some numeical examples ae pesented in this section in ode: to demonstate the flexibility and the accuacy of the analytical method used to evaluate the teletaffic pefomance of the UMTS systems. to study how vaious paametes affect the pefomance metics of inteest. to quantify the benefits of the poposed taffic management stategy by calculating the maximum numbe of uses that a UMTS cellula netwok can handle fo a given GoS unde cetain system conditions, such as netwok topology, elevant cell scenaio, ealistic use/taffic distibution pofile, etc. We conside the UMTS netwok, suppoting two taffic classes: speech calls with 12.2 kbps and video steam with 128 kbps and taget E b /N 0 of 4 and 1.5, espectively. The maximum numbe of channels in one cell is 315, the othecell intefeence facto i =0.55 and the offeed taffic is ρ=35 Elang. We model the othe-cell intefeence as a lognomal andom vaiable as poposed in [5]. Unde ealistic assumptions, the mean othe-cell intefeence powe is about E{ I othe 2 } =5 10 15 mw with a vaiation coefficient of 1, coesponding to medium taffic load. This has been veified by using a UMTS adio netwok simulato. The most impotant paametes and equiements fo diffeent sevice-classes of the system consideed hee ae summaized in Table 1. Type of the sevice k Speech Data Data ate, d k [kbps] 12.2 128 E b /N 0 [db] 4 1.5 Coesponding numbe of equied channels, m k 1 10 Chip ate of speading signal [Mchip/s] 3.84 Maximum numbe of channels/cell 315 Table 1: Main simulation paametes. It is evident that in systems with a complete shaing policy the sevice classes with low tansmission ates pofit fom a lowe blocking pobability in compaison to high ate sevice classes. A vey undesiable situation is, howeve, when the blocking pobability of high ate sevice classes exceeds the GoS constaint, while the blocking pobabilities of othe classes ae much lowe than the pescibed constaints. In the following we examine the effect of applying diffeent admission citeia to diffeent taffic steams to give satisfactoy pefomance in tems of both, blocking pobability and satisfactoy use tansmission ate fo each taffic class. Specifically, blocking pobability fo the speech sevice class and blocking pobability fo the data sevice class ae expected to be in the ange of 1-2% and 10-11%, espectively. Fo this, we assume that speech is a constant-bit-ate sevice, while data calls allow adaptive data ates, depending on the load of the system. Each new use of the data class has a esouce equiement d k =128 kbps that will be satisfied if J kl, (in ou case l=1) and a educed ate d kl =64 kbps, othewise. By applying the TRRP into the system unde consideation, it is necessay to bea in mind, that loweing the tansmission ates has two conflicting impacts. On the one hand, less esouces ae needed at a given time fo handling the connections
whose tansmission ates wee educed. On the othe hand (and in contast to eal time connections), the tansmission duation of such calls becomes longe, since the amount of infomation to be tansmitted does not depend on the tansmission ate, which leads to use dissatisfaction. Thus, a tade-off between system pefomance and use satisfaction must be found. The Fig. 2 demonstates the system pefomance vesus the theshold fo a sevice-mix scenaio with 70% speech calls and 30% data calls. If no theshold is applied, the blocking pobability fo the algoithm of [2] without TRRP is obtained (ight points of the cuves). This is the wost case utilization. Fom the figues one can see that setting the theshold to a low value leads to a decease of sevice pobability of speech calls and to unnecessay degadation of tansmission ate fo the data sevice class, wheeas a gain in tem of sevice pobability fo these calls is not so ponounced. In tun, the oveestimation of the theshold leads to a negative impact on the blocking pobability of data calls. The optimal theshold fo above mentioned scenaio is J=200, coesponding to a tansmission ate of a data call of about 77 kbps. In Table 2 Blocking pobability fo speech (70%) and video steaming (30%) 0.3 speech data 0.25 Blocking Pobability [%] 0.2 0.15 0.1 0.05 Data ate fo Video Steam [kbps] 0 0 25 50 75 100 125 150 175 200 225 250 275 300 325 Theshold [numbe of channels] Data ate fo Video Steam (N=315, ρ= 35 Elang) 130 125 120 115 110 105 100 95 90 85 80 75 70 65 60 0 25 50 75 100 125 150 175 200 225 250 275 300 325 Theshold [numbe of channels] Fig. 2. Blocking pobability and data thoughput fo speech and video steam vesus the theshold. we compae the pefomance of the algoithm of [2] with that of ou algoithm. The esults indicate the significant system pefomance impovement which can be attained by TRRP. With incopoation of a esevation scheme in ou algoithm we can fulfill the specified GoS demands. It is clea, that a deceased blocking pobability of data calls is achieved at the expense of some degadation of sevice pobability fo speech taffic; due to the esevation data calls get moe access to the available esouces. The maximum numbe of uses a UMTS cell can handle using the above mentioned taffic management stategy is about 1075 fo speech and 462 fo data sevice. Diffeent B k, % system scenaios Speech Video Steam Without TRRP 2.27 27.13 With TRRP 2.27 15.16 With TRRP and channel esevation 2.83 11.45 Table 2: System pefomance with/without TRRP and esevation. V. CONCLUSIONS AND FURTHER WORK In this pape we have developed a fast and accuate ecusion fo efficient computation of GoS fo a multi-sevice scenaio in the UMTS system, by taking into account the dynamic behaviou of use tansmission ate. The algoithm pesented in this pape is not only useful fo the pefomance pediction of the UMTS systems, but also fo obtaining a suitable admission contol policy fo optimal usage of the system esouces. The following conclusions can be dawn fom the numeical esults: The blocking pobability of a new call depends both on the state of the system and the esouce equied. The incease of aival ate of data sevice classes inceases the blocking pobability fo all sevice classes in the system. Incoect taffic diffeentiation in tem of adjusted theshold esults in unacceptable call blocking pobability fo all sevice-classes o leads to dissatisfaction of aleady accepted uses (too long tansmission duation of the calls with educed tansmission ate). In case of a popely selected admission policy we obseved a educed data call loss fom about 28% fo the no-theshold case to about 11%. The algoithm allows to study pefomance of UMTS systems in thei natual envionments with appopiate adio inteface chaacteistics and ealistic taffic volumes. Although ou esults have been pesented fo a two-class system, with ou algoithm we can easily extend ou model fo moe sevice classes, as well as fo multiple admission thesholds. The poposed method enables to analyse system pefomance unde consideation of othe impotant UTRAN featues, such as channel esevation, genealization of taffic aival chaacteistics, handove, as well as packet-sevice delay, which is unde cuent investigation. REFERENCES [1] N. Hassanein, A. Olive, N. Nasse, and E. Elmallah. Uplink QoS-Awae Admission Contol in WCDMA Netwoks with Class-Based Shaing. In Poc. of QSHINE, Oct. 2004. [2] V. Ivesen, V. Benetis, N. Ha, and S. Stepanov. Evaluation of Multisevice CDMA Netwoks with Soft Blocking. In Poc. of ITC 16, Spec. Sem., Antwep. Belgium, Sept. 2004.
[3] J. Kaufman. Blocking in a Shaed Resouce Envionment. IEEE Tans.Commun., 29, 1981. [4] I. Moscholios, M. Logothetis, and G. Kokkinakis. Connection-dependent theshold model. An Inten. Jounal Pefomance Evaluation, 48, 2002. [5] D. Staehle and A. Maede. An analytic appoximation of the uplink capacity in a UMTS netwok with heteogeneous taffic. In Poc. of ITC 18, Belin, Aug./Sept. 2003. [6] M. Stasiak, A. Wisniewski, and P. Zwiezykowski. Blocking pobability calculation in the uplink diection fo cellula systems with WCDMA adio inteface. In Poc. of MMB, Desden, Septembe 2004.