0911 kim

15
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003 911 Dynamic Random Access Code Assignment for Prioritized Packet Data Transmission in WCDMA Networks Dong In Kim, Senior Member, IEEE, Ekram Hossain, Member, IEEE, and Vijay K. Bhargava, Fellow, IEEE Abstract—In this paper, we propose a measurement-based dynamic random access (RA) code assignment procedure for prioritized packet data transmission in wideband code-division multiple access (WCDMA) networks. This dynamic adaptation process is based on analytical performance results derived for random packet access under Rayleigh fading in WCDMA net- works. The performance of the proposed measurement-based RA code assignment procedure with three different adaptation methods is evaluated by using computer simulations. The per- formance of the proposed scheme is compared with those of a retransmission control-based and static channel allocation-based prioritized packet access scheme. An integrated (physical layer and link layer) delay-throughput performance model is presented for finite population RA WCDMA systems. The proposed dynamic RA code assignment procedure can be used in an adaptive quality of service (QoS) framework for dynamically adjusting the QoS of prioritized RA data traffic in the evolving WCDMA-based differentiated services wireless Internet protocol networks. Index Terms—Measurement-based resource allocation, quality of service (QoS), wideband code-division multiple access (WCDMA), wireless Internet protocol (IP). I. INTRODUCTION W IDEBAND direct-sequence code-division multiple ac- cess (DS-CDMA) is being considered as one of the candidate radio transmission technology for IMT-2000, the third-generation wireless communication system ([1], [2]) which is expected to provide a variety of broadband mobile services in- cluding high-speed Internet access. Evolving wideband CDMA (WCDMA)-based wireless Internet protocol (IP) networks will be required to handle prioritized data traffic in the WCDMA air-interface. To meet the diverse quality of service (QoS) re- quirements of different data traffic flows (which are inherently bursty) and at the same time to achieve high resource uti- lization, dynamic resource management would be required. In Manuscript received August 14, 2001; revised February 28, 2002; accepted May 15, 2002. The editor coordinating the review of this paper and approving it for publication is Q. Bi. This work was supported in part by the Brain Korea 21 Research Project in 2000 and in part by a Strategic Project Grant from the Natural Sciences and Engineering Research Council (NSERC) of Canada. D. I. Kim was with the University of Seoul, Seoul 130-743, Korea. He is now with the School of Engineering Science, Simon Fraser University, Burnaby, BC V5A 1S6, Canada (e-mail: [email protected]). E. Hossain is with the Department of Electrical and Computer Engi- neering, University of Manitoba, Winnipeg, MB R3T 5V6, Canada (e-mail: [email protected]). V. K. Bhargava is with the Department of Electrical and Computer Engi- neering, University of British Columbia, Vancouver BC V6T 1Z4, Canada (e-mail: [email protected]). Digital Object Identifier 10.1109/TWC.2003.816795 this paper, we are concerned with the dynamic management of random access (RA) code channels in WCDMA networks. We apply the concept of measurement-based dynamic resource management to provide predictive QoS 1 to prioritized uplink common-channel data traffic in WCDMA systems. Noteworthy is the fact that devising efficient resource management mecha- nisms in the presence of highly bursty packet data traffic (e.g., self-similar wireless web traffic [4], [5]) is a big challenge. In a WCDMA network, packet data transmission from a mo- bile station (MS) is based on multicode spread slotted ALOHA. Each packet (or RA burst) transmitted within a frame period (which is further divided into time offsets/time slots) consists of a fixed preamble part and a variable length data part. The preamble part of a packet is transmitted using a preamble code or RA code while the data part is spread and modulated using a scrambling code which is selected based on the randomly chosen preamble code and the randomly chosen time offset. Dif- ferential services to the different data traffic flows can be provi- sioned through prioritized access to RA code channels. In addition to resource underutilization, fixed assignment (FA) of RA code channels may cause prolonged QoS degra- dation to high-priority data packets. Therefore, we consider a dynamic RA code assignment process which is based on the traffic load-sensing in the base station (BS). In this paper, we consider data traffic with two different priorities. The QoS measure that we consider here is , the success rate for the high-priority RA packets, which is measured as the ratio of high-priority RA packet throughput ( ) and high-priority RA packet load ( ). For a certain traffic load, this metric is in turn related to the average network access delay performance for the high-priority packets. Although achieving high packet success rate ( ) and at the same time high system throughput ( ) (i.e., total number of successful RA attempts per slot) are conflicting networking goals, an efficient RA code assignment mechanism is expected to provide a reasonably good performance tradeoff between total throughput ( ) and high-priority packet success rate ( ). Optimal code assignment would require the BS to have exact knowledge of the uplink data traffic, which may not be realiz- able in the case of bursty packet data transmission. Therefore, heuristic-based code channel assignment policies are sought for which may not necessarily result in an optimal solution but some suboptimal solutions (from the viewpoint of total throughput and QoS tradeoff). 1 It is better than the best-effort QoS from the viewpoint of realization of QoS [3]. 1536-1276/03$17.00 © 2003 IEEE

Upload: mishel1980

Post on 25-Apr-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003 911

Dynamic Random Access Code Assignmentfor Prioritized Packet Data Transmission in

WCDMA NetworksDong In Kim, Senior Member, IEEE, Ekram Hossain, Member, IEEE, and Vijay K. Bhargava, Fellow, IEEE

Abstract—In this paper, we propose a measurement-baseddynamic random access (RA) code assignment procedure forprioritized packet data transmission in wideband code-divisionmultiple access (WCDMA) networks. This dynamic adaptationprocess is based on analytical performance results derived forrandom packet access under Rayleigh fading in WCDMA net-works. The performance of the proposed measurement-basedRA code assignment procedure with three different adaptationmethods is evaluated by using computer simulations. The per-formance of the proposed scheme is compared with those of aretransmission control-based and static channel allocation-basedprioritized packet access scheme. An integrated (physical layerand link layer) delay-throughput performance model is presentedfor finite population RA WCDMA systems. The proposed dynamicRA code assignment procedure can be used in an adaptive qualityof service (QoS) framework for dynamically adjusting the QoSof prioritized RA data traffic in the evolving WCDMA-baseddifferentiated services wireless Internet protocol networks.

Index Terms—Measurement-based resource allocation,quality of service (QoS), wideband code-division multiple access(WCDMA), wireless Internet protocol (IP).

I. INTRODUCTION

WIDEBAND direct-sequence code-division multiple ac-cess (DS-CDMA) is being considered as one of the

candidate radio transmission technology for IMT-2000, thethird-generation wireless communication system ([1], [2]) whichis expected to provide a variety of broadband mobile services in-cluding high-speed Internet access. Evolving wideband CDMA(WCDMA)-based wireless Internet protocol (IP) networks willbe required to handle prioritized data traffic in the WCDMAair-interface. To meet the diverse quality of service (QoS) re-quirements of different data traffic flows (which are inherentlybursty) and at the same time to achieve high resource uti-lization, dynamic resource management would be required. In

Manuscript received August 14, 2001; revised February 28, 2002; acceptedMay 15, 2002. The editor coordinating the review of this paper and approvingit for publication is Q. Bi. This work was supported in part by the Brain Korea21 Research Project in 2000 and in part by a Strategic Project Grant from theNatural Sciences and Engineering Research Council (NSERC) of Canada.

D. I. Kim was with the University of Seoul, Seoul 130-743, Korea. He is nowwith the School of Engineering Science, Simon Fraser University, Burnaby, BCV5A 1S6, Canada (e-mail: [email protected]).

E. Hossain is with the Department of Electrical and Computer Engi-neering, University of Manitoba, Winnipeg, MB R3T 5V6, Canada (e-mail:[email protected]).

V. K. Bhargava is with the Department of Electrical and Computer Engi-neering, University of British Columbia, Vancouver BC V6T 1Z4, Canada(e-mail: [email protected]).

Digital Object Identifier 10.1109/TWC.2003.816795

this paper, we are concerned with the dynamic managementof random access (RA) code channels in WCDMA networks.We apply the concept of measurement-based dynamic resourcemanagement to provide predictive QoS1 to prioritized uplinkcommon-channel data traffic in WCDMA systems. Noteworthyis the fact that devising efficient resource management mecha-nisms in the presence of highly bursty packet data traffic (e.g.,self-similar wireless web traffic [4], [5]) is a big challenge.

In a WCDMA network, packet data transmission from a mo-bile station (MS) is based on multicode spread slotted ALOHA.Each packet (or RA burst) transmitted within a frame period(which is further divided into time offsets/time slots) consistsof a fixed preamble part and a variable length data part. Thepreamble part of a packet is transmitted using a preamble codeor RA code while the data part is spread and modulated usinga scrambling code which is selected based on the randomlychosen preamble code and the randomly chosen time offset. Dif-ferential services to the different data traffic flows can be provi-sioned through prioritized access to RA code channels.

In addition to resource underutilization, fixed assignment(FA) of RA code channels may cause prolonged QoS degra-dation to high-priority data packets. Therefore, we consider adynamic RA code assignment process which is based on thetraffic load-sensing in the base station (BS). In this paper, weconsider data traffic with two different priorities. The QoSmeasure that we consider here is , the success rate for thehigh-priority RA packets, which is measured as the ratio ofhigh-priority RA packet throughput ( ) and high-priority RApacket load ( ). For a certain traffic load, this metric is in turnrelated to the average network access delay performance for thehigh-priority packets. Although achieving high packet successrate ( ) and at the same time high system throughput ( ) (i.e.,total number of successful RA attempts per slot) are conflictingnetworking goals, an efficient RA code assignment mechanismis expected to provide a reasonably good performance tradeoffbetween total throughput ( ) and high-priority packet successrate ( ).

Optimal code assignment would require the BS to have exactknowledge of the uplink data traffic, which may not be realiz-able in the case of bursty packet data transmission. Therefore,heuristic-based code channel assignment policies are sought forwhich may not necessarily result in an optimal solution but somesuboptimal solutions (from the viewpoint of total throughputand QoS tradeoff).

1It is better than the best-effort QoS from the viewpoint of realizationof QoS [3].

1536-1276/03$17.00 © 2003 IEEE

912 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

The motivation of our work in this paper is also due toanother issue of concern in the design of WCDMA-basedwireless packet data networks. It is the understanding of theimpacts of different physical layer parameters on higher layerprotocol [e.g., radio link control (RLC)/medium access control(MAC)] performance which would be required for protocolstack performance optimization. We derive some analyticalresults for performance evaluation of uplink common channelpacket access in terms of different physical layer parametersin a WCDMA system. An integrated (physical layer and linklayer) delay-throughput performance model based on thoseresults and some exact Markov analyses are also presented.The performance of the proposed measurement-based dynamicRA code allocation procedure is evaluated through computersimulation. The performance of the proposed scheme is alsocompared with those of a retransmission control-based and astatic allocation-based prioritized packet access schemes.

The organization of the rest of the paper is as follows. In Sec-tion II, we describe the proposed dynamic RA code assignmentprocedure. RA packet access performance in the WCDMAradio interface is evaluated in Section III assuming multipathRayleigh fading channels. The simulation results for the pro-posed measurement-based RA code assignment procedure withthree different adaptation methods are presented in Section IV.In Section V, we present an integrated performance modelbased on exact Markov analyses to evaluate the delay-throughutperformance under different static channel assignments and/orfixed probability retransmission control and to evaluate theimpact of different physical layer parameters on RLC/MAClayer performance. In Section VI, conclusions are stated.

II. DYNAMIC RA CODE ASSIGNMENT PROCESS

To realize the measurement-based dynamic RA code as-signment, traffic load is sensed at the BS over a measurementwindow of time slots. In general, traffic load in this casecan be measured based on the channel throughput (i.e., averagenumber of RA requests successfully received) since throughputis a function of the offered load (i.e., average number of RAattempts) on the uplink common channel [i.e., RA channel(RACH)]. The dynamic RA code assignment can be imple-mented through a multistate system operation as describedbelow. The system state can be described by the number of RAcode channels allocated to high–low priority data traffic. In theconventional case where each MS can randomly choose anyone of the RA code channels irrespective of the priority ofthe RA packet, the system operates in a single-state mode.

Under the proposed dynamic RA code assignment scheme,the system can be in any one of the states, where

depending on the value of the parameter . Theparameter here denotes the minimum number of RA codesallocated to high-priority packets.

The system operates in state ( ) until overloadis sensed in the BS receiver after which it switches to someother state which is determined based upon the sensed total of-fered load, the parameter , and the proportion of high-pri-ority traffic load in the total offered load measured in the ob-servation window . Note that, such a scheme does not re-

quire any identification mechanism for the packet type (i.e.,high-priority/low-priority) in the data part of a transmitted framesince the receiver at the BS can identify the packet type basedon the RA code associated with the packet. Identification ofpacket type, say, by appending only identification bit(s) in theRA burst might become difficult due to wireless channel er-rors. Again, identification of packet type, say, by using specificscrambling code in the data part depending on the packet type,would introduce additional overhead for load estimation at theBS receiver. Therefore, the proposed dynamic code assignmentscheme would not incur much overhead from the implementa-tion viewpoint.

The load estimation is based on an exponential weightedaverage estimate of the correctly received packets at the BSreceiver. The parameters of the measurement process are thelength of the measurement window , low-pass filter gainsused for the measurement process, the state switching threshold

which denotes the desired success rate for the high-pri-ority RA packets, and the guard-value for the state-switchingthreshold ( ) or “guard-load” ( ).

One way of dynamic adjustment of RA codes is to performproration of RA codes in a memoryless fashion based upon themeasured load on the presently allocated RA codes for high- andlow-priority RA packets. That is, RA code reassignment is doneafter each measurement period based on the proportion of loadof each traffic type during the measurement period. This methodis referred to as the proportional adjustment (PA) method.

Another way is linear adjustment (LA) in which case thenumber of RA codes allocated for high-priority packet accesscan be incremented–decremented depending on the estimatedQoS. We refer to this method as the LA method. Due tothe CDMA soft capacity [6], the degradation in RA packetthroughput performance would presumably not become verysignificant in the case of occasional traffic overload. In fact,this soft capacity allows for a gradual adaptation in resourceallocation rather than a more aggressive and abrupt resouceadaptation. Therefore, the LA method is expected to offerreasonably good performance.

A hybrid of the PA and LA methods is one in which RAcode allocation for high-priority traffic is increased in propor-tion to corresponding measured load while the decrement incode channel is performed in a linear fashion. We refer to onesuch method described in this paper as the proportional-incre-ment linear-decrement (PL) method.

The LA and PL methods may provide a better isolation ofRA code resources for high-priority RA packets compared withthat in the case of PA method especially when the RA trafficload due to low-priority packets increases significantly (e.g., dueto some “hot-spot” traffic and/or good channel condition). Thealgorithms for these three methods will be presented later in thispaper.

The modified assignment of the RA codes and , where, is broadcast by the BS so that the RLC/MAC

mechanism in an MS can control the RA procedure dependingon the traffic priority. In the case of a high (low)-priority RApacket, an MS will choose one of those codes randomlyto encode the preamble part of the packet. It is to be noted that,in the conventional case, which we refer to as the FA scheme

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 913

Fig. 1. Receiver structure for preamble detection.

in this paper, an MS can choose any one of the RA codesirrespective of the data traffic priority.

The measurement-based RA code assignment procedure asdescribed above requires estimation of the network offered QoS( ) from the measured system load ( , ). Theanalytical results derived in the following section enable us tomake such an estimation.

It is to be noted that, the above methods can be extendedfor dynamic code channel allocation among packet flows withmore than two priority levels. But a good tradeoff between QoSand channel utilization would become increasingly difficult toachieve as the number of priority levels increases.

III. RELATIONSHIP AMONG SYSTEM LOAD, RA THROUGHPUT,AND RA SUCCESS RATE IN FADING CHANNELS

We consider a WCDMA uplink RACH where the frame du-ration is five time slots and the length of each time slot is 2 ms.An RA packet has the preamble part of symbols which arespread by a common RA short code of length specified bythe BS in a cell. A set of orthogonal preamble patterns is gen-erated using symbols from the set of symbols and an MSchooses one among these patterns randomly. The preamblepart of an RA packet can be followed by some other informa-tion such as the source address all of which are assumed to fitinto the slot period of 2 ms. This “padded” information can beencoded by one of the scrambling codes which can be selectedbased on the chosen preamble pattern.

Before transmitting an RA packet, an MS has to acquire theframe timing from the broadcast channel and the informationregarding the cell-specific RA short code and the preamble pat-terns. Next, an open-loop power adjustment is carried out toeliminate the effects of shadowing and path loss so that the trans-mission is affected only by fast fading. A mobile terminal se-lects the preamble pattern randomly depending on the operationmode and traffic type and subsequently transmits the RA packet.

We assume a fast Rayleigh fading channel with resolvablepaths. It is also assumed that the fade variation on each path re-mains unchanged during the preamble part of the RA packet andthat the scattering is uncorrelated with equal average path power.An RA packet is assumed to be correctly received at the BS ifthere is only one transmission attempt in an RA code channeland sufficient signal-to-interference plus noise ratio (SINR) is

achieved during the preamble detection (Fig. 1). In addition, wemay introduce a small additional random delay to properly re-solve those transmissions using the same RA code at the sametime in the presence of multipath, and the details of the pre-amble detection in such a case can be found in [7]. Here, themultipath delays associated with different users will be distinctso that collisions can be resolved, but a conservative assump-tion on the success event (i.e., only single transmission in anRA code channel) is made in this paper.

An RA packet is processed at the BS to find the correspondingpreamble pattern by correlating with the candidate patternswhich are orthogonal to each other. The BS sends a positiveacknowledgment if the correlated output exceeds a prespecifiedthreshold. Otherwise, the RA attempt is declared to be failedand the MS attempts retransmission after a random delay.

A. Preamble Detection

Preamble detection is performed using a single matched filterfollowed by -preamble pattern correlators with a RAKE-typesquare-law combiner, as shown in Fig. 1.

After despreading (matched filtering), the output signal canbe modeled as

(1)

where denotes the Rayleigh-faded path gain of the first (de-sired) user in the first RA code channel that is collision-free,

is the first preamble symbol pattern, and indi-cate the path delay and phase, is a triangular pulse over

with and symbol time , themultiple access interferences (MAIs) and rep-resent the RA/traffic channel interferences incurred by and

simultaneous transmissions, respectively ( implies theself-noise from desired signal), , ( denotesthe integer part of ), and is the background noise.

By using an RA short code of period , it follows thatis correlated with , while there exists

914 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

no correlation in because of long codes in the trafficchannel. We note that and correspond to two partial corre-lation values after despreading in case of mismatched symbols.

Due to the processing gain of per symbol, it turns out thatthe effective interference plus noise can be approximated as in-dependent complex Gaussian variable with zero mean and vari-ance (except the self-noise), as follows [8]:

(2)

where refers to the frequency reuse factor in a cellular CDMAsystem.

We have assumed above that ( denotes theexpectated value), , and per-path signal energy

( is symbol energy). With , we have thefactor , instead of , in the first term of (2) because one oftwo partial correlation values turns out to be zero after the firstpreamble correlator due to orthogonality.

Then, the decision statistics after correlation over thesymbol periods is equivalent to

(3)

where is assumed and. Hence, the second-order moment

, given ( denotes the com-plex conjugate), becomes

(4)

where the first term is due to the self-noise (since, it is ignored), and .

Therefore, the decision variable in a RAKE combiner canbe defined by which has the noncentral dis-tribution with degrees of freedom and noncentral parameter

[9].Proposition 1: The preamble detection probability can be

defined by and can be derived as

(5)

where the effective SINR is defined by , and thethreshold can be adjusted with the false-alarm rate

being specified. It is interesting to note that.

Proof of Proposition 1: See Appendix A.

Note that (5) gives a conservative estimate of the preambledetection probability assuming a single-shot RA probe, wherethe MAI statistics of the preamble detection include the auto-and cross-correlation effects among the preamble sequencesas well as the data traffic channels, unlike that addressed in[8]. If the RA follows a power-ramping procedure in accessattempts (i.e., if the first probe fails, the second probe willbe transmitted with higher power to increase the successprobability), the preamble detection probability increases withincreasing transmitter power. However, analysis of the pre-amble detection probability with this power-ramping procedureis mathematically intractable because of strong correlation inMAI statistics between successive RA probes, especially in thepresence of correlated fading.

B. RACH Throughput

Let us denote by the probability that exactly RApackets are received without collision when there are RA at-tempts on the RA code channels.

Proposition 2:

(6)

where and .Proof of proposition 2: See Appendix B.

If we assume that the aggregate RA traffic (i.e., both the newand the backlogged RA packets) can be modeled by a Poissondistribution with arrival rate , then the channel throughputcan be evaluated as

(7)

where . For a given value of , the averagenumber of collision-free RA packets measured at the BS is re-lated to the traffic load by (7).

Fig. 2 shows typical variations in throughput and RA packetsuccess rate with offered load for some typical values of thesystem parameters (e.g., for , ,

dB, , , ,, , ). The number of active traffic

channels per carrier per cell is assumed to be 50 (i.e.,).2 The frequency reuse factor is assumed to be 1.5 (i.e.,

), since, typically, the other-cell interference is approx-imately equivalent to about 50% of the in-cell interference [10].Also, the diversity reception is typically based on two-fingercoherent RAKE receiver [11]. Therefore, as a typical value,

is assumed. The -preamble patterns are generatedfrom the Hadamard sequences of . The correlation values

are calculated using the Hadamard sequences of with. It is to be noted that reliable RA preamble detection

can be assured with the false-alarm rate of .

2In a typical cellular CDMA system, for example, in IS-95, there are 64 trafficchannels per 1.25-MHz carrier.

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 915

Fig. 2. Variations in RA throughput and RA success rate with RA traffic load (for Poisson model).

Therefore, as a typical value, is assumed in thispaper.

If the traffic load increases beyond some threshold value, theRA packet success rate falls below the desired value. For ex-ample, with , as exceeds 3.6, falls below 0.75. In thecase of dynamic RA code assignment, the state transition can betriggered at this point. It is to be noted that for a particular valueof , a decrease in the value of (which is assumed to be theQoS parameter, in this paper) implies an increase in the averagepacket access delay for high-priority packets.

The above formulation enables us to evaluate the effects ofdifferent physical layer parameters on RLC/MAC layer perfor-mance in the case of RA packet transmission in a WCDMAsystem. This would be required for system design with protocolstack performance optimization in WCDMA-based wireless IPnetworks.

IV. MEASUREMENT-BASED RA CODE

ASSIGNMENT PROCEDURE

As has already been described in Section II, in the case ofdynamic measurement-based code allocation that we considerin this paper, the system can be in any one of the states (

). corresponds to the state where andall the RA attempts corresponding to the high-priority packetsare made using the code channels.

If the system state is and the measured load on thecode channels exceeds some threshold corresponding

to the (i.e., , andrefer to the “threshold-load” and “guard-load” correspondingto and , respectively), or equivalently, if theachieved QoS ( ) falls below the QoS threshold ( ) [i.e.,

], then in the case of PA method of RA

code channel assignment, the system switches to statewhere . Here, and refer to themeasured load corresponding to high-priority and low-prioritytraffic, respectively, during an observation window. The esti-mated value of corresponding to a value of may be obtainedby some customized software routine based on the analyticalresults on relationship derived in Section III.

In the case of the LA method of operation, the RA code re-assignment is done in an additive fashion. That is, if

, the system switches to state , and if, the system switches to state .

In the case of the PL method of operation, the number of RAcodes for high-priority packets is incremented in proportion tothe measured high-priority load. But the decrement operation isdone linearly. In all three cases, even if the newly calculatedfalls below , the system state is retained in . Systemevolution in other states occurs in a similar manner.

Let be the measurement window size (in time slots).Then, the measurement-based RA code assignment procedureperformed in the three cases (i.e., PA, LA, and LP) can bedescribed through -like code, as follows.

In the case of PA-based RA code assignment, the adaptationprocedure can be described as follows:

1. , , ,2. , , ,Measure and

3.If ( ) {a)b)

916 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

c) If ( or)

d) Switch to state} /* If */Else go to step 2.

The average load and the average success probability are mea-sured using a low-pass filter.3 To capture the short term increasein load resulting from the bursty nature of RA data traffic, thevalue of the time constant of the low-pass filter (i.e., , ) ischosen to be relatively large (e.g., 0.8).

In the case of the LA method, Step 3(c) in the above procedurewill be as follows:

If ( ) {

} /* If */Else if ( )

.

In the case of the LP method, Step 3(c) goes as follows:

If ( )

Else if ( ).

A. Simulation Model

To analyze the performance of the proposed measurement-based RA code channel assignment procedure described aboveand to identify the suitable method of adaptation, we simulatethe system behavior under two different traffic generation sce-nario referred to as “Poisson Model” and “Pareto Model,” asdescribed below.

1) Poisson Model: In this model, the number of RA packetsgenerated per time slot is Poisson distributed with mean . Thevalue of is kept constant during the entire simulation period.Of the total packets generated in a slot, fraction of them arechosen to be high-priority RA packets while the rest of themare considered as low-priority RA packets. Under this model,typical variations in RA packet throughput ( ) with RA packetload ( ) in a conventional system are as shown in Fig. 2.

2) Pareto Model: In this scenario, the number of RA packetsgenerated in a time slot is Pareto distributed with mean . Paretodistribution is characterized by slowly decaying tail and highvariability. Therefore, it induces a strongly bursty RA packetarrival process in our model. The probability density function

3Similar exponential smoothing technique is used for estimation of variousnetwork parameters such as the round-trip time estimate for a transmission con-trol protocol (TCP) connection and the average queue-size in an Internet router.

(pdf) and the expected value of a Pareto-distributed random vari-able are given by [5]

(8)

where is called the Pareto shape parameter and the parameterrepresents the smallest possible value of the random variable.

If , the distribution has infinite variance, and if , thedistribution has also infinite mean.

It is to be noted that, although the traffic generation processfollows a Pareto distribution, for estimation of , we use thesame relationship defined by (7). If the values of the Paretoshape parameter and the location parameter were known,the Pareto model-based standard relationship could beobtained in this case using (7) (with

substituted for ).A -coded time-driven simulator is used for performance

evaluation. For each point in the simulation result set, the sim-ulator run-time is taken to be sufficiently large (e.g., 5 000 000time slots) so that accurate performance measures are obtainedby averaging the results from each time slot. The assumed valuesof in this paper are 0.8 and 0.6 for Poisson and Pareto trafficmodels, respectively. The corresponding values of are as-sumed to be the same as the values of . The assumed valuesfor the other simulation parameters are as follows: ,

, , , ,dB, , , ,

, , .In contrast with the Poisson model, in the case of Pareto

model for RA packet generation, the value of ismuch less even for moderately low values of (Fig. 3). Forexample, with , , when

and , respectively, while for , thecorresponding values are , respec-tively. While simulating the RA code assignment procedure,we assume similar range of values for for both the trafficmodels.

B. Performance Results for the Dynamic Code AssignmentScheme

1) Poisson Model: Simulation results show that in theconventional case of RA code channel assignment (i.e., withFA scheme), falls off rapidly with increase in the value of(Fig. 4). For the assumed values of the simulation parameters,we observe that , as long as in the conven-tional RA procedure. For as large as 10.00, the success ratefor RA packets is about 0.5.

With a PA-based RA code assignment, system performanceis more or less similar to that in the case of the FA method whileonly at relatively large values of (e.g., 10) small improvementin is observed. But with the LA method, improves signifi-cantly at higher system loads. The performance improvement interms of is achieved at the expense of decreased throughput ofthe low-priority RA packets with a consequent decrease in totalRLC/MAC layer throughput . For example, with the assumedsimulation parameters, for , in case of conventional RA

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 917

Fig. 3. Variations in RA throughput and RA success rate with RA traffic load (for Pareto model).

Fig. 4. Performance comparison of FA, LA, PA, and PL methods for Poisson model (for � � �).

code assignment, and the resulting , whilewith the dynamic LA-based RA code assignment,and when . Therefore, improvement inhigh-priority RA packet success rate of about 21% is achievedin this case at the expense of 24% decrease in total RA packetthroughput. This tradeoff improves at higher values of .

Simulation results show that the performance of PL-based dy-namic RA code adjustment is comparable with that of the LAscheme and when is relatively large (e.g., 0.75, 0.80) thetwo schemes perform almost identically. For a particular ,

the effects of variation in on the total throughput ( ) andthe packet success rate for high-priority packets ( ) are illus-trated in Fig. 5. Variations in and with are observedto be more sensitive for LA and PL-based adaptation methodscompared with those for PA-based adaptation, and that for rela-tively large values of , better success rate (for high prioritypackets) is achieved with LA- and/or PL-based adaptation. Dif-ferences among performance results at higher values of(e.g., 0.8, 0.85) are observed to be negligibly small. PA-basedadaptation is found to be not much sensitive to variation in .

918 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

Fig. 5. Effect of � on the performance of LA, PA, and PL methods for Poisson model (for � � �, � � ���).

It is observed that for the assumed traffic model and theload estimation procedure, for the assumed dynamic adaptationinterval (100 time slots), the value of does not have signifi-cant impact on the system throughput and high-priority packetsuccess rate performance in the case of all three adaptationmethods. For dynamic adjustment of QoS, would be moreeffective compared with under the proposed adaptationframework.

As the value of is decreased, the packet success rateand throughput corresponding to high-priority packets increase,but the overall system throughput falls due to reduction inthroughput corresponding to low-priority packets. Therefore,the value of can be chosen based on the desired tradeoffbetween and .

It is to be noted that, although the desired value of maynot be achieved, clearly a service differentiation between thelow-priority and high-priority RA packets can be achieved withreasonably high channel utilization with LA/PL-based RA codeassignment in the proposed adaptation framework over a widerange of system load and QoS values.

2) Pareto Model: Simulation results show that when thevalue of , the Pareto shape parameter, is relatively low (e.g.,1.1), PA method may even perform poorer than FA scheme interms of both and while consistently better is achievedfor the LA and PL methods at the cost of degradation in(Fig. 6). For example, with , when , about 32%improvement in is observed while decreases by about 16%in the case of LA-based adaptation. Similar to the case withPoisson model, tradeoff improves as is increased. Forexample, with , improves by about 22% at the expenseof about 10% reduction in .

Interestingly, as increases, depending on the value of ,the dynamic RA code assignment methods outperform the FAscheme in terms of both and . For example, when ,

this performance crossover occurs at and forLA/PL and PA methods, respectively (Fig. 7). As increases,the crossover points move left in the axis; for example, with

, the crossover occurs at and for LA/PLand PA methods (Fig. 8), respectively.

We observe that fluctuations in and with may becomeirregular with increase in in the case of PL-based adaptation.In contrast, the variations in and with under the sameconditions are more structured (and hence, more predictable)with LA-based adaptation. Therefore, the latter is a more robustadaptation method.

C. Performance Comparison With Retransmission Controland Static Code Channel Allocation-Based Prioritized PacketAccess

In a WCDMA system, service differentiation in the case ofuplink common channel packet access can also be achievedby assigning different retransmission probabilities and/or a dif-ferent number of code channels corresponding to the differentpacket types. Here, we compare the performances of fixed prob-ability retransmission control-based and static allocation-basedprioritized packet access with the performance of the dynamicRA code channel assignment-based prioritized packet access interms of total RA packet throughput and success rate for high-priority packets. Both the Poisson and the Pareto traffic modelsare considered where the aggregate number of RA packets gen-erated during each time slot (i.e., newly generated packets andbacklogged packets) is Poisson–Pareto distributed with mean .

In the case of fixed probability retransmission control, for anaverage traffic load , the actual channel load during a timeslot which depends on the retransmission control parameters –

, and the backlog – and corresponding to high andlow-priority packets, respectively, can be expressed as follows:

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 919

Fig. 6. Performance results for Pareto model (for � � �, � � ���, � � ���).

Fig. 7. Performance results for Pareto model (for � � �, � � ���, � � ���).

, for .Here, an immediate first transmission mode is considered forthe newly generated packets (i.e., packets from nonbackloggedterminals). For , is assumed to be the same as

.In the case of static channel allocation the actual channel load,

which is distributed between the two sets of preassigned codechannels corresponding to the high-priority and the low-prioritydata packets, is the same as the average traffic load .

Performance results in Figs. 9 and 10 indicate that betterpacket success rate for the high-priority packets can be achievedwith LA-based dynamic code channel assignment. For the re-transmission control-based scheme, as the retransmission con-trol parameters are decreased, system throughput ( ) decreasesand high-priority packet success rate ( ) increases. For the staticassignment case, as the number of code channels allocated forthe high-priority traffic ( ) is increased, increases but at thecost of decreased . For the Pareto model, depending on the

920 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

Fig. 8. Performance results for Pareto model (for � � �, � � ���, � � ���).

Fig. 9. Performance comparison among various prioritized packet access schemes for Poisson model (SA � static assignment, RC � retransmission control).

value of the Pareto shape parameter, better system throughputand at the same time better packet success rate for the high-pri-ority data packets can be achieved with the proposed LA-baseddynamic code channel assignment scheme.

V. AN INTEGRATED DELAY-THROUGHPUT PERFORMANCE

MODEL FOR A FINITE POPULATION SYSTEM

In this section, we pesent an integrated delay-throughputperformance model for uplink packet access in a finite pop-

ulation WCDMA system. It is a programmable performancemodel based on exact Markov analyses that can be used toassess the impact of the different physical layer parameters onthe link-layer performance measures and, hence, the higherlayer protocol (e.g., TCP) performance can also be evaluated.The number of code channels, along with the retransmissionprobabilities corresponding to the high- and low-prioritypackets, can be varied. This model is also useful in the caseof dynamic code channel allocation in a time-scale when eachstate stabilizes before switching to another state.

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 921

Fig. 10. Performance comparison among various prioritized packet access schemes for Pareto model.

Suppose that the system is operating at a certain instant withthe RA code assignment, as follows: preamble codes forhigh-priority traffic and preamble codes for low-prioritytraffic and the total number of users is . Each user generatesan RA packet with probability and the probability of thatbeing a high-priority data packet is . A blocked user retrans-mits the RA packet with probability or depending on thepriority of the RA packet and it does not generate a new RApacket until the current one is successfully transmitted.

Let the random variables ( , ) represent the number ofblocked users at time with high-priority and low-priority datatraffic, respectively. Then, the system evolution can be describedby a Markov chain with state space consisting of the set of in-tegers , where if and , aninteger in this set can be related to and , as follows:

(9)

Since and , . For a giventhe numbers ( , ) are uniquely determined, as follows:

(10)

(11)

Using these definitions, the one-step state transition probabil-ities from the state ( , ) can be expressed as follows:

(12)

where is as defined in (9) and .Proposition 3: The state transition probabilities

can be described by (13), at the bottom of the

next page, whereProof of Proposition 3: See Appendix C.

The Markov chain described by the above state transitionprobabilities is irreducible, and hence, a stationary proba-bility distribution exists which can becomputed by solving the following set of simultaneous linearequations:

(14)

with [12].The steady-state throughput is given

by (15), at the bottom of the next page, where

, ( ).Then, the steady-state throughput for both high-priority and

low-priority data traffic is

(16)

Referring to in (10) with ( , ) in (11), the expectednumber of blocked users is

(17)

922 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

Finally, applying Little’s result [13], the average delay in theblocked state is

(18)

On the other hand, in the conventional case, all access re-quests can be transmitted using any one of preamble patterns,and then the Markov chain is described only by the total numberof blocked users with state space as the set ofintegers . Similarly, the one-step state transitionprobability is derived as in (19), where

(19)

The steady-state channel throughput can then be ob-tained using

(20)

where and the mean number of blockedusers is . Therefore, the average delay in theblocked state is .

Some typical results are shown in Fig. 11 (for ,, , , dB, , ,

, , ,). Specific performance behaviors (rather than average

performance in the case of ) can be assessed fromthe presented theoretical framework.

It is to be noted that using parameter (instead of using twoparameters and ) enables us to avoid manipulation of atwo-dimensional Markov chain. Therefore, the complexity ofthe delay-throughput performance evaluation is reduced.

VI. CONCLUSION

Analytical results have been presented for performance eval-uation of RA packet access in WCDMA networks in a Rayleighfading environment. A measurement-based dynamic RA codeassignment procedure for prioritized RA packet transmissionhas been proposed. Simulation results have revealed that forshort-range dependent traffic arrival patterns, a good perfor-mance tradeoff between the total system throughput and therequired QoS can be achieved with the LA/PL-based adapta-tion methods in the proposed dynamic RA code assignmentframework. In addition to differentiated QoS among high- andlow-priority RA packets, higher system throughput may alsobe achieved within the proposed dynamic RA code assignmentframework in the case of long-range dependent traffic arrivalpatterns. The LA-based adaptation is found to be the most ro-bust among all the methods described in this paper.

An integrated (physical layer and link layer) delay-throughput performance model based on exact Markovanalyses has been presented which can be used to assess theimpact of different physical layer parameters and link layerparameters (e.g., retransmission control parameters, channelallocation parameters) on higher layer protocol performancefor prioritized uplink packet access in WCDMA systems.

(13)

(15)

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 923

Fig. 11. Delay-throughput performance under specific RA code assignment (for � � ��, � � �).

APPENDIX APROOF OF PROPOSITION 1

First, if the effective Gaussian, the decision vari-able has the pdf [9]

(21)

where is the modified Bessel function of the first kind andorder . Under uncorrelated scattering and equal average pathpower, (noncentral parameter) is central dis-tributed with degrees of freedom, and the pdf is

(22)

where . Therefore, the probability condi-tioned on RA attempts is given by (23), at the bottom of thepage, where we have assumed the changes of variables

and , and also the order of integration has beenchanged.

Using the following relationship:

(24)

equation (23) is simplified to

(25)

Further, applying the following relationsip:

(26)

along with the change of variable , we obtain the prob-ability in (5). With a chip-rate preamble processing, thefalse-alarm rate before a RAKE is

, where the self-noise in(4) would be increased with replaced by .

APPENDIX BPROOF OF PROPOSITION 2

Let us define the index sets and asand , which rep-

resent the RA messages attempted in a slot and the RA codesbeing used for the message encoding, respectively. Then, themapping of is equivalent to the collection of allpossible ways in which RA messages will select any one of

(23)

924 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 5, SEPTEMBER 2003

RA codes with equal probability. For this, we define a setwhere order counts , where .

To apply the principle of inclusion and exclusion from com-binatorics [14], we define a set as

has the property that the RA message

is collision-free against other messages

Then, the number of elements satisfying exactly properties is

(27)

where denotes the number of properties satisfied by anelement , and is any possible subset of the index setwith size . Since each property is equally likely tooccur, it turns out that

(28)

Now, it remains to find the number of occurrences of the eventthat at least of RA messages are collision-free, whichis given by

(29)

with . Hence, combining the above results, theprobability distributionis derived as (6).

APPENDIX CPROOF OF PROPOSITION 3

Given the number of blocked users ( , ),thinking users transmit with probabilitywhile out of RA packets correspond to high-priority packetwith probability . Also, ( , ) blocked users transmitwith probability , which produces atotal and RA attempts on the corresponding

and RA codes. The number of successful attempts on theand RA code channels is considered to follow binomial

distribution with the detection probability , con-ditioned on the total RA attempts.

Now, if and RA attempts succeedon the and code channels, respectively, then the systemmoves to state ( , ). Thus, the numberof collision-free packets ( , ) on the code channels shouldbe and

. Under S-ALOHA with perfectcollision channel assumption, success of an RA attempt requiresthat there is only one attempt on the corresponding RA code, andhence, can be derived as (13).

REFERENCES

[1] E. Dahlman, P. Beming, J. Knutsson, F. Ovesjö, M. Persson, and C.Robol, “WCDMA – the radio interface for future mobile multimediacommunications,” IEEE Trans. Veh. Technol., vol. 47, pp. 1105–1118,Nov. 1998.

[2] “Concept Group Alpha – Wideband Direct-Sequence CDMA, Evalua-tion Document (3.0), Part 1: System Description & Performance Eval-uation, UMTS Terrestrial Radio Access (UTRA),” ETSI SMG2, draftdocument (3.0), 1997.

[3] J. Kurose, “Open issues and challenges in providing quality of ser-vice guarantees in high-speed networks,” ACM SIGCOMM Comput.Commun. Rev., vol. 23, no. 1, pp. 6–15, Jan. 1993.

[4] Z. Harpantidou and M. Paterakis, “Random multiple access of broadcastchannels with Pareto distributed packet interarrival times,” IEEE Pers.Commun., vol. 5, no. 2, pp. 48–55, Apr. 1998.

[5] M. Oguz Sunay, S. Tekinay, and S. Zorlu Özer, “Efficient allocation ofradio resources for CDMA-based wireless packet data systems,” in Proc.IEEE GLOBECOM’99, vol. 1B, Dec. 1999, pp. 638–643.

[6] R. Kohno, R. Meidan, and L. B. Milstein, “Spread spectrum accessmethods for wireless communications,” IEEE Commun. Mag., vol. 33,pp. 58–67, Jan. 1995.

[7] D. I. Kim and J. C. Roh, “Performance of slotted asynchronous CDMAusing controlled time of arrival,” IEEE Trans. Commun., vol. 47, pp.454–463, Mar. 1999.

[8] M. B. Pursley, “Performance evaluation for phase-coded spread-spec-trum multiple-access communication – Part I: System analysis,” IEEETrans. Commun., vol. 25, pp. 795–799, Aug. 1977.

[9] A. D. Whalen, Detection of Signals in Noise. New York: Academic,1971.

[10] A. J. Viterbi, A. M. Viterbi, and E. Zehavi, “Other-cell interference incellular power-controlled CDMA,” IEEE Trans. Commun., vol. 42, pp.1501–1504, Feb./Mar./Apr. 1994.

[11] R. Esmailzadeh and M. Gustafsson, “A new slotted ALOHA basedrandom access method for CDMA systems,” in Proc. IEEE ICUPC’97,vol. 1, Oct. 1997, pp. 43–47.

[12] L. Kleinrock and S. S. Lam, “Packet switching in a multiaccessbroadcast channel: Performance evaluation,” IEEE Trans. Commun.,vol. COM-23, pp. 410–423, Apr. 1975.

[13] L. Kleinrock, Queueing Systems. New York: Wiley, 1975, vol. I and II.[14] D. A. Cohen, Basic Techniques of Combinatorial Theory. New York:

Wiley, 1978.

Dong In Kim (S’89–M’91–SM’02) received theB.S. and M.S. degrees in electronics engineeringfrom Seoul National University, Seoul, Korea, in1980 and 1984, respectively, and the M.S. and Ph.D.degrees in electrical engineering from the Universityof Southern California (USC), Los Angeles, in 1987and 1990, respectively.

From 1984 to 1985, he was with the KoreaTelecommunication Research Center as a Re-searcher. During 1986–1988, he was a KoreanGovernment Graduate Fellow in the Department

of Electrical Engineering, USC. In 1991, he was with the University ofSeoul, Seoul, Korea, leading the Wireless Communications Research Group.He is now with Simon Fraser University, Burnaby, BC, Canada, where heis an Associate Professor of the School of Engineering Science. He was aVisiting Professor at the University of Victoria, Victoria, BC, Canada, during1999–2000. He has performed research in the areas of packet radio networksand spread-spectrum systems since 1988. His current research interests includespread-spectrum systems, cellular mobile communications, indoor wirelesscommunications, and wireless multimedia networks.

Dr. Kim has served as an Editor for the IEEE JOURNAL ON SELECTED AREAS

IN COMMUNICATIONS: WIRELESS COMMUNICATIONS SERIES and also as a Di-vision Editor for the Journal of Communications and Networks. Currently, heserves as an Editor for the IEEE TRANSACTIONS ON COMMUNICATIONS and theIEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.

KIM et al.: DYNAMIC RA CODE ASSIGNMENT FOR PRIORITIZED PACKET DATA TRANSMISSION 925

Ekram Hossain (S’98–M’01) received the B.Sc.and M.Sc. degrees in computer science and engi-neering from Bangladesh University of Engineeringand Technology (BUET), Dhaka, Bangladesh, in1995 and 1997, respectively, and the Ph.D. degreein electrical engineering from the University ofVictoria, Victoria, BC, Canada, in 2000.

He is an Assistant Professor in the Department ofElectrical and Computer Engineering at the Univer-sity of Manitoba, Winnipeg, MB, Canada. He was aFellow with the University of Victoria. His main re-

search interests include radio link control and transport layer protocol designissues for the next-generation wireless data networks.

Dr. Hossain currently serves as an Editor for the IEEE TRANSACTIONS ON

WIRELESS COMMUNICATIONS.

Vijay K. Bhargava (S’70–M’74–SM’82–F’92)received the B.Sc., M.Sc., and Ph.D. degrees fromQueen’s University, Kingston, Canada in 1970,1972, and 1974, respectively.

He is a Professor and Chair of Electrical andComputer Engineering at the University of BritishColumbia, Vancouver, Canada. He is a coauthor ofthe book Digital Communications by Satellite (NewYork: Wiley, 1981) and coeditor of the IEEE PressBook Reed–Solomon Codes and Their Applications(New York: IEEE, 1994). His research interests are

in multimedia wireless communications.Dr. Bhargava was the President of the IEEE Information Theory Society

in 2000, Co-Chair for IEEE ISIT’95, Technical Program Chair for IEEEICC’99, and was the Chair of IEEE VTC 2002 Fall. He is a Fellow of theB.C. Advanced Systems Institute, Engineering Institute of Canada (EIC), andthe Royal Society of Canada. He is a recipient of the IEEE Centennial Medal(1984), IEEE Canada’s McNaughton Gold Medal (1995), the IEEE HaradenPratt Award (1999), and the IEEE Third Millennium Medal (2000).