PERFORMANCE EVALAUTION OF UPLINK CAC FOR 3G WCDMA WIRELESSNETWORKS WITH MULTISERVICES

Salman A. AlQahtani, AshrafS. Mahmoud, andAsrar U. Sheikh

King Fahd University of Petroleum & Minerals, Dhahran 31261, Saudi Arabia{salmanq, ashraf, asrarhaq}@kfupm.edu.sa

ABSTRACT

The wide-band code division multiple access(WCDMA) based 3G and beyond cellular mobile wirelessnetworks are expected to provide a diverse range ofmultimedia services to mobile users with guaranteedquality of service (QoS). Call admission control is a veryimportant measure in WCDMA system to guarantee thequality of the communicating links. Two throughput-based admission control strategy with multi-services,referred to herein as the complete partitioning CAC (CP-CAC) and the queuing priority CAC (QP-CAC), areanalyzed and compared. The main contribution of thispaper is the development of an analytical model for theQP-CAC algorithm which can be easily extended andused for CP-CAC. We also develop a simulation tool totest and verify our results. Finally, we present numericalexamples to demonstrate the performance of the proposedCAC algorithms and we show that analytical andsimulation results are in total agreement.

1. INTRODUCTION

3G and beyond wideband code-division multiple access(WCDMA) wireless networks are expected to provide adiverse range of multimedia services to mobile users withdifferent quality of service (QoS). The Universal MobileTelecommunication System (UMTS), a WCDMA basedsystem, is required to support a wide range of applicationseach with its own specific QoS requirements. Four distinctQoS classes are specified: namely, conversational,streaming, interactive and background [1]. Each class hasits own QoS specifications such as delay and bit error rate(BER). One of the main challenges in 3G and beyondwireless networks is to guarantee QoS requirements whiletaking into account radio resource limitations. Calladmission control is one method to manage radioresources while optimizing the overall networkperformance [1-3].The main contributions of this paper can be stated asfollows. We consider a throughput-based CAC where therelative load estimate as in [ 1 ] can be used for calladmission decision then we extend two uplink CACalgorithms for operation in 3G WCDMA networks. These

adapted CAC schemes are the complete partitioning CAC(CP-CAC) and the queuing priority CAC (QP-CAC). InCP-CAC, each call class has its own queue and resourcepartition whereas in QP-CAC, each call class has its ownqueue and all classes share the available resources. Themain contribution of this paper is the development of ananalytical model for the QP-CACAC algorithm specifiedin this study which can be easily used for CP-CAC. Wealso develop a simulation tool to test and verify ourresults.The rest of this paper is organized as follows. Section 2describes the system model. Section 3 explains theproposed schemes in details while section 4 presents theperformance analysis. Section 5 presents the obtainedresults, as well as the discussion. Finally, the paper isconcluded in section 6.

2. SYSTEM MODEL

The system under consideration is a 3G and beyondWCDMA cellular network supporting heterogeneoustraffic. We only consider a system with homogenous cells.Thus, an equal cell load, same traffic patterns, andsymmetric directions of handoff calls are assumed foreach cell. The capacity of 3G WCDMA cell is defined interms of the cell load where the load factor, 71, is theinstantaneous resource utilization upper bounded by themaximum cell capacity, 77max Instantaneous values forthe cell load ij range from 0 to 1. We assume two types oftraffic 1) Real time traffic (RT) ( such as voice) which isdelay sensitive 2) Non-Real time traffic (NRT) ( such asWWW, FTP) which is non-delay sensitive. Each type hastwo call classes 1) handoff calls and 2) New calls. Also,each traffic type has its own queue: QI and Q2, with finitecapacities M, and N respectively. The requests for bothqueues are served based on first-in-first-out (FIFO)manner when channels are free. A call class request isplaced in its corresponding queue if it cannot be servicedupon its arrival and assigned a resource when availablebased on its calculated priority. Therefore, the priority isdivided into two classes 1) RT traffic (including new andhandoff calls) and 2) NRT traffic (including new andhandoff calls).

1-4244-0411-8/06/$20.00 C 2006 IEEE.

Depending on how this loading limits is controlled andhow priority of among queues are sets, we devise QoS-aware CAC algorithms for WCDMA based networkswhere we have the following CAC schemes: 1) thequeuing priority CAC (QP-CAC), and 2) the completepartitioning CAC (CP-CAC) with isolated load partitionassigned for each class. These CAC schemes are depictedpictorially in Fig. 1, Fig. 2, and Fig. 3, respectively. Theoperation of these CAC schemes is detailed in the nextsection.

Total Cell LoadHandoff RT Highest Pnority "max

Anl l T ou RT CompletionNew Calls RT

Handoff NRT / : 2Qh22 0 ......... FNRT Completion

h2

anr Time out \ \ ......New Calls NT\\:-:::

Lotwest Pnority

Figure 1: QP-CAC scheme Model

Handoff RT Tota

Anl -1-l Time out

New Calls RT

Handoff NRT

ih12 02

An2 ,, j// o2 Time out

New Calls NRT

Figure 2: CP-CAC scheme Model

a Cell Load

LIP

Lp1

2......

RT Completion

NRT Completion

A/k = (1 + f)I Vkl+ Gklek

(1)

where Gk = W/Rk is the processing gain for the kth MS,Ri is the bit rate associated with the kth MS, and W is thechip rate of the WCDMA system. ek is the bit-energy tonoise-density (Eb/NO) figure corresponding to the desiredlink quality. f is the factor accounting for interferencefrom other cells and is defined as the ratio of inter-cellinterference to the total interference in the referenced cell,whereas vi is the average traffic activity factor of kth MS.Using the load factor increment definition, the currenttotal load, 7C/. for such an interference system is the sumof the load factor increments brought by all B active MSs.Therefore,

C/c = k=Al/k ( Vk=1 + G ./ < max (2)

In our systems we will have two different load incrementfactors for RT and NRT connections. By including thesetwo factors, we will have two types of load increments inour system, namely;

A77, = (1I+ f)) VI1) for RT traffic

2) for NRT traffic A52 = (1 +P V2+

Hence, the WCDMA capacity bound supporting RT andNRT traffic in uplink direction shown in (2) is re-expressed as;A77i + A/2 j < /max (3)Where i andj are the number of connected calls ofRT andNRT, respectively.

3.1. QP-CAC Schemes

3. CALL ADMISSION CONTROL SCHEHMES

The developed algorithms attempt to manage resource

allocations amongst the different call classes, and toefficiently utilize the resources while satisfying the QoSrequirements. Similar to studies [3-7], only the uplinkdirection is considered in this study where it is assumedthat whenever the uplink channel is assigned the downlinkis established. In addition, the study assumes perfectpower control operation where a mobile station (MS) andits home base station (BS) use only the minimum neededpower in order to achieve the required performance. Theadmission control for WCDMA systems first estimates thetotal current cell loads and the new load increment andthen employs them in the decision process of accepting or

rejecting new connections. Considering the load on theuplink, the load factor increment Al/k for a new request kcan be estimated as;

The QP-CAC scheme's model is shown in Fig. 1. In QP-CAC, each call class has its own queue and all classesshare the available resources. However when all resourcesare occupied, the prioritization is implemented using onlythe queuing techniques where the queued calls with higherpriority (i.e., RT in our case) are served first. That is, thequeued NRT request will be served only when the queueofRT requests is empty. Thus, this scheme guarantees themaximum system utilization and improve the performanceofRT calls but severely degrades the performance ofNRTcalls. When a call request i arrives, it will be admitted bythe system if the following criteria is satisfied:

Aqi + .1c-max (4)Any call class is deleted from its queue if it exceeds thequeuing time limit.

3.2. CP-CAC Scheme

In CP-CAC, each call class has its own queue andresource partition as shown in Fig. 2. The CP-CAC can

guarantee the resource commitment, and therefore theQoS, to each traffic class, but may underutilize theresources. This scheme is explained as follows. The cellload is divided into two non overlapping load partitions,Li (i= 1, 2), such that

22 i=lLi < 7max (5)

The capacity size or load partition, Li, of each division isselected based on the traffic characteristics and thepredefined QoS requirements of each service class [2].The total current usage load occupied by each connectedcalls of class i, Oi, is defined as,

1, if i+sj=C-s0, otherwise

I 1,ifi+sj=C-s+l{O, otherwise

1,if i + sj <C-s{0, otherwise

Ix {O,hifx > Ox 0, otherwise

I fl,ifi+sj=C2 otherwise

1ifi+sj<C-s, otherwise

I = l,if i+sj<C6 0O otherwise

k s- 1,if i > s- 1li, otherwise

(6)

Where Bi is the number of currently connected class i call.The arrived class i call is admitted if and only if thefollowing criterion is satisfied:

Aqi+O 1.L (7)Any call is deleted from its queue if it exceeds the queuingtime limit.

4. PERFORMANCE ANALYSIS

In this section we present the analytical results for QP-CAC and show how it can be adapted for CP-CAC. Weassume the arrival processes of new and handoff calls,belonging to each type, are Poisson occurring with rates,

'hl Ah2' n1 An2' for RT handoff, NRT handoff calls,RT new calls and NRT new calls, respectively. The totalarrival rate of the system is Poisson with rateA = Ahl + 2h2 + 'nl + 2n2 * The total arrival rate of RTand NRT calls are l= Ahl + Anl and '2 = -Zh2 + -In2 ,

respectively. The channel holding time for RT and NRTcalls is exponentially distributed with mean u, 1 andu2-,respectively. The queuing time limit is exponentiallydistributed with mean al,' and a2-' for RT and NRTrespectively.Based on (1) and (3), and since NRT (such as FTP or

WWW) mostly has higher bit rate and thus loadincrements than RT ( such as voice), we assume that theload needed by one call of type NRT is equivalent to thetotal loads required by s calls oftype RT, where the factors is defined as:

S A772j Such that A772 .A71

Then we can rewrite (3) as

i+sj< LlmaxA

We also consider the following indicators:

(10)

If we define C _/max such that C is the total numberLA7 jof basic channels within a cell equivalent to C RTconnected calls such that C * A771 < "max ' then (10) canbe also rewritten as

i+sj.C (11)In our system analysis , the RT and NRT calls can havedifferent bit-rate, channel holding time , time out, andEb/No requirements. This scheme is analyzed using fourdimensional Markov chain. Let (i, j, m, n) be a vectorrepresenting the system state where the state variables iand j denote the number of RT and NRT calls in thesystem, and m and n indicate the number of RT and NRTcalls in the respective queues. The state space of thescheme can be expressed as5 (i, j,m,n)I i0, j.0,i+sj<C, (12)

U0 < m < Ml O < n < N, and s <MMSumming all the probabilities of the states reaches

<Pi,j,m,n = 1 (13)(i,j,m,n)e SUsing the above indicators, we will consider six cases toexploit the relationship among state probabilities. All non-feasible states (states that do not included in the systemstate) should be set to zero.

when 0 <= i+sj<C, m=0,n=0(iy1 + A1 + j1/2 + A2 )pi, j,0,0= IiAlPi 1,j,O,O + (i + 0jUIPi+I,j,o,o+ IjI522Pi, j-I,,0 + (I + 1)f2pi,j+1,o,o+ IN(II2 + of )pi,j,O,l

k+ IIis_ (j + 1),U2 Pi-l, j+1,,0

1=1

oi = (I+ f)yBj vi <- Lii=l I + Gi lpi

When C-s<i+sj<C, m=O, O<n<=N

(iyl +21 + jt2 + noa 2 + IN-nA2 )Pi,j,O,n= i2Pi,j,O,n - + Iiu1Pi-l,j,O,n+ I'Nn (12 + (n + 1)a2 )Pi,j,O,n+l+ (i + l) IPi+l,j,O,n

k

+ IM Ii's_ (j + 1)/12 Pi-l,j+l,l,n/1=

When i+sj=C, m=O ,n=O

(ijly + IMA1 +jI2 + INA2 )Pi,j,,oI-Alpi-,,, + Im(igl + al)Pi,j,,,O=Ij1-1,j,O,O +'(Y

+ ((1- Is-1) + Is-13)}I'N(O + )Plpi+l,j-1,0,1+ Ij2Pi,j-11,0,0 + IN[1U2 + °2 ]Pi,j,o,l+ I'IMIM-s+l(j + 1)Y2Pi-s,j+1,s,O

When i+sj=C, m=0,0 <n <= N

(I + 'Ml + j1/2 + NIN-n2)Pi,j,O,n= >22Pi,j,O,n -1 + IM (ip/1 + a1 )Pi,X,l,+ (I1 I-Is_)IN_nIj 0 + ')PtlPi+l,j-l,O,n+l+ I'Nn U(12 + (n + I)a2)Pi,j,O,n+l+ Iiimim-s+m(j +<Ml2,pi-s,j+l,s,n

When i+sj=C, O < m<= M, n=O

(i,ul + ma, + IM m'm +j12 + INA2)pi,j,m,O= UPiPj,m -1,0 + I'M m(ip/ + (m + I)Da)Pi,j,m+i,o+ I 2Pi, j,m,1 + IiIM m-s+I(j + 1)/12 i-s,j+l,m+s,O

When i+sj=C, 0 < m<= M,O <n <= N(i/ul+ m al + IM_Ai+ n a2 + jI/2 + IN-nA2)pi,j,m,n= >IPi,j,m -,n + L2Pi,j,m,n-1 + IM-m (ii/l+ (m + 1)al)Pi,j,m+ln + I'Nn(n + 1)a2pi,j,m,n+l+ IiIM m_s+l (j + l)/2Pi-s, j+l,m+s,n

From the balance equations, steady state can be resolved.The total blocking probability of class i in the system is

Where BP, and TPI are the blocking and time outprobabilities, respectively and defined as

BPIj- Pi,j,m,n (15)V(i,j,m,n)e Q,

E Qi aiPi,j,m,nTP V(i,j,m,n)e Q2 (16)

Z ARi (1 - BPi )Where Q = {all (i,j,m,n) such that Qi is full) and Q2 = {all(ij,m,n) such that Qi > 0). In case of CP-CAC, if we setthe parameters of one class to zero, we can get theperformance of others.

5. NUMERICAL RESULTS

The Bit rates (Kbps) , Eb/No and vi are 12 Kbps, 5dBand 0.4 for RT and 256 Kbps, 2dB and 1.0 for NRT,respectively. The arrival rates are Ahl=0.1, Ah2 =0.2,

Anln=0.3, An2 =0.4 call/sec. the queue size are 5 for bothclasses. The cell load is 0.8 and the call duration is 100sec and the queuing time is 5 sec. In Fig. 3, the totalblocking probability for RT services using simulation andanalytical model is shown. As seen the two results arematched.The system utilization provided by each scheme as afunction of the offered traffic load of all classes isdepicted in Fig. 4. One can note that utilization of QP-CAC is higher than CP-CAC. On the other hand, asexpected CP-CAC has the lowest system utilization. Thehigher resource utilization of QP-CAC is due to anincrease of statistical multiplexing gain which is not existsin CP-CAC.In Fig. 5, the resources that are utilized individually byNRT traffics using QP-CAC and CP-CAC are depicted.We assume that the RT traffics constitute 70% of the totaltraffic and the NRT traffics constitute 30% of the totaltraffic. At high system loads, the resources that areutilized by NRT in case of QP-CAC are very smallbecause they are overwhelmed by the high traffic of RTwhile in case of CP-CAC there is no effect. So, CP-CACprovides resource protection at high traffic conditions.

TBPi = BPi + (1 - BPi)TPi (14)

TBP-RT0.95

Analt.0.9 Sim.

~I0.85

~~I~~ ~~I0.8 - I- L

/1 1~II~I0.75 - - - 4 - - I- - - - -

0.7

0.5---/ - --1-- ---- r-- - ----- -- ----0.6

0.55

0.50.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Figure 3: the total blocking probability ofRT

0.8

n r0

N

D 0.4

0.2

0

0 5 10 15 20Offered traffic (Erl.)

Figure 4: Total system utilization

0.71

25 30

+ QP-CAC-NRT-< CP-CAC-NRT0.6

0

N 0.5

,) 0.410(I)

I 0.3

0.2

z

0. 1

0 20 40 60 80Offered traffic (Erl.)

Figure 3: Total NRT resource utilization

100 120

6. CONCLUSION

Call admission control is a very important measure inWCDMA system to guarantee the quality of thecommunicating links. In future wireless networksmultimedia traffic have different QoS requirements. twothroughput-based admission control strategy with multi-services are analyzed and compared in this paper. A multi-dimensional Markov model is used to evaluate the systemperformance with varying capacity. The analytical resultsand simulation results are matched. In future study, how todynamically design the admission control to improvingperformance will be researched.

7. REFERENCES

[1] H. Holma and A. Toskala (Editors), "WCDMA for UMTS:Radio Access for Third Generation Mobile Communications,"John Wiley & Sons, Ltd, England, 2000.

[2] Mohamed H.," Call admission control in wireless networks: acomprehensive survey", IEEE Communications Surveys &Tutorials, pp. 50-69 First Quarter 2005.

[3] F. Gunnarson, E. Geijer-Lundin, G. Bark, and N. Winberg,"Uplink admission control in WCDMA based on relative loadestimates," in Proceeding of ICC 02, Vol. 5, New York, NY,USA, pp. 3091-3095, Apr. 2002.

[4] Song Liu; Zhisheng Niu; Dawei Huang;" Performanceanalysis of voice message service in CDMA cellular systems",International Conference on Communication TechnologyProceedings, ICCT 2003 Vol. 2 , pp. 891 - 895, 9-11 April2003.

[5] Fantacci, R.; Mennuti, G.; Tarchi, D.;" A priority basedadmission control strategy for WCDMA systems", IEEEInternational Conference on Communications, vol. 5, pp. 3344 -3348. 16-20 May 2005.

[6] Yu, O.; Saric, E.; Anfei Li;" Adaptive prioritized admissionover CDMA", IEEE Wireless Communications and NetworkingConference Vol. 2, PP. 1260 - 1265, 13-17 March 2005.

[7] Kuenyoung Kim; Youngnam Han, "A call admission controlwith thresholds for multi-rate traffic in CDMA systems,"Vehicular Technology Conference Proceedings, 2000. VTC2000-Spring Tokyo. 2000 IEEE 51st, Volume: 2 , pp. 830 - 834May 2000.

[8] B. Epstein and M. Schwartz," Reservation strategies formultimedia traffic in a wireless environment", in Proceeding ofIEEE Vehicular Technology Conference, pp. 165-169, July1995.