high capacity indoor and hotspot wireless systems in shared spectrum: a techno-economic analysis

IEEE Communications Magazine • December 2013102 0163-6804/13/$25.00 © 2013 IEEE

1 For the remainder of thearticle , an operator refersto any business entity thatowns and operates its ownnetwork. In this context,the facility owners can beconsidered operators.

INTRODUCTION

EMERGING LOCAL OPERATORS INSHARED SPECTRUM

Over the last few years, mobile and nomadicbroadband access has achieved tremendous suc-cess. Innovation in mobile handsets (e.g., smart-phones and tablets) has caused a virtual datatsunami, leading to severe capacity problems formany operators. The dramatic surge in traffic isexpected to continue in upcoming years. Sincethe majority of the traffic is likely to be generat-ed inside buildings, significant investment inindoor network deployment is foreseen.

Some of the investment will be made byincumbent mobile network operators (MNOs)for deploying heterogeneous networks. Anothertype of investment, which is the focus of thisarticle, will be provided by new market actors,such as facility managers, real estate owners, and

private companies [1]. The main driver for themis to deploy and operate dedicated networks forwireless Internet access inside their buildings toincrease their attractiveness to tenants or cus-tomers (e.g., hotels or office space providers).The private networks may also serve those MNOcustomers that happen to be in the buildings (i.e.provide offloading). This is a business modelsimilar to the one seen in current fixed networkaccess: public operators provide connection tobuildings, whereas in-building networks aredeployed and managed by the facility owners.An additional problem compared to fixed net-work access is that such local operators1 do nothave their own dedicated spectrum. One possi-ble solution is using shared spectrum, where thesharing takes place between neighboring indooroperators.

REGULATORY TRENDS IN SHARED SPECTRUMThere are regulatory initiatives worldwide aim-ing to promote shared access to new spectrumfor fostering more competition and innovationin a wireless access market. The national regu-lators of Sweden and the Netherlands recentlyannounced that a portion of the spectrumaround 1800 MHz was opened for cellulartechnologies with indoor usage in an unli-censed or a preregistration manner [2, 3]. Inthe United Kingdom, 1781.7–1785 MHz pairedwith 1876.7-1880 MHz band was allocated to12 operators with shared licenses in 2006 forlow-power indoor networks [4]. The l ightlicensing of nationwide 3650–3700 MHz wasalso adopted by the United States in 2007 [4].An overview of regulatory initiatives can befound in [5, 6].

CONTRIBUTION OF THIS ARTICLELocal wireless access operation in shared spec-trum presents new research challenges from atechno-economic perspective. In this article, weprovide an analysis framework that effectivelynavigates and compares potential deploymentstrategies of the operators. We define a strategyspace that integrates technology and businessaspects. Then a legacy cost model for a singleoperator is reformulated by introducing newinter-operator cost factors. The validity of ourframework is demonstrated by numerical com-

ABSTRACT

Predictions for wireless and mobile Internetaccess suggest an exponential traffic increase,particularly in in-building environments. Non-tra-ditional actors such as facility owners have agrowing interest in deploying and operating theirown indoor networks to fulfill the capacitydemand. Such local operators will need spectrumsharing with neighboring networks because theyare not likely to have their own dedicated spec-trum. Management of internetwork interferencethen becomes a key issue for high capacity provi-sion. Tight operator-wise cooperation providessuperior performance, but at the expense of highinfrastructure cost and business-related impair-ments. Limited coordination, on the other hand,causes harmful interference between operators,which in turn will require even denser networks.In this article, we propose a techno-economicanalysis framework for investigating and compar-ing indoor operator strategies. We refine a tradi-tional network cost model by introducing newinter-operator cost factors. Then we present anumerical example to demonstrate how the pro-posed framework can help us to compare differ-ent operator strategies. Finally, we suggest areasfor future research.

RADIO COMMUNICATIONS

Du Ho Kang, Ki Won Sung, and Jens Zander, KTH Royal Institute of Technology

High Capacity Indoor and HotspotWireless Systems in Shared Spectrum: A Techno-Economic Analysis

KANG_LAYOUT_Layout 12/3/13 2:58 PM 02

IEEE Communications Magazine • December 2013 103

parisons of selected strategies in a two-operatorexample. Finally, we outline important researchareas to be addressed for future studies.

DEPLOYMENT CHALLENGES FORHIGH CAPACITY

WIFI MAY NOT BE ENOUGH INUPCOMING YEARS

The local operators certainly need a low-costsystem to provide high-capacity services in thenew shared spectrum. WiFi naturally seems thefirst candidate because a substantial amount ofindoor traffic is already offloaded to WiFi in anindustrial, scientific, and medical (ISM) band.However, it has been reported that WiFi maynot cope with very large traffic loads due to the“performance wall” caused by the underlyingcarrier sense multiple access with collision avoid-ance (CSMA/CA) mechanism [7]. In these situa-tions, we need to consider cellular-type systemswith interference coordination capabilities.

TRADE-OFF ININTERFERENCE MANAGEMENT OPTIONS

In shared spectrum, interference managementbetween adjacent operators is one of the keyissues. The easiest option would be not to coop-erate with neighbors. However, this may inflictmutual interference, which may in turn lead topoor performance. Alternatively, a cellular tech-nology developed in a single operator contextcan be applied. The simplest form would be tra-ditional interference avoidance techniques withstatic resource partitioning (e.g., frequency plan-ning), which, however, comes at a significantperformance loss. More advanced techniques(e.g., interference cancellation, joint multicell

processing, or coordinated scheduling) can befurther exploited to enhance system capacity.

Improvement in technical performance isobviously expected from tighter interferencecoordination. However, additional cost and vari-ous business constraints are the hidden barriersthat should not be overlooked. The coordinationmay require better infrastructure and extra net-work resources for reliable informationexchange, which incurs a higher cost. Further-more, a strategic cooperation agreement in abusiness domain needs to be made for the inter-operator coordination. An operator may be lim-ited in his/her business strategies due to formingan alliance with its neighbor because he/she maylose competitive advantages by limitations innetwork operation and network upgrades [8, 9].

NEED FOR A TECHNO-ECONOMICANALYSIS FRAMEWORK

As summarized in Table 1, the system design inshared spectrum is inherently a multidimension-al problem where technology, business, and reg-ulatory issues are intertwined. The problem ofchoosing an operator strategy brings up severalresearch challenges that have rarely beenaddressed in the literature. First, the analysis ofoperator-wise competition or cooperation isnontrivial due to the business complexityinvolved, although competition between individ-ual users inside a network has been extensivelystudied [10]. Second, the operators need to beable to compare different levels of technicalcoordination in addition to their associated busi-ness complexity. There is a lack of systematicevaluation methodology for this. Third, a propercost model should be in place for a cost-perfor-mance trade-off analysis. There are someattempts to model the network cost of a conven-

Table 1. Changing business landscape and paradigm shift on system design in shared spectrum.

Traditional wide area single-operatorsystem Local area system in shared spectrum

Businesslandscape

Who and why? MNOs: revenue generation from serviceprovisioning

• MNOs: data offloading• Facility owners: complements facility services• Hotspot operators: new revenue generation in niche markets

Where? Large-scale public outdoors Mainly private/public indoors controlled by facility owners

Inter-operatorrelation Service/price competition in markets

• Service/price competition in markets• Cooperation/competition for internetwork interferencecoordination

Major network-related cost

• Network cost• Spectrum cost

• Network cost• Inter-operator cost

Systemdesign

Design problem Minimizing network cost at a giventraffic demand

Minimizing network cost + inter-operator cost at a giventraffic demand

Decision domain Mainly technology Both technology and business

Main decisionmaker MNO Both operators and a regulator

Coordinationtarget Nodes (e.g., BSs) Networks

KANG_LAYOUT_Layout 12/3/13 2:58 PM Page 103

IEEE Communications Magazine • December 2013104

tional single operator network [11]. However,the cost model in a multi-operator context hasnot been covered yet.

DEFINING THE SOLUTION SPACE: A CONCEPTUAL FRAMEWORK

A fundamental network design problem that theoperators face is minimizing their total networkcost while satisfying their users’ demands. Inorder to achieve this, the cost of differentdeployment options should be compared withrespect to the performance (e.g., capacity)requirement. This comparison becomes morecomplicated in a shared spectrum environmentbecause both cost and performance are heavilyaffected by neighboring operators.

Since cooperation intrinsically involves strate-gic decisions in the business domain, a means toassess the combined effects of a technologychoice and business aspects is needed. In thissection, we provide a conceptual framework toeffectively categorize potential strategies. Asshown in Fig. 1, various operator strategies arecharacterized on one hand by the strategic deci-sion of the operators, i.e., cooperation and com-petition, and on the other hand by the technicalsolution expressed as the level of coordination.

STRATEGIC DECISION: COOPERATION OR COMPETITION

The strategic decision in the business domaininfluences the way neighboring networks arecoordinated. In this article, we model coopera-tion and competition by using different technicalobjective functions that each operator aims tomaximize. The two types of strategies are: • Cooperation: The operators aim at maxi-

mizing a common objective function agreedbetween them.

• Competition: Each operator aims at maxi-mizing its own objective function in a self-ish manner.

Cooperating operators who synchronize theirtechnical behaviors form a network alliance.Reaching an agreement on a common objectivefunction is in itself challenging, particularly whenthe partners have chosen different criteria fortheir quality of service (QoS), for example, aguaranteed-rate video service vs. a best effort ser-vice. From a radio resource allocation perspec-tive, the network alliance behaves as aconventional single-operator network. The coor-dination issues turn into a problem of internalresource management for the combined network.

Competition between the operators requiresa non-traditional system design. Regulators mustprovide guidelines for operator behavior by issu-ing coexistence rules (“etiquettes”), which cancoordinate the network rivalry to some extent.For the case of network-wise competition, inter-network interference is coupled with intranet-work interference control, which creates newchallenges.

TECHNICAL SOLUTIONS: COORDINATION BETWEEN NETWORKS

The technical solutions of the operators imple-menting their strategic decisions directly affectthe network’s performance. We define coordina-tion in a technical domain as the process of shar-ing relevant information. The level ofcoordination is measured by the amount of infor-mation shared in the process. The informationrelevant to the interference coordination can bestatistical or instantaneous traffic load, or pathgains between all the involved access points(APs) and user terminals, usually referred to aschannel state information (CSI).

More accurate and frequent informationexchange increases the global knowledge of thewhole system. A system with complete informa-tion sharing can be interpreted as a centralizednetwork (whether decisions are made centrallyor not). Such a system is desirable from a per-formance perspective since it can provide real-time resource allocation (e.g., beamforming orcoordinated scheduling [12]). On the other hand,only slow-varying information (average propaga-tion conditions and the number of users per cell)might be shared. This requires considerably lesssophisticated equipment, and the informationmay be exchanged using existing IP connections.

SHARED SPECTRUM COST MODELRecall that the network design objective is tofind the operator’s strategy that enables them tooffer the lowest total cost for the required capac-ity. The cost in a shared spectrum environmenthas an additional element with regard to techni-cal inter-operator relations. In addition, there is“strategic cost” representing the business uncer-tainty caused by the decision to cooperate. Inthis section, we recap a traditional single-opera-tor cost model, and highlight new inter-operatorcost items.

TRADITIONAL SINGLE-OPERATOR COST MODELFor a legacy operator in a wide area, the cost ofa wireless network mainly consists of two parts,infrastructure and spectrum, as described in [11]:

Figure 1. A conceptual framework to define a strategy space and navigate theoperator strategies.

Strategy IIICentralized

network alliance

Strategy IIDistributed

network alliance

Strategy IVCentralized

network rivalry

Strategy IDistributed

network rivalry

Tight coordination

Loose coordination

CooperationCompetition

Operator B

Operator A



Ctot = Cinfra + Cspectrum = NCr + WCw (1)

where N and Cr are the number of deployedAPs and the normalized cost per AP (€/AP),respectively. W represents the allocated spec-trum (MHz), and Cw the cost per unit of spec-trum (€/MHz). Here, Cr includes all capitalexpenditure (CAPEX) and operational expen-diture (OPEX) aspects. CAPEX is mostly relat-ed to all one-time investments (e.g., APs orcore network equipment, site installation/build-out, and antenna systems). Cost during opera-tion (e.g., backhaul transmission, site rental,operation and maintenance [O&M], and elec-tricity) is categorized as OPEX. OPEX is typi-cally discounted to present value assumingexpected annual running cost and network life-time.

Indoor and hotspot systems may have as impler cost s tructure than convent ionalmacro networks due to the small physicalsize of equipment. For instance, cost relatedto s i te instal lat ion/bui ld-out , s i te rental ,antenna systems, and O&M may be free orignorable , whereas expenses for the APequipment and the new backhaul installationmay be dominant . However , we can s t i l lemploy a linear model as a function of thenumber of APs as in Eq. 1. By sharing spec-trum, spectrum cost is also likely to becomenegligible.

INTER-OPERATOR COST IN SHARED SPECTRUM

(Invisible) Strategic Cost — Operators havetraditionally been reluctant to share networks orto cooperate across business boundaries mainlydue to the limitations and uncertainties they per-ceive regarding their marketing strategies. Coop-eration takes place only if large economic gainsare foreseen (e.g., mobile broadband in ruralareas where a small customer base cannot sup-port multiple parallel networks). In the businessliterature, such barriers against strategic alliancewith competitors have been widely studied (e.g.,see [13]). Similar issues have also been discussedin [9] in the context of outdoor network sharing.Some key network-related obstacles are:• Management overhead: Decision making on

network deployment/upgrade can bedelayed because it requires an agreementwith the cooperation partner.

• Limited network controllability: An opera-tor may lose control over the deploymentand operation of its own network, whichcan restrict individual network dimension-ing and make its service indistinguishablefrom its competitors.

• Risk of information leakage: The coordina-tion may reveal customer statistics andknow-how on network optimization to theother operator.

• Lack of trust: An operator may suspect that

Figure 2. Examples of Ccoord according to the cooperation level.

Using existing IP connection

Installingdedicated

fibers/copperlines

Installing acentral unit and

a dedicated inter-network line

Installingdedicated

fiber/copperlines

Using existing IPconnection

Negligible costfor loose coordination

Cost for tightcoordination

Cost for moderatecoordination Installing a

central unitUsing existing IP

connection

Operator A Both Operator B

Main Ccoorddrivers

Who pays for what?

CAPEX is mostly

related to all one-

time investments,

e.g., APs or core net-

work equipment, site

installation/build-out,

and antenna sys-

tems. Cost during

operation, e.g.,

backhaul transmis-

sion, site rental,

operation and main-

tenance (O&M), and

electricity, is catego-

rized as OPEX.



a partner is delivering false information totake advantage of the coordination whenthere is no trustable intermediate coordina-tion entity.One way for decision makers to handle uncer-

tainty or risk is to present a risk margin whencalculating the expected profits of the operation.In this article, we do this by introducing an addi-tional fictitious cost, denoted by the strategic costCstex. Notice that Cstex may not be strictly mea-surable because it may be related not only toobjectively computable risk but also to perceiveduncertainty about the future. This is not a newconcept in strategic decision making in the busi-ness domain, where one often considers anuncertainty margin, such as using a fictitious“required rate of return” or “hurdle rate” whencomparing investment options [14].

Coordination Cost — From a technical pointof view, information sharing between networksrequires additional complexity and infrastructurecost. The coordination cost Ccoord is spent foracquiring relevant information for resource allo-cation. Depending on the amount of informationto be shared, Ccoord can be negligible or signifi-cant. As exemplified in Fig. 2, the extra costmainly emanates from installing a dedicatedbackhaul and/or an intermediate entity to coor-dinate the interference between networks. Forinstance, real-time interference coordinationmay necessitate expensive dedicated backhaulboth inside and between buildings to allow forreliable low-latency information sharing. In addi-tion, an internetwork coordinator should be putin place for fast and synchronized resource allo-cation. While in-building backhaul would bepaid by an individual operator, the common cost(i.e., inter-building backhaul or the intermediateequipment) can be shared. In a moderate coor-dination scenario, each network may use its

existing in-building IP connection to control itsown APs. Even in this case, intermediate coordi-nation equipment may need to be introduced bya third party due to a regulatory constraint (e.g.,spectrum broker) or a trust reason in the busi-ness domain. The cost for extra coordinationequipment can also be saved by directly exchang-ing information between neighboring operators,resulting in a distributed network architecture.In general, a lower level of coordination requiresmore APs to satisfy required traffic demand [7].Thus, finding a proper level of coordination togive the lowest total cost is an important issue.

By including the two inter-operator cost fac-tors, the total cost model of an operator inshared spectrum is extended to

Ctot = Cinfra + Cspectrum + Ccoord + Cstex (2)

In the following section, we demonstrate how touse our proposed techno-economic modelingapproach in a typical scenario.

COMPARISON OFOPERATOR STRATEGIES

TWO-BUILDING SCENARIOLet us consider a scenario with two nearby build-ings as illustrated in Fig 1. Two indoor operatorswithout exclusive spectrum want to deploy net-works in their respective buildings. We assumethat the regulator has arranged a shared fre-quency band in order to foster such local deploy-ments under a light licensing regime. This bandonly requires cost-free preregistration to avoiduncertain interference from end users. Bothoperators run their networks in this frequencyband. In this scenario, two operators want tofind the most economic deployment strategy.

CANDIDATE OPERATOR STRATEGIESA multitude of solutions for operators can beenvisaged that combine technology and businessaspects. However, we choose three candidatestrategies as follows for illustrative purposes.

Strategy I (no cooperation): Neither operatorwants to cooperate due to the potential limita-tions of strategic alliance. Instead, they chooseto deploy a CSMA/CA network imposed by aregulator as a coexistence mechanism unlesscooperation between operators is implemented.

Strategy II (loose cooperation): The opera-tors decide on joint deployment of cellular tech-nology based on a mutual contract. Althoughthis improves network performance compared tostrategy I, it increases dependence on the neigh-bor operator. They choose a traditional picocellsystem across the two buildings employing con-ventional frequency planning that can be imple-mented without further investment ininfrastructure.

Strategy III (tight cooperation): The opera-tors jointly deploy a cellular network. This time,they want to use a system with an advancedmulti-cell joint processing technique, zero-forc-ing (ZF) coordinated beamforming, in order tohave higher system capacity with fewer APs. Inthis strategy, they essentially need to invest inintra/inter-building optical fiber infrastructure.

Figure 3. Comparison of the total cost between no cooperation and loosecooperation; the economic strategy differs depending on traffic demand level(Cn

r = 1, Clr = 2, Lw = dB, 95 percent coverage requirement).

Traffic demand λ (Gbytes/mo/user)2010

10

0

Cto

t

20

30

40

50

60

70

80

30 40 50 60 70 80 90

No cooperationLoose coop. w/estimated Cstex = 0Loose coop. w/estimated Cstex = 10Loose coop. w/estimated Cstex = 20



COST COMPARISON OF STRATEGIES

Evaluation Assumptions — For analysis sim-plicity, we assume that each local propagationcondition and average traffic demand l (giga-bytes per month per user) are symmetric. Twoneighboring single-story buildings with the sizeof 50 m × 100 m accommodate 500 employeeseach. Wall loss of Lw dB exists both inside of abuilding and between buildings. The CSMA/CAnetwork assumes channel bonding (aggregation)to fully exploit available frequency range. Perfectcarrier sensing without redundant idle APs isassumed to have optimistic CSMA/CA perfor-mance. Loose cooperation uses the best staticfrequency reuse subject to a given outage con-straint. The tight cooperation assumes ideal ZFwithout any CSI at the transmitter (CSIT) errorand feedback delay in an uncorrelated channel.Some of the important simulation parametersused are summarized as follows: • Path loss exponent and transmission power:

3 and 20 dBm• Maximum link spectral efficiency: 6.67

b/s/Hz• System bandwidth and signal-to-interfer-

ence-plus-noise (SINR) outage threshold:60 MHz and 3 dB

• Carrier sensing threshold for the CSMA/CAnetwork: –72 dBmMore detailed system modeling and simula-

tion parameters can be found in [7]. Hereafter,the three strategies (i.e., no cooperation, loosecooperation, and tight cooperation) will bedenoted by superscripts n, l, and t, respectively.

No Cooperation vs. Loose Cooperation —We assume a unit price for an AP with no coop-eration (Cn

r = 1) without loss of generality. Fig-ure 3 illustrates and depending on l . Thestrategic cost Cstex can have a heavy influence onthe lowest cost strategy. If two operators arevery reluctant to cooperate, Cstex will be high,potentially significantly enough to prevent coop-eration, as depicted by the dotted lines in Fig. 3.

From an analysis perspective, we can quantifythe condition for Cstex given that Nl (l)Cl

r + Cstex≤ Nn(l)Cn

r. Then let us define maximum accept-able cooperation risk (MACR) for a givendemand l:

MACR = max(Nn (l)Cnr – Nl (l)Cl

r, 0). (3)

This quantitatively provides us with the upperbound of risk the operators are willing to takewhen choosing their cooperation strategy.MACR of zero means that cooperation is worth-less because Cl

r is already too high. On the con-trary, high MACR indicates that operators needto cooperate even if they perceive a high risk ofcooperation. As shown in Fig. 4, MACR can beexplicitly plotted as a function of l and Cl

r for agiven unit cost Cn

r = 1 in order to aid an opera-tor’s decision making. This only requires estimat-ing Nn (l ) and Nl (l ) for varying l . In thisparticular example, we can observe from Fig. 4that MACR rapidly increases after a certaindemand level because the CSMA/CA mecha-nism cannot satisfy such a high demand. Thisindicates that operators will have to rely on the

cooperation in the end as the traffic demandcontinues to grow.

Loose Cooperation vs. Tight Cooperation— Tight cooperation additionally incurs the costof fiber installation between all APs (or remoteradio head) and a central baseband processor.Let Cfiber denote the fiber installation cost perAP. Then, unlike Cstex, Ccoord is assumed to beapproximately linear to the number of placedAPs: Ccoord = Nt (l) Cfiber. We assume the costsfor the central processor and inter-building fiberare relatively negligible and Cstex is same forboth strategies. Then we can assess the conditionfor Cfiber. Similar to the MACR, let us definebreak-even fiber cost (BEFC) per AP:

Break-even fiber cost per AP (BEFC)= max (h(l)Cl

r – Ctr, 0) (4)

where

represents coordination efficiency indicated bythe relative difference in the required deploy-ment density between two system solutions. TheBEFC provides an upper bound on fiber costper AP to make tight cooperation more econom-ic than loose cooperation. When the actual fibercost per AP is larger than the BEFC, the systemwith advanced coordination does not yield a costbenefit even with its superb performance. Thezero value of BEFC suggests that the tight coop-eration cannot be economic regardless of fibercost due to too high equipment cost Ct

r. BEFC isquantitatively shown in Fig. 5 according to wallloss at a given traffic demand. When wall loss issmall, the tight cooperation provides more tech-nical gain by cancelling a significant amount ofinterference. Thus, h(l) becomes large so that

η λ λλ

= N

N( )

( )

( )

l

t

Figure 4. The maximum acceptable cooperation risk depending on trafficdemand (Cn

r = 1, Lw = 0 dB, 95 percent coverage requirement).

Traffic demand λ (Gbytes/mo/user)3025

0Max

imum

acc

epta

ble

coop

erat

ion

risk

(M

AC

R)

10

20

30

40

50

60

70

35 40

Increasing cooperation incentiveat high traffic demand

45 50

Loose coop. w/Clr=1

Loose coop. w/Clr=2

Loose coop. w/Clr=3



tight cooperation is preferred for given Cfiber.Nevertheless, closed environments with higherwall loss do not bring about total cost benefit bytight cooperation. In this situation, a moresophisticated system may not be desirable due tothe marginal performance benefit compared tothe fiber investment.

FUTURE RESEARCH AREASAlthough substantial work has to be done torealize affordable high-capacity provisioning,research on the design of local operator net-works in shared spectrum is still in its earlystage. In the following, we suggest high-levelresearch areas to be addressed.

TRADE-OFF BETWEENDIFFERENT LEVELS OF COORDINATION

Various levels of intra-/inter-network coordina-tion result in different system performance aswell as coordination costs. Therefore, the quanti-tative comparison of different forms of coordina-tion needs to be done to identify the most viablecoordination strategy. The main tasks are cate-gorizing relevant information to be shared andquantifying network performance in variouslocal environments. Performance improvementby coordination can be converted into cost sav-ing in terms of the reduced number of APs.Then operators can examine if this compensatesfor the associated coordination cost.

IMPACT OF NETWORK SEPARATIONIndoor and hotspot networks will be locatedclose to each other, particularly in dense urbandistricts, inflicting mutual interference. However,they are usually separated by walls and geo-graphic distance. Their internetwork interfer-ence can be outweighed by intra-networkinterference if the separation between the net-

works is large enough. This means that the bene-fit of internetwork coordination relies on net-work separation, which requires thoroughinvestigation. We scratched the surface in [15].

ASYMMETRY BETWEEN OPERATORS IN ADEMAND AND DEPLOYMENT ENVIRONMENT

It is likely that nearby networks have differentuser demands, QoS requirements, or in-buildingpropagation conditions. Such asymmetry indemand and physical environments may lead tounequal incentives or even negative cooperationgain to one of the participants. Moreover, find-ing a proper common objective function for thecooperation is a challenging task, especiallywhen the partners aim at different QoS require-ments specific to local services.

MULTITUDE OF INTERFERINGNETWORKS/COOPERATION PARTNERS

It is common that a multi-story building accom-modates several networks. At the same time, thebuilding is surrounded by multiple neighboringbuildings. Thus, the number of networks generat-ing interference can be substantial. As the num-ber of potential cooperation partners increases,the performance gain would be higher sincemore interference could be controlled. However,it would come at the expense of inflating businesscomplexity. Therefore, the relation betweencooperation incentive and the number of involvedpartners needs to be further explored.

IMPACT ON EXISTING MNOSAs more investments are made by local opera-tors, subscribers of existing MNOs can havemore opportunities to enjoy high-speed access.Such investments indirectly relieve the soaringtraffic burden in wide area networks owned byMNOs. In addition, new local infrastructure canbe shared by MNOs via roaming agreements.This can provide MNOs with competitive advan-tages over other competitors. While most of theliterature focuses on the technical performanceof a single operator’s heterogeneous network,the impact of local operators on existing MNOshas to be studied further in both the technicaland business domains.

CONCLUSIONWe have proposed an analysis framework toexplore the trade-off between performance ben-efits of operator cooperation in shared spectrumand involved technical and strategic costs. Thekey application in mind has been indoor wirelessaccess networks that could potentially offloadwide area cellular networks. In this framework,the implications of various operators’ strategicdecisions to cooperate or compete, and therelated costs of internetwork coordination levelare explicitly modeled. We have also validatedour framework by showing a quantitative analy-sis example in which we compare the total costof the candidate deployment strategies. Futureresearch challenges involve assessing differentoperator strategies in various business situationsand physical environments by using the frame-

Figure 5. The break-even fiber cost per AP for different indoor environments(Cl

r = 2, l = 200 Gbytes/mo/user, 95 percent coverage requirement).

Wall loss Lw (dB)

20

0

Brea

k-ev

en f

iber

cos

t pe

r A

P (B

EFC

)

2

4

6

8

10

12

14

4 6 8 10 12

Tight cooperation can be economicat open area

14 16 18 20

Tight coop. w/Ctr=2

Tight coop. w/Ctr=3

Tight coop. w/Ctr=4



work, with the objective of finding the strategythat provides the lowest cost satisfying perfor-mance requirements.

ACKNOWLEDGMENTThe authors would like to acknowledge the finan-cial support of Wireless@KTH, project MBB++.We also thank Mats Nilsson and Dr. Jan I Mark-endahl at KTH for valuable discussions on indoorcellular systems and business aspects.

REFERENCES[1] J. Zander and P. Mähönen, “Riding the Data Tsunami in

the Cloud: Myths and Challenges in Future WirelessAccess,” IEEE Commun. Mag., vol. 51, no. 3, 2013, pp.145–51.

[2] Swedish Post and Telecom Authority (PTS), “PTS Spec-trum Orientation Plan,” Oct., 2012, http://www.pts.se/en-GB/Industry/Radio/Spectrum-Policy-and-Spectrum-Orientation-Plan/.

[3] Analysis Mason, “Future Regulation of Wireless Accessin the 790MHz–3400MHz Spectrum Bands,” rep. forBIPT, Feb. 2010, available: www.bipt.be.

[4] K. Berg, M. A. Uusitalo, and C. Wijting, “SpectrumAccess Models and Auction Mechanisms,” Proc. IEEEDySpan, Bellevue, WA, Oct. 2012.

[5] S. Forge, R. Horvitz, and C. Blackman, “Perspectives onthe Value of Shared Spectrum Access-Final Report forthe European Commission,” Feb. 2012.

[6] J. S. Marcus et al., “Inventory and Review of SpectrumUse: Assessment of the EU Potential for Improving Spec-trum Efficiency,” WIK-Consult final report, Sept. 2012.

[7] D. H. Kang, K. W. Sung, and J. Zander, “Cost Efficient HighCapacity Indoor Wireless Access: Denser Wi-Fi or Coordi-nated Pico-cellular?,” http://arxiv.org/abs/1211.4392.

[8] “Successful Network Sharing,” White Paper, Ericsson,Sept. 2010.

[9] I. Chatzicharistou and F. Berkers, “Overview of Regula-tors and Operators Views on Resource Sharing (initial),”EU FP7 Project SAPHIRE, Deliverables D7.1a, June 2010,http://www.saphyre.eu/publications/.

[10] J. Zander, “Competitive Wireless Multi-access — Impli-cations and Research Issues,” Proc. IEEE PIMRC, Helsin-ki, Finland, Sept. 2006.

[11] J. Zander, “On the Cost Structure of Future WidebandWireless Access,” Proc. IEEE VTC, Phoenix, AZ, May1997.

[12] D. Gesbert et al., “Adaptation, Coordination, and Dis-tributed Resource Allocation in Interference-LimitedWireless Networks,” Proc. IEEE ICC, Glasgow, Scotland,June 2007.

[13] M. Bengtsson and S. Kock, “Coopetition in BusinessNetworks — To Cooperate and Compete Simultaneous-ly,” Industrial Marketing Management, vol. 29, 2002,pp. 411–26.

[14] A. K. Dixit and R. S. Pindyck, Investment Under Uncer-tainty, Princeton Univ. Press, 1994.

[15] D. H. Kang, K. W. Sung, and J. Zander, “Cooperation andCompetition between Wireless Networks in Shared Spec-trum,” Proc. IEEE PIMRC, Toronto, Canada, Sept. 2011.

BIOGRAPHIESDU HO KANG [S’10] ([email protected]) received his B.S.degree in electrical and electronics engineering fromPohang University of Technology and Science, Korea, in2008. In 2010 he got his M.S. degree in electrical and elec-tronics engineering from Seoul National University, Korea.He is pursuing his Ph.D. degree at the Communication Sys-tems Department, KTH Royal Institute of Technology, Swe-den. He has been participating in EU FP7 project METISsince 2012. His research interests include future low-costultra-high-capacity wireless access network design andspectrum options.

KI WON SUNG [M’10] ([email protected]) is a Docentresearcher in the Communication Systems Department atKTH Royal Institute of Technology. He is also affiliated withthe KTH Center for Wireless Systems (Wireless@KTH). Hereceived a B.S. degree in industrial management, and M.S.and Ph.D. degrees in industrial engineering from KoreaAdvanced Institute of Science and Technology (KAIST) in1998, 2000, and 2005, respectively. From 2005 to 2007 hewas a senior engineer at Samsung Electronics, Korea,where he participated in the development and commercial-ization of a mobile WiMAX system. In 2008 he was a visit-ing researcher at the Institute for Digital Communications,University of Edinburgh, Scotland. He joined KTH in 2009.He served as an assistant project coordinator of EuropeanFP7 project QUASAR. He also served as a track chair forCROWNCOM 2012 and a TPC member for several interna-tional conferences. His research interests include dynamicspectrum access, energy-efficient wireless networks, cost-effective deployment and operation, and future wirelessarchitecture.

JENS ZANDER [S’82, M’85] ([email protected]) is a full professor aswell as co-founder and scientific director of the KTH Cen-ter for Wireless Systems (Wireless@KTH).at KTH RoyalInstitute of Technology.He was past project manager ofthe FP7 QUASAR project assessing the technical and com-mercial availability of spectrum for secondary (cognitiveradio) use. He is on the board of directors of the SwedishNational Post and Telecom Agency (PTS) and a memberof the Royal Academy of Engineering Sciences. He wasthe Chairman of the IEEE VT/COM Swedish Chapter(2001–2005) and TPC Chair of the IEEE Vehicular Technol-ogy Conference in 1994 and 2004 in Stockholm. He is anAssociate Editor of ACM/Springer Wireless Networks Jour-nal. His current research interests include architectures,resource and flexible spectrum management regimes, aswell as economic models for future wireless infrastruc-tures.

Future research chal-

lenges involve assess-

ing different

operator strategies in

various business situ-

ations and physical

environments by

using the frame-

work, with the

objective of finding

the strategy that

provides the lowest

cost satisfying perfor-

mance requirements.


http://www.bipt.be

http://arxiv.org/abs/1211.4392

http://www.saphyre.eu/publications/

mailto:[email protected]



http://www.pts.se/en-GB/Industry/Radio/Spectrum-Policy-and-Spectrum-Orientation-Plan/