578 IEEE TRANSACTIONS ON WIRELESS … Multi-Carrier Traffic ... (WCDMA) access technology ... Purpose of current research is to find a generalized solution to multi-band traffic

578 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2006

A Class of Call-Fail-Safe and Distribution-EffectiveMulti-Band Multi-Carrier Traffic Allocation

Methods for 3GB Wireless SystemsParthasarathy Guturu, Senior Member, IEEE, and Abdennaceur Lachtar

Abstract— For better capacity and higher availability, presentday third generation (3G) wireless systems based on the CodeDivision Multiple Access(CDMA) technology are evolving tooperate on multiple carriers (frequencies) spread over multi-ple bands. In order to provide better quality of service, the3GB (3rd Generation and Beyond) systems need to distributecalls equitably to different carriers on different base-stationsaccessible to the mobiles irrespective of the bands or carrierson which those mobiles initiated their calls. However, there isa risk of call failure when a call originated on a carrier ina band is migrated to another carrier in a different band,particularly because of the differences in the radio coverageof the base-stations operating in different bands. This paperpresents a class of methods that offer equal robustness againstcall failures and varying degrees of call distribution effectiveness.For call distribution, these methods employ an enhanced carriercapacity measure (ECM) proposed in this paper. ECM augmentsthe gross capacities of the carriers (to house calls) with pre-configured biases specific to the mobile users. We develop herean intuitively appealing distribution-effectiveness measure basedon the ECM for comparing the methods. Relative performancesof the proposed methods with respect to call failure rate anddistribution effectiveness are established by means of simulationresults for calls originating anywhere in the cell coverage areaas well as calls originating exclusively near the cell boundaries.The latter results help to study the effect of mobility on theperformances of the algorithms.

Index Terms— Code Division Multiple Access (CDMA), CDMA2000, 3GB systems, Wideband CDMA (WCDMA), UMTS, 1x-RTT, 1xEVDO, load balancing, multi-band traffic distribution,multi-carrier systems, call-fail-safe algorithms, wireless trafficdistribution, call failure robustness, call distribution effectivenessmeasure.

I. INTRODUCTION

W ITH ever increasing demand for wireless voice anddata services, the wireless systems of the third gen-

eration and beyond (3GB) are required to implement highcapacity solutions. An important aspect of the capacity man-agement is effective usage of the air interface. In the 3GBCode Division Multiple Access (CDMA) systems, this is

Manuscript received September 15, 2003; revised September 16, 2004and November 24, 2004; accepted December 23, 2004. The associate editorcoordinating the review of this paper and approving it for publication was G.Cao.

Parthasarathy (Partha) Guturu is with the Department of Electrical Engi-neering, College of Engineering, University of North Texas, P.O. Box 310470,Denton, Texas 76203-0470 USA (email: [email protected]).

Abdennaceur Lachtar is with the Core RF Engineering Department, NortelNetworks Corporation, 2201 Lakeside Boulevard, Richardson, TX 75082 USA(email: [email protected]).

Digital Object Identifier 10.1109/TWC.2006.03015

achieved by implementing efficacious load balancing methodsthat distribute call traffic by dynamically selecting optimal(usually, least loaded) carriers (frequencies) for call setup froma pool of carriers in the spectrum allocated to the serviceprovider owning the wireless network. As demand for capacityincreases, service providers must necessarily acquire largerand larger spectra, but difficulties in procuring a large enoughspectrum in a single band force them to operate in spectraallocated to them in different bands. The latest band-classspecification for CDMA 2000 spread spectrum systems [1]provides for up to 13 bands of operation. Some of the currentlydeployed 1xRTT (1 times Radio Transmission Technology)systems based on CDMA 2000 standard [2], employ the state-of-the-art dual band (with 800 and the 1900 MHZ bands)multi-carrier access technology. Efforts are in progress toestablish in some countries tri-band systems that operate onan additional 450 MHZ band. In the evolving 1xEVDO (1xEvolution Data Only, also called High Rate Packet Data) [3]systems supporting high data rates (with average speeds of300-500 kbps and infrequent burst speeds beyond 2.4 mbps),and the CDMA 1xEVDV (1x Evolution Data and Voice) [4]systems, much higher bandwidth and quality of service (QOS)requirements enforce them to implement optimal algorithmsfor effective utilization of the available pool of carriers fromdifferent bands.

While CDMA based technologies are gaining ground inNorth America, Universal Mobile Telephone Systems (UMTS)[5] based on the IMT-2000 standard [6] , are gaining popu-larity, for this kind of high speed 3GB wireless applications,in Europe and some other parts of the world. UMTS systemsuse the Wide-band CDMA (WCDMA) access technology andoperate in the 1.9 and 2.1 GHZ bands. Though multi-bandmulti-carrier functionality is not implemented in the currentlydeployed UMTS systems, their resource utilization can bedefinitely toned up by using the traffic distribution algorithmsas proposed in this paper.

Effective utilization of a larger carrier pool of carriersavailable from multiple bands facilitates working out of thehigh capacity solutions with acceptable QOS. At the sametime, the diversity of the carrier pool brings in the technicalcomplications associated with the differences in radio cover-age characteristics of different bands due to frequency band-dependent fading effects. On the business front, higher costsof sophisticated radio circuitry required for handling multiplefrequency bands in the mobile station (MS) and the base

1536-1276/06$20.00 c© 2006 IEEE

GUTURU and LACHTAR: CLASS OF CALL-FAIL-SAFE AND MULTI-BAND MULTI-CARRIER TRAFFIC ALLOCATION METHODS 579

station (BS) need to be addressed, but this is an inevitableconsequence of availability of large enough spectrum only inchunks spread over multiple bands.

Efficacious methods for wireless traffic distribution ontocarriers from the above mentioned heterogeneous (multi-band) pool need to address both distribution effectivenessand robustness against call failures without sacrificing quality.Increased capacity can be achieved by setting up a number ofcalls with different access codes on each carrier in the pool.However, as indicated by Nishith and Sarvesh [7], forwardlink power decreases with the number of calls on a carrierand thereby affects the quality of transmission (quantifiedby frame error rate) for voice/data. A. He [8] demonstratedwith the help of simulation results that distribution algorithmsbased on Walsh code or power usage on the carriers farebetter in terms of the utilization of air-interface capacitythan those based on the round robin policy. Thus it is clearfrom these results that, in order to ensure a good qualityof service (QOS) while addressing system capacity issues,the number of calls on individual carriers must necessarilybe limited. The number of calls that can be supported on acarrier without any perceptible deterioration in quality may beconsidered as a gross measure of carrier capacity. In this paper,we present an enhanced capacity measure that incorporatesvarious customer defined constraints as well. A good calltraffic distribution method must necessarily perform dynamicload balancing on the carriers in the pool by taking intoconsideration the capacities (quantified by enhanced capacitymeasures) of individual carriers available for call setup.

In order to support fault tolerance and high availability,present day wireless networks maintain cell sites containingmultiple Base-station Transceiver Systems (BTSes) each usinga number of carriers. The set of co-located BTSes at a siteis termed as a 1-1 overlay system. While initiating a call,a mobile station (MS) tunes to a carrier determined by itsinternal hashing algorithm and gets access to a cell site whoseradio coverage area includes the mobile. Now this call canbe seamlessly migrated within the cell site to some othercarrier with least load (i.e. maximum available capacity). Apatented [9] algorithm is available for such dynamic loadbalancing and is known as the Multi-carrier Traffic Allocation(MCTA) algorithm. The superiority of the MCTA algorithmover the random carrier assignment (RCA) algorithm, in termsof enhanced capacity, lesser power requirements per user andreduction of call blocking probability has been establishedby Nishith and Sarvesh [7] by means of simulation resultsfor IS-95 based mobility and fixed wireless networks andIS-2000 based mobility networks. Harris et al [10] presenta distribution algorithm based on call hold time estimatesfor individual carriers and demonstrate with theoretical andsimulation results its advantage in terms of spectral efficiency.However, these algorithms consider only a single band multi-carrier situation. They do not consider the differences in radiocoverage characteristics of different bands and consequent callfailure risks associated with call distribution across bands. Inthis paper, we present an innovative method (with a pending

patent application 1 [11]) that provides distribution efficiencywhile addressing these issues. This method is a refined andmore generalized version of our dual band traffic allocationalgorithm presented by us in [12]. It is applicable to morethan two bands.

Rest of this paper is organized as follows. Section II con-tains an exposition of the problem associated with migrationof a call originated on a carrier in a band to another carrierin a different band and our general solution to circumventthis problem. In Section III, we present an enhanced carriercapacity measure for the purpose of comparing and choosingthe best carrier for call set-up on the basis of the capac-ity information of the individual carriers as well as somecustomer-configured load-balancing criteria. In Section IV,we utilize this capacity measure and develop an intuitivelyappealing load distribution effectiveness measure. Section Vcontains a formal presentation of a number of algorithms forcall allocation across bands and their implementation within acall processing framework. We establish, with the help of thesimulation results in Section VI, the relative performances ofthe proposed algorithms in terms of call failure robustness anddistribution effectiveness. This section also presents simulationresults showing the combined effect of mobility and mobiledensity in the cell, on the performances of the algorithms.Summary and conclusions are presented in Section VII.

II. PROBLEM STATEMENT AND OVERVIEW OFTHE SOLUTION

Purpose of current research is to find a generalized solutionto multi-band traffic allocation problem that allocates eachcalls initiated, by origination or termination or multi-modehard handoff, on a carrier (called access carrier) of a cell siteto a possibly different carrier of the same or different band. Anaive extension of the single band traffic distribution strategysuch as the MCTA to the multi-band situation is to migrateeach call dynamically from its access carrier to a carrier withbest available capacity among all the carriers used in the accessBTS as well as its neighbors, irrespective of the band of thecarrier. However, this approach has some practical difficulties.First, the cell planning in different bands has been doneindependently. In future also, it should be done in the sameway based on the diversities in radio coverage characteristicsof the bands of operation. Thus, it is not always possible havethe 1-1 overlay systems of different bands deployed at thesame location. Next, even if the BTSes supporting differentbands are co-located, their coverage areas differ dependingupon the bands being supported. Hence, the position of anmobile station (MS) in the range of a cell site operating in aband with higher coverage can be guaranteed to be coveredonly by multiple cell sites (so called non 1-1 overlay systemof BTSes) operating in a different band of lesser coverage. Forexample, it is known that three or more spatially disjoint BTScell sites using 1900 MHz carriers are required for coveringthe radio coverage area of an 800 MHz BTS cell site. It isalso expected that a 450 MHz BTS can cover 1.5 to 2 times asmuch as the area covered by an 800 MHz BTS. In Fig. 1, we

1Anybody interested in licensing this patent may email to [email protected].


Fig. 1. An example depicting differences in the coverage areas of differentbands. (Cell sites with higher coverage are shown with larger size icons withthicker and darker line types and marked with letters having larger font size.Coverage area of A is shown to be covered by cell sites B (with hatched linecoverage boundary) and C (with dotted line coverage boundary) operating ina different band. B, in turn, is covered by D, E and F. Similarly, C is coveredby G, H and I. All cell sites D though I operate in yet another band. MS canaccess the cell sites A, B and D only).

depict an example scenario based on this information. Here,the coverage area of a co-located system (A) of BTSes of the450 MHz band is minimally covered by 2 spatially disjointco-located systems (B and C) of the 800 MHz band. Thecoverage area of B, in turn, is covered by spatially disjointco-located systems D, E and F operating in the 1900 MHzband. C is similarly covered by the 1900 MHz cell sites G, Hand I. In this multi-band cell deployment scenario, if an MS,at any time, initiates a call on a carrier used in the cell siteA, it only implies that is located at that time anywhere in thecoverage area of A (enclosed with thick and bold boundarylines in the figure). However, for its position shown in thefigure, only one cell site in each band, namely A in the 450MHz, B in the 800 MHz and D in the 1900MHz bands, isaccessible to it for call set up. Using the simplistic extensionof MCTA for load distribution, we may now choose to migratethe call to the carrier with the highest available capacity amongall the carriers of the sites A through I because they are thecell sites covering A in different bands, but it could result ina call failure in a typical cell site loading situation such asfollows: Cell site I has the highest capacity and hence the callis redirected to that band. Assuming that mobile can operatein all the three bands, mobile initiates the call in the 1900MHz band, but cannot get access to cell site I because it isnowhere in its range. It can, however, get access to the cellsite D operating in the same 1900 MHz band and the cellsite D does not have sufficient resources to support the call.Hence the call fails. This kind of situation could occur quitefrequently when the wireless network is overloaded.

From the above example, it is clear that all the carriers ofdifferent bands with diverse coverage characteristics cannotbe put on equal footing as in the MCTA algorithm whenthe system is required to support Multi-Band Traffic Allo-cation (MBTA). Intuitively, any multi-band traffic distributionalgorithm seeking to achieve robustness against call failuresshould give due consideration to the following elements of the

problem solution:

1) If the mobile does not have the capability to operatein a band (as per an earlier provisioned informationor the information elicited from the mobile during thecall flow before the application of the traffic distributionalgorithm), all the cell sites (and the carriers) of thatband should be eliminated from consideration.

2) If an MS is within the range of a cell site, the best carrierof that co-located system can be allocated to it for callset up. Hence, it is necessary to make a cell-site-wisepartitioning of all the carriers, considering all the cellsites (of different bands) that cover the range of the cellsite of call access and find the best carrier of each cellsite.

3) In case multiple cell sites of a different band cover therange of the cell site operating in a band on which acall initiates, the worst carrier among the best carriersof the cell sites of that different band should be found.If that carrier does not have the capacity for call setup,call redirection to that band should be precluded. Thisis because, in the worst case scenario, mobile could beso positioned as to access in that band, only that cellsite, none of whose carriers have capacity and the callwould eventually fail. Thus, this step ensures robustnessagainst call failures.

4) After elimination of certain bands using steps 1 and 3above, the best carriers of different cell sites of differentbands may be used for determination of the band forcall set up. Carrier for call setup is also determinedif it is decided to retain the call in the band of callaccess (henceforth, called as the inband). In case thedetermined band happens to be a band other than theinband, mobile is required to re-initiate the call in theother band and a re-run of the algorithm will determinethe carrier for call set up in that band. It will be thebest carrier of that cell site accessed by the mobile onre-initiation.

5) Information whether a call is initiated or re-initiatedmust be available to the algorithm from call flow mes-saging. On re-initiation, the algorithm should eliminateall other bands except the band of call re-initiation fromconsideration. This is to prevent prolonged call setuptimes and call failure possibilities due to repeated re-initiations that can occur when load in different bandsswings in the call setup time window.

Many algorithms can be designed with above components ofproblem solution, particularly based on how the best carriersof different cell sites can be utilized. In Section IV, we presenta number of algorithms that incorporate these considerationsfor call-fail-safeness along with an algorithm that focusesexclusively only on distribution efficiency.

III. AN ENHANCED CARRIER CAPACITYMEASURE FOR CARRIER-TO-CARRIER

COMPARISON

As indicated in Section I, it is required to limit the numbersof calls on a carrier. Hence, number of new calls (in additionto existing ones) a carrier can house at any point in time may


ND−Bit SC−Bit RC−Bit RB−Bit BPR−Bits CPR−Bits Capacity Magnitude Bits CAC−Bit IBC−Bit

Fig. 2. Bit pattern defining individual measurements of the EnhancedCapacity Measure.

be considered as a gross measure of its available capacity atthat time. In this section, we present an Enhanced CapacityMeasure (ECM) that incorporates many other factors such asindividual carrier capabilities to support calls without down-grading (from 3G-quality to 2G-quality) and preconfiguredbiases on them. These factors or attributes of ECM canbe ranked and carrier-to-carrier comparison can be done bymatching the corresponding carrier attributes one by one in thedescending order of their rank till there is no tie or all attributeshave been considered. Mathematically, ECM can be computedby the function

∑i wiai where wi is the weight of the i-th

attribute ai so chosen that a higher value of this attribute fora carrier always results in a higher value of ECM no matterwhat the values of the lower order attributes are. As long as theweights are so chosen to satisfy this condition, actual valuesof the weights do not matter since a unique ordering of thecarriers based on the attribute values can be achieved. Oneapproach to choose such weights is to arrange the attributesin the order of their rank in a computer word so that eachweight wi automatically turns out to be 2p where p is the bit-position of the i-th attribute in the word and ECM is given by∑

i 2pai. A typical ECM measure constructed with differentcapacity attributes constituting the bit-pattern of a capacityword is shown in Fig. 2. As indicated above, higher orderbits of this word characterize attributes of higher importance.This ensures that only when there are ties on the higher ordercharacteristics, progressively lower order attributes will beinfluential in tilting balance towards one of two carriers undercomparison. Various attributes that need be considered whileallocating a call to a carrier and their significance may beunderstood from the following description of the individualbits of the capacity word:

• First of all, when gross available capacity (measured interms of calls of the incoming call type that can befurther supported) of a carrier is zero, call cannot be setupirrespective of the preconfigured biases in favor of thiscarrier and hence all the bits must be set to zero.

• ND-Bit (Non-Downgraded-status bit): A carrier is said tobe downgraded if it has only 2G resources to support anincoming 3G-Voice call. This bit is set 0 for such carriersand to 1 for others. In case of a data calls or a 2G voicecall, downgraded mode of operation does not make senseand hence this bit is set on if the carrier at all has capacityto support a call of the corresponding type. Supportingcalls without downgrading them is of utmost importanceand accordingly this bit is made the highest ordered bit.

• SC-Bit (Spare Capacity bit): A part of the total capacity(i.e. the capacity with no calls on) of a carrier could bereserved for soft-handoff calls and may not be availablefor call distribution unless a call failure is imminent.Different carriers can be configured to have differentamounts of reserved capacity and the reservations canbe implemented using thresholds preconfigured by the

service provider. In this reservation scheme, only carri-ers with spare capacity over and above their respectivethreshold capacities are considered for load distributionunless none have spare capacity. Hence, a carrier withspare capacity is considered superior to another onewithout the same even if the latter has more availablecapacity. The SC bit is set to 1 for such spare capacitycarriers and zero for others.

• RC-Bit (Retain Carrier bit): This bit implements a biasor preference preconfigured on a carrier by the serviceprovider to accommodate a mobile users preference fornot migrating the call to another carrier and therebyincurring the risk of call failure or longer call setuptime This can be done because the access carrier forcall setup is a carrier selected by the hashing functionin a mobile users MS. If this access carrier has beenpreconfigured with retain carrier preference, migration toanother carrier is not attempted unless the other carrieris more dominant in terms of either non-downgradedor spare capacity status. Hence, this RC-bit is set to 0usually, but set to 1, if all of the following conditions aretrue: i) Retain carrier criterion is configured on the mobileuser’s access carrier. ii) Carrier under consideration is anaccess carrier.

• RB (Retain Band): This bit is similar to the retain carrierbit, but is related to a mobile users preference to retainthe call in inband for identical reasons. It is set to 0usually, but set to 1, if all of the following conditions aretrue: i) Retain band criterion is configured on the mobileuser’s access carrier ii) Carrier under consideration isan inband carrier. The retain carrier and retain bandpreferences can be set up independently of each other.In case both are configured on a carrier, retain carrierpreference dominates the other.

• BPR-Bits (Band preference bits): These bits encode thelevel of preference for loading a carrier of a particularband. In case all other higher priority bits are same fortwo carriers, a carrier of a higher preference band ischosen between two carriers. If SC bit is zero, these bitscan be set to zero (the lowest possible value) in case itis required to enforce that carriers with no spare capacityare treated on an equal footing.

• CPR-Bits (Carrier preference bits): These bits encode thelevel of preference for loading a carrier relative to carriersof the same band or another band of equal preference. Away to use these preferences effectively is to configuremore privileged users access carriers with lesser prefer-ence values so as to enforce lesser encroachment on theirresources for traffic distribution purposes. Like BPR bits,CPR bits may also be set to 0 if a carrier does not havespare capacity.

• Magnitude Bits: These bits set to magnitude of sparecapacity, if the carrier has spare capacity, otherwise, to themagnitude of available capacity. This helps to comparetwo carriers equal in terms of spare capacity status andother aspects considered earlier, using the relative mag-nitudes of their spare capacities. Similarly, the carrierswith no spare capacity could also be compared.

• CAC-bit (Call Access Carrier bit): This bit is set only for


the access carrier so that it may be chosen in preferenceto another carrier which is equal in all respects.

• IBC-bit (Inband Carrier Bit): This is set only on theinband carriers so as to resolve ties between them andother band carriers in favor of them.

Here the capacity attributes are weighted based on theirunique position in the computer word. This ensures that theECM is computed in a unique way for a set of attribute valuesand thereby facilitates an invariant assessment of relativeperformances of various algorithms that distribute calls amongdifferent cell sites based on the following call distributioneffectiveness measure employing the ECM-values of the cor-responding cell-site carriers.

IV. A CALL DISTRIBUTION EFFECTIVENESS MEASURE

In a single band situation, if an MS accesses a cell site, anideal distribution scheme would choose the carrier with thehighest ECM in that site for call setup in order to achieve loadbalancing as per requirements stipulated in the Section III.Thus, the ECM of a cell site may be equated to that of itsbest carrier. In a multi-band situation, there will be multiplecell sites operating in multiple bands that provide, in theirrespective bands, minimal coverage for the geographic areacovered by the call access cell site. Let this total set of cellsites be denoted by T and the subset of cell sites amongthem that belong to band i by Bi. Thus T is partitionedinto mutually exclusive subsets of sites belonging to differentbands. Multi-band traffic distribution algorithms are requiredto perform load balancing among these heterogeneous groupsof sites in T utilizing two distribution criteria- i) how well thecalls are distributed between bands ii) how well the calls aredistributed among the sites in each band. The first criterionmay be termed as the inter-band distribution criterion and thesecond one, the intra-band distribution criterion. For evaluatingthe first criterion, we need to compute the ECM of each band.This can be computed as the mean value of the ECMs of thecell sites of that band as follows:

ECM(Bi) =1

n(Bi)

∑Sj∈Bi⊆T

ECM(Sj) (1)

where n(Bi) is the cardinality of the set Bi and Sj is acell site whose ECM , as discussed above, is given by themaximum value among all its carriers (such as Ck):

ECM(Sj) = maxCk∈Sj

ECM(Ck) (2)

We can also define ECM of T as follows:

ECM(T ) =1|T |

∑Bi⊆T

ECM(Bi) (3)

where |T | is the number of bands in T . Now, the inter-banddistribution efficiency may be defined by the between-banddispersion measure

DB =1

|T |.ECM(T )

√ ∑Bi⊆T

‖ ECM(Bi) − ECM(T ) ‖2

(4)Naturally, lesser the value of DB , better the ECM distributionacross bands. ECM(T ), the average capacity across bands,

is used in the denominator of the above equation as a normal-ization factor that makes the dispersion measure invariant tothe level capacity in the network. Since zero value ECM(T )implies that no site in T has capacity, DB is defined to be zeroin that case. The within-band dispersion measure DW maybe defined, as an average of ECM dispersions in individualbands, as follows:

DW =1

n(T )

√√√√ ∑Bi⊆T

∑Sj⊆Bi

∥∥∥∥ECM(Sj) − ECM(Bi)ECM(Bi)

∥∥∥∥2

(5)Here, n(T ) is the cardinality of T . Once again, the bandaverage ECM(Bi) is a normalization factor that is used formaking the dispersion within that band invariant to the levelof capacity in that band. If ECM(Bi) is zero, the innersummation for that band is also considered to be zero sinceall sites in that band have uniformly zero capacity.

Now, it is possible to define an overall dispersion measureD as a weighted sum of these dispersion measures as follows:

D = (W1.DB + W2.DW )/(W1 + W2) (6)

W1 and W2 in this equation are pre-selected weights basedon the importance of each type of dispersion in definingthe q.uality of distribution. Lesser the value of D, better thedistribution. Since ideal loading balancing implies equal loaddistribution among all sites irrespective of their bands, weuse the D-measure with W1 = W2 = 1 in our simulationexperiment to compare the different algorithms suggested inthe following section.

V. CALL ALLOCATION ACROSS BANDS-ALGORITHMS AND IMPLEMENTATION DETAILS

A. Theoretical Formulation of the Algorithms

In this section, we follow the same notation as above. Fur-ther, since the sets Bi and Bj are subsets of sites belonging todifferent bands, it is assumed throughout that Bi

⋂Bj = ∅.

Now, if Band(Bi) denotes the band to which the cell sitesin Bi belong, a simplistic approach to Multi-Band TrafficAllocation (MBTA) may be stated by the rule:

NewcallAssignTo

=⇒Band(Bi)

∣∣∣∣∀Bi,Bj⊆T

[ECM(Bi) � ECM(Bj)∧

ECM(Bi) �= 0

](7)

If no band has capacity, the above condition fails for all Bi

and hence the call is failed immediately. This algorithm maybe termed as an Optimistic MBTA as it considers it to beunlikely that a call would fail because of mobile’s position inthe exclusive range of a zero capacity site in the best capacityband. The general cross-band call allocation strategy presentedin Section II, on the other hand, does not share this optimism.It offers robustness against failures by disallowing migrationof calls to a band including a zero-capacity site. This is calleda min-max criterion for a band because it computes, using (2),capacity of a site as that of its best carrier and checks whetherthe worst site in a band has zero capacity by computing:

MinMaxECM(Bi) = minSj∈Bi

ECM(Sj) (8)


Since only one cell site is required to be considered in the bandof call initiation, the inband best carrier may be considered asthe min-max carrier of that band and the min-max criterion canbe applied uniformly to all bands. Within the framework of thiscall-fail-safe strategy, a number of min-max MBTA algorithmscan be constructed. In [12], we proposed, in the context of adual band situation, a conservative algorithm which considersthe other band only if the inband has no capacity. An extensionof this method to a multi-band problem would result in theFirst-Fit MBTA algorithm which considers the bands in a pre-determined order starting with the inband first and assigns thenew call to the first band with non-zero MinMaxECM ascomputed by (8) and applies the optimistic algorithm if nosuch band is found. Another min-max MBTA algorithm is anextension of the dual band min-max algorithm proposed byus in [12]. This generalized min-max MBTA algorithm firstcomputes the ECM of best min-max carrier using

MaxMinMaxECM(T ) = maxBi⊆T

MinMaxECM(Bi)

(9)and applies the rule:

NewcallAssignTo

=⇒ Band(Bi)

∣∣∣∣∣∣[MinMaxECM(Bi) =MaxMinMaxECM(T )]∧

[MinMaxECM(Bi) �= 0](10)

Since this algorithm looks for the best min-max band, thisalgorithm may be termed as the Best-Fit Min-Max algorithm.In case MaxMinMaxECM(T ) = 0, the above rule willnot be applicable to any band. Hence this algorithm, like allthe min-max algorithms to be discussed in the sequel, resortsto the optimistic algorithm to save the call from prematurefailure in the event of favorable positioning (near non-zerocapacity site) of the mobile. In general, all min-max algorithmsconstruct a subset E (of T ) with groups of sites in a band thatneed be eliminated (because of risk of call failure) as follows:

E =

⎧⎨⎩⋃

Bi|MinMaxECM(Bi) = 0iff MaxMinMaxECM(T ) �= 0

∅, Otherwise(11)

When E = ∅, it implies that all min-max carriers of differentbands under consideration are of zero capacity and hencethe optimistic algorithm need be applied. Motivated by thedistribution measure developed above, we propose hereintwo more MBTA algorithms with min-max flavor. The firstalgorithm, based on the intuitive consideration that allocationof the call to a band with best capacity will at least contributeto reduction of the inter-band dispersion, applies the rule:

NewcallAssignTo

=⇒ Band(Bi)

∣∣∣∣∣∣[ECM(Bi) =max∀Bi,Bj⊆T −E ECM(Bj)]∧

[ECM(Bi) �= 0](12)

We term this as the Best-Band Min-Max MBTA algorithm. Itmay be noted that the above rule subsumes the rule given by(7) for the optimistic algorithm. Non-applicability of this ruleimplies that none of the bands have capacity. Call must befailed immediately in this case. The second new algorithmproposed here assumes that a call can be assigned withequal probability to any site in a band and computes post-call assignment band dispersion measures, for different bands

satisfying the min-max criterion, as averages of the post-call assignment dispersion measures of their respective sites.Then it assigns the call to the band with minimal value of socomputed dispersion measure. In essence, it applies the rule:

NewcallAssignTo

=⇒ Band(Bi)∣∣∣∀Bi,Bj�ED(Bi) � D(Bj)

(13)where

D(Bi) =1

n(Bi)

∑Sk∈Bi⊆T

D(Sk) (14)

Here, D(Sk) is computed after a hypothetical assignment ofthe call to the site Sk. For computation of the same for a differ-ent site, original load distribution is restored and the measureis again computed after assigning the call to the new site in thatloading condition. In situations where no band has capacityor no min-max ECM is non-zero, it behaves exactly like theother min-max algorithms. For obvious reasons, this algorithmis termed as the Best-Distribution Min-Max MBTA algorithm.Because of the safeguard against possible call failures, allmin-max algorithms are expected to perform identically withrespect to call failures though they are expected to exhibitvariations in their distribution efficiencies. Further, because ofthe exhaustive search for the best possible distribution effectivecall assignment, the Best-Distribution Min-Max algorithm iscomputationally intensive.

All the above algorithms can be assessed for call failurerates using a simulation experiment. However, with respect todistribution efficiency, they need a baseline ideally-distributiveMBTA (BIDMBTA) algorithm to compare with. We define thedistribution efficiency of such an algorithm as follows:

DBIDMBTA = minSk∈T

D(Sk) (15)

This hypothetical algorithm computes the D-measure for Tafter assigning the new call to a site irrespective whether themobile can actually access it. In case call cannot be assignedto that site due to zero capacity, it returns a very high value forthe site. The minimum value among the D-values so computedusing assignments to different sites will be the D-value for thealgorithm. This value should be smaller than a value that canbe obtained by call assignment to a band (in other words, toa site accessible to the mobile in the band assigned) using apractical algorithm.

B. Implementation Details

We present in Fig. 3 a sequence diagram depicting thesequence of inter-system interactions leading to invocation ofthe MBTA algorithm at BSC and in Fig. 4, the inter-systeminteractions involved in the implementation of the algorithm.A call may be initiated in two possible ways. In mobileoriginated calls, the BTS accessed by the Mobile Station (MS)will pass on the call initiation (origination) request to mobileswitching center (MSC). In case of mobile terminated calls,an MSC will, in response to call initiation from a land linephone or a mobile at the other end, will page the mobile (via aBTS) which will respond indicating its readiness for call setupthrough that BTS. In either case, the MSC triggers a call setup action on the BSC connected to that BTS and providesthe cell site identification and access carrier information. The


Fig. 3. Sequence diagram of inter-system interactions leading to MBTAalgorithm invocation at BSC, from either mobile originated or terminatedcalls.

TABLE I

A TYPICAL CELL SITE DATA TABLE. (For clarity’s sake, the information

for each one of the individual BTSes in the cell site is not shown. It should

contain the BTS address and a list containing the information regarding

different carriers supported by the BTS. Individual carrier information, in

turn, includes its priority, reserved capacity threshold and

retain-carrier/band preferences).

Cell Cell Site Data

Site Current Cell Site Data Other band cell

No. sites covering the

original site range

Band Band BTS Data 450 800 1900

Pref. BTS1 . . . BTS-M

1 450 0 . . . . . . . . . - 3,5 2,4,6

2 1900 2 . . . . . . . . . 1 3 -

3 800 1 . . . . . . . . . 1 - 2,4

4 1900 2 . . . . . . . . . 1 5 -

5 800 1 . . . . . . . . . 1 - 2,6

6 1900 2 . . . . . . . . . 1 3 -...

......

......

......

......

BSC uses this information to index into a pre-configured cellsite information table such as the one depicted in Table Iand obtains there from the access carrier specific information(e.g. retain carrier/band criteria). It also obtains the prioritiesand reservation thresholds for all the carriers used in differentcell sites of T . Using the addresses of different BTSes inT availed from the table, it then sends, as shown in Fig. 4,the capacity requests to different BTSes and, based on theinformation provided in their responses about different carriersin conjunction with the access and other carrier informationprocured earlier, computes, as discussed in Section III, theECMs for their carriers and finds out the best carrier of eachcell site by comparing corresponding cell site carriers. Thisinformation regarding best carriers of different cell sites issufficient for decision making for call allocation to a bandusing any of the above discussed algorithms and subsequentcall setup or re-origination (re-initiation) as indicated in thesequence diagram of Fig. 4.

Fig. 4. Sequence diagram of inter-system interactions involved in theimplementation of the MBTA algorithms.

C. Order of Complexity of the Algorithms

In all of the proposed MBTA algorithms, different BTSesoperate in parallel to process their capacity requests, determinetheir respective best carriers and send responses. At the BSCend, ECM computations are done as and when capacityresponses are received and individual site best carriers areupdated in an incremental mode. Determination of the ECMof a carrier involves negligible processing effort requiredfor setting individual bits of a computer word. Differentalgorithms determine the best band for call setup by usingdifferently the information regarding the best carriers of indi-vidual sites and min-max carriers of individual bands. Most ofthis computational effort, except in case of the best-distributionalgorithm, involves finding the largest or the smallest of afew numbers and hence is negligible. The best distributionalgorithm involves an additional effort for computing disper-sion measures for different possible call allocations. In all thealgorithms, the worst-case time is dominated by the configuredtime (MBTA timer value) for forceful convergence of all thecell site best carrier determination processes and MBTA timeris usually set to a value slightly higher than the capacityrequest and response transit time for the farthest BTS. Thus,order of the worst case complexity is a constant for all thealgorithms.

D. Considerations for Efficient Implementation

All the MBTA algorithms, prior to deciding the band (orcarrier in case of inband decision) for call setup, need todetermine the best carriers of individual cell sites in T . Ata point of time, the incremental Best Carrier DeterminationProcess (BCDP) of a cell site is said to be convergent if thebest carrier of that site, as assessed from the BTS responsesreceived till then, is determined to be unequivocally the best(based on the way ECM is defined) irrespective of other carrier


capacity information that would be available from any pendingBTS responses. There are also two extreme cases of BCDPconvergence- i) Trivial convergence- that occurs only when allthe responses from the BTSes of the site have been processed.ii) Forceful convergence- that takes place MBTA timer expiresbefore some BTS responses are received. In the latter case,it is assumed that the corresponding BTSes are incapable ofhousing any calls and hence their carriers are virtually (if notreally) of zero capacity.

In the sequel, we refer to the BCDP convergence of a cellsite as local convergence. We call the convergence of theoverall process of cell site capacity information gathering forMBTA decision making as global convergence. For globalconvergence, it is not always necessary to wait for localconvergence of the BCDPs of all cell sites. For example, ifretain carrier constraint is configured on the access carrierand the corresponding BTS indicates that the access carrier isof high grade in the sense of having non-downgraded statusand spare capacity, call can be set up on this carrier withoutawaiting further responses because the access carrier will standout among all carriers with a higher value of ECM due tothe higher order ND, SC and RC bits. In this case, globalconvergence and local convergence of BDCP for the inbandsite occur simultaneously. Similarly, in case access carrieris configured with retain band constraint, local convergenceof the inband BCDP with a high grade inband best carrierwill also signal global convergence. Early determination oflocal convergence of the BCDP of any site can also bedone in the same way if all the responses regarding carrierswith a preference value or higher have been received fromBTSes of a site and the current best carrier (determinedusing the responses received so far) of the site happens tobe of high grade. The efficiency of the MBTA algorithmscan be toned up by properly flagging all these conditions thatresult in possible early global convergence. Incorporating theseconsiderations for early convergence of the MBTA informationgathering processes, we may mathematically state the globalconvergence condition GloblConvrgnce as follows:

GloblConvrgnce =

⎧⎨⎩RCConvrgnce

∨RBConvrgnce

∨∧Si∈T LoclConvrgnce(Si)

(16)

Here, Retain Carrier Convergence (RCConvrgnce) is satis-fied when retain carrier constraint is on and the access BTSresponse indicates that the access carrier is of high grade asdiscussed before. In other words,

RCConvrgnce = ∃Ci∈Inband

∣∣∣∣∣∣NDBit(Ci) = ON

∧SCBit(Ci) = ON

∧RCBit(Ci) = ON

(17)It may be noted here that, as discussed in Section III, RCBitcan be set only on the access carrier. Retain Band Convergence(RBConvrgnce), as given below, is similarly satisfied when

the conditions on the right side of the equation are satisfied:

RBConvergence = ∃Cj∈Si=Inband

∣∣∣∣∣∣∣∣∣∣∣∣

NDBit(Cj) = ON∧

SCBit(Cj) = ON∧

RBBit(Cj) = ON∧

LoclConvrgnce(Si)= ON

∧ECM(Cj) ≥

MaxECMSoFar(Si)(18)

Now the Local Convergence (LoclConvrgnce) of the BCDPof a cell site is given by

LoclConvrgnce(Si) =

⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩∀BTSj∈SiResponded(BTSj)∨

MBTATimerExpired∨

CPConvrgnce(Si)∨

(Si ∈ Inband∧

RCConvrgnce)(19)

The first two clauses in (19) signify the extreme scenarioswhere decision can be made only when either all the BTSes ofa site respond or timely responses are not received from someBTSes. The last clause relates to the best case scenario for theinband BCDP. The third clause in (19) refers to a fair situationof early convergence where the pending responses are relatedonly to carriers with lower priority than the current best carrierwhich is of high grade. In this case, the pending responses willin no way influence the best carrier determination process asthe current best carrier will eventually turn out to be the bestcarrier of the site from the way the ECM is defined. Basedon this discussion, mathematical formulation of the CarrierPriority Convergence (CPConvrgnce) criterion will be asfollows:

CPConvrgnce(Si) =

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(∀Cj∈Y etToRespondBTSes∈Si

Priority(Cj) <

Priority(CBC(Si)

))∧(

NDBit(CBC(Si)

)=

ON)∧(

SCBit(CBC(Si)

)= ON

)∧(Si � Inband

∨AccessBTSResponded

)(20)

where CBC is a short-hand notation for theCurrent Best Carrier The last clause in the aboveformulation is to ensure that, for the inband cell site, apremature wrong decision to mark some other inband carrierto be better than an access carrier is not taken before theinformation regarding the access carrier (possibly better thanthe current best because of the RC-bit and other higher orderbits) is available.

From the above formulation, it is clear that we can tone upthe efficiency of the MBTA algorithms by taking advantage ofcustomer configuration (e.g. RC/RB-constraints) on the accesscarrier or the BTS response pattern. If applicable, RetainCarrier Convergence results in the fastest global convergence.Retain Band Convergence is the next most favorable conditionfor global convergence because it is based solely on theinformation about the carriers in the inband cell site. It ispossible to have Carrier Priority Based Convergence when


carriers are of unequal preference for loading. Flagging of thiscondition helps the detection of early convergence of differentcell site BCDPs and consequently leads to a quicker globalconvergence. It is useful when the retain carrier and retainband convergence criteria do not apply.

E. Applicability of the MBTA Framework

The above call processing framework for implementationof an MBTA algorithm, though presented in the context of aCDMA network, it is equally applied to the WCDMA-UMTSsystems currently operating only in two bands. Particularlyin case of Time-Division Duplex (TDD) WCDMA systemswhere different time slots of the same carrier is used for uplinkand down-link transmission, dual-band versions of the abovealgorithms can be separately used to determine uplink anddownlink carriers. In the Frequency Division Duplex (FDD)WCDMA systems, on the other hand, a dual band algorithmmay be applied to determine uplink and downlink carrierpair for call setup just as in case of CDMA systems. In thecorresponding call flow sequence diagram with the UMTSjargon, User Equipment (UE), Node B and Radio Networkcontroller (RNC) represent Mobile (or Mobile Station), BTSand BSC, respectively.

VI. SIMULATION RESULTS

Considering robustness against call failures and distributionefficacy as two performance attributes of an MBTA algorithm,we compare, by means of the following simulation experi-ments, earlier defined five MBBTA algorithms- i) OptimisticMBTA ii) First-fit Min-Max MBTA iii) Best-Fit Min-MaxMBTA iv) Best-Band Min-max MBTA v) Best-DistributionMin-max MBTA and vi) the Single-band MCTA algorithm inwhich call is invariably allocated to the best carrier of theband of call origination.

A. Experimental Setup

In our experiments, we consider the cell site topology ofFig. 1. In this, T consists of one 450MHZ cell site, two 800MHZ cell sites and six 1900 MHz cell sites. Without loss ofgenerality, the mobile is assumed to be located in the exclusiveaccess ranges of the sites A, B and D in the three bands, asshown in the figure. Since it is difficult to experiment with allpossible combinations of pre-configurable biases on bands andcarriers, we deal only with a case of fair competition in whichall carriers compete on an equitable basis without any of thosebiases listed in Section III. Accordingly, we do not considerthe priority, CAC, IBC, RC and RB-bits and construct a bytesized capacity word with only 7 magnitude bits (bits 0 through6) assuming that a maximum of 35 calls are supported by acarrier. All carriers are assumed to support calls the same waywith respect to downgrading and hence the non-downgradedbit is also not included in the word. Reserved capacity forsoft-handoffs is assumed to be 7 (20% of total capacity of35) for all the carriers and the bit-7 of the capacity word isdesignated as the Spare Capacity bit. All calls are assumedto be of the same type (in terms of data rate/quality) so thatavailable capacities on carriers can be expressed in terms ofnumber of calls.

Fig. 5. Variation of Call Failures with Availability of Resources.

B. Study on Relative Performances of Algorithms with RadioResources Availability

In our first experiment, we consider the situation wherecalls could originate in any part of the cell boundary. In thissituation, effect of mobility is expected to be imperceptiblebecause of the negligible distance (compared to the cellradius), on the average, traversed by an MS in the call setuptime frame. This conjecture has also been established byour next experiment. The first experiment seeks to measurethe performances of the above algorithms in evenly balancednetwork loading configurations where the available capacities(gross values, not ECMs) of best carriers of individual sitesare the same but for statistical variations caused by a uniformdistribution of on-going calls. We treated the maximum valueof the available capacity at a site in the topology underconsideration as a parameter of experimentation and varied itfrom 1 to 35. Further, in order to compare the algorithms on astatistical basis, we generated 100,000 random configurationsfor each value of this maximum available capacity (at sites)parameter so that the actual available capacity at a site in aconfiguration is a uniform random number varying between0 and the parameter value. By applying the above algorithmssuccessively in each one of the random configurations, weobtained the number of call failures for those algorithms andcomputed the call failure rates by computing the averagesover the 100,000 configurations. Fig. 5 depicts the variationof call failure rate with average call resources for the al-gorithms under consideration. These results indicate that allthe four min-max algorithms perform equally, but outperformthe optimistic algorithm with respect to robustness againstcall failures. The single-band MCTA algorithm is found tobe inferior to all the multi-band algorithms in this respect.


The kink in the vicinity of the available resources value of7 for the curve of the optimistic algorithm may be ascribedto the discontinuous nature (caused by the spare capacity bit)of the call distribution function (ECM) at that value. Thisdiscontinuity is not observed in case of the single-band MCTAalgorithm by the very non-distributive nature of the algorithmwhich invariably allocates a call to the site of its origination.The kink, though present in case of the min-max algorithms, isnot perceivable because of the negligibly small values of thecall failure probability when available call resources exceedfive.

We also obtained the deviations of the algorithms from idealdistribution efficacy by using the BIDMDTA algorithm asa benchmark and measuring ΔAlgorithm = DAlgorithm −DBIDMBTA for successful calls. We averaged these devia-tions over the successful calls and plotted in Fig. 6 their vari-ation, for different algorithms, with the number of resources.These results indicate that the first-fit min-max algorithm is theworst among the MBTA algorithms. The performance of theoptimistic algorithm is inconsistent; it lags behind the Best-distribution initially, outperforms the latter for low congestionvalues and then again falls behind. The Best-Band algorithmstarts similarly with a larger deviation value and catches upwith the Best-Distribution algorithm. For higher values ofcapacity (above the threshold), the Best-Band, the Best-Fit andthe optimistic algorithms perform equally, though their resultsare slightly shy of those of the Best-Distribution algorithm.As expected, the single-band MCTA algorithm proved to beinferior to the all the multi-band algorithms with respect todistribution efficacy as well.

In case of Fig. 6 also, the peaks in the vicinity of theavailable resources value of 7 can be explained by the fact thatthe site ECM values change drastically near that value becauseof the spare capacity bit and thereby yield large deviationvalues. Even though the single-band algorithm is basicallynon-distributive in nature, ECM comes into picture in thiscase also while computing the dispersion measure. The peakobserved at the available resources value of 2 can be ascribedto the way we compute the deviation as an average over onlysuccessful calls. Since all but a few calls fail all the algorithmsfor resources value 1 and the cases where the optimal andactual allocations differ are still fewer, the deviation value islow and almost the same for all the algorithms. When theavailable resources increased to 2, there is a steep rise in thecall success rate of all the algorithms (as demonstrated by thesteep fall in the failure probability depicted in Fig. 5). Sincemany of the allocations of the successful calls in the individualalgorithms deviated from the ideal allocation, the cumulativedeviation substantially increased resulting in a peak in thedeviation measure for the available resources value of 2. Withfurther increase in site resources, differences between idealand non-ideal allocations became less pronounced as mostof the allocations resulted in the decrease in the dispersionmeasure. The same explanation can be offered for decreasein the deviation function for different algorithms after thepreviously explained peak at the resource value of 7.

Fig. 6. Variation in Distribution Efficacy of Algorithms with Call Resources.

C. Study of Impact of Mobility and Call Volume on Perfor-mance of the Algorithms

In order to study the joint effect of mobility of the MSand mobile density (or equivalently call volume) in the cellon relative performances of the algorithms, we carried outour next experimentation with altogether 9 scenarios including3 cases of mobility in conjunction with 3 cases of mobiledensity. The three mobility cases are- i) High mobility case:MS moving with an average velocity equal to 100 timesradius of the coverage area of the base station per hour (Fora cell with 2 km radius of coverage, this implies a mobilespeed of around 200 km/hour in a random direction fromthe position of call origination. ii) Medium mobility case:MS velocity is around 10 times the cell coverage radius perhour iii) Low mobility: MS velocity is such that it coversnearly the cell radial distance in an hour. In each case, weconsidered the standard deviation of the velocity to be 10%of the corresponding average value. The three mobile densitysituations considered are- i) Heavily loaded network: Here themobile density (or, on-going call volume) is over 95% of themaximal cell site resources ii) Moderately loaded network: Inthis case, the mobile density is in the range of 75-85% of themaximal cell site resources iii) Lightly loaded network whereMS density is in the range of 40-60% maximal call resources.First two cases correspond to the two peaks observed in Fig. 6and the third case represents the situation when contention forpooled resources eases out. In our experimentation, we useda practical value of 100 ms as the worst case time neededfor completion of the MBTA algorithms in situations whenone or more BTSes fail to respond with their carrier capacityvalues. To this figure, we added 100 more milliseconds as anoverhead of subsequent mobile to base-station communication


and computed the new position an MS occupies in the totalof 200 ms starting in a random direction from a randomposition in the cell range in each one of the above 3 mobilitysituations. As in the earlier experimentation, 100000 mobilepositions are considered for assessing the call failure ratesand distribution efficiencies for different algorithms on astatistical basis. Call failures were assessed based on whetherthe mobiles new position is within the range of a cell sitewith non-zero capacity in the band of call allocation. Thedeviation measures for different algorithms are computed byaveraging out on successful calls exactly the same way as inthe first experiment. However, in the simulation with new callsoriginating randomly anywhere in the cell range, the effect ofmobility is imperceptible obviously because of the extremelylow probability of the MS going out of the cell range with therelatively small distance it could travel in 200 ms time frame.Hence, we changed the course of our simulation experimentand considered the case where each one of the 100000 callsoriginates at a random position within 5% of distance of thecall access cell radius from the corresponding cell boundary.Call failure probabilities and deviation measures of differentalgorithms for the above mentioned 9 cases are estimatedin the same way as before. The bar graphs of Figs. 7(a),7(b), and 7(c) depict the effect of mobility on the call failureprobabilities of different algorithms for the heavily, moderatelyand lightly loaded wireless network conditions, respectively.We found that the mobility has no significant impact on thedeviation measures of the algorithms in any of the networkloading situations. Hence, we have recorded only the effectof network loading on the deviation measures of differentalgorithms and plotted the same in the bar graph of Fig. 8.

Results in Fig. 7(a) for the heavy network loading situationindicate that the call failure probabilities of different algo-rithms are nearly the same for the low and medium mobilitycases and increase considerably for the high mobility case.In all cases, the single-band MCTA algorithm is considerablyworse than the rest of the algorithms. The optimistic algorithmhas the next highest call failure probability. The first-fitalgorithm fares the best with a marginally better performanceover the best-band algorithm. The best-distribution and best-fit algorithms tie up on the next best performance. Results inFigs. 7(b) and 7(c) for the moderate and light network loadingsituations show a dramatic improvement in the performanceof all the algorithms with steep fall in their error probabilities.In both these loading situations, the differences in the perfor-mances of the algorithms for the low mobility case are toosmall to be noticed. For the medium and high mobility cases,the first-fit and the single-band MCTA algorithms tie up for thebest performance. The best-fit algorithm takes the next placefollowed by the optimistic, best-band and best distributionalgorithms with nearly equal performance, on the whole.

Fig. 8 presents a comparative view of the performancesdifferent algorithms with respect to distribution of calls orig-inating at cell boundaries by bench-marking the algorithmswith the optimal distribution algorithm in the low, moderateand heavy loading situations. In lightly loaded networks,all algorithms prove to be near optimal with low deviationvalues with negligibly small differences among themselves. Asindicated by the higher deviations, the distribution efficacies of

(a) Case of Heavy Network Loading

(b) Case of Moderate Network Loading

(c) Case of Light Network Loading

Fig. 7. Effect of mobility on call-fail-safeness of different algorithms fordifferent network loading conditions


Fig. 8. Effect of network loading on boundary-call distribution efficacies ofdifferent algorithms

all the algorithms turned out to be worse in moderately loadednetwork compared to a heavily loaded one. Only the single-band MCTA algorithm has equal performance and it wasthe worst among all algorithms in both the situations. Theseresults are consistent with the relative sizes of the peaks shownin Fig. 5 at around 95% and 80% network loading values.Hence, the same arguments used for explaining the resultsof Fig. 5 hold good for these apparently anomalous resultsalso. From these results, it is clear that the single-band MCTAalgorithm is the least efficacious with respect to distributionof calls originating at cell boundaries. The first fit algorithmties up with this MCTA algorithm in moderately loadednetworks and turns out to be the next worst the heavily loadednetworks. The best-distribution algorithm, as expected, givesthe best performance in both the loading situations. The best-fit algorithm is slightly inconsistent in its performance takingthe second place in the heavily loaded networks and a spotafter the optimistic and best-band algorithms in the moderatelyloaded networks. The best-band and optimistic algorithmsperform on par with each other in both the loading situationsand fare better than the MCTA and first-fit algorithms withsmaller deviations.

VII. SUMMARY AND CONCLUSIONS

In this paper, we consider call failure risk associated withunequal radio coverage of different bands, for multi-bandtraffic distribution in 3GB wireless networks and present aclass of min-max algorithms to address this problem. Fordistribution and setup of calls originating at random positionsin the coverage area of a cell site, all these min-max algorithmsoutperform the single-band MCTA algorithm in all respectsand fare equally among themselves with respect to call-fail-safeness, but offer varying degrees of distribution efficiency.Though the Best-Distribution Min-Max MBTA algorithm is,

on the whole, the best in distribution efficacy, it is notrecommended because of its exhaustive search strategy andconsequent adverse effect on real time performance duringcall setup. The First-Fit Min-Max algorithm, though efficientfor call setup because of early decision making, is not rec-ommended as well because of its inferior distribution ineffi-ciency. We can also rule out the optimistic algorithm on thegrounds of high call failure risk in heavily loaded networks.Its performance with respect to distribution is also not veryconsistent. The Best-Fit and Best-Band Min-Max algorithmsare robust against call failures and are, at the same time,ideal compromises for both distribution and computationalefficiencies. Either could be chosen, but the Best-Fit algorithmis preferable because of its simplicity and better distributionefficacy when the network is heavily loaded.

Results of experimentation with calls susceptible to mobilityeffects because of their origination at cell boundaries also leadto identical conclusions. The single band MCTA and the opti-mistic algorithms can be ruled out because of their relativelylow standing in either or both of our performance criteria. Thebest distribution algorithm with expectedly high distributionefficacy and low-to-fair level of call-fail-safeness also goesout of reckoning because of its computational intensivenessand a low level of call-fail-safeness in lightly and moderatelyloaded networks. Among the remaining algorithms, the firstfit algorithm is only marginally better in call-fail-safenessthan the best-fit and best algorithms in the low and mediummobility situations. Though it exhibits better call fail-safenessin high mobility situations, such situations are not commonbecause, in practice, outdoor mega-cells have a radial coverageof 35 km and above and mobile velocities never reach up to3500 km/hour. Further, as indicated in Fig. 8, the other twoalgorithms outperform this algorithm in distribution efficiency.Finally, among the best-fit and best-band algorithms, theformer seems to be a better choice because of its simplicity,over-all better call-fail-safeness and comparable distributionefficacy.

As mentioned in the introductory section, the results pre-sented here for the CDMA systems are equally applicable toWCDMA systems, though the latter systems do not seemto have, to the best of authors knowledge, any current orplanned deployments with multi-band operation. However,from the results presented here, it may be concluded thatCDMA/WCDMA system call-fail-safeness and distributionefficacy can be considerably improved using the multi-bandalgorithms proposed herein or their variants for effective usageof pooled radio resources.

ACKNOWLEDGMENT

Authors wish to thank Sanjay Rajasekhar for the constantsupport provided throughout the course of this work. They alsolike to thank the anonymous reviewers for their constructivecriticism that helped to improve the presentation substantially.

REFERENCES

[1] “Band class specification for cdma2000 spread spectrum systems,” 3rdGeneration Partnership Project, 3GPP2 C-S0057, rev. 0, ver. 1.0, Feb.2004.


[2] “The CDMA2000 family of standards for spread spectrum systems,”TIA/EIA/IS-2000.1.A -2, Feb. 2002.

[3] “Recommended minimum performance standards for cdma2000 high ratepacket data access terminal,” 3rd Generation Partnership Project, 3GPP2C-S0033, rev. 0, ver. 2.0, Dec. 2003. (Also categorized as TIA-866-1standard.)

[4] “Introduction to cdma2000 standards for spread spectrum systems,”Release C, 3rd Generation Partnership Project, 3GPP2 C.S0001-C, ver.1.0, May 2002. (This document contains initial release of CSVS.)

[5] “General UMTS architecture,” T1.3GPP.23.101V310, Dec. 2000.[6] “IMT-2000 CDMA DS and TDD radio interface specifications,” T1.715-

2000, June 2000.[7] Nishith D. Tripathi and Sarvesh Sharma, “Dynamic load balancing in

a CDMA system with multiple carriers,” in Proc. IEEE 54th VehicularTechnology Conference, vol. 2, pp. 1010-1014, Oct. 2001.

[8] A. He, “Performance comparison of load balancing methods in multiplecarrier CDMA systems,” in Proc. 11th IEEE Symposium on Personal andIndoor Mobile Radio Communications, vol. 1, pp. 113-118. Sep. 2000.

[9] Sarvesh Sharma and Ahmad Jalali (Nortel Networks), Traffic allocationand dynamic load balancing in a multiple carrier cellular wirelesscommunication system. US Patent 6, 069, 871, filed March 6, 1998,issued May 30, 2000.

[10] J. M. Harris, D. T. Chen, and P. J. Fleming, “Load balancing in mixedservices CDMA system with delayed information,” in Proc. IEEE 53rdVehicular Technology Conference, vol. 2, pp. 850-860, May 2001.

[11] Parthasarathy Guturu and Abdennaceur Lachtar (Nortel Networks), Acall-fail-safe method for wireless traffic distribution across bands. USPatent Application No: 10/194329, filed July 12, 2002.

[12] Parthasarathy Guturu and Abdennaceur Lachtar, “An efficacious methodfor dual band multi-carrier traffic allocation in CDMA wireless systems,”in Proc. IEEE GLOBECOM Conference, vol. 1, pp. 10-14, Dec. 2003.

Parthasarathy Guturu is currently a faculty mem-ber of the newly founded Electrical Engineeringdepartment of the University of North Texas (UNT),Denton. Prior to his recent return to academia, he hashad experience of more than 7 years in corporate R& D and over ten years in teaching and academicresearch abroad. While in industry, he contributedto the areas of Advanced Intelligent Networks and3G Wireless Systems with architecture, design anddevelopment of complex real-time systems. His in-novations in the corporate world culminated in two

US patents and a third patent application which is under review. In hisearlier stint in academia at Indian Institute of Technology, Kharagpur, afterbachelor’s though PhD (Eng.) degrees from the same place, he directed4 PhD dissertations and a large number of graduate and undergraduatetheses. He published over 35 papers in international journals and conferencesand contributed to disparate areas of Electrical and Computer Engineeringincluding Pattern Recognition, Computer Vision/Image Processing, ArtificialIntelligence, and solultions of ill-posed and combinatorial optimization prob-lems with Neural Networks/Genetic Algorithms. Dr. Guturu plans to integratehis past experience in Computational Intelligence and the latest experience inWireless Networks and embark upon an ambitious research program in thearea of Wireless Sensor Networks and Systems.

Abdennaceur Lachtar is a senior engineer in theCore RF Engineering group of Nortel Networks.He has been working in the area of CDMA systemdevelopment for more than 7 years and has two U.S.patents (pending)- one on ”Call-Fail-Safe Methodfor Wireless Traffic Distribution Across Bands” andthe other one on ” Enhancements to MCTA to Com-bat Access Failures and to Work Across Bands.”Abdennaceur’s current responsibilities include per-formance evaluation and optimization for wirelesspacket data applications, new features performance

verification in live air testing, and priming product development supportfor the new product releases. This includes recommendation of enhancedfunctionality and essential new Operational Measurements (OMs) to enablethe evaluation of the performance of the new features through meaningfulmetrics. Abdennaceur holds an M.S. in Electrical Engineering from theUniversity of Florida.

Documents

578 IEEE TRANSACTIONS ON WIRELESS … Multi-Carrier Traffic ... (WCDMA) access technology ... Purpose of current research is to find a generalized solution to multi-band traffic