On Spectrum Sharing Among Micro-Operators in 5G
Tachporn Sanguanpuak, Sudarshan Guruacharya, Ekram Hossain, Nandana Rajatheva, Matti Latva-aho
aa r X i v : . [ c s . N I] A ug On Spectrum Sharing Among Micro-Operators in5G
Tachporn Sanguanpuak ∗ , Sudarshan Guruacharya † , Ekram Hossain † , Nandana Rajatheva ∗ , Matti Latva-aho ∗∗ Dept. of Commun. Eng., Univ. of Oulu, Finland; † Dept. Elec. & Comp. Eng., Univ. of Manitoba, Canada.Email: { tsanguan, rrajathe, matla } @ee.oulu.fi; { Sudarshan.Guruacharya, Ekram.Hossain } @umanitoba.ca Abstract —The growing demand in indoor small cell networkshas given rise to the concept of micro-operators (MOs) for localservice delivery. We model and analyze a spectrum sharingsystem involving such MOs where a buyer MO buys multiplelicensed subbands provided by the regulator. Also, all small cellbase stations (SBSs) owned by a buyer MO can utilize multiplelicensed subbands at the same time which are also used byother MOs. A deterministic model in which the location of theSBSs are known can lead to unwieldy problem formulation,when the number of SBSs is large. Subsequently, we adopt astochastic geometric model of the SBS deployment instead of adeterministic model. Assuming that the locations of the SBSs canbe modeled as a homogeneous Poisson point process, we find thedownlink signal-to-interference-plus-noise ratio (SINR) coverageprobability and average data rate for a typical user (UE) servedby the buyer MO in a spectrum sharing environment. In order tosatisfy the QoS constraint, we provide a greedy algorithm to findhow many licensed subbands and which subband for the buyerMO to purchase from the regulator. We also derive the coverageprobability of the buyer MO for interference the limited system.
Index Terms —Micro-operator, spectrum sharing, stochasticgeometry, coverage probability, average data rate.
I. I
NTRODUCTION
In recent years, the concept of network infrastructure andspectrum sharing has been investigated to address the resourcesharing problem for the network operators. On one hand, withthe increasing of demand for mobile services, the under utiliza-tion of licensed spectrum auctioned off to the mobile networkoperators has become a bottleneck for the future growth ofthe industry [1]. On the other hand, in rural areas, where thedemand can be low, the high cost of network infrastructureforces the network operators to charge high prices to theircustomers, this makes the service unaffordable to most people[2]. One of the key aspects of the fifth generation (5G) mobilecommunication networks is to maximize the usage of existingnetwork resources in terms of spectrum, infrastructure, andpower while simultaneously minimizing the cost of purchasingresource, and reducing the energy consumption of the mobiledevices [2], [3].Nowadays wireless mobile service is given by typical mo-bile network operators which we refer to here as MNOs whosebusiness model is to offer services with very high infrastruc-ture investments and long investment period [3]. Regardingthe research works based on the MNOs’ points of view, in[4], the concept of neutral host network deployment wasproposed where the MNOs deploy cells in the best positionswith optimal tuning to satisfy the quality-of-experience (QoE). The authors also considered the sharing of other resources suchas spectrum, rate, power adaptation, edge caching, and loadbalancing, which can be done across different virtual MNOs.In order to facilitate the local licensing models and to constructin high frequency band, the new innovations for mobileedge computing, network slicing, software defined networking,massive MIMO and wireless backhauling was proposed in [5].In [6], hardware demonstration of the benefit of inter-operatorspectrum sharing was demonstrated. Resource sharing in thecontext of heterogeneous network and cloud RAN conceptswas proposed [7].Regarding stochastic geometry modelling of cellular sys-tems owned by the MNOs, in [8], the point processes thatmodel the spatial characteristics of the base stations (BSs)belonging to multiple MNOs was empirically studied, usingthe data from field surveys in Ireland, Poland, and UK. Theauthors conclude that the log-Gaussian Cox process is the bestfit for the deployment patterns of the BSs. In [9], the authorsconsidered a single buyer–multiple seller BS infrastructuremarket as a Cournot oligopoly market. They modeled thelocations of the base stations as a homogeneous Poisson pointprocess and obtained the downlink signal-to-interference-plus-noise ratio (SINR) coverage probability for a user servedby the buyer MNO in an infrastructure sharing environment.However, since the high volume of traffic densities comes fromindoor environment such as hospitals, campuses, shoppingmalls, sport arenas, it leads to the problem that the traditionalmacro cellular networks become insufficient when the buildingpenetration losses limit the indoor connectivity [10]. Hence,in the future, the business model, which is dominated by theMNOs, will become inadequate and various services cannotdevelop unless the wireless systems can response rapidly tothe specific local traffic requirements.One possible paradigm to address the above issue is to usethe concept of micro operator (MO) to serve the specific localconnectivity as firstly started in [11]. The authors identifiedthe business model for the new MO concept. In [10], the MOconcept with the relation between MO to other stakeholderswas proposed. Also the new spectrum regulation for MOnetwork was provided. In this paper, we consider the scenariowhere one MO buys multiple licensed subbands from theregulator. In the spectrum sharing deployment, all the SBSs ofthe buyer MO can utilize multiple subbands. Also, the buyerMO allows the other MO who has low activity of UEs toutilize each subband at the same time. As such, for downlinkransmission, each typical UE of the buyer MO experiencesinterference from the SBSs belonging to the other MO whois occupying that particular subband. We use results fromstochastic geometric analysis of large-scale cellular networksto evaluate SINR outage probability and the average data ratefor such a spectrum sharing system. In order to satisfy theQoS constraint in terms of coverage and the minimum requiredrate, we provide a greedy algorithm to find how many licensedsubbands and which subbands the buyer MO has to purchasefrom the regulator. Then, in the simulation results, we showthat spectrum sharing for MO network is beneficial for bothcoverage and average data rate.II. S
YSTEM M ODEL , A
SSUMPTIONS , AND Q O S A. System Model and Assumptions
Consider a system with the set of licensed spectrum subband L = { L , . . . , L , . . . , L J } owned by the regulator. We con-sider a system with K +1 micro-operators (MOs) given by theset K = { , , . . . , K } , where each MO serves different localarea such as, university, hospital and supermarket. Let MO- denotes the buyer who wants to buy the multiple licensedsubbands from the regulator and MO- k , where k ∈ K\{ } is the other MO who is occupying the subband L j , where L j ∈ L . We assume that each MO- k , where k ∈ K\{ } ,has low level of UEs’ activity. Let the set of small cell basestations (SBSs) owned by the MO- k be given by F k , where k ∈ K . Each of the SBSs and UEs is assumed to be equippedwith a single antenna. The maximum transmit power of eachSBS is p max also a UE subscribing to an MO associates to thenearest SBS. The SBSs owned by different MOs are spatiallydistributed according to homogeneous Poisson point processes(PPPs). Let the spatial intensity of SBSs per unit area ofMO- k be denoted by λ k , where k ∈ K . We can denote theoverall net BSs intensity of all MOs as sum of all λ k by λ + P k ∈K\{ } λ k .The licensed subbands are orthogonal and hence there is nooverlap between any two licensed subbands. We study whenthe buyer MO- buys multiple spectrum from the regulatorand it allows other MO- k , where k ∈ K\{ } , which has alow level of UEs’ activity, to use each subband. For spectrumsharing among multiple MOs, we assume that the followingassumptions hold: Assumption 1.
The MO- serves each typical UE of MO- itself using its own infrastructure while buying the licensedspectrum from the regulator. The typical UE of MO- asso-ciates with the nearest SBS in the set F owned by the MO- .Since each SBS can utilize multiple subbands, this implies thatin each subband L j ∈ L , the net SBS intensity that a typicalUE of MO- can associate itself with is λ A = λ ( L j ) . (1) Assumption 2.
When the SBSs of MO- use the subband L j ∈L , MO- allows at most one k th MO, where k ∈ K\{ } , touse the same subband simultaneously. Also, each SBS of MO- is assumed to use multiple subbands. As for the downlink transmission from one SBS to each typical UE, each UE willreceive transmissions from multiple subbands (channels) at thesame time. In each subband L j , the typical UE of MO- willexperience interference from SBSs F \{ } of MO- and F k ofMO- k , where k ∈ K\{ } . When all the SBSs of MO- use thelicensed subband L j ∈ L , we have the intensity of interferingSBSs in each subband L j as λ I ( L j ) = λ ( L j ) + ν k λ k ( L j ) , for k = 0 . (2) Here ≤ ν k ≤ denotes the level of UEs’ activity of MO- k in the subband L k . Fig. 1. The MO- buys only spectrum while using its own infrastructure. Fig. 1 illustrates the scenario when the MO- buys spectrumfrom the regulator. At the same time, there is another MO- k , where k ∈ K\{ } , which uses the same subband. In eachsubband L j , where L j ∈ L , MO- allows at most one MO- k toutilize the same subband. In this figure, MO- buys subbands L and L to serve its UE- which is associated with SBS- .Since the SBSs of MO- can use multiple subbands (accordingto Assumption 2 ), in this figure, we assume that SBS- ofMO- uses all three subbands L , L , and L to serve itsUEs.We see that in the subband L , the transmit signal from theSBS- of MO- creates interference to the typical UE- ofMO- . Similarly, in subbands L and L , the transmit signalfrom the SBS- of MO- and the transmit signal from theSBS- of MO- cause the interference to the typical UE- and the typical UE- of the MO- , respectively. B. SINR Coverage and Average Rate
Without loss of generality, we assume a typical UE of MO- located at the origin and associates with the nearest SBS ofMO- from the set given by F . For the typical UE of MO- ,we will denote the nearest SBS from F as SBS- .We assume that the message signal undergoes Rayleigh fad-ing with the channel power gain given by g . Let α > denotethe path-loss exponent for the path-loss model r − α , where r is the distance between the typical UE and SBS- . Let σ denote the noise variance, and p denote the transmit power ofll the SBSs in MO- , including SBS- . The downlink SINR at the typical UE of MO- is SINR = g r − α pI + σ , (3)where I is the interference experienced by a typical UEfrom the SBSs that operate on the spectrum L j where L j ∈ L . These are the SBSs that belong to MO- k ,where k ∈ K\{ } and MO- . Thus, the interference I = P i ∈F k ∪F \{ } ψ i,j g i r − αi p . Here g i is the co-channel gainbetween the typical UE and interfering SBS- i , and r i is thedistance between the typical UE and the interfering SBS- i ,where i ∈ F k ∪ F \{ } . The transmit power of each SBS is < p ≤ p max . Then, we assume ψ i,j ∈ { , } as a binaryvariable indicating whether the SBS- i is active (if ψ i,j = 1 )or inactive (if ψ i,j = 0 ) in spectrum subband L j .For a given threshold T , if SINR < T the UE is said toexperience an outage (i.e., outage probability P outage ( T ) =Pr(SINR < T ). Likewise, if SINR > T , then the UE issaid to have coverage (i.e., coverage probability, P c ( T ) =1 − P outage ( T ) = Pr(SINR ≥ T ). Given the SINR coverageprobability, the average downlink transmission rate for atypical UE can be computed as E [ R ] = Z ∞ P c ( e T − dT. (4)We consider both the SINR coverage probability and a mini-mum average rate as the QoS metrics for a typical user.III. A NALYSIS OF
SINR C
OVERAGE P ROBABILITY
A. SINR Coverage Probability When MO-0 Uses a SingleBand
Following to [12, Theorem 1], conditioning on the nearestBS at the distance r from a typical UE, the coverage proba-bility averaged over the plane is P c = Z r> Pr(SINR > T | r ) f r ( r ) dr, (5)where the probability density function (PDF) of r can beobtained as [12], f r ( r ) = e − πλr πλrdr . Using the fact thatthe distribution of the channel gain follows an exponentialdistribution, a formula for a coverage probability of the typicalUE when the BSs are distributed according to a homogeneousPPP of intensity λ is derived in [12, Eqn.2]. By observation,we can express the coverage probability in the most generalform in terms of three components which are noise, interfer-ence and user association while each BS employs a constantpower p = 1 /µ as follows: P c = Z z> e − µT z α/ σ | {z } noise e − π ( λ I ( β − z | {z } interference e − λ A πz πλ A | {z } user association dz, (6)where λ I is the BS intensity causes interference to a typicalUE, the UE associates with the closest BS (where the BSintensity is λ A ), the path-loss exponent is denoted as α , and β is given by β = 2( T /p ) /α α E g [ g /α (Γ( − /α, T g/p )) − Γ( − /α )] . (7) In particular, the general expression of the coverage proba-bility in (6) can be expressed as [12, Theorem 1] P c = πλ A Z ∞ exp {− ( Az + Bz α/ ) } dz, (8)where A = π [( λ I ( β − λ A ] and B = T σ p . When theinterfering links undergo Rayleigh fading, β = 1 + ρ ( T, α ) ,where ρ ( T, α ) = T /α Z ∞ T − /α (1 + u α/ ) − d u. (9)For this special case, we see that β is independent of transmitpower. Except for α = 4 , the integral for P c cannot beevaluated in closed form. Nevertheless, a simple closed-formapproximation for the general case, where α > , and whereboth noise and intra-operator interference are present, can begiven as [13, Eqn.4] P c ≃ πλ A " A + α B /α Γ (cid:0) α (cid:1) − , (10)in which Γ( z, a ) = R ∞ z x a − e − x dx is the upper incompleteGamma function. B. SINR Coverage Probability Under Spectrum Sharing
In our spectrum sharing model, the regulator sells thelicensed subband to the MO- while some of the SBSs ofthe MO- k , where k ∈ K\{ } , are using the same subband.Also, we consider that all SBSs of MO- are using L licensedsubbands (where | L | = L , | A | denotes the cardinality of aset A ) at the same time. Due to the fact that MO- buys onlyspectrum, the UEs of MO- always associates to the SBS- ,where { } ∈ F . For our system, since the SBSs of MO- utilize multiple subbands at the same time, we have to modifythe formulas (6) and (8) and show that a more general coverageformula is given as follows: Proposition 1.
Under
Assumption 1 and
Assumption 2 , thecoverage probability of a typical UE of MO- is P c = X L j ∈L P c ( L j ) Pr( L j ) , (11) where P c ( L j ) denotes the coverage probability of the UE ofMO- using band L j and P r ( L j ) is the probability of thetypical UE of MO- using band L j . By using the conditional probability, we obtain the coverageprobability P c ( L j ) of the MO- when it uses the band L j given the probability of using the band L j , Pr( L j ) , in which, Pr( L j ) ∈ { , } indicating whether the MO- buys and usesthe licensed band L j (if Pr( L j ) = 1 ) or the MO- does not usethe subband L j (if Pr( L j ) = 0 ). Then, we take the summationover the bands L , where | L | = L .Let us consider he case when λ I ( L j ) = λ ( L j )+ ν k λ k ( L j ) and ν k = 1 . Since all SBSs F of MO- using the licensedband and in each aubband L k , while there is MO- k who haslow activity UEs occupying that band. We can denote ν k = 1 n (2). The intensity of interfering SBSs in the band L k is λ I ( L k ) = λ ( L k ) + λ k ( L k ) , where k ∈ K\{ } . Proposition 2.
Under
Assumption 1 and
Assumption 2 , thecoverage probability of a typical UE of MNO- using the band L j , where L j ∈ L , is given by P c ( L j ) = πλ A Z ∞ exp {− ( A z + Bz α/ ) } dz, (12) where A = π (( λ ( L k ) + λ k ( L k )) β − λ k ( L k )) , and by Assumption 1 , we can assume λ A = λ . Also, β and B aregiven by (7) and (8), respectively. Then, we can approximate P c ( L j ) in (12) using (10) as P c ( L j ) = πλ ( L j ) A + α B /α Γ ( α ) . (13) Without loss of generality, we can assume
Pr( L j ) = 1 ,where L j ∈ L . Hence, we obtain P c as P c = X L j ∈L πλ ( L j ) A + α B /α Γ (cid:0) α (cid:1) . (14) Proof:
We obtain P c ( L j ) in (12) by using an expressionin (8) after that evaluating (12) by using a closed formapproximation in (10), where B and β are the same as (8).Then, we can express P c by substituting P c ( L j ) in (11) whileassuming Pr( L j ) = 1 .Next, we consider the scenario when the system becomes “interference-limited” , which occurs when σ → . Proposition 3.
The coverage probability for interference-limited case when the MO- using the subband L j , where L j ∈ L , can be expressed as P c = X L j ∈L β + ( β − λ k ( L j ) λ ( L j ) , (15) in which, the λ k ( L j ) and λ ( L j ) are the SBS intensity ofMO- k and MO- using the band L j , respectively.Proof: Let L = |L| . For interference-limited case, i.e., B → in (14), and let C = α B /α Γ (cid:0) α (cid:1) as such we canneglect C when the system becomes interference limited. Aftersimplifying (14) while assuming C = 0 , we have the requiredresult.IV. A MOUNT OF S PECTRUM B ANDS R EQUIRED TO S ATISFY THE Q O SThe expected rate can be derived using the closed formapproximation of coverage probability from (10) with theassumption that interference is Rayleigh fading where β = 1 + ρ ( ˆ T , α ) , given in (9), with ˆ T = e T − . The expected rate E [ R ] for the general case can be given by E [ R ] = πλ A Z ∞ (cid:20) π (cid:0) λ I ˆ T /α Z ∞ ˆ T − /α (1 + u α/ ) − du + λ A (cid:1) + α B /α Γ(2 /α ) (cid:21) − d ˆ T , = πλ A Z ∞ (cid:20) π (cid:0) λ I Tα − F (cid:0) , − α ; 2 − α ; 1 − ˆ T (cid:1) + λ A (cid:1) + α B /α Γ(2 /α ) (cid:21) − d ˆ T , (16)where Γ( z ) , and F ( a, b, c, z ) are the Gamma function, andthe Hypergeometric function, respectively. The average rate in(16) is valid for any real values of α > , T > and can beevaluated by numerical integration techniques. Proposition 4.
The expected rate of a typical UE of MO- using multiple subbands L , while in each subband L j ∈ L ,there is other MO- k who has low activity UE using the samesubband is E [ R ] = X L j ∈L πλ ( L j ) Z ∞ (cid:20) π (cid:8) ( λ ( L j ) + λ k ( L j )) Tα − ! × F (cid:0) , − α ; 2 − α ; 1 − ˆ T (cid:1) + λ (cid:9) + α B /α Γ(2 /α ) (cid:21) − d ˆ T . (17)
Proof:
As the average rate is obtained by taking anintegration of P c with respect to ˆ T from zero to infinity, and weuse P c from (14). Following Assumption 1 and
Assumption2 , when MO- uses the band L j to serve its UE, we substitute λ A = λ ( L j ) and λ I = λ ( L j ) + λ k ( L j ) in (16). We obtainthe result in (17).Let us further assume that the MO- wants to ensure thatthe coverage probability of a typical UE satisfies the QoSconstraint P c ≥ − ǫ, (18)where < ǫ < is some arbitrary value.In order to satisfy the QoS constraint in (18), the buyerMO- will select the number of licensed subbands needed,at minimum cost, such that it can serve its UEs guaranteeingsome QoS. Due to the fact that, there is a cost associated witheach subband denoted as q L k , then we use greedy algorithm tofind which subband and how many of them the MO- will buyto satisfy the QoS of its UE. Let N = P l ∈L N l denote theminimum number of licensed subbands needed for the MO- . For the QoS condition of the minimum rate requirementneeded at each UE of the buyer MO- denoted by R min to befeasible, the minimum number of licensed subbands neededmust satisfy N × E [ R ] ≥ R min , (19)Let E [ R ] denote the expected rate at the UE of MO- obtainedfrom (16). For MO- , the maximum number of bands requiredn order to satisfy both SINR coverage and the minimum raterequired at it’s typical UE is L max = max { N, M } , (20)where M = P l ∈L M l is the number of licensed subbandsneeded to satisfy the rate constraint. Let us now proposea simple greedy algorithm in [14, Chap 17.1] to select forthe MO- to select which licensed subband and how manylicensed aubbands that the MO- will purchase from theregulator. The greedy algorithm is provided in Algorithm1 . The idea behind this greedy algorithm is as follows: Wefirst sort the licensed subbands L j ∈ L according to the costper subband q L j in an ascending order. After using the greedyalgorithm, we obtain L max number of licensed subbands inorder to satisfy both of the coverage QoS and the minimumrate needed at the UE. Algorithm 1
Greedy Algorithm Sort the subbands by q L j in ascending order such that q π ≤ q π · · · ≤ q π LJ for i = 1 to L J do Set N = { π , . . . , π L j } where |N | = N if P l ∈N πλ ( l ) A + α B /α Γ (cid:0) α (cid:1) ≥ − ǫ then Compute P l ∈N M l = M . end if if P l ∈L N l × E [ R ] ≥ R min then Compute P l ∈N N l = N . Terminate end if
Compute L max = max { N, M } . end for Compute P c using (14) with L = L max . Compute E [ R ] using (16) with L = L max .V. N UMERICAL R ESULTS
We assume that the SBSs are spatially distributed accordingto homogeneous PPP inside a circular area of meter radiusfor all K + 1 MOs. The MOs are assumed to have the sameintensity of SBSs per unit area. The maximum transmit powerof each SBS is p max = 10 dBm. The path-loss exponentis α = 4 , and noise power σ = − dBm. Each SBSfrom all MOs transmits at the maximum power. The coverageprobability is obtained from (14) and the average data rateis plotted accordingly. We illustrate the simulation results forthe case when the buyer MO- purchases multiple licensedsubbands while assuming that the cost of each subband isequal. A. Effect of Changing the Average Number of SBSs of MO- per Unit Area In Fig. 2 and Fig. 3, the simulation parameters are asfollows: the SINR threshold at each typical UE of MO- isset to T = 10 dB. We consider when the regulator sells twolicensed subbands and each subband has one MO- k , where Average no. of BSs of MO-0 per area π *500 C o v e r age P r obab ili t y ( P c ) λ k = [4,6]/( π *500 ) λ k = [8,10]/( π *500 ) λ k = [12,14]/( π *500 )L = 2 Fig. 2. The coverage probability of the typical UE of MO- while increasingthe number of MO- per unit area. π *500 A v e r age D a t a R a t e ( b i t s / s e c ) λ k = [4,6]/( π *500 ) λ k = [8,10]/( π *500 ) λ k = [12,14]/( π *500 )L = 2 for k ≥ Fig. 3. The average data rate of the typical UE of MO- while increasingthe number of MO- per unit area. k ∈ K\{ } is occupying the subband. The MO- is usingits own infrastructure to serve its UE. We consider the caseswhen the MO- buys two licensed subband, it means that eachSBS of MO- utilizes two licensed spectrum at the same time.Fig. 2 plots the coverage probability when the average numberof SBSs of MO- per unit area π × is increased. Whenaverage number of SBSs of MO- increases, the coverageprobability of MO- is also increased. By increasing the SBSsintensity of the MO- k , where k ∈ K\{ } in each subband L k ,we see that the coverage probability of MO- decreases. Thisis not surprising since the SBS intensity of MO- k in which k ∈ K\{ } causes interference in each band L k .In Fig. 3, the average data rate of MO- is shown. When theaverage number of SBSs of MO- increases, the average datarate of MO- is increased. We also show when the averagenumber of SBS of MO- is very high such that it tends toinfinity, the average data rate tends to saturate at one value.Also when the SBS intensity of MO- k , where k ∈ K\{ } increases, the average data rate decreases. B. Effect of Changing the SINR Threshold
In Fig. 4 and Fig. 5, we illustrate the coverage probabilityand the average rate of MO- when the increasing of SINRthreshold ( T ) at each UE of MO- . The SBS intensity of MO- Threshold (T) C o v e r age P r obab ili t y L=2, λ k = [4,6]/( π *500 )L=4, λ k = [4,6,8,10]/( π *500 )L=6, λ k = [4,6,8,10,12,14]/( π *500 ) λ = 10/( π *500 ) Fig. 4. The coverage probability of the typical UE of MO- when increasingthe SINR threshold (T).
10 20 30 40 50 60 70
Threshold (T) A v e r age D a t a R a t e ( b i t/ s e c ) L=2, λ k = [4,6]/( π *500 )L=4, λ k = [4,6,8,10]/( π *500 )L=6, λ k = [4,6,8,10,12,14]/( π *500 ) λ = 10/( π *500 ) for k ≥ Fig. 5. The average data rate of the typical UE of MO- with increasing theSINR threshold ( T ) at the UE of MO- . is set to λ = 10 / ( π ∗ ) . We see that when T increases,the SINR coverage probability of MO- decreases. We alsoconsider when the MO- buys two, four and six licensedbands. In Fig. 4, we see that when the number of licensedsubbands increases, the coverage probability also increases.Although for the case of the MO- buys six licensed subbandswith high SBSs intensity of MO- k , the coverage of MO- stillincreases. In Fig. 5, we see that the average data rate remainsconstant when the SINR threshold increases. This is because,we first calculate the coverage probability for each threshold( T ). Then, integrate the coverage probability P c with respect tothe threshold, the expression of the average data rate becomes(17). As such, the average data rate remains constant withthe changing of T . However, the average data rate increasessignificantly when MO- buys more bands.VI. C ONCLUSION
We have studied the problem of spectrum sharing amongmultiple micro-operators (MOs) using stochastic geometry, where the buyer MO buys multiple subbands from a regu-lator. Also, the buyer MO allows other MOs to utilize thesame subband. We have first analyzed the downlink coverageprobability for a typical user served by a buyer MO, andsubsequently, we have derived the average data rate. Boththe SINR coverage and a minimum rate requirement areconsidered as the QoS metrics. In order to satisfy the QoSconstraints of the typical user served by the buyer MO, wehave provided a greedy algorithm to find how many subbandsand which subbands for the buyer MO to purchase from theregulator. Both the coverage and the average data rate of thebuyer MO increase when the buyer MO buys more licensedsubbands. However, when the average number of SBS per unitarea of the buyer MO increases and approaches infinity, theaverage data rate for a typical user served by the buyer MOsaturates to a single value.A
CKNOWLEDGMENTS
For this work, the authors would like to acknowledge thesupport from Finnish Funding Agency for Technology andIn- novation (TEKES), Nokia Networks, Anite Telecoms,Broadcom Communications Finland, Elektrobit Wireless Com-munications and Infotech Oulu Graduate School. Also, supportfrom the Natural Sciences and Engineering Research Councilof Canada (NSERC) is acknowledged.R
EFERENCES[1] Cisco, “Visual Networking Index: Global Mobile Data Traffic ForecastUpdate,” Cisco, San Jose, CA, USA, Feb. 2014, 2013–2018.[2] International Telecommunications Union, “Mobile infrastructuresharing,”
ITU News Magazine . [Online]. Accessed: 28 Oct. 2016.Available: .[3] M. Matinmikko et al. , “Spectrum sharing using licensed shared access:the concept and its workflow for LTE-advanced networks,”
IEEE WirelessCommun. vol. 21, no. 2, pp. 72–79, 2014.[4] 5G America, “Multi-operator and neutral host small cells, Drivers,architectures, planning and regulation,” 2016.[5] C. Dehos, et al. , “Millimeter-wave access and backhauling: the solutionto the exponential data traffic increase in 5G mobile communicationssystems?.”
IEEE Commun. Magazine , vol. 52, pp. 88–95, 2014.[6] E.A. Jorswieck et al. , “Spectrum sharing improves the network efficiencyfor cellular operators,”
IEEE Commun. Mag. , vol. 52, no. 3, pp. 129–136,2014.[7] M. Antonio Marotta, et al. ,“Resource sharing in heterogeneous cloudradio access networks,”
IEEE Wireless Commun. , vol. 22, pp. 74–82,2015.[8] J. Kibida, B. Galkin, and L. A. DaSilva, “Modelling multi-operator basestation deployment patterns in cellular networks,”
IEEE Trans. on MobileComputing , vol. 15, no. 12, pp. 3087–3099, Dec. 1 2016.[9] T. Sanguanpuak et al. , “Inter-operator infrastructure sharing: Trade-offsand Market,”
Proc. IEEE Int. Conf. Commun. (ICC) , to appear.[10] M. Matinmikko et al. , “Micro operaors to boost local service deliveryin 5G,”
Wireless Personal Communications , to appear.[11] P. Ahokangas et al. , “Future micro operators business models in 5G,”
Proc. Int. Conf. on Restructuring of the Global Economy (ROGE) , 2016.[12] J.G. Andrews, F. Baccelli, and R.K. Ganti, “A tractable approach tocoverage and rate in cellular networks,”
IEEE Trans. on Commun. , vol.59, no. 11, pp. 3122–3134, Nov. 2011.[13] S. Guruacharya, H. Tabassum, and E. Hossain, “Integral approximationsfor coverage probability,”
IEEE Wireless Commun. Letters , vol. 5, issue1, pp. 24–27, Feb. 2016.[14] B. Korte and J. Vygen,