A self-organized resource allocation scheme for ...epubs.surrey.ac.uk/807410/1/A self-organized resource allocation sc… · Femtocell overlaid on existing macro-cell network. interference

WIRELESS COMMUNICATIONS AND MOBILE COMPUTINGWirel. Commun. Mob. Comput. 2016; 16:330–342Published online 19 August 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.2518

RESEARCH ARTICLE

A self-organized resource allocation scheme forheterogeneous macro-femto networksMahima Mehta1*, Nirbhay Rane1, Abhay Karandikar1, Muhammad Ali Imran2 andBarry G. Evans2

1 Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai, India2 CCSR, University of Surrey, Guildford, U.K.

ABSTRACT

This paper investigates the radio resource management (RRM) issues in a heterogeneous macro-femto network. The objec-tive of femto deployment is to improve coverage, capacity, and experienced quality of service of indoor users. The locationand density of user-deployed femtos is not known a-priori. This makes interference management crucial. In particular,with co-channel allocation (to improve resource utilization efficiency), RRM becomes involved because of both cross-layerand co-layer interference. In this paper, we review the resource allocation strategies available in the literature for hetero-geneous macro-femto network. Then, we propose a self-organized resource allocation (SO-RA) scheme for an orthogonalfrequency division multiple access based macro-femto network to mitigate co-layer interference in the downlink transmis-sion. We compare its performance with the existing schemes like Reuse-1, adaptive frequency reuse (AFR), and AFR withpower control (one of our proposed modification to AFR approach) in terms of 10 percentile user throughput and fairnessto femto users. The performance of AFR with power control scheme matches closely with Reuse-1, while the SO-RAscheme achieves improved throughput and fairness performance. SO-RA scheme ensures minimum throughput guaranteeto all femto users and exhibits better performance than the existing state-of-the-art resource allocation schemes. Copyright© 2014 John Wiley & Sons, Ltd.

KEYWORDS

co-layer interference; femtocell; macrocell; OFDMA; resource allocation

*Correspondence

Mahima Mehta, Department of Electrical Engineering, Indian Institute of Technology Bombay, Mumbai, India.E-mail: [email protected]

1. INTRODUCTION

With the rapid increase in varied wireless applications andthe advent of smart-phones, personal digital assistants, etc.,there is a tremendous proliferation in indoor voice anddata traffic. It is envisaged that in future, about 50% ofvoice traffic and 70% of data traffic will originate fromindoor wireless users [1]. However, there are large penetra-tion losses and attenuation indoors due to which the indoorusers often suffer from quality of service (QoS) degra-dation [1]. To meet the high data rate requirement, effi-cient mechanisms for resource allocation and interferencemitigation indoors are needed. Femto Base Station (BS)deployment is one such mechanism to meet these objec-tives. It is a short-range, user-deployed, and low-powernode operating in the licensed spectrum. It connects mobiledevices to a cellular operator’s network using residentialdigital subscriber lines/wired broadband connections [2].The purpose of femto deployment is to improve capacity

(by achieving higher rates due to the proximity to indoorusers and increasing reuse of resources) and coverage(by covering the dead zones formed because of insuffi-cient macro signal penetration) in the indoor environment.Because of power-efficient transmission, femtocell net-work improves battery life and contributes to greenercommunication. Moreover, femto offloads indoor trafficfrom the macrocell, which increases capacity of macroBS and reduces capital expenditure (CAPEX) and oper-ational expenditure (OPEX) of network operator. Thus,macro-femto networks are beneficial to both operator andsubscribers.

Femtocell network is an overlay deployment (Figure 1)by the indoor users and resource allocation is doneindependently by macro and femto BSs. This makesradio resource management (RRM) in a macro-femto net-work challenging. As macrocells and femtocells share theavailable radio resources, it may cause cross-layer inter-ference (between femtocell and macrocell) and co-layer

330 Copyright © 2014 John Wiley & Sons, Ltd.

M. Mehta et al. SO-RA for HetNets

Figure 1. Femtocell overlaid on existing macro-cell network.

interference (between neighboring femtocells) [3]. Ideally,orthogonal resource allocation alleviates cross-layer inter-ference but results in poor resource utilization efficiency.Therefore, co-channel allocation is preferred. Variouscross-layer interference reduction schemes are available inthe literature [4–6]. However, limited literature is avail-able to address the problem of co-layer interference. In[7] and [8], authors have proposed schemes, which beginwith an orthogonal allocation amongst the co-layer fem-tocells. Then, they apply different variants of adaptivereuse schemes to increase resource utilization efficiencybased on either power control or coordination betweenneighboring femto BSs.

In general, cross-layer interference mitigation has beenaddressed sufficiently well in the literature. Therefore, weassume in this paper that the cross-layer interference is mit-igated by using one of the well illustrated schemes [9] andfocus on co-layer interference mitigation only. However,our proposed self-organized resource allocation (SO-RA)scheme reduces cross-layer interference in implicit man-ner as described in Section 3.3. Reviewing the differentschemes available in the literature (discussed in Section 2),it is realized that coordination between neighboring femtoBSs is essential to adapt the resource allocation strategyintelligently (according to the changes in interference lev-els). With this motivation, we propose a SO-RA scheme forheterogeneous macro-femto network, which mitigates co-layer interference. The distinct feature of our scheme is thatin addition to reduced co-layer interference, it ensures fair-ness and improves 10 percentile throughput performancefor femto users. It is to be noted that although we illustrateour proposed scheme in the framework of an orthogonalfrequency division multiple access (OFDMA)-based LongTerm Evolution (LTE) network, it is applicable to anycellular system in general. Also, our proposed scheme pro-vides a generalized framework of self-organized resourceallocation, which can be applied to any small-cell net-work. The only distinction would be the interface used toexchange information between the small cells.

The outline of this paper is as follows: Section 2 dis-cusses few significant approaches available in the literaturefor interference management in macro-femto networks.Section 3 explains the system model and describes ourproposed resource allocation schemes. One is the adaptivefrequency reuse (AFR) with power control, which is ourproposed modification to AFR approach. Other is the SO-RA scheme. Section 4 discusses the simulation results andinferences. Finally, we conclude in Section 5.

2. RELATED WORK

An overview of interference analysis and resource alloca-tion approaches in a macro-femto network is given in [3].The fundamental trade-off in achieving interference man-agement in macro-femto networks is to ensure two things:(i) interference due to femto BS does not severely affectmacro user equipments (MUEs) and neighboring femtoUEs (FUEs) and (ii) the transmit power of femto BS mustbe sufficiently high to ensure that the rate requirement ofFUEs are met.

A centralized approach is one of the ways to achievethis trade-off. In this approach, a centralized controlleruses information from femto BSs and FUEs to mitigatecross and co-layer interference. However, because of therandom variations in topology and large number of femtoBSs, centralized approach may not be scalable. Anotherapproach could be coordination-based, where intelligentdecisions for resource allocation and interference mitiga-tion are based on information exchange between macroand femto BSs. We briefly review such coordination-based approaches available in the literature for interferencemanagement.

� Resource Partitioning based methods: The simplestof the resource partitioning mechanism is to allo-cate all resources to all femtocells. This is knownas Frequency Reuse-1 or simply Reuse-1. Suchresource allocation improves the resources utiliza-tion. However, it may increase the co-layer inter-ference significantly when the density of femtocellsincreases. In [9–11], authors suggest Fractional Fre-quency Reuse (FFR) for OFDMA-based macro-femtonetwork. They deploy hybrid spectrum allocationwhere orthogonal allocation is deployed for innerfemtocells (located close to macro BS) and sharedallocation for the outer femtocells (located away frommacro BS). These schemes ensure cross-layer inter-ference mitigation to the FUEs located near macroBS. Dynamic resource partitioning for cross-layerinterference avoidance is proposed in [6], wherefemto BSs are denied access (via wired backhaul)to those resources that are assigned to nearby macroUEs. In [12], authors propose a low complexity ran-domized interference avoidance method for femto-cells. Each femtocell is allocated a random subsetof resources considering the fact that neighboringfemtocells are unlikely to use the identical resources

Wirel. Commun. Mob. Comput. 2016; 16:330–342 © 2014 John Wiley & Sons, Ltd. 331DOI: 10.1002/wcm

SO-RA for HetNets M. Mehta et al.

consistently. In [13], location based resource man-agement algorithm is proposed which allows femto-cell to reuse macrocell resources to increase spatialreuse. Authors provide a hashing scheme basedresource allocation for femtocells which does notrequire coordination. In [7,8,14,15], coordination-based co-layer interference avoidance and adaptivefrequency reuse (AFR) algorithms are proposed. Itbegins with orthogonal resource allocation to femto-cells and then, coordination-based resource reuse isdeployed to improve spectrum efficiency. The coor-dination between femtocells ensures that co-layerinterference remains below the acceptable level. Dueto the initial orthogonal allocation, such schemes suf-fer from poor resource utilization in the initial phaseof algorithm.

� Transmit power control based methods: In [16]and [17], a power control method is suggested toachieve constant femto BS coverage while ensur-ing no adverse impact on the macrocell throughput.In [18], a reward (signal to interference ratio) andpenalty (interference) based objective function is for-mulated for femto BSs in which the interfering femtoBSs reduce their transmit power to mitigate cross-layer interference. Similar adaptive power controlalgorithms to mitigate cross-layer interference arediscussed in [4] and [5]. In [19], two joint power con-trol and resource allocation schemes are discussed,one is centralized while other is a coordination-baseddistributed scheme.

� Cognition based methods: Femtocells may determineinterference pattern and resource utilization of net-work cognitively [20]. Authors consider femto BSsas secondary users, determine the available chan-nels cognitively and design autonomous algorithms

for cross-layer interference management. The bene-

fits of cognitive approach depend on the spectrum

occupancy of primary macrocell UEs.

� Self-organized and learning based methods: In [21],

we have proposed a self-organized resource alloca-

tion algorithm to mitigate intercell interference in a

macro-relay network. Focussing on macro-femto net-

work in [22], resource allocation algorithm to avoid

co-layer interference is executed at the backhaul after

each femtocell identifies its neighboring femtocells.

Authors in [23] propose sensing and tuning phase

to minimize interference and maximize the system

performance. One method is based on information

exchange between femtocells and other, on measure-

ment reports from users. Both schemes give improved

performance compared to random allocation policy.

Authors describe a self optimization framework to

jointly optimize spectrum assignment and transmis-

sion power in [24] and Q-Learning based distributed

interference control scheme for self-organized fem-

tocell network in [25] to mitigate cross-layer

interference.

In this paper, we propose a scheme, which mitigates co-

layer interference, while improving the minimum rate (10

percentile throughput) achieved by FUEs and ensuring

fairness to them. This issue of improved minimum rate

achieved by FUEs and fairness along with interference mit-

igation has not been addressed in the literature to the best

of our knowledge.

Figure 2. Network layout.

332 Wirel. Commun. Mob. Comput. 2016; 16:330–342 © 2014 John Wiley & Sons, Ltd.DOI: 10.1002/wcm


3. SYSTEM MODEL ANDPROPOSED RESOURCEALLOCATION SCHEMES

3.1. System model

We consider the downlink transmission scenario in anOFDMA-based macro-femto network. Our system modelconsists of seven macrocells with macro BS located at thecenter of each macrocell. L femtocells are overlaid in thecentral macrocell (Figure 2). In an OFDMA-based LTEnetwork [26], the system resources are divided along fre-quency (subcarriers) and time slots. These resources arescheduled in units of physical resource blocks (PRBs).Each PRB (bandwidth D 180 kHz) consists of 12 sub-carriers. We assume that N PRBs are available for bothmacrocells and femtocells. To compute SINR, we use thepath loss models for these links: between FUE and femtoBS when FUE is in the same/different apartment as femtoBS, and between FUE and macro BS when FUE is insidethe apartment as specified in [27].

Next, we describe both of our proposed resource alloca-tion schemes.

3.2. Proposed AFR with powercontrol scheme

In this section, we illustrate our proposed modificationto the existing AFR scheme, and we call it adaptive fre-quency reuse (AFR) with power control. The existing AFRscheme [7,8,14,15] achieves coordination-based co-layerinterference avoidance, where the neighboring femtocellsuse nonoverlapping resources initially. Later, femtocellsattempt to reuse the resources by coordinating with neigh-bors to improve resource utilization efficiency. It is ensuredthat the co-layer interference does not exceed the tolerablethreshold. However, its limitations are as follows: (i) ini-tially, system performance is low due to inefficient resourceutilization, and (ii) this scheme does not consider the reuseof resources at lower power.

Adaptive frequency reuse with power control is our pro-posed modification to the existing AFR scheme. Whena femto BS cannot use a particular PRB at full trans-mit power, it is likely that the same PRB may be usedwith reduced power level without causing significant inter-ference to its neighboring femtocells. We exploit thisconcept in AFR with power control scheme, which is atwo-step interference coordination algorithm. In the firststep, resources are shared between interfering femtocellsin an orthogonal manner. Then in the second step, reuse ofresources is facilitated with power control. When a femtoBS is not allowed to use a PRB at full transmit power due tointerference concerns, it checks for the feasibility of usingthe same PRB at half of the transmit power. This feasi-bility is determined by a-priori interference measurement,which may be caused to neighboring femtocells if this PRBwas used. If this interference is less than the acceptablethreshold, femto BS is allowed to use that PRB at half the

transmit power. Note that the idea behind using half thetransmit power is to exploit the possibility of maximiz-ing throughput of femtocells by transmit power variation,which was not explored in the AFR scheme.

3.3. Proposed self-organized resourceallocation scheme

There is an inevitable trade-off between aggressiveresource reuse and co-layer interference. Our proposedself-organized resource allocation scheme meets this trade-off by coordination between femtocells. It is a two-stepalgorithm to reduce co-layer interference between fem-tocells, while ensuring rate requirement satisfaction andfairness to FUEs.

In Step-1, interfering neighbor set for each femtocell isidentified and Reuse-1 is employed. In Step-2, each femtoBS identifies PRB, which offers minimum SINR and per-forms two levels of a-priori check to ensure that droppingof that PRB does neither cause any degradation in systemperformance nor in its own performance. The femto BSdrops that identified PRB to reduce co-layer interference.This is performed iteratively for all femto BSs. This self-organized resource allocation algorithm mitigates co-layerinterference by exchanging information with neighboringfemtocells locally and achieves an overall improvementin system performance. Figure 3 gives the flowchart ofSO-RA algorithm.

Step-1: Interfering neighbor set discovery and initialReuse-1 allocation

In accordance with the LTE standard [28], reference signalreceived power measurement is performed by UEs for pathloss estimation between UE and BS [26]. Based on thesemeasurements made by FUE, femto BS computes path lossfrom neighboring femto BSs, which are then comparedwith specific threshold value PLth to determine whetherthey may cause interference or not. Thus, each femto BSdetermines a set of neighboring femto BSs that are likely tocause significant interference to its FUEs. The interferingneighbor set of femto BS l is given by,

Il D˚Femto BSj j PLj,l � PLl,l > PLth

�,

j D 1, 2 : : :L(1)

where PLj,l is pathloss between FUE l and femto BS j andPLl,l is pathloss between FUE l and femto BS l. Note thatsame index l is used for femto UE and femto BS becausewe consider only one femto UE per femtocell. Note thatthe underlying assumption in determining threshold valuePLth is that only those femtocells present in the vicinitycause co-layer interference.

After interfering neighbor set discovery, resourceallocation is to be performed. Most of the existing schemesperforms orthogonal allocation initially ([7,14,15]). How-ever, this increases the signaling overheads requiredfor coordination in the initial stage, while in SO-RA,



Figure 3. Self-organized resource allocation algorithm.

employing Reuse-1 eliminates the need of coordinationbetween femto BSs at the initial stage and signaling over-head reduces.

Step-2: Coordinated resource drop

In this step, each femto BS identifies and drops thePRB with minimum SINR, such that neither femtocellthroughput reduces below the threshold nor the overall

system performance deteriorates. We assume that mes-sages required for coordination are exchanged betweenfemto BS and its interfering neighbors via backhaul. TheSINR for FUE l on PRB n is given by,

SINRnl D

Pl,nFtx � PLl,l

Ifemto C Imacro C No(2)



where

Ifemto D

jjIljjX

jD1,j¤l

Pj,nFtx � PLj,l � xj,n (3)

and

Imacro D

7X

kD1

Pk,nMtx � PL0k,l � yk,n. (4)

Pj,nFtx and Pk,n

Mtx denote the transmit power of femto BS j andmacro BS k on PRB n, respectively. PL0k,l is the pathlossbetween FUE l and macro BS k. xj,n is a variable indicatorthat denotes whether PRB n is used by femto BS j or not.Similarly, yk,n indicates whether PRB n is used by macroBS k or not. No is additive white Gaussian noise.

To determine which PRB to drop, femto BS l calcu-lates the SINR experienced on all PRBs that are used byits FUEs. Then, it chooses a candidate PRB n on whichit experiences minimum SINR. It is likely that minimumSINR is due to high amount of interference on that PRB.If such PRB is dropped, it may reduce the interferencecaused to the neighboring femtocells, at the cost of reducedserving femtocell throughput. To ensure that this penaltyis minimal, we select the PRB with minimum SINR, sothat its dropping results in minimum rate loss and eventu-ally co-layer interference reduces. However, to ensure thatthe decision of dropping PRB (taken in coordination withthe localized neighborhood) does not adversely affect theglobal system performance, femto BS l performs two levelsof a-priori checks before actually dropping that PRB.

(1) Level-I a-priori check:Here, we analyze the impact of dropping PRB n onthe performance of serving femtocell l. For femtoBS l, the achievable throughput on PRB n is given by,

Rnl D B � log2

�1C SINRn

l

�(5)

and total throughput of femto BS l is given by,

Rl D

NX

nD1

B � log2�1C SINRn

l

�� xn,l, (6)

where B is the PRB bandwidth.Femto BS l calculates its new throughput value

Rl,new assuming that it has dropped PRB n as,

Rl,new D Rl � Rnl (7)

Further, it compares the new throughput Rl,new withthe specified threshold thptth. If

Rl,new < thptth, (8)

then femto BS defers the decision to drop PRB nand algorithm repeats for the next femtocell. Other-wise, when (8) is not satisfied, it implies that Level-I

a-priori Check results in favor of dropping PRBn. Only then, femto BS l initiates Level-II a-prioriCheck, as discussed next.

(2) Level-II a-priori check:Here, we analyze the impact of dropping PRB non the performance of neighboring femtocells bycoordination. Femto BS l requests all its neighbor-ing femto BSs to report the gain in their individualthroughputs assuming femto BS l 2 Il has droppedPRB n. Each femto BS m 2 Il calculates its newSINR and throughput as follows,

SINRnm,new D

Pm,nFtx � PLl,l

Ifemto � Pl,nFtx � PLm,l C Imacro C No

(9)where

Ifemto D

jjImjjX

jD1,j¤m

Pj,nFtx � PLj,m � xj,n. (10)

Rm,new D

NX

nD1

B � log2�1C SINRn

m,new

�� xn,l (11)

Then, femto BS m computes gain in throughput as,

�Rm DX

m2Il

.Rm,new � Rm/ (12)

On receiving �Rm from all neighboring femto BSs,femto BS l calculates the total throughput gain of itsneighbors as,

�R DX

m2Il

�Rm (13)

To observe the impact of dropping PRB n on theoverall system throughput, we compare,

Rnl < �R (14)

where Rnl represents the loss in throughput due to

dropping PRB n and �R represents the net gain insystem throughput.

If dropping PRB n at femto BS l results in increas-ing system throughput (3.4), femto BS l takes a finaldecision to drop the PRB n. Otherwise, Step-2 isrepeated for the next femto BS.

Level-I check provides minimum throughput guarantee toeach Femto cell, while Level-II check allows the femtocellthroughput to increase further (above the minimum thresh-old) to the extent that the neighboring femtocells are notadversely affected.

Step-1 of the algorithm is triggered periodically after apredetermined number of OFDMA frames. This repetitionperiod can be configured based on the system dynamics.



Step-2 of the algorithm gets implemented iteratively foreach femto BS successively. The algorithm stops when anyfurther dropping of PRB becomes infeasible due to viola-tion of minimum throughput guarantee to the femto BS.

In case any femto BS gets deactivated during Step-2, itwill release its PRBs, and they will become available forreuse at its neighboring femto BSs immediately. However,Step-2 of our algorithm does not consider reallocation ofresource and therefore, until Step-1 is revoked, resourcesmay remain underutilized. However, due to the periodicrepetition of Step-1, underutilization of resources is likelyto happen only for a small time duration.

The distinct features of SO-RA scheme are thefollowing:

� It ensures minimum throughput guarantee to theFUEs by applying Level-I a-priori check.

� In Level-II a-priori Check, each femto BS decides todrop PRB by comparing the impact of dropping thatPRB on self and neighbors. This distributed decisionmaking based on localized interaction between neigh-boring femto BSs makes algorithm self-organized.

� Our algorithm iteratively drops the PRB with mini-mum SINR such that the throughput requirement ofFUEs is ensured and overall gain is not compromised.It also helps in reducing interference to MUEs locatedeither close to femto BS or in the overlapping cov-erage area. Thus, SO-RA scheme reduces cross-layerinterference implicitly.

3.4. Comparative summary

In this section, we highlight the key aspects of both of ourproposed schemes, proposed AFR with power control andSO-RA, in comparison with the state-of-the-art, Reuse-1and AFR scheme. Note that we focus only on the resourceallocation for femtocell users.

In Reuse-1, all resources are available to all femtocells,and there is no restriction on their usage. As mentionedbefore, this improves the resource utilization, while com-promising on the co-layer interference. In AFR scheme,resource allocation begins with assignment of disjointresources to each femtocell and then, reuse of resourcesis permitted with appropriate coordination amongst theneighboring femtocells. The decision for reuse of resourceis based on whether the co-layer interference lies below theacceptable threshold limits or not.

In our proposed modification to the AFR scheme, wego a step further and give consideration to the reuse ofresource at lower power level, if the reuse is not feasible atthe maximum transmit power. We perform a-priori inter-ference measurement to ensure that the co-layer interfer-ence remains below the acceptable threshold limits whenthe resources are reused at lower power level.

In case of our proposed SO-RA scheme, we intend tomaximize the resource utilization, like Reuse-1 schemeand reduce co-layer interference simultaneously. The allo-cation strategy is to begin with reuse-1 amongst the most

interfering set of femtocell neighbors. Then, the next stepis to identify that resource on which femtocell user is expe-riencing the worst SINR, i.e., identify the most interferingPRB. Now, we drop this PRB after ensuring that droppingdoes not impact the femtocell performance and the overallnetwork performance. This is performed iteratively for allfemtocells in the network.

Thus, in case of our proposed modification to AFR,we attempt to gradually maximize the resource reuse andexploit the power dimension to reduce co-layer interfer-ence. In contrast, with SO-RA, we begin with aggressiveReuse-1 and then, selectively drop the most interferingPRBs, in a iterative manner to reduce co-layer interference.The interesting aspect of SO-RA is that with the change inuser statistics like new users joining the network or existingones exit. However, the algorithm of our proposed SO-RAscheme introduces computational complexity. Also, thisalgorithm takes some finite time to converge. The conver-gence time analysis can be investigated as future work.

4. SIMULATION RESULTS

To evaluate the performance of our proposed schemes, wehave performed system level simulations in MATLAB. Thesimulation parameters are given in Table I. We have carriedout performance analysis for the users located in the centralcell. For the FUEs, we consider the interference from allseven macro cells and the neighboring femtocells.

We consider dual stripe model for dense urban femto-cell deployment where each femtocell block has two stripesof apartments [29]. Each stripe has 2�10 apartments eachof size 10m�10m. To ensure sufficient separation betweenfemto BSs from different stripes, there is a 10 m widestreet in between and a 10 m wide space around the stripe(Figure 4). Each femtocell can have random number offloors F within the range 1 to 10. So each femtocell blockhas a 40�F apartments. Each apartment may not have

Table I. Simulation parameters.

Parameter Value

Inter site distance 500 mTotal bandwidth 10 MHz with 50 PRBsPRB bandwidth 180 KHzMacro users per macrocell 10Number of femtocell blocks 3Deployment ratio ˇ 0.2Active ratio � 0.1 to 1Macro BS TX power 46 dBmFemto BS TX power 20 dBmMacro BS antenna gain 14 dBiFemto BS antenna gain 0 dBiUE noise figure 9 dBWall losses Low , Low,1 20 dB, 5 dBPathloss threshold PLth 30 dBFemto BS throughput 7 Mbpsthreshold thptth

Averaging interval 1000 ms



Figure 4. Dual Stripe Model for Femtocell Deployment.

active femto BS in it. To model this scenario, two ratiosare defined: deployment ratio (ˇ) indicates the fraction ofapartments with femto BS, and active ratio (�) gives thefraction of active femto BSs.

We assume that each apartment can have at most oneactive femto BS in it, located at the center of apartment.Both FUEs and MUEs are dropped uniformly in the apart-ment and macrocell, respectively. We assume only onefloor per femtocell block, that is, F D 1. We consider onlyone FUE per femtocell and therefore, use same indices forFUE and femto BS. Also, we assume femto BSs to oper-ate in closed access mode (i.e., only a set of users areallowed association with femto BS). We consider four fem-tocell blocks with deployment ratio ˇ D 0.2 and activeratio � varying from 0.1 to 1. This implies that there are4� .20�0.2/ D 16 femtocells deployed in the network, ofwhich a set of femtocells will be active based on the activeratio. To model the scenario of randomly deployed femto-cells, we have performed the following. For a given activeratio, we have randomly varied the femto block position,femtocell user location in that block and macro UE loca-tions. Then, all simulations are averaged over an interval of1000 ms.

We review the significance of femtocell deployment bycomparing the throughput performance of users with andwithout femtocell deployment in Figure 5. We observe thatabout 80% of the users experience an improvement of morethan 5 Mbps in their throughput and 30% of the usersexperience an improvement of more than 10 Mbps, whenfemtocells are deployed in the network. To investigate theimpact of increased femtocell density (i.e., active ratio) onthe cell throughput performance of macro and femto cells,we consider four femtocell blocks with deployment ratioˇ D 1 and active ratio � varying from 0.1 to 1. We observefrom Figure 6 that with an increase in the femto cell den-sity, the femtocell users experiences increased co-layer

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

User throughput (Mbps)

CD

F

Figure 5. Performance comparison—with and withoutfemtocell.

interference while macrocell users experience increasedcross-layer interference. As a consequence, the averagethroughput in both reduces by about 57%.

Figure 7 compares the performance of SINR per PRBfor both of our proposed schemes, AFR with power con-trol, and SO-RA scheme with that of Reuse-1 and AFRscheme. We observe that the SINR performance of AFRscheme is the best of all. This happens because of the factthat resource allocation in AFR begins with orthogonalallocation, which is followed by co-channel reuse sub-ject to condition that the co-layer interference remains lessthan the acceptable threshold. because of this constrainedresource allocation, majority of the allocated PRBs arelikely to be orthogonal with respect to those PRBs that areallocated to other femtocell users. However, this reduces



0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

10

15

20

25

30

35

Active Ratio (ρ)

Ave

rage

Tho

ughp

ut (

Mbp

s)

Figure 6. Impact of active ratio on throughput.

−30 −20 −10 0 10 20 30 400

0.2

0.4

0.6

0.8

1

SINR per PRB (dB) for Femtocell Users

CD

F

Figure 7. SINR per PRB for Femtocell users.

the resource utilization by about 60% in case of AFRscheme as shown in Table II. We also observe that theperformance of Reuse-1 and proposed AFR with powercontrol is almost similar. This happens because we increasethe reuse of resources in proposed AFR by relaxing the fullpower allocation constraint of the AFR scheme. As a result,we observe 89% resource utilization in proposed AFR withpower control compared to 41% achieved in case of AFRscheme. Note that the cummulative ditribution function(CDF) of SINR per PRB for the proposed SO-RA schemefalls in between Reuse-1 and AFR. Also, the resource uti-lization in proposed SO-RA scheme is 83.7%, which isalmost two times of that obtained using AFR scheme. Thisindicates that our proposed SO-RA scheme achieves trade-off in maximizing user throughput and improving resourceutilization.

We compare the performance of our proposed SO-RAscheme with Reuse-1, AFR, and AFR with power control

Table II. Resource utilization efficiency of different schemes.

Frequency allocation Resource utilization

scheme efficiency

Reuse-1 100%AFR 41.3%Proposed AFR 89%with power control (41.3% resources with full

power and 47.7% resourcesat half power)

Proposed SO-RA scheme 83.7%

0 10 20 30 40 50 60 70 800

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Femtocell UE Throughput (Mbps)

CD

F

Reuse−1Proposed SO−RA SchemeAFRProposed AFR withpower control

Figure 8. CDF of femtocell UE throughput.

Table III. Throughput comparison of different schemes.

FUE throughput (Mbps)

Frequency allocation Avg. 10 percentilescheme throughput throughput

Reuse-1 28.83 13.05AFR 29.57 10.6Proposed AFRwith power control 29.21 13.18Proposed SO-RA scheme 29.92 13.79

scheme [7]. Figure 8 compares the cumulative distribu-tion functions of average throughput of FUEs. For betterunderstanding, Table III gives the average and 10 per-centile throughput comparison of all four schemes. Thereis a slight improvement in the average throughput perfor-mance of SO-RA scheme compared to all other schemes.It is observed that SO-RA achieves 30% improvement inthe 10 percentile throughput performance of FUEs com-pared to AFR scheme. This happens due to the followingreason - in SO-RA scheme, the femto BS experiencingsevere interference does not drop PRBs simply if it dete-riorates its own performance (ensured by Level-I a-priori



0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 12

4

6

8

10

12

14

16

18

Active Ratio (ρ)

10 P

erce

ntile

Thr

ough

put (

Mbp

s)

Figure 9. 10 percentile Throughput comparison of differentschemes.

check). Rather, SO-RA strategy ensures that the loss inthroughput performance of femto BS is lesser than the netgain in throughput performance of neighboring femto BSs.Thus, global system performance improvement in SO-RAscheme is ensured. On the contrary, to increase the sys-tem throughput performance, the severely interfered femtoBS in AFR scheme may allow neighboring femto BSs toreuse PRBs, thereby increasing co-layer interference. InAFR with power control scheme, reuse efficiency improvesat the cost of reduced throughput. However, the averagethroughput performance of AFR with power control isclose to that of AFR without power control, but the 10 per-centile throughput performance shows an improvement ofabout 24% relatively.

Figure 9 shows the impact of increased femtocell den-sity on the 10 percentile throughput of FUEs for differentschemes. SO-RA scheme offers the best 10 percentilethroughput, even with increased femtocell density. The per-formance of AFR with power control is close to Reuse-1with marginal improvement when the femtocell densityincreases.

Further, we investigate the fairness performance(Figure 10) by using Gini fairness index. It is given by,

I D1

2L2R

LX

lD1

LX

mD1

jRl � Rmj (15)

where R D

LPlD1

Rl

L . Gini fairness index lies between 0and 1. A scheme is perfectly fair if Gini index is 0 andunfair if 1. We observe that SO-RA scheme outperforms interms of fairness to FUEs. AFR with power control exhibitssimilar performance behavior as for the 10 percentilethroughput (Figure 9).

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.25

0.3

0.35

0.4

0.45

0.5

Active Ratio (ρ)

Gin

i Ind

ex

Reuse−1AFRProposed AFR withpower controlProposed SO−RA Scheme

Figure 10. Comparison of Gini fairness index.

Finally, we compare the schemes based on the resourceutilization efficiency metric (Table II). We define resourceutilization efficiency (�ru) as the ratio of used PRBs to theavailable PRBs and is given by,

�ru D

PFnumjD1 PRBj

PRBtot � Fnum(16)

where Fnum gives the count of active femtocells in the net-work. PRBj and PRBtot denotes the number of used PRBsin jth femtocell and total number of PRBs available in thesystem, respectively.

The resource utilization efficiency is the lowest for AFR.�ru in AFR with power control scheme is close to Reuse-1. However, an important observation is that only 41.3%of resources are used with full transmit power and remain-ing 47.7% resources are used at half of the transmit power.In SO-RA scheme, �ru gets almost doubled compared toAFR, with all resources being used at full transmit power.

In a nutshell, our results indicate improved 10 percentilethroughput and fairness performance in SO-RA schemecompared to AFR with power control, AFR, and Reuse-1 schemes. Thus, SO-RA scheme offers a reasonabletrade-off in achieving improved throughput performance ofseverely affected FUEs, fairness to all FUEs, and improvedresource utilization efficiency.

5. CONCLUSIONS

Femtocell deployment provides capacity and coverageimprovement to indoor users. However, intelligent andself-organized resource allocation is required to ensureimproved performance with minimal interference and QoSguarantees. In this paper, we have proposed two resourceallocation algorithms. First, proposed AFR with powercontrol scheme performs at par with that of Reuse-1in terms of average throughput and resource utilization



efficiency. Second, proposed self-organized resource allo-cation algorithm reduces co-layer interference in the down-link scenario, while ensuring throughput performanceimprovement to the severely affected FUEs and improve-ment in resource utilization and fairness to all FUEs simul-taneously. Thus, SO-RA scheme achieves feasible trade-offbetween 10 percentile throughput, fairness and resourceutilization efficiency, compared to other schemes availablein the literature. The two levels of a-priori check in SO-RA operate in a self-organized manner to ensure that theemphasis is not on the localized performance improve-ment of an individual femtocell, but on the global systemperformance improvement.

ACKNOWLEDGEMENT

This project is being carried out under the India-UKAdvanced Technology of Centre (IU-ATC) of Excel-lence in Next Generation Networks project and fundedby the Department of Science and Technology (DST),Government of India, and UK EPSRC Digital EconomyProgramme.

REFERENCES

1. Chandrasekhar V, Andrews JG, Gatherer A. FemtocellNetworks: A Survey. IEEE Communications Magazine2008; 46(9): 59–67.

2. Zhang J, de la Roche G. Femtocells: Technologies andDeployment. John Wiley & Sons: United Kingdom,2010.

3. Perez L, Valcarce D, Roche AD, Zhang GJ. OFDMAFemtocells: A Roadmap on Interference Avoidance.IEEE Communications Magazine 2009; 47(9): 41–48.

4. Kim MS, Je HW, Tobagi F. Cross-tier Interfer-ence Mitigation for Two-tier OFDMA Femtocell Net-works with Limited Macrocell Information. In IEEEGlobal Telecommunications Conference (GLOBE-COM), Miami, FL, 2010; 1–5.

5. Morita M, Matsunaga Y, Hamabe K. Adaptive PowerLevel Setting of Femtocell Base Stations for Mitigat-ing Interference with Macrocells. In IEEE VehicularTechnology Conference, Ottawa, ON, 2010; 1–5.

6. Bharucha Z, Saul A, Auer G, Haas H. Dynamicresource partitioning for downlink femto-to-macrocellinterference avoidance. In EURASIP Journal on Wire-less Communications and Networking, 2010; 77–88.

7. Ruihong A, Xin Z, Gen C, Ruiming Z, Lin S. Interfer-ence avoidance and adaptive fraction frequency reusein a hierarchical cell structure. In IEEE Wireless Com-munications and Networking Conference (WCNC),Sydney, Australia, 2010; 1–5.

8. Hu F, Zheng K, Lei L, Wang W. A distributed inter-cellinterference coordination scheme between femtocells

in LTE-advanced networks. In IEEE Vehicular Tech-nology Conference, Yokohama, 2011; 1–5.

9. Bai Y, Zhou J, Liu L, Chen L, Otsuka H. ResourceCoordination and interference mitigation betweenmacrocell and femtocell. In IEEE Personal Indoorand Mobile Radio Communications Conference, TokyoJapan, 2009; 1401–1405.

10. Guvenc I, Jeong MR, Watanabe F, Inamura H. Ahybrid frequency assignment for femtocells and cov-erage area analysis for co-channel operation. In IEEECommunication Letters, vol. 12, 2008; 880–882.

11. Juang RT, Ting P, Lin HP, Lin DB. Interference man-agement of femtocell in macro-cellular networks. InWireless Telecommunications Symposium, Tampa, US,2010; 1–4.

12. Chandrasekhar V, Andrews J. Spectrum allocation intiered cellular networks. IEEE Transactions on Com-munications 2009; 57(10): 3059–3068.

13. Sundaresan K, Rangarajan S. Efficient resource man-agement in OFDMA femtocells. In Mobile Ad hocNetworking and Computing (MobiHoc), New Orleans,LA, USA, 2009; 33–42.

14. Cao G, Yang D, Ruihong A, Xuan Y, Zheng R, ZhangX. An adaptive sub-band allocation scheme for densefemtocell environment. In IEEE Wireless Communica-tions and Networking Conference, Cancun, QuintanaRoo, 2011; 102–107.

15. Zheng K, Hu F, Lei Lei, Wang W. Interference coordi-nation between femtocells in LTE-advanced networkswith carrier aggregation. In International ICST Con-ference on Communications and Networking in China,Beijing, China, 2010; 1–5.

16. Jo HS, Mun C, Moon J, Yook JG. Self-optimizedcoverage coordination in femtocell networks. IEEETransactions on Wireless Communications 2010;9(10): 2977–2982.

17. Claussen H. Performance of macro- and co-channelfemtocells in a hierarchical cell structure. In Per-sonal, Indoor, Mobile Radio Communications, Athens,Greece, 2007; 1–5.

18. Chandrasekhar V, Andrews J, Muharemovic T, ShenZ, Gatherer A. Power control in two-tier femtocellnetworks. IEEE Transactions on Wireless Communica-tions 2009; 8(8): 4316–4328.

19. Cao G, Yang D, Ye X, Zhang X. A downlink jointpower control and resource allocation scheme for co-channel macrocell-femtocell networks. In IEEE Wire-less Communications and Networking Conference,Cancum, Quintana Roo, 2011; 281–286.

20. Cheng SM, Ao WC, Tseng FM, Chen KC. Design andanalysis of downlink spectrum sharing in two-tier cog-nitive femto networks. IEEE Transactions on VehicularTechnology 2012; 61(5): 2194–2207.



21. Mehta M, Aliu OG, Karandikar A, Imran MA. A self-organized resource allocation using inter-cell inter-ference coordination (icic) in relay-assisted cellularnetworks. ICTACT Journal on Communication Tech-nology 2011; 2(2): 300–313.

22. Liang YS, Chung WH, Ni GK, Chen IY, Zhang H, KuoSY. Resource allocation with interference avoidance inOFDMA femtocell Networks. IEEE Transactions onVehicular Technology 2012; 61(5): 2243–2255.

23. Lopez-Perez D, Ladanyi A, Juttner A, Zhang J.OFDMA femtocells: A self-organizing approach forfrequency assignment. In IEEE Personal, Indoor andMobile Radio Communications, Tokyo, Japan, 2009;2202–2207.

24. Bernardo F, Agusti R, Cordero J, Crespo C. Self-optimization of spectrum assignment and transmissionpower in OFDMA femtocells. In Advanced Interna-tional Conference on Telecommunications, Barcelona,2010; 404–409.

25. Galindo-Serrano A, Giupponi L. Distributed Q-learning for interference control in OFDMA-basedfemtocell networks. In IEEE Vehicular TechnologyConference, Taipei, 2010; 1–5.

26. Agilent. 3GPP long term evolution: system overview,product development, and test challenges, June 2009.

27. 3GPP TR 36814. Further advancements for E-UTRAphysical layer aspects V900, March 2010.

28. 3GPP TR 36921. FDD home eNodeB (HeNB) radiofrequency (RF) requirements analysis V1000, April2011.

29. Femtoforum. Interference management in OFDMAfemtocells. http://smallcellforum.org [accessed onMarch 2010].

AUTHORS’ BIOGRAPHIES

Mahima Mehta received the B.E.degree in Electronics and Communi-cation Engineering from GovernmentEngineering College, Bhopal, India,in 2001; the M.E. degree in Elec-tronics and Communication Engineer-ing from the Shri Govindram SeksariaInstitute of Technology and Science,

Indore, India, in 2007; and the Ph.D. degree in ElectricalEngineering from the Indian Institute of Technology (IIT)Bombay, Mumbai, India, in 2014. She has worked as aLecturer and then as an Assistant Professor for about eightyears and has research experience of about five years. Sheis currently a project Research Scientist with IIT Bombay.She has been actively involved in a DST-funded project:India-UK Advanced Technology Centre (IU-ATC) and col-laborates with researchers in UK and India. She is also a3GPP RAN2 delegate. Her research interests include radio

resource and mobility management in LTE heterogeneouscellular networks and LTE-WLAN interworking.

Nirbhay Rane received his B.E. Deg-ree in Electronics and Telecommu-nication from Mumbai University in2006. He completed his M.Tech. inCommunication Engineering from theIndian Institute of Technology (IIT)Bombay in 2012. He worked asResearch Associate student in ’Tata

Indicom Center of excellence in Telecommunication Lab’,IIT Bombay’ from 2009-2012. His research interest liesmainly in Wireless networking, Mobile communicationsystems. He is currently working as Network ConsultingEngineer in market leading networking company.

Abhay Karandikar earned his MTechand PhD degrees in Electrical Engi-neering from the Indian Institute ofTechnology, Kanpur in 1988 and1994, respectively. He joined theDepartment of Electrical Engineering,IIT Bombay in April 1997, wherehe is currently Professor and Head

of the Department. He is also the coordinator of TataTeleservices IIT Bombay Center for Excellence in Tele-com where he has initiated many projects in the area ofbroadband wireless networks. Dr Karandikar has severalpatents issued/pending, contributions to IEEE standards,contributed chapters in books and large number of papersin international journals and conferences to his credit. Hehas also served as technology advisor and consultant tomany companies. He has served on the working groupof telecom for 12th plan of Government of India. DrKarandikar is a two time recipient of award for excellencein teaching at IIT Bombay- in 2006 and 2011.

Muhammad Ali Imran received theM.Sc. (with Distinction) and Ph.D.degrees from Imperial College Lon-don, London, U.K., in 2002 and 2007,respectively. He is currently a Readerwith the Centre for CommunicationSystems Research (CCSR), Universityof Surrey, Guildford, U.K., where he

is leading a number of multimillion international researchprojects encompassing the areas of energy efficiency, fun-damental performance limits, sensor networks, and self-organizing cellular networks. He is also leading the newphysical-layer work area for 5G innovation center at Surrey(the 5G innovation center and an outdoor cellular test bed isbeing developed at Surrey University with a research grantof above 35 million). He has a global collaborative researchnetwork spanning both academia and key industrial playersin the field of wireless communications. He has supervised17 successful Ph.D. graduates and published over 150 peer-reviewed research papers, including more than 25 IEEEJournals. His research interests include the derivation of


http://smallcellforum.org


information-theoretic performance limits, energy-efficientdesign of cellular systems, and learning/self-organizingtechniques for the optimization of cellular system oper-ation. Dr. Imran is a Fellow of the Higher EducationAcademy, U.K. He received the 2014 IEEE Communica-tions Society Fred W. Ellersick Prize.

Barry G. Evans received the B.Sc.and Ph.D. degrees. Up until 2009,he was the Pro Vice Chancellor forResearch and Enterprise with the Uni-versity of Surrey, Guildford, U.K., andthe Director of the Centre for Com-munication Systems Research, whichis a 150-strong postgraduate center.

He now runs a number of research projects in satellite

communications with the European Space Agency (ESA)and industry and leads the Surrey part of the UK-India net-work of excellence in next-generation networks. He is alsothe Director of a Surrey spin-off company MulSys Ltd. Hehas personal research background in satellite and mobilecommunications, working on speech coding, radio propa-gation, advanced physical layers, and cognitive radio withover 700 research publications and three textbooks to hisname. Dr. Evans is the Editor of the International Journalof Satellite Communications and the Chair of the Steer-ing Committee of SatNEx, which is an EU/ESA networkof excellence, and the R&D strategy group of the NET-WORLD2020 satellite working group. He is a Fellow ofthe Royal Academy of Engineering, the Institution of Engi-neering and Technology, and the Royal Society of Arts andis a Chartered Engineer.


Documents

A self-organized resource allocation scheme for ...epubs.surrey.ac.uk/807410/1/A self-organized resource allocation sc… · Femtocell overlaid on existing macro-cell network. interference