A distributed dynamic resource allocation for a hybrid TDMA/CDMA system

1162 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 47, NO. 4, NOVEMBER 1998

A Distributed Dynamic Resource Allocation for aHybrid TDMA/CDMA System

Lauro Ortigoza-Guerrero,Student Member, IEEE,and A. Hamid Aghvami,Senior Member, IEEE

Abstract—A distributed dynamic resource allocation (DDRA)strategy for a hierarchical cellular structure (HCS) is proposed.In the DDRA, resources are shared not only between cellsof the same hierarchy, but between layers. The proposedDDRA strategy is evaluated using the hybrid time-divisionmultiple-access/code-division multiple-access (TDMA/CDMA)proposal made in the future radio wide-band multiple-accesssystem (FRAMES) Project Mode 1 (FM1) as a case study.A mixed environment is suggested for the evaluation of theDDRA, which consists of Manhattan-like microcells coveredby hexagonal-shaped umbrella cells (macrocells). Users areclassified according to their speed asslow- and fast-moving usersand are attended to by the most suitable layer of the hierarchyaccording to their speeds. Two types of real-time circuit-switchedservices are considered in the evaluation: speech and data atdifferent rates. The DDRA is compared with the fixed resourceallocation (FRA) strategy with overflow and with FRA withoverflow, handdown, and channel reallocations (FRAHR).

Index Terms—Hybrid TDMA/CDMA systems, resource alloca-tion strategies.

I. INTRODUCTION

T HE MAIN objectives and technical challenges in theuniversal mobile telecommunication systems (UMTS’s)

are to provide higher spectral efficiency than existing systemsand to have resource flexibility to accommodate multiplenetworks and traffic types within a given frequency band[1]. To reach these goals, wide-band time-division multipleaccess (WB-TDMA) with and without spreading feature [WBTDMA/code-division multiple access (CDMA)] has been iden-tified as one of the major multiple-access schemes for UMTS:its main contender is WB-CDMA.

A hybrid TDMA/CDMA system is based on a TDMA framestructure. The basic way of supporting multiple bit rates is toassign multiple slots to a user. The number of slots insidea frame can be varied dynamically in order to adapt to thetransmission of changing service needs. For operation withspreading, multiple codes on a slot can also be assigned. Thebit rate for a specific service is finetuned by selecting anappropriate combination of slot length, burst type, number ofslots, and coding rate.

It has been shown in the literature that the fixed resourceallocation (FRA) is a strategy that does not fully satisfy the

Manuscript received January 19, 1998; revised June 12, 1998. This workwas supported by CONACyT-MEXICO under Grant 67257/110660.

The authors are with the Center for Telecommunications Research,King’s College, London, WC2R 2LS, U.K. (e-mail: [email protected]).

Publisher Item Identifier S 0018-9545(98)08269-3.

requirements of a mobile network, particularly with unevenand time-varying traffic distributions. Therefore, a dynamicresource allocation (DRA) strategy will certainly be needed tomanage all the available resources to increase the capacityof a UMTS network. Some work relevant to DRA strate-gies can be found in the literature. For instance, in [2] aset of DRA strategies applied to a real global system formobile communication (GSM) network is presented. In [3],a complete selection of DRA, channel borrowing, hybrid,and reuse partitioning strategies are described. However, theirdescriptions are confined to a single layer. Nevertheless, in [4]a DRA for hierarchical cellular structure’s (HCS’s) that makesuse of reuse partitioning is presented. Due to its nature, thisalgorithm is completely decentralized. This algorithm is oneof the first dealing with the dynamic allocation of channels inan HCS. In [5], a channel assignment strategy making use ofhanddown and handup is presented. This strategy is so far themost complex strategy for HCS described in the literature, butthere is no dynamic assignment of resources.

This paper describes a distributed dynamic resource allo-cation (DDRA) strategy for an HCS that can be applied toa hybrid TDMA/CDMA system, to a CDMA system [WB-CDMA with time-division duplex (TDD)] or to a pure TDMAsystem. Our contribution is the extension and evaluation of aDDRA strategy suitable for UMTS. The dynamic nature of thisstrategy permits its adaptation to uneven and changing traffic,while the distribution of the decision making process betweenthe cells reduces the required computation and communicationbetween base stations (BS’s). This strategy is evaluated usingthe future radio wide-band multiple-access system (FRAMES)Project Mode 1 (FM1) proposal [1], [6] in a practical HCS thatcould appear in a future UMTS. Two types of circuit-switchedservices are considered in the evaluation: voice at 8 Kbpsand data transmissions at 32, 64, 144, and 384 Kbps. Also,an analytical method to assess the blocking probability in anHCS using FRA with different types of users and service ispresented. The remainder of this paper is organized as follows:Section II describes the system on which our study is done,Section III presents the DDRA strategy and two strategiesused as references for comparison purposes, and Section IVdiscusses the evaluation environment. A teletraffic model toassess the blocking probability for each type of service inan HCS using only FRA is presented in Section V, andsimulations results are presented in Section VI. Section VIIpresents some concluding remarks.

0018–9545/98$10.00 1998 IEEE

ORTIGOZA-GUERRERO AND AGHVAMI: DYNAMIC RESOURCE ALLOCATION FOR TDMA/CDMA SYSTEM 1163

II. SYSTEM DESCRIPTION

We start with the definition of the basic concepts that areused throughout the paper and the operations involved in theallocation strategies.

A. Resources

In a hybrid WB-TDMA/CDMA system, users are separatedorthogonally into time slots, and within each time slot anadditional separation by spreading codes is used. The methodto be used depends on the service and radio conditions handledby the link adaptation. The unit of transmission is one slot.This unit is then divided into smaller units, either subslots orby spreading codes if the spreading feature is being used. Inthis way, a smaller granularity is achieved. Given that a usermight require several resources, a DRA scheme could assignseveral spreading codes to the same time slot (multiple codeoption), several channels in different time slots (multiple slotoption), or a combination thereof. In this work, it is proposedmoreover, that several time slots/codes from different carriersmay be combined to deal with a given type of service. Theresource allocation strategies presented will be applied to aportion of a time slot or to a code associated with a timeslot, which will hereafter be called simplyresources. The goalof the allocation strategy is that all available time slots areused with approximately the same number of active spreadingcodes [6].

B. Hierarchical Cellular Structure Description

Fig. 1 depicts a large geographical area covered by a set ofcontiguous Manhattan-like microcells. Every microcells areoverlaid by only one large hexagonal-shaped macrocell, whichfulfills the role of umbrella cell. The microcells constitute thelower layer, and the macrocells form the upper layer of atwo-layer HCS. It is assumed that there is a large numberof users randomly traversing the covered area, each of themwith a different speed. Users are classified asfast- and slow-moving users, and they do not change their speed class duringa service. They are only tracked in the microcellular layersince this can give all the information required as to whento initiate a handoff, which neighboring cell it is moving to,etc. Each layer in the network is given a subset of physicalchannels by the spectrum partitioning strategy [7] derived fromthe total available spectrum. Then, every cell in the HCS isassigned a set of resources. All physical channels are sharedamong new calls and handoff calls, i.e., no prioritization ofhandoff calls by means of reserved physical channels is used.No resource reservation applies for any of the different typesof service available in the system, hence, all of them can makeuse of any of the resources as long as they are available.

C. FRAMES Proposal Mode 1

Within the framework of the Advanced CommunicationTechnologies and Services (ACTS) Program, the FRAMESproject is in charge of defining a multiple-access schemefor the third-generation UMTS, based on adaptive radiointerface concepts. The evaluation has led to the FRAMES

Fig. 1. Evaluation environment.

multiple-access (FMA) concept, which has been presented tothe European Telecommunications Standards Institute (ETSI)for consideration in the UMTS standardization process. Twomultiple-access techniques have been identified, namely,WB-TDMA/CDMA with spreading feature and TDD (WB-TDMA/CDMA-TDD) and W-CDMA, which can both meetthe UMTS requirements. They are based on FM1 andFRAMES project mode 2 (FM2), respectively [1], [6]. Thesemodes are harmonized with each other and also supportcompatibility with GSM. Recently, WB-CDMA-FDD andWB-TDMA/CDMA-TDD have been chosen for operationin a paired and unpaired spectrum, respectively. FM1 withspreading facilitates the application of a variety of dynamicchannel allocation (DCA) strategies, where the allocation ofchannels depends on the current traffic load and/or the currentinterference conditions [8]. The evaluation of the proposedresource allocation strategies presented here is done in asystem using the FM1 proposal.

III. D ESCRIPTION OFSTRATEGIES

Three different strategies are considered: classic flexibleresource allocation (FRA) with overflow procedure, FRA withresource reassignments and handdown procedure (FRAHR),and the DDRA. In the following, we describe the operationof each of them.

A. Fixed Resource Allocation

The FRA strategy involves the following system operations.

1) New requests for service and new handoff requests(voice or data transmission) originated by aslow-movinguser will be directed toward the preferred microcell (apreferred cell in each layer is that cell that receives thestrongest signal from the mobile in the particular HCSlevel). They will be attended to if there are enoughidle resources, otherwise, they will be overflowed tothe preferred macrocell. The overflowed requests will beaccepted by the macrocell if the number of idle resourcesis equal to or larger than the required number, otherwise,the call will be either blocked or dropped (forced toterminate), whichever is applicable.


2) A new request for service originated by afast-movinguser is directed to its preferred macrocell and will beaccepted if there are enough idle resources to serve it.Otherwise, the call will be blocked. When handoff isperformed either by afast-moving useror by a slow-moving userin the macrocell layer, the target macrocellwill try to accommodate the handoff request assigningenough resources. If the number of available resourcesis not enough, the call will be dropped.

B. Fixed Resource Allocation with Handdown Procedureand Resource Reassignments (FRA-HR)

Since the resources in the macrocell layer are scarce, ahanddown procedure (HDP) has been added to the FRAstrategy to increase the possibility of successful resourceacquisitions byfast-moving usersin the macrocell layer. HDPis similar to the take back request described in [9] andconsists of sending back aslow-moving userattended to ina macrocell to its preferred microcell to release the resourcesin the macrocell. This is possible because aslow-moving userattended to by a macrocell continuously monitors the microcellit is traversing. The way HDP is applied to FRA to createFRA-HR is described as follows.

1) Resource Acquisition: New requests for service andhandoff requests originated by afast-moving userin amacrocell will be attended to as long as there are enoughidle resources. When this is not the case, HDP takesplace (providing that there is at least oneslow-movinguserbeing attended to in the macrocell). When a singleexecution of the HDP does not release the requirednumber of resources, it may be repeated more than onetime to get the desired resources. If there are severalslow-moving usersbeing attended to by the macrocell,the HDP will select thatslow-moving userwhich wouldput in the idle state a number of resources equal toor closest to the desired number. Note that the moremicrocells that are overlaid by the macrocell, the morethe chances of the HDP being successful. This sameprocess is applied when requests originated byslow-moving usersoverflow from the microcell layer (eithernew requests or handoff requests) to the macrocell layer.

2) Release of Resources: When a slow-moving userfin-ishes its connection to the network in a microcell, allthe resources it was occupying are released. A set ofcandidateslow-moving usersbeing attended to by theoverlaying macrocell is then found. If the number ofresources just released are enough to take back one oftheseslow-moving usersfrom the macrocell layer, HDPis performed.

C. Distributed Dynamic Resource Allocation

In [10], a channel assignment strategy was proposed thatmakes use of channel ordering in each cell. In [11], a modi-fication of this scheme was presented. In this last reference, acarrier ordering rather than a channel ordering was used. TheDDRA strategy here presented and applied to an HCS makesuse of the simple dynamic channel allocation (SDCA) strategy

presented in [11], but modified and complemented to accountfor HCS scenarios, the different types of services offered andusers with different speed. What makes the DDRA schemesuitable for use in an HCS is the fact that the resource alloca-tion and the release policy for both FRA and FRA-HR are usedin this strategy. An important feature added to the DDRA isthe capability of sharing resources between layers and actuallyto change the spectrum partitioning according to different ortime-varying conditions of traffic (e.g., propagation, trafficjams, etc.).

When a request for service arrives at a cell in a particularlayer of the HCS, the DDRA gets the required (or part of the)resources from that cell. If this procedure does not satisfy theusers requirements, then the DDRA finds the required (part ofthe, or the rest of the) resources by:

1) borrowing a channel (according to SDCA) in the samehierarchical level;

2) applying the HDP;3) borrowing resources from the upper hierarchical layer.

Point 3) deals with the spectrum sharing between layers inthe HCS and can be seen as the ultimate step in the allocationof resources by a call and is avoided whenever possible tominimize blocked calls in the upper layers. In a two-layeredHCS, the use of borrowed resources from a lending macrocellwill be forbidden in the macrocell layer while they are in usein the microcell layer, and as soon as they are released, theywill be returned to the original macrocell. Note that the use ofresources borrowed from a macrocell in a microcell does notimply restricting their use in cochannel macrocells because thereuse constraints will still be fulfilled.

Similar to that stated in [11], the DDRA strategy does notneed an exchange of information within the interference neigh-borhoods. The busy/idle status of carriers can be determinedby passive nonintrusive monitoring at each BS of a particularhierarchy level. In fact, this strategy is suboptimal since everycell has only access to partial information. However, thedistributed computation and the reduced communication makeit feasible for hierarchical cell structures, where the signalingload among the BS is required to be minimum.

IV. EVALUATION ENVIRONMENT

In our simulations, a two-layered environment consisting ofa microcell and a macrocell layer is assumed. The microcel-lular environment is formed of a (regular) rectangular grid ofintersecting streets (referred to as Manhattan-like microcells),as shown in Fig. 1. The BS’s are at street-lamp height andplaced at the center of the crossroads (clover leaf cells.)In a real system, this would ease the problems related tohandoff since users do not experience a sudden drop of signalwhen turning at a street corner [12]. BS’s are separated by200 m. Users are only tracked at the microcell level, anda wraparound topology is used so that the boundary effectsin the simulations are eliminated, as depicted in Fig. 2. Theinterference neighborhoods of each cell also wrap around. Themacrocell layer is formed of hexagonal cells overlaying themicrocells, and their BS antenna height is above the averagerooftop height. According to [1], a cluster size in the range


TABLE ISERVICE CLASSES AND EXAMPLES FOR MIXED SERVICES FOR FMA1 WITH SPREADING

Fig. 2. Wraparound topology.

of three–nine has to be supported for FRAMES1. In oursimulations, a reuse pattern of four was used for both themicrocell and the macrocell layer, as depicted in Fig. 3. Sixcarriers of 1.6 MHz were assumed to be available in frequency-division duplex (FDD) approach: three carriers for the uplinkand three more for the downlink. Hence, the total number of1/64 time slots or codes available is 192 (we are assumingthat up to eight bursts can be transmitted within one timeslot only, even if the bursts are allocated to a single user). Anapproximate partitioning of 60% of resources for the microcelllayer and 40% resources for the macrocell layer is used [7]. Asa result, a total of 28 1/64 time slots or codes are given to eachmicrocell and a total of 20 to each macrocell in the system.Five types of real-time service are considered. Speech serviceis available at 8 Kbps as well as data transmissions at 32, 64,144, and 384 Kbps. All the services are circuit switched with100% activity. Users are classified asfast- and slow-movingusersas described in previous sections, with the probabilityof a user being of the slow type equal to 0.9 and, therefore,with probability 0.1 of being afast-moving user. Table I showthe types of service considered along with all their associatedrequirements. Some of the parameters used in the simulationsare shown in the Table II.

Fig. 3. Cell reuse pattern for the microcellular layer.

In general, the models used with this evaluation environ-ment, explained below, are different to those used within theETSI Special Mobile Group (SMG) 2 evaluation process [13].Hence, the results presented here may not be the same asthose obtained according to [13].

A. User Mobility

The user mobility model is highly related to the Manhattan-like structure defined early. With this environment, mobilesmove along the streets and may turn around at cross streetswith a given probability as shown in Fig. 4. The turn proba-bility Turn Prob is equal to 0.6666.

B. Propagation Model for the Microcellular Layer

For the microcell layer, a three-slope model is adopted[14]. The model is given mathematically by the followingexpressions for the line of sight (LOS) condition:

(1)


TABLE IITRAFFIC PARAMETERS USED IN THE SIMULATION

and for the no LOS (NLOS) condition:

(2)

Each BS is a street lamp so that the buildings act aswaveguides. A breakpoint, located at a distance from

the transmitter, marks the separation between the two LOSsegments. The second of these LOS segments predicts largerlosses than the first one. For the case of NLOS, the lossesare given by three components: the LOS component (as if themobile were at a distance from the transmitter), withthe additional losses caused by turning at a crossroad and bythe proper NLOS condition. In the equations,stands for thedistance from the transmitter to the receiver measured alongthe street path. is the distance from the transmitter to


TABLE IIISUMMARY OF THE PHYSICAL PARAMETERS USED IN THE SIMULATION

Fig. 4. Mobility parameters in a crossroad point.

the corner, in the case of NLOS. The rest of the parametersinvolved are

(3)

(4)

(5)

where and are the BS and MS antenna heights,is thewavelength, and is the street width. The shadowing effectis modeled as , where is a Gaussian variable withzero mean and a standard deviation of 4 dB. The parametersinvolved in the simulation using this propagation model arepresented in Table III. In our simulations, there are no reuseconstraints on NLOS cochannel cells because of the large-signal attenuation around a corner.

C. Propagation Model for the Macrocellular Layer

For the macrocellular layer, the path losses are calculatedaccording to the following [13]:

(6)

where

(7)

is the difference between the mean building heightand the mobile antenna height, andis the horizontal distancebetween the mobile and the diffracting edges. In the practicalcase, when m, m, and m, typicalin urban and suburban environments with an average buildingheight of four stories, the path-loss expression above reducesto the following [13]:

dB (8)

with the distance from the transmitter to the receiver(inkilometers), the BS antenna height measured from the rooftop

(m) and frequency (in megahertz). The shadowingeffect is modeled as , where is a Gaussian variablewith zero mean and a standard deviation of 6 dB.

D. C/I Threshold

Only the downlink carrier-to-interference ( ) ratio iscalculated since it is assumed that this will be worse than theuplink and thus provide the limit to system performance. The

is calculated according to the corresponding propagationmodel, and the multiple-access scheme used to establish aconnection: TDMA with or without spreading. In the casewithout spreading, only intercell interference is taken intoaccount (cochannel interference), while for the other case,intercell and intracell interference are considered. Once the

is calculated accordingly, this ratio is matched to theappropriated , as specified in Table IV [15] (a moresophisticated simulation should take into account differentthresholds for each different data transmission).

The relation between the average signal-to-noise ratio(SNR) ( is the energy per bit and is the noisepower spectral density) and the ratio, with denoting


TABLE IVEb/No THRESHOLDS FOR THEMULTIPLE SERVICES IN THE CDMA/TDMA SYSTEM [15]

the interference power, is given by [16]

(9)

where is the rate of the channel encoder (depending onthe service), is the size of the data symbol alphabet (4 forQPSK and 16 for 16 QAM), is the user bandwidth (1.6MHz), is the number of chips per symbol (16), and isthe chip duration (0.4615 s).

E. Teletraffic Model

As already stated, users are only tracked in the microcelllayer. This allows the active–dormant Markov model to beused in our simulations to account for even and uneventeletraffic distributions in the proposed scenario. This modelwas originally presented in [17] and in a modified way tomodel handoff queuing in [11]. In order to account for severaltypes of service, we apply this teletraffic model independentlyto each type of service available in our system. A completedescription of the model can be found in [11].

The total offered load to each microcell is given by

(10)

where to are the offered loads produced by each ofthe five types of service available and are given by

(11)

where is the proportion of the total offered load submittedto microcell by type users (see Table II).

V. MATHEMATICAL APPROACH

In this section, an analytical method to assess the blockingprobability for each type of user in the hierarchical networkusing pure FRA is presented. Neither handup nor handdown isconsidered. The expressions for the handoff failure probabilityare found as well. We assume that the system is homogeneous,so that all cells in the same hierarchical level are statisticallyidentical. In the equilibrium state, the overall system can beanalyzed by focusing on only one cell in each level [18]. Letus define anArea consisting of one macrocell overlaying

Manhattan-like microcells. The totaloffered load to theArea is given by

(12)

where is the offered load submitted to a single microcelland is the offered load submitted to the overlayingmacrocell. From the total offered load to anArea, a fraction

is sent to the microcell layer, and a fraction is sentto the macrocell. So we can write the offered load to a singlemicrocell and to the macrocell as follows:

(13)

where and . and couldbe understood as the probability that the speed of usersis, respectively, smaller or larger than a predefined speedthreshold to classify users asslow- or fast-moving users(notealso that , with defined in the previoussection). The following memoryless assumptions allow theproblem to be cast in the framework of a multidimensionalbirth and death process [18]. It should always be borne inmind that there are five types of service users in the systemunder consideration.

1) New call arrival processes offered to a given cell arePoisson processes for all five types of service. The meannew call arrival rates fromslow-moving usersto eachmicrocell for each type of service are

and , and the mean new call arrival rates fromfast-moving usersto each macrocell for each type ofservice are given by and .

2) The handoff call arrival processes due to the motion ofslow-moving usersin each of the microcells are alsoconsidered to be Poisson processes with mean handoffarrival rates and , respec-tively, for each type of service and in the macrocell layerwith mean handoff arrival rates

and , respectively. The handoff arrival pro-cesses due to the motion offast-moving usersin themacrocell layer are considered to be Poisson processeswith mean handoff arrival rates

and , respectively, for each type of service.3) The calls from eitherslow- or fast-moving usershave

an unencumbered call duration according to a negativeexponential distribution with parameter.

4) The cell dwell time is the time spent by a mobilestation (MS) in a cell independent of being engaged ina call, is a random variable approximated by a negativeexponential probability density function (pdf) (clearly,this assumption would not be valid forfast-moving users


Fig. 5. Possible state transitions diagram for the case where there are threeresources available for two different types of users. One of them requires oneresource for the communication, and the other one requires two (representedby the first and the second of the pair of numbers in each state, respectively).

in microcells, but is valid forslow-moving users). Forslow-moving userstraveling across a microcell, the celldwell time has a mean equal to and, while travellingin a macrocell, a mean equal to . On the other hand,fast-moving users travelling in a macrocell have a celldwell time with a mean equal to .

A. Microcell Layer

In a birth and death process, the probability of a uniquetransition in a very small interval of time is directly relatedto the interval duration, while the probability of a transitioninvolving more than one state in a very small interval oftime is equal to zero. This means that, in a given time, therecan only be a birth, a death, or the possibility of remainingin the same state. Hence, there can only be transitions toneighboring states. The case where a user needs more than onechannel to establish a connection cannot be solved directly bymatching a birth and a death with the occupation or release ofa single channel in the system, but with a transition to the stateproduced by the arrival or departure of a certain type of user.

The transitions produced by a birth or a death will be madeto a neighboring state when the user requires only one channel,but nonneighboring states will be reached when more thanone channel is required by the user. However, if we representthe number of users in the transition diagram instead of thenumber of busy channels in the system, an exact solution couldbe found as long as the number of users occupies less than orequal to the number of available channels. An example ofa state transition diagram is shown in Fig. 5 for the case oftwo types of users, with one and two channels required perconnection, respectively.

Since there are five types of service available in the system,is formed of five different streams of traffic produced by

slow-moving usersand therefore can be expressed as follows:

(14)

with

(15)

Every microcell has channels with guarded channelsneither for handoff calls nor for new calls of any typeof service. Clearly, this system should be solved by a-dimensional birth and death process whereis equal to the

number of services available—in this case five. The process fordrawing the equilibrium state equations is obviously a tedioustask, even when each type of user makes use of only onechannel per connection. However, since each type of useris independent and since each type of service is requestedindependently, the joint probability that there is a combinationof or fewer users of any type in the system, can be found asthe multiplication of each of the marginal distributions. Thatis, each marginal distribution is given by

! ! (16)

with , where is the number of servicesavailable. is a constant. Then, the probability that the systemis in the state is given by

(17)

remembering that

(18)

where to represent the number of channels requiredby each type of service. Then

(19)

to should be understood as the integer part of. Finally

(20)

subject to the very important condition:.

The blocking probability for each type of user is given bythe summation of all the valid states for which the condition

is fulfilled. This is

(21)

with . The overall blockingprobability in the microcell layer is then given by

(22)

Since there is no channel reservation, then the handofffailure probability for each type of user is equal to theblocking probability of each type of user, so . Thesame can be said about the overall handoff failure probability,hence, . Beginning with an initial guess of handoff


Fig. 6. Blocking probability for each type of service in an HCS with different types of users and different types of service using the methoddescribed in Section V.

arrival rates for each of the services available, the equationsare solved for the state probabilities, which are then used todetermine the average handoff departure rates for each typeof service in much the same way as described in [19]. A bigdifference here is that five types of handoff arrival rate arebeing considered in the calculation.

B. Overflow Traffic and Macrocell Layer

The composite overflow traffic is modeled by an IPPprocess. The way to determine the IPP parameters canbe found in [18]. represents the intensity of the modulatedPoisson process, is the mean on time, and is the meanoff time of the random switch that modulates the interruptedPoisson process. There are ten different streams overflowingfrom the microcell layer to the macrocell layer: there are newand handoff calls from five types of services produced byslow-moving usersonly. The on-time and off-time randomvariables follow a negative exponential probability densitydistribution. Once the parameters are determined, the statesequations can be easily formulated in the usual way [18].This process is rather complicated because in the macrocelllayer there are also fast-moving users. The interested readeris referred to [18].

A numerical example of the application of this analyticalmethod under uniform traffic is given in Fig. 6 when appliedto a system like the one described in previous sections. Theblocking probability for each type of user is plotted againstthe total offered load per microcell. As can be seen, there isclose agreement between the numerical and simulation results(dashed and solid lines, respectively).

VI. EXAMPLES AND DISCUSSION

Simulations results are generated for the FRA, FRAHR,and DDRA strategies. The purpose of these simulations isto compare the performances of FRA, FRAHR, and DDRAfor different types of service in an HCS. All the simulationresults are applicable to the FM1 proposal. Figs. 7–12 comparethe probability of new call (or data transmission) blockingfor the three strategies when plotted versus the offered loadper microcell. Fig. 7 shows a substantial reduction in theoverall new call blocking probability for the DDRA strategyin comparison with the other pair of strategies. DDRA clearlyoutperforms FRA and FRAHR in all the evaluation rangesin terms of the overall blocking probability for each type ofservice, as shown in Figs. 8–12. For all the cases shown, witha fixed blocking probability of 1%, the DDRA strategy alwayspresents an increase in capacity of at least 30% in comparisonwith the FRAHR strategy. With the use of a DRA rather thana FRA strategy, a smaller blocking probability is experiencedin the HCS for new calls from all types of service, both forslow- and for fast-moving users. It can also be noticed that,even when the offered load increases, the advantages that theFRAHR strategy has in comparison with the FRA strategyare almost constant. This means that the use of the HDPalways gives the FRA strategy capacity gain for all kind oftraffic loads (light and heavy). As Fig. 11 shows, the serviceat 144 Kbps has good performance, and the system can copewith it without major problems. However, the data serviceat 384 Kbps experiences a very high blocking probability(see Fig. 12) due to the limited number of resources in themicrocell layer and in the system in general. The system


Fig. 7. Overall blocking probability in the HCS.

Fig. 8. Blocking probability in the HCS for a user with a service of type 1 (speech).

requires the use of more resources to improve this service’sperformance. From this part of the evaluation it is observedthat the DDRA is the strategy that makes the best balancebetween channel usage in the macrocells and their role as

overflow channels. Sharing channels not only between cellsof the same hierarchy, but between cells of different layersmakes the DDRA strategy a powerful scheme to increasecapacity in the HCS.


Fig. 9. Blocking probability in the HCS for a user with a service of type 2 (data at 32 Kbps).

Fig. 10. Blocking probability in the HCS for a user with service of type 3 transmitting data at 64 Kbps.

On the other hand, Figs. 13–17 show and compare theforced-termination probability for the FRA, FRAHR, andDDRA strategies. Observe from Fig. 13 that a considerablereduction in the overall forced-termination probability of the

system is achieved when DDRA is used rather than a FRAbased strategy. Very small forced-termination probabilitiescan be expected for all types of service with a DRA, asshown in Figs. 14–17, where the total forced-termination


Fig. 11. Blocking probability in the HCS for a user with service of type 4 transmitting data at 144 Kbps.

Fig. 12. Blocking probability in the HCS for a user of type 5 transmitting data at 384 Kbps.


Fig. 13. Overall forced-termination probability.

probability for each type of service (except type 5) is plottedagainst the offered load in each microcell. As seen in thisset of graphs, the advantages of using the DDRA are keptthroughout the evaluation range, and the DDRA performsbetter than the FRA strategies. The HDP always helps theFRAHR strategy to decrease the number of dropped callsthroughout the evaluation range when compared to the FRAstrategy. The forced-termination probability experienced byfast-moving userswith service types 1–4, when the DDRAis being used in the system, is nil. Therefore, the forced-termination probability shown in Figs. 14–17 for the DDRAstrategy, is the one experienced byslow-moving usersonly.Due to a high blocking probability for the type of serviceof data transmission at 384 Kbps, the number of users en-gaged in a conversation in the HCS is very small, andthe forced-termination probability for this type of serviceis also nil.

From the system operator point of view, it is importantto know some figures that could affect the signaling loadof the system. The number of handoffs in the lifetime of aconversation is an important parameter to consider. Fig. 18shows the average number of handoffs necessary in the lifetimeof a conversation for aslow- and a fast-moving user(forany type of service). Clearly, under the DDRA strategy, thisnumber is large compared to the FRA and FRAHR strategies,but it is a clear tradeoff with the number of blocked calls,particularly in the case offast-moving users. Finally, in Fig. 19the average number of reallocations that a call suffers ispresented. Obviously, FRAHR has the smallest number ofreallocations, as compared to DDRA, nevertheless, it is worth

having this increase in the system signaling load because ofthe advantages already explained.

VII. CONCLUDING REMARKS

This paper has suggested a DRA strategy suitable foroperation in an HCS. This strategy makes use of channelreallocations, handup, and handdown to decrease the blockingprobability in the system as well as the forced-terminationprobability. The DDRA only requires that each BS has limitedinformation about resource usage, thus making this strategydistributed. This last feature decreases the communicationbetween BS and therefore the signaling load in the systemwhen compared with centralized DRA strategies. This strategy,have been successfully evaluated by means of simulationsusing a hybrid TDMA/CDMA system proposed for a futureUMTS. From the simulation results, we can say the following.

1) DDRA performs much better than the FRA strategy andeven better than the FRA strategy with HDP and channelreallocations.

2) The forced-termination probability (and therefore thenumber of dropped calls) in the HCS is significantlysmaller when using DDRA rather than FRA. This istrue for the types of service studied in this work and forall types of users.

3) Although service type 5 experiences a high blockingprobability, different figures for this parameter can beobtained by using more resources in the evaluationor a different number microcells that each macrocell


Fig. 14. Forced-termination probability for users with a service of type 1 in the HCS.

Fig. 15. Forced-termination probability for users of type 2 in the HCS.

overlays. This may change when using TDD (symmetricand asymmetric).

4) Basically, all types of service can be held in the HCSwith the system proposed by ETSI (FM1). However,


Fig. 16. Forced-termination probability for users with service of type 3 in the HCS.

Fig. 17. Forced-termination probability for users with a service of type 4 in the HCS.

if the ratios of users requesting each type of servicechanges, then results could vary considerably.

Implementation of DDRA is worthwhile because there areenough advantages in doing so, such as an increase of the


Fig. 18. Average number of handoffs an engaged call is expected to do during its lifetime. The dashed lines representfast-moving users, and solidlines representslow-moving users.

Fig. 19. Average number of channel reallocations an engaged call might experience during its lifetime in the HCS with different resource assignmentsschemes.


load that the system can carry to meet a specific quality ofservice, reduction of dropped calls, reduction of signaling loadcompared to other DCA strategies, etc.

Even when we have tried to consider as many parametersas possible in the evaluation to make simulations similar toreal case conditions, results may change in practical systemsdue to the limitations of the computer simulation model. Onlycircuit-switched services have been taken into account, whichis not a real case. For future work, packet-switched servicesmust also be considered, and their impact on the DDRA shouldbe investigated to add some new characteristics that can copewith these types of service.

REFERENCES

[1] FRAMES Multiple Access Scheme Proposal for the UMTS RadioInterface-SMG2 Workshop on UMTS Radio Interface Technologies,Sophia-Antiopolis, France, Dec. 16–20, 1996.

[2] F. D. Priscoliet al., “Application of dynamic channel allocation strate-gies to the GSM cellular network,”IEEE Trans. Veh. Technol., vol. 15,no. 8, pp. 1558–1567, 1997.

[3] I. Katzela and M. Naghshineh, “Channel assignment scheme for cellularmobile telecommunication systems: A comprehensive survey,”IEEEPersonal Commun., vol. 3, no. 3, pp. 10–31, 1996.

[4] M. Scheibenbogen, S. Clausen, and A. Guntsch, “Dynamical channelallocation in hierarchical cellular system,” inProc. IEEE VTC’96, pp.721–725.

[5] K.-R. Lo et al., “A combined channel assignment strategy in a hier-archical cellular systems,” inProc. IEEE ICUPC’97, San Diego, CA,Oct. 1997, pp. 651–655.

[6] A. Klein et al., “FRAMES multiple access mode 1-wideband TDMAwith and without spreading,” inProc. PIMRC’97, Helsinki, Finland, pp.37–41.

[7] L. Ortigoza-Guerrero and A. H. Aghvami, “On the optimum spectrumpartitioning in a microcell/macrocell cellular layout with overflow,” inProc. IEEE GLOBECOM’97, Nov. 1997, pp. 721–725.

[8] ETSI SMG2, Wideband TDMA/CDMA,Evaluation Rep.—Pt. 2, 1997.[9] B. Jabbari and W. F. Fuhrmann, “Teletraffic modeling and analysis

of flexible hierarchical cellular networks with speed sensitive handoffstrategy,” IEEE Trans. Veh. Technol., vol. 15, no. 8, pp. 1539–1548,1997.

[10] S. Elnoubi et al., “A new frequency channel assignment algorithmin high capacity mobile communication systems,”IEEE Trans. Veh.Technol., vol. VT-31, pp. 125–131, Aug. 1982.

[11] K. A. West and G. L. Stuber, “An aggressive dynamic channel as-signment strategy for a microcellular environment,”IEEE Trans. Veh.Technol., vol. 43, no. 4, pp. 1027–1038, 1994.

[12] M. Frullone et al., “Analysis of optimum resource management strate-gies in layered cellular structures,” inProc. ICUPC’94, pp. 371–375.

[13] “Universal mobile telecommunications systems (UMTS): Selection pro-cedures for the choice of radio transmission technologies of the UMTS,”Tech. Rep., Sophia-Antiopolis, France, 1997.

[14] M. Pizzarroso and J. Jimenez, “Common basis for evaluation of ATDMAand CODIT system concepts,” Rep. R2020/TDE/CA/DS/L/SIG5-1/al,1995.

[15] I. D. Hedouville and H. Azad, “Performance evaluation of multi-ratehybrid TDMA/CDMA systems,” Internal Rep., Center for Telecommu-nications Research, King’s College, London, U.K., Sept. 1997.

[16] J. Blanzet al., “Performance of a cellular hybrid C/TDMA mobile radiosystem applying joint detection and coherent receiver antenna diversity,”IEEE Trans. Veh. Technol., vol. VT-12, no. 4, pp. 125–131, 1982.

[17] S. Nanda and D. J. Goodman, “Dynamic resource aquisition: Distributedcarrier allocation for TDMA cellular systems,” inThird GenerationWireless Information Networks. New York: Kluwer, 1992, pp. 92–124.

[18] L.-R. Hu and S. S. Rappaport, “Personal communication system usingmultiple hierarchical cellular overlays,”IEEE Trans. Veh. Technol., vol.13, no. 2, pp. 406–415, 1995.

[19] Y. B. Lin, A. R. Noerpel, and D. J. Harasty, “The sub-rating channelassignment strategy for PCS handcuffs,”IEEE Trans. Veh. Technol., vol.45, no. 1, pp. 122–130, 1996.

Lauro Ortigoza-Guerrero (S’96) received theB.Sc. degree in electronic and communicationsengineering in 1993 and the M.Sc. degree inelectrical engineering in 1996, both from theNational Polytechnic Institute, Mexico City,Mexico. He is currently working toward thePh.D. degree at the Center for TelecommunicationResearch, King’s College, London, U.K.

His area of interest is mobile cellular commu-nication networks with a concentration in dynamicchannel allocation strategies in HCS’s with hybrid

multiple-access schemes for UMTS, traffic analysis, and performanceevaluation.

A. Hamid Aghvami (M’87–SM’91), for a photograph and biography, seethis issue, p. 1161.

Documents

A distributed dynamic resource allocation for a hybrid TDMA/CDMA system