Distance [m]

Distance[m]

10 20 30 40 50 60 70

35

30

25

20

15

10

Power

[dBm]

−100

−90

−80

−70

−60

−50

−40

−30

−20

−10

0

Figure 1: Simulation scenario.

Table 1: System-Level Simulation ResultsDeployed

Scheme Rand. MNL IMOur

users Method

4 users/cellOutage 15.09% 8.62% 2.59% 0.86%

8RB networkTx.-ing Users 31.13 33.11 35.20 35.75

Mbps 7.65 8.24 8.77 8.83

8 users/cellOutage 19.87% 19.77% 15.52% 3.19%

8RB networkTx.-ing Users 58.36 58.59 60.47 69.63

Mbps 14.40 14.48 14.98 16.78

minχu,k,r

U∑u=1

K∑k=1

R∑r=1

pfu,k,r · χu,k,r (2a)

subject to: U∑u=1

R∑r=1

χu,k,r ≤ 1 ∀k (2b)

R∑r=1

ρu,r ≤ 1 ∀u (2c)

χu,k,r ≤ ρu,r ∀u, k, r (2d)

K∑k=1

R∑r=1

Θ · effr · χu,k,r ≥ TP requ ∀u (2e)

ρu,r ∈ {0, 1} ∀u, r (2f)

χu,k,r ∈ {0, 1} ∀u, k, r (2g)

where χu,k,r (2g) is a binary decision variable that is equal to1 if user u uses MCS r in RB k, or 0 otherwise, ρu,r (2f) is abinary decision variable that is equal to 1 if user u makes useof MCS r, or 0 otherwise, constraint (2b) makes sure thatRB k is only assigned to at most one user u, constraints (2c)and (2d) together guarantee that each user is allocated toat most one MCS, and constraint (2e) makes sure that eachuser u obtains a throughput no less than its demand TP req

u .Average channel quality in the form of instantaneous wu,k

averaged over tens of user MRs is utilised to compute pfu,k,r,

thus avoiding rapid fluctuations of pfu,k,r due to fast-fading,which may result in fast variations of resource assignments.One possible approach to solve (2) is to apply integer lin-

ear programming (ILP) techniques readily available in soft-ware packages. However, although ILP solvers are able tosolve (2) up to optimality, their running times are unpre-dictable (exponential in the worst case), which renders theminappropriate for real-time use at femto BSs. Hence, to solve(2) in short time, we propose a two-level decomposition ap-proach based on an heuristics search and a minimum costnetwork flow problem. Assuming that a MCS ru has been as-signed to user u ∀u, i.e., ρu,r∀u∀r is known a priori as part ofthe input, (2) can be reduced to a simpler form, i.e., a user-

to-RB assignment problem weighted by pfu,k,r∀u∀k. Thisassignment problem can be formulated as a minimum costnetwork flow problem, an optimally solved using a networksimplex algorithm. Thereafter, a heuristic search can beperformed over the MCS assignment solution space, where

1 2 3 4 5 6 7 80

2

4

6

8

1 2 3 4 5 6 7 80

2

4

6

8

Pow

er (m

W)

1 2 3 4 5 6 7 80

2

4

6

8

RBs

Femtocell 2

Femtocell 1

Femtocell 3

Figure 2: RB allocation.

for each MCS assignment to cell users, the optimal RB andpower allocation can be found by using our network sim-plex algorithm. Since network simplex runs fast, RB andpower allocations to users can be updated on a millisecondbasis (dealing with fast channel variations), while the MCSassignment can be updated in a half-second basis (dealingwith traffic load and mobility).

5. PERFORMANCE AND CONCLUSIONSThe performance of the proposed self-organisation rule

has been investigated in an enterprise femtocell scenario(Fig. 1). Simulation results show that compared to exist-ing radio resource management techniques in the literature,i.e., random [3], network listening mode (NLM) [4] and in-terference minimisation (IM) [5], our approach is able tosignificantly decrease the number of user outages, increasethe average number of simultaneously transmitting users inthe network (around 15%), and enhance the average networksum throughput (around 12%) (Tab. 1). This is because ourapproach (self-organization rule) achieves inter-cell interfer-ence coordination without the need of assigning orthogonalRBs among neighbouring cells. Instead, it allows all cells toallocate all RBs to their users in an intelligent manner: A cellthat minimises its own transmit power assigns less power tothose RBs allocated to users having good channel conditionsor with lower data-rate demands. Thus, neighbouring cellswill ‘see’ low interference in such RBs and will allocate themto users having bad channel conditions or with large data-rate demands, thereby improving spatial reuse. To illustratethis self-organising feature, Fig. 2 shows the transmit powerallocated by 3 neighbouring femtocells at a given time in8 available RBs. We can see that each femtocell tends toallocate higher power levels in RBs in which neighbouringfemtocells assign lower power levels and vice versa. In thisway, a same RB can be dynamically reused in neighbouringfemtocells. Therefore, our proposed sefl-organisation ruleand resource allocation model effectively introduce an im-plicit coordination among independently operating femto-cells through distributed dynamic radio resource allocations.

6. REFERENCES[1] V. Chandrasekhar et. al. Femtocell Networks: A

Survey. IEEE Comm. Mag., 46(9):59–67, Sep. 2008.

[2] E. Dahlman et. al. 3G Evolution. HSPA and LTE forMobile Broadband. Elsevier, 2 edition, Aug. 2008.

[3] V. Chandrasekhar et. al. Spectrum Allocation in TieredCellular Networks. IEEE Transactions onCommunications, 57(10):3059–3068, Oct. 2009.

[4] J. Ling et. al. On Resource Allocation in DenseFemto-Deployments. In IEEE COMCAS, Nov. 2009.

[5] J. Zhang et. al. Femtocells: Technologies andDeployment. John Wiley and Sons, Jan. 2010.

411