uplink scheduling in lte systems using distributed base stations

EUROPEAN TRANSACTIONS ON TELECOMMUNICATIONSEur. Trans. Telecomms. 2010; 21:532–543Published online 1 February 2010 in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/ett.1408

Mobile Networks

Uplink scheduling in LTE systems using distributed base stations

Elias Yaacoub* and Zaher Dawy

Department of Electrical and Computer Engineering, American University of Beirut, Beirut, Lebanon

SUMMARY

Uplink scheduling in Long-Term Evolution (LTE) systems with distributed base stations (DBSs) isconsidered. In the proposed model, users are connected to remote radio heads (RRHs) deployed throughoutthe cell and connected to a central base station. Centralised and distributed scheduling with a DBS arecompared to centralised scheduling with a central base station. Monte Carlo simulation results show thatLTE systems with DBSs achieve better throughput than systems with conventional base stations whileallowing edge users better chances of accessing the resources. Copyright © 2010 John Wiley & Sons, Ltd.

1. INTRODUCTION

The UMTS Long-Term Evolution (LTE) standard willstretch the performance of 3G technology, in order tomeet user expectations in a 10-year perspective andbeyond [1]. Goals for the evolved system include supportfor improved system capacity and coverage, high peakdata rates (100 Mbps downlink and 50 Mbps uplink),low latency (10 ms round-trip delay), reduced operatingcosts, multi-antenna support, flexible bandwidth operations,and seamless integration with existing systems. To meetthese requirements, LTE is based on orthogonal frequencydivision multiple access (OFDMA) due to its immunity tointer-symbol interference and frequency selective fading[1]. However, for the LTE uplink, single carrier frequencydivision multiple access (SCFDMA), a modified formof OFDMA, is used. Although it has similar throughputperformance and essentially the same overall complexityas OFDMA, its principal advantage is its lower peak-to-average power ratio (PAPR) [2]. In this paper, we investigatethe role of distributed base stations (DBSs) in the frameworkof LTE uplink scheduling.

* Correspondence to: Elias Yaacoub, Department of Electrical and Computer Engineering, American University of Beirut, P.O.Box 11-0236/ECEDepartment, Riad El-Solh/Beirut 1107 2020, Lebanon. E-mail: [email protected].

The concept of DBSs and remote radio heads (RRHs)emerged to increase the coverage and capacity of wirelessnetworks in a cost effective way. It consists of a centrallylocated BS enclosure connected to RRHs via fiber opticcables [3]. In the existing literature, the terms DBS anddistributed antenna system (DAS) are used interchangeably.DBSs were initially proposed to enhance indoor coverage ofcellular systems where a building is treated as a single cellwith several distributed antennas rather than either multiplepico cells each with a dedicated antenna or as a single cellwith one central antenna [4]. The DBS approach allowsavoiding excessive handovers in the first case and significantfading in the latter. The coverage and capacity of DBSs in anindoor WCDMA system were investigated in Reference [5]for several types of antennas. In a multi-cell CDMA systemwith DBSs, it was shown that maximum ratio combining(MRC) in the uplink achieves a considerable capacityand coverage enhancement, but simultaneous transmissionin the downlink reduces performance since it increasesthe intercell interference [6]. A solution for this problemwas proposed in Reference [7], where it was found thatselecting only the RRH with best channel to the user

Received 18 February 2009Revised 24 September 2009

Copyright © 2010 John Wiley & Sons, Ltd. Accepted 24 November 2009

UL SCHEDULING IN LTE SYSTEMS USING DBS 533

ensures the best downlink performance with DBSs. Asimilar conclusion was reached in Reference [8] wheretransmitting from the RRH with the best channel was shownto outperform the case of using the RRH as a relay whiletransmitting the signal directly from the BS. These resultswere validated from an information theoretic standpoint inReferences [9] and [10], where selective transmission (fromonly the RRH with best channel to user) was comparedto maximum ratio transmission (using all the RRHs). InReference [11], it was shown that selection combining(SC) in the uplink provides considerable enhancement overcentralised BSs and constitutes a good tradeoff betweenperformance and complexity when compared to MRC.A generalisation of the concept of DBSs was presentedin Reference [12], where each RRH consists of severalantennas ensuring microdiversity, and the set of RRHscontributes to macrodiversity.

It should be noted that, in a practical scenario, installingRRHs at desired locations (e.g. equidistant along thecell boundary) might not be possible. Therefore, theperformance of random placement of RRHs throughoutthe cell was investigated in References [13] and [14], interms of outage probability, as a lower bound on the actualperformance. Interestingly, it was found that as the numberof RRHs increases, the performance converges to that ofregularly deployed RRHs. In fact, in the case of bothfixed and random RRH locations, the gains achieved bya DBS system were shown to increase with the numberof RRHs up to a certain limit where the gain obtainedafter using an additional RRH is negligible. This limit wasconsidered to be four and seven RRHs in References [7]and [9], respectively, for the regular RRH positions, andseven in Reference [14] for the random RRH positions.Hence, we will select our investigated scenarios withinthese limits. In Reference [15], a solution is presentedfor DBS cooperation via block-diagonalisation and dual-decomposition to maximise the weighted sum networkcapacity under per-antenna power constraint. The solutionis a trade-off between intercell interference mitigation,spatial multiplexing and macro diversity. In Reference [16],a protocol to manage multiuser interference in a DBSsystem is proposed. The scheme is not specific to a particularmultiple access scheme. It is assumed in Reference [16]that users may connect to more than one RRH using thesame resources, which leads to an increase in interference.The protocol of Reference [16] deals with the problem ofassigning channels to mobiles having dynamically changingsets of RRHs linking each of them to the network with amacro-diversity gain. Conversely to References [15] and[16] where RRHs are considered to use the same resources

without centralised control, the RRHs in this paper areconnected to a single BS, and the resources allocated todifferent RRHs are orthogonal subsets of the resourcesavailable at the central BS. Hence, contrarily to References[15] and [16], interference is not an issue within a singlecell.

Distributed BSs were implemented in a variety ofwireless systems. In Reference [17], they were proposedfor LMDS. In Reference [18], the commercial deploymentof distributed BSs for WCDMA/HSPA was announced, andin Reference [19] it was announced for cdma2000/EvDO.Although many of the DBS results in the existing literatureapply to OFDMA, there is no investigation on the impactof DBSs on user scheduling and resource allocationin OFDMA-based systems (e.g. the UMTS long-termevolution (LTE)).

Hence, it would be interesting to apply the concepts ofdistributed BSs to resource allocation in OFDMA systemsin order to investigate the achievable performance gains.The novelty in this work consists in applying the conceptsof DBSs in the context of LTE scheduling. We showthat considerable throughput and fairness enhancementscan be achieved without any advanced techniques such asdiversity with MRC through the antennas of the differentRRHs, or selecting the optimal multiple input multipleoutput (MIMO) weights. We investigate a centralisedscheduling algorithm initially proposed for centralised BSsin Reference [20] and apply it in the case of DBSs. Thisalgorithm was shown to outperform other algorithms inthe literature in Reference [21]. We also propose a noveldistributed algorithm for scheduling in the case of DBSs,and show that it has considerably less complexity than thecentralised algorithm.

The paper is organised as follows. A review of uplinkscheduling in OFDMA and SCFDMA is presented inSection 2. The system model is presented in Section 3.The scheduling algorithms are described in Section 4. Thesimulation results are presented and discussed in Section 5.Finally, conclusions are drawn in Section 6.

2. REVIEW OF SCHEDULING INOFDMA-BASED SYSTEMS

Resource allocation in OFDMA-based systems has beenwidely investigated in the downlink (e.g. [22–26]). Thesolution is generally divided into two parts: subcarrierallocation and power allocation. The solution described inReference [22] consists of allocating each subcarrier to the

Copyright © 2010 John Wiley & Sons, Ltd. Eur. Trans. Telecomms. 2010; 21:532–543DOI: 10.1002/ett

534 E. YAACOUB AND Z. DAWY

user with the best channel condition on that subcarrier andallocating power by water-filling over the subcarriers.

The scheduling problem in the uplink is more challengingthan that in the downlink due to the distributed powerconstraint: in the downlink, the power has a centralisednature because power allocation is done at a centralentity, the BS, whereas in the uplink, the power has adistributive nature and should be considered on a peruser basis. Scheduling in the uplink was investigated inReferences [27–30]. The allocation problem is divided intotwo subproblems in Reference [27]. A greedy algorithmis proposed with water-filling used to allocate power foreach user on its allocated subcarriers. Then using themarginal functions, an optimal (user, subcarrier) pair isfound. Steps are repeated until all subcarriers are allocated.In Reference [28], fairness is added to the approachof Reference [27] by allocating subcarriers to a givenuser until its required rate is reached then the user isexcluded from the allocation of the remaining subcarriers.The algorithm proposed in Reference [29] has similarsteps to that of Reference [27], but differs in that itperforms water-filling for each user on all unallocatedsubcarriers in addition to the subcarriers allocated to thatuser before searching for the optimal (user, subcarrier)pair. The algorithms in References [27–29] are suboptimal.In Reference [30], instantaneous sum-rate maximisationis formulated into a convex optimisation problem andsolved using a dual decomposition approach, then a setof suboptimal algorithms are presented and compared, dueto the prohibitive complexity of implementing the optimalsolution. It was found in Reference [27] that equal powerallocation over the assigned subcarriers leads to almostthe same results as water-filling. Similar uplink schedulingalgorithms were developed in Reference [31], where theutility functions to maximise are the sum of user throughputor the sum of the logarithm of user throughput in order toprovide better fairness.

LTE uplink transmission is based on SCFDMA whichis a modified form of OFDMA. As in OFDMA,the transmitters in SCFDMA use different orthogonalfrequencies (subcarriers) to transmit information symbols.However, they transmit the subcarriers sequentially, ratherthan in parallel in order to reduce the PAPR problem [2].High PAPR is problematic for uplink transmission wherethe mobile transmission power is usually limited. Due tothe high PAPR value, the mean transmit power will belimited by the linear range of power amplifiers. The requiredbackoff reduces the average transmitted power significantlyand hence reduces the link budget in the uplink. This isvery crucial for cell edge users due to their large path

loss. For PAPR reduction, 3GPP-LTE agreed on usingSCFDMA transmission with cyclic prefix in the uplinkwhere frequency domain generation of the signal by aDFT precoding followed by an IFFT structure was assumed[32]. Relative to OFDMA, SCFDMA reduces considerablythe envelope fluctuations in the transmitted waveform.However, in cellular systems with severe multipathpropagation, the SCFDMA signals arrive at a base stationwith substantial intersymbol interference. The base stationemploys adaptive frequency domain equalisation to cancelthis interference [2]. This arrangement reduces the burdenof linear amplification in portable terminals at the cost ofcomplex signal processing (frequency domain equalisation)at the base station. SCFDMA has two types of sub-carrier mapping [31]: Localised FDMA (LFDMA) andInterleaved FDMA (IFDMA). In LFDMA, the schedulerassigns consecutive subcarriers to convey informationfrom a particular user. In IFDMA, users are assignedsubcarriers that are distributed over the entire frequencyband in order to avoid allocating adjacent subcarriersthat are simultaneously in a deep fade. IFDMA was notincluded into the LTE standard due to slight performancedisadvantages caused by the requirements of channelestimation accuracy [33]. Hence, in the remainder of thispaper, when referring to SCFDMA, we will only considerLFDMA. In LTE, the available spectrum is divided intoresource blocks (RBs) consisting of 12 adjacent subcarriers.The duration of a single RB is 1 ms, agreed to be the durationof one transmission time interval (TTI) [33, 34].

SCFDMA is combined with frequency dependentscheduling in Reference [35] and high spectral efficiency isachieved in moderate and high signal to noise ratio (SNR)conditions. It was also shown in Reference [35] that for celledge users with relatively low SNR, SCFDMA increasesthe cell edge throughput. A search-tree based channel awarepacket scheduling algorithm is proposed in Reference [36]and its performance is evaluated in terms of throughput andnoise rise distributions. In Reference [37], an SCFDMAresource allocation algorithm is derived based on a purebinary-integer program called the set partitioning problem.In addition, a suboptimal greedy heuristic that performsclose to the optimal algorithm with lower complexity isproposed.

In this paper, LTE uplink scheduling in the presenceof DBSs is investigated. A novel distributed schedulingalgorithm for distributed BS scenarios is proposed inSection 4. The algorithm is inspired from a centralisedalgorithm developed in Reference [20] and shown inReference [21] to outperform several other algorithms inthe literature.



Figure 1. Examples of possible deployment scenarios.

3. SYSTEM MODEL

The model is composed of a single central BS connectedto several RRHs distributed throughout the cell area. TheBS controlling the RRHs could be co-located with any ofthe RRHs or in a separate location. Although other typesof media are possible, it is mainly connected to RRHs viafiber optic cable. Connection topologies include star, chain,tree, and ring topologies [3]. Each RRH consists mainly ofa remote antenna connected to the central BS. This allowscentralised control to be performed by the BS as in theconventional case while the RRHs allow extended coverageand/or more user capacity. In addition, for fixed coverageand user capacity, the RRHs provide the users with betterquality of service (QoS) since the distance from a user tothe nearest RRH will be smaller than the distance to thecentral BS antenna in the conventional case, which leads toa higher SNR. Figure 1 shows examples of distributed BSdeployment scenarios. In Figure 2, an example of six usersconnected to the five RRHs of deployment scenario (c) isshown.

In this work, distributed BSs will not be investigatedin terms of increasing the cell coverage and/or capacity.We consider a single cell scenario, and we compare theperformance of the LTE scheduling algorithms discussedin Section 4 in terms of throughput and fairness. In thecomparisons, we consider the same coverage area and thesame number of users in the cell in the case of a centralisedBS and distributed BS. We also consider the same numberof subcarriers in both cases. Hence, the RRHs are not usedto ensure more frequency reuse, but rather to make a moreefficient and fair use of the available subcarriers.

6

1

2

3

4

5

Figure 2. Example of user association to RRHs in the case ofdeployment scenario (c).

The investigated and compared scenarios consist of thefollowing:

� Centralised BS and centralised scheduling (CBS): asingle central BS with omnidirectional antenna coveringthe whole cell. A centralised scheduling algorithm(Algorithm 1 proposed in subsection 4.1) is run atthe central BS location. This scenario is illustrated inFigure 3.

� Distributed BS and centralised scheduling (DBS-C): acentral BS with several RRHs throughout the cell. A

Figure 3. CBS scheduling scenario.



Figure 4. DBS-C scheduling scenario.

centralised scheduling algorithm (Algorithm 1 proposedin subsection 4.1) is run at the central BS locationto allocate resources to the users. The presence ofthe RRHs contributes in enhancing the channel statesof the different users by providing each user with anantenna that is closer to it than the central BS antenna.Hence, the scheduling operation is similar to the previousscenario but with better channel state information (CSI)available at the BS for the different users. This scenariois illustrated in Figure 4.

� Distributed BS and distributed scheduling (DBS-D): acentral BS with several RRHs throughout the cell. Adistributed scheduling algorithm (Algorithm 2 proposedin subsection 4.2) is run at the locations of the RRHsto allocate resources to the users. The presence ofthe RRHs contributes in enhancing the channel statesof the different users as in the previous case (DBS-C). Furthermore, the complexity of implementing thescheduling algorithm at the various locations is less thanimplementing it at the central location for the whole cell.This scenario is illustrated in Figure 5. It should be notedthat in the (DBS-D) case, the scheduler for each RRHdoes not have to be co-located with the correspondingRRH. A single scheduler in the central BS location couldbe used to perform scheduling operations separately foreach RRH and communicate the scheduling informationvia the medium connecting the BS to the RRHs (e.g.dedicated fiber optic cables).

Figure 5. DBS-D scheduling scenario.

4. SCHEDULING ALGORITHMS

In this section, we present a novel distributed schedulingalgorithm for distributed BS scenarios. The algorithmis inspired from a centralised algorithm developed inReference [20] and shown in Reference [21] to outperformother algorithms in the literature. Therefore, we will presentthe centralised algorithm briefly first, then we will proposethe novel distributed algorithm applicable to the DBSscenario. Finally, we will show that the distributed algorithmhas considerably less complexity. The performance of thetwo algorithms is compared in Section 5.

We will denote by IRB,k the set of RBs allocated touser k, Isub,k the set of subcarriers allocated to user k,N the number of RBs, Nsub the number of subcarriers,Pk the instantaneous transmission power of user k, Pk,maxits maximum transmission power, and Rk its achievablethroughput. U(Rk|IRB,k) is the utility of user k as a functionof the throughput Rk given the allocation IRB,k. Theutility function depends on the throughput, and could varydepending on the different services and QoS requirements.Letting the utility equal to the throughput, Algorithm 1leads to a maximisation of the sum-throughput of thecell. However, in this case, users close to the BS will beallocated most of the resources, whereas edge users willgenerally suffer from starvation. To solve this problem,utility functions providing proportional fairness (PF) aredesired. In References [22] and [23], it was proven thatthe logarithmic utility function is associated with the



proportional fairness for the utility-based optimisation.Hence, letting U = ln(R) provides proportional fairness,where ln represents the natural logarithm.

4.1. Algorithm 1

The centralised algorithm proposed in Reference [20]consists of allocating RB n to user k in a way to maximisethe following difference:

�n,k = U(Rk|IRB,k ∪ {n}) − U(Rk|IRB,k) (1)

where the marginal utility, �n,k, represents the gain in theutility function U when RB n is allocated to user k, comparedto the utility of user k before the allocation of n. Algorithm 1is described as follows:

� Consider the set of available RBs Iavail RB ⊆{1, 2, . . . , N}. At the start of the algorithm, Iavail RB ={1, 2, . . . , N}.

� Step 1: Find an RB-user pair which has the highestmarginal utility defined in Equation (1) among allavailable RBs and users. For each available user k andRB n, find the pair:

[n∗, k∗] = arg maxn∈Iavail RB,k

�n,k (2)

� Step 2: Allocate RB n∗ to user k∗: IRB,k∗ = IRB,k∗ ∪{n∗}

� Step 3: Delete the RB from the set of available RBs:

Iavail RB = Iavail RB − {n∗} (3)

� Repeat Steps 1, 2, and 3, until all RBs are allocated.

4.2. Algorithm 2

Algorithm 2 consists of two allocation phases. In thefirst phase, RBs are allocated to RRHs, with the amountof RBs allocated to each RRH being proportional tothe number of users associated with this RRH (Thisassociation will be discussed in subsection 5.5). In thesecond phase, Algorithm 1 is applied for each RRHseparately. Algorithm 2 is described as follows:

� After the users are connected to the network, each subsetof users is associated to a certain RRH, with user kassociated to RRH a if the association would lead tothe best SNR for user k. We will assume that the users

have low mobility such that they remain associated tothe same RRH throughout the transmission time.

� Step 1: Allocate Na = �NKa/K� RBs to each RRH awith Ka users connected to it.

� Step 2: Due to the floor �� operation, some RBs willremain unallocated. Allocate sequentially one RB perRRH until all RBs are allocated.

� Step 3: Apply Algorithm 1 sequentially to each RRH;i.e. for RRH a, apply Algorithm 1 with K = Ka andN = Na.

4.3. Complexity analysis

It was shown in Reference [20] that Algorithm 1 hasapproximately O(N2K) complexity. As for Algorithm 2,the first phase has a complexity of the order of thenumber of RRHs. Since in most cases (and particularly theones investigated here), the number of RRHs is negligiblecompared to N2K, it is the complexity of the second phasethat will dominate. With uniform distribution of the usersinside the cell, and assuming the RRHs are distributedsuch that they cover equal areas in the cell, it is logicalto assume that, as the number of users increases, we willapproximately have Na = N/A and Ka = K/A, with Abeing the number of RRHs. Hence, the complexity of thesecond phase of Algorithm 2 is given by

O(

A∑a=1

N2aKa

)≈ O

(A∑

a=1

N2KA3

)

= O(

N2KA2

) (4)

Consequently, the complexity of the distributed algorithmdecreases with the square of the number of RRHs. Thisresult is interesting, since it shows that the complexity canbe reduced by around an order of magnitude with only threeRRHs.

4.4. Throughput calculations

For the throughput calculations, we consider the followingexpression:

R(Pk, Isub,k) =Nsub∑i=1

αi,k

B

Nsub· log2(1 + βγi,k) (5)

where B is the total bandwidth, αi,k = 1 if subcarrier i ispart of an RB allocated to user k (i.e. i ∈ Isub,k), and β iscalled the SNR gap. It indicates the difference between the



SNR needed to achieve a certain data transmission rate for apractical M-QAM system and the theoretical limit (Shannoncapacity). It is given by [38]:

β = −1.5

ln(5Pb)(6)

where Pb denotes the bit error rate (BER). Each user isassumed to transmit at the maximum power (Pk = Pk,max),and the power is assumed to be subdivided equally amongall the subcarriers allocated to that user. Hence, the SNRover a single subcarrier, γi,k, is given by:

γi,k =Pk

|Isub,k |Hk,i

σ2i

(7)

where Hk,i is the channel gain over subcarrier i allocated touser k, and σ2

i is the noise power.Subdividing the power equally over the subcarriers is

justified in Reference [39] by the fact that the achievedgains are negligible compared to the increase in complexitywhen optimal power allocation is performed. In addition,it was shown in Reference [27], via simulations, thatoptimal power allocation using water-filling and equalpower allocation over subcarriers lead to approximately thesame results. In fact, in the uplink scenario, the maximumtransmission power of mobile users is limited, contrarilyto the downlink case where the BS has considerably morepower and where the variation in user distances from the BSallows the latter to achieve gains by optimising the powerallocation.

5. RESULTS AND DISCUSSION

This section presents the simulation results obtained byapplying the algorithms presented in Section 4. The resultsinclude plots of the cell throughput, in addition to adiscussion of fairness.

5.1. Simulation model

The simulation model consists of a single cell with a BSequipped with an omnidirectional antenna, or consistingof several RRHs, each consisting of an omnidirectionalantenna. The investigated RRH deployment models areshown in Figure 1. Deployment scenario (a) consists of theconventional centralised single BS. Deployment scenario(b) consists of four RRHs: one located at the cell centreand three located at a distance of 2Rc/3, with Rc being the

cell radius considered to be 1 km. The angular separationbetween these three RRHs is 120◦. Deployment scenario(c) consists of five RRHs: one located at the cell centreand four located at a distance of Rc/2, with 90◦ angularseparation between them. Finally, deployment scenario (d)consists of seven RRHs: one located at the cell centre and sixlocated at a distance of 2Rc/3, with 60◦ angular separationbetween them. In each deployment scenario, we investigatethe performance of the three scheduling cases: CBS, DBS-C, and DBS-D.

The throughput is averaged over 200 TTIs, with theduration of a TTI being 1 msec. Then the simulation isrepeated over 50 iterations. The total bandwidth consideredis B = 5 MHz, subdivided into 25 RBs of 12 subcarrierseach [33]. We consider a target BER of 10−6. The maximummobile transmit power is considered to be 125 mW [23]. Allmobiles are assumed to transmit at the maximum power,and the power is subdivided equally among all subcarriersallocated to the mobile. The channel gain over subcarrier icorresponding to user k is given by:

Hk,i,dB = (−κ − λ log10 dk) − ξk,i + 10 log10 Fk,i (8)

In Equation (8), the first factor captures propagation loss,with κ a constant chosen to be 128.1 dB, dk the distancein km from mobile k to the nearest RRH in a distributedBS scenario, or to the BS in a centralised BS scenario, andλ the path loss exponent, which is set to a value of 3.76.The second factor, ξk,i, captures log-normal shadowing withan 8 dB standard deviation, whereas the last factor, Fk,i,corresponds to Rayleigh fading with a Rayleigh parameterb such that E[b2] = 1. Perfect channel state information(CSI) estimation is assumed at the BS. Results from theround robin (RR) algorithm are presented for reference.The implementation of the RR algorithm is as follows: Thenumber of RBs is fixed, and the RBs are allocated to userson the basis of one RB, selected randomly, per user.

5.2. Throughput results

Throughout this section, the results corresponding toRR will be denoted by (RR), and those correspondingto the proposed algorithms with throughput utilityand logarithmic throughput utility will be denotedby (U = R) and (U = ln(R)), respectively. Figures 6–8 compare the results of the different algorithms inthe case of deployment (a), to deployments (b–d),respectively.

These figures show that for (U = R), the case DBS-C outperforms the case CBS, which in turn outperforms



0 10 20 30 40 50 60 700

20

40

60

80

100

120

Number of Users

Sum

−Thr

ough

put (

Mbp

s)

CBS: RRCBS: U=RCBS: U=lnRDBS−C: RRDBS−C: U=RDBS−C: U=lnRDBS−D: RRDBS−D: U=RDBS−D: U=lnR

Figure 6. Sum-Throughput comparison in the case of deploymentscenario (b).

the case of DBS-D. This is due to the allocation of RBsto each RRH proportionally to the number of users inAlgorithm 2 (case DBS-D), which hinders the concept ofmax throughput scheduling that would allocate most RBs tousers nearest to RRHs (case DBS-C), not proportionally toeach RRH. For (U = ln(R)), the case DBS-C outperformsthe case DBS-D, which in turn outperforms the case CBS. Itshould be noted that the performance of DBS-C and DBS-Dscheduling is the same with RR, since each user is allocateda single RB in both cases. In addition, these figures showthe expected result that in terms of throughput, schedulingwithU = Routperforms scheduling withU = ln(R), whichin turn outperforms RR scheduling. However, it should benoted that Figures 6–8 show the sum throughput, but do notshow the effect of the distance from the BS or RRH on thethroughput of the different users. With (U = R), users close

0 10 20 30 40 50 60 700

20

40

60

80

100

120

Number of Users

Sum

−Thr

ough

put (

Mbp

s)


Figure 7. Sum-Throughput comparison in the case of deploymentscenario (c).

0 10 20 30 40 50 60 700

20

40

60

80

100

120

Number of Users

Sum

−Thr

ough

put (

Mbp

s)


Figure 8. Sum-Throughput comparison in the case of deploymentscenario (d).

to the BS or RRH are expected to receive most of the RBs,preventing edge users from fair access to resources most ofthe time. These issues will be discussed in the followingsection.

Comparing Figures 6–8, we can see that as the numberof RRHs increases, the performance of RR increases.Although the number of RBs allocated to each user isthe same, the sum throughput increases since each user’schances of finding a nearby RRH increase with the numberof RRHs. This leads to an increase in SNR and thusthroughput. In addition, with U = R, the superiority ofthe scenarios with more RRHs decreases as the numberof users increases. In fact, with a high number of users, theprobability of some users being near a RRH and receivingmost of the resources increases, even with a low numberof RRHs. With U = ln(R), the performance enhancementincreases with the number of RRHs for both Algorithms 1and 2.

5.3. Fairness analysis

To obtain an indication about the fairness of the differentinvestigated algorithms in different cases, we consider theexample of Figure 2, which consists of six users locatedat fixed positions in the case of scenario (c). We expresstheir positions in polar coordinates, i.e. a distance andan angular position from the origin, taken to be the cellcentre. The coordinates of the six users are shown inTable I.

Figure 9 shows the throughput results of each userin the case (U = R), and Figure 10 shows the averagenumber of subcarriers allocated per TTI to each of the usersof Figure 9. With the CBS case and U = R, almost allresources are allocated to User 1, the closest to the BS,



Table I. User positions with respect to the cell centre.

User 1 2 3 4 5 6

Distance (km) 0.15 0.9 0.85 0.95 0.5 1Angle (deg.) 190 0 80 160 200 290

which achieves the highest throughput. User 5, the secondnearest user, receives a very small amount of resources,whereas all other users suffer from starvation. In the DBSscenarios, the resource allocation process is clearly morefair. Cases DBS-C and DBS-D with U = R allow allusers except User 1 to receive more resources and achieveconsiderably higher throughput than the centralised case(CBS).

Figure 11 shows the throughput results of each user inthe PF case (U = ln(R)), and Figure 12 shows the averagenumber of subcarriers allocated per TTI to each of the usersof Figure 11.

With U = ln(R), Cases DBS-C and DBS-D clearlyoutperform the CBS case in terms of throughput andsubcarriers allocated, except for User 1. Contrarily to theU = R case, the performance of cases DBS-C and DBS-Dwith U = ln(R) is comparable, with a slight superiority forDBS-C. In addition, scheduling with U = ln(R) is clearlymore fair than with U = R. The most striking example canbe seen in the case of Users 4 and 5, which are associatedwith the same RRH. Figures 9 and 10 show that cases DBS-C and DBS-D with U = R clearly favour User 5, the nearestto the RRH, whereas Figures 11 and 12 show that casesDBS-C and DBS-D with U = ln(R) are considerably morefair towards User 4.

1 2 3 4 5 60

5

10

15

20

25

30

User No.

Ave

rage

Thr

ough

put (

Mbp

s)

CBS: U=RDBS−C: U=RDBS−D: U=R

Figure 9. Throughput achieved by each user: max throughputscheduling.

1 2 3 4 5 60

50

100

150

200

250

300

User No.

Ave

rage

Num

ber o

f Sub

carr

iers

CBS: U=RDBS−C: U=RDBS−D: U=R

Figure 10. Average subcarrier allocation: max throughputscheduling.

1 2 3 4 5 60

5

10

15

User No.

Ave

rage

Thr

ough

put (

Mbp

s)

CBS: U=ln(R)DBS−C: U=ln(R)DBS−D: U=ln(R)

Figure 11. Throughput achieved by each user: PF scheduling.

1 2 3 4 5 60

20

40

60

80

100

120

User No.

Ave

rage

Num

ber o

f Sub

carr

iers

CBS: U=ln(R)DBS−C: U=ln(R)DBS−D: U=ln(R)

Figure 12. Average subcarrier allocation: PF scheduling.



Figure 13. Optimised positions of the RRHs of scenario (d).

5.4. Location optimisation

In this section, we investigate the effect of optimising thelocations of the RRHs on the overall cell throughput. Weconsider scenario (d) as an example. In subsection 5.2, weinvestigated deployment scenario (d) with one RRH locatedat the cell centre and six RRHs located at a distance of2Rc/3, with 60◦ angular separation between them. In thissection, we place the RRHs such that the areas covered byeach RRH are equal. In fact, with a uniform user distributionthroughout the cell and with the RRHs having the samecharacteristics (transmit power, antenna, etc.), dividing theload equally among the RRHs is the best solution. Hence,fixing one RRH at the cell centre, we divide the cell intoseven equal coverage areas and place one RRH at thecentroid of each area, as shown in Figure 13. We comparethe results to those of scenario (d) in subsection 5.2, forboth the DBS-C and DBS-D cases, with utilities U = R andU = ln(R). The results are displayed in Figure 14. Althoughthe optimised positions yield better results as expected, theenhancement is less than five per cent between the optimised

0 10 20 30 40 50 60 7020

30

40

50

60

70

80

90

100

110

120

Number of Users

Sum

−Thr

ough

put (

Mbp

s)

DBS−C: U=RDBS−C: U=lnRDBS−D: U=RDBS−D: U=lnRDBS−D−Opt: U=RDBS−D−Opt: U=lnRDBS−C−Opt: U=RDBS−C−Opt: U=lnR

Figure 14. Sum-Throughput comparison in the case ofdeployment scenario (d): Optimised versus non-optimised RRHlocations.

distance and the positions selected in subsection 5.2.Furthermore, the enhancement increases with the numberof users with the utility U = R, but it is negligible withthe utility U = ln(R). Hence, in a practical deploymentscenario, deviating slightly from the optimised positionsdoes not lead to a noticeable degradation in performance.This agrees with the results of Reference [14] in thatthe position of the RRHs does not severely affect theperformance, as long as the number of RRHs is sufficient.

5.5. Mobility considerations

In the simulations of the previous sections, the mobile userswere considered with limited or no mobility. It would seemlogical to expect that in higher mobility scenarios, handoverrates between the different RRHs and the correspondingsignalling overhead would reduce the performance of thesystem. However, this is not the case. In the proposed model,the presence of the RRHs is transparent to the users whoact as if there was only a single central BS in the cell. Thesignalling and allocation of resources to RRHs take placeat the central BS location. Below is a proposed algorithmfor this allocation.

� At the beginning, the BS allocates the RBs equallyamong all RRHs, with an equal separation between RBsto ensure frequency diversity.

� The BS extracts the necessary CSI from the soundingreference signal (SRS) transmitted by the mobile in orderto perform channel dependent scheduling [32].

� From the measured CSI, if the estimated SINR of a givenuser k averaged over the subcarriers allocated to RRH jis higher than the SINR averaged over the subcarriersallocated to other RRHs, then user k is associated toRRH j.

� After determining the number of users linked to eachRRH, The BS allocates the RBs to RRHs appropriately(e.g. proportionally to the number of users in case DBS-D). Naturally, as the number of users increases, theuniform user distribution assumption will eventuallylead to a nearly equal allocation of subcarriers in thecase of optimal RRH positions.

� The association of users to RRHs can be modifieddepending on the changes in the SRS measurements.

Hence, the distributed BS system appears as a single BSto the user. The RRH acts as a simple BS antenna locatedcloser to the user position. The communication between thecentral BS and the RRHs is via fiber optic (or microwavelinks having a non-overlapping spectrum with LTE) andconsequently does not consume any LTE radio resources.



The mobility of a user heading from the area of RRH Ato the area covered by RRH B is translated in terms ofdecreasing channel quality on the subcarriers of RRH A andbetter channel quality on the subcarriers of RRH B. Hence,the user is allocated a subset of the subcarriers of RRH Bsimilarly to what would happen in a single BS scenariowhen channel quality deteriorates on certain subcarriersand improves on others. The presence of RRHs and theactual handover occurrence are oblivious to the user. As formobility between cells, the handover rules apply as in thecase of two cells with a central BS in each.

6. CONCLUSIONS

The concept of DBSs was investigated in the context ofLTE uplink. Resource allocation with a DBS was comparedto resource allocation with a centralised base station. Inthe DBS case, both centralised and distributed schedulingalgorithms were compared. Utility functions used in thescheduling algorithms included the throughput and thelogarithm of the throughput. Monte Carlo simulationsshowed that centralised scheduling with a DBS achievesmore throughput and fairness than centralised schedulingwith a centralised base station. In addition, in the DBS case,it was shown, via simulations, that distributed schedulingachieves almost the same results as centralised schedulingwith a proportional fair utility, with considerably reducedcomplexity.

ACKNOWLEDGMENTS

The authors thank the anonymous reviewers for their commentsthat helped in enhancing the quality of the paper.

This work was supported by the American University of Beirut(AUB), the AUB Research Board (URB), Dar Al-Handassah(Shair & Partners) Research Fund, and the Rathman (Kadifa)Fund.

REFERENCES

1. Ekstrom H, Furuskar A, Karlsson J, et al. Technical Solutions for the3G Long-Term Evolution. IEEE Communications Magazine 2006;44(3):38–45.

2. Myung HG, Goodman DJ. Single Carrier FDMA: A New Air Interfacefor Long Term Evolution. Wiley, 2008.

3. CPRI Specification. Common Public Radio Interface (CPRI);Interface Specification V2.0. Technical Report October 2004.

4. Saleh AAM, Rustako AJ, Roman RS. Distributed Antennas for IndoorRadio Communications. IEEE Transactions on Communications1987; 35(12):1245–1251.

5. Schuh RE, Sommer M. WCDMA Coverage and Capacity Analysis forActive and Passive Distributed Antenna Systems. Proceedings of theIEEE Vehicular Technology Conference-Spring 2002, Birmingham,Alabama, USA.

6. Dai L, Zhou S, Yao Y. Capacity with MRC-based Macrodiversity inCDMA Distributed Antenna Systems. Proccedings of IEEE Globecom2002, Taipei, Taiwan.

7. Dai L, Zhou S, Yao Y. Capacity Analysis in CDMA DistributedAntenna Systems. IEEE Transactions on Wireless Communications2005; 4(6):2613–2620.

8. Hasegawa R, Shirakabe M, Esmailzadeh R, Nakagawa M. DownlinkPerformance of a CDMA System with Distributed Base Station.Proceedings of the IEEE Vehicular Technology Conference-Fall 2003,Orlando, Florida, USA.

9. Choi W, Andrews JG, Yi C. The Capacity of Multicellular DistributedAntenna Networks. Proceedings of the International Conference onWireless Networks, Communications and Mobile Computing 2005,Maui, Hawaii, USA.

10. Choi W, Andrews JG. Downlink Performance and Capacity ofDistributed Antenna Systems in a Multicell Environment. IEEETransactions on Wireless Communications 2007; 6(1):69–73.

11. Clark MV, Willis TM, Greenstein LJ, Rustako AJ, Ercegt V, RomanRS. Distributed versus Centralized Antenna Arrays in BroadbandWireless Networks. Proceedings of the IEEE Vehicular TechnologyConference-Spring 2001, Rhodes, Greece.

12. Roh W, Paulraj A. Outage Performance of the Distributed AntennaSystems in a Composite Fading Channel. Proceedings of the IEEEVehicular Technology Conference-Fall 2002, Vancouver, Canada.

13. Zhang J, Andrews JG. Cellular Communication with RandomlyPlaced Distributed Antennas. Proceedings of IEEE Globecom 2007,Washington, DC, USA.

14. Zhang J, Andrews JG. Distributed Antenna Systems withRandomness. IEEE Transactions on Wireless Communications 2008;7(9):3636–3646.

15. Hadisusanto Y, Thiele L, Jungnickel V. Distributed Base StationCooperation via Block-Diagonalization and Dual-Decomposition.Proceedings of IEEE Globecom 2008, New Orleans, Louisiana,USA.

16. Leroux P, Roy S, Chouinard J-Y. A Multi-Agent Protocol to ManageInterference in a Distributed Base Station System. Proceedingsof the International Conference on Advanced Technologies forCommunications (ATC) 2008, Hanoi, Vietnam.

17. Highsmith WR. An Investigation into Distributed Base Station Designfor LMDS Systems. Proceedings IEEE Southeastcon 2002, Columbia,South Carolina, USA.

18. Alcatel-Lucent. Alcatel-Lucent expands 3G W-CDMA/HSPA portfo-lio with new distributed base station that offers increased deploymentflexibility and lowers power requirements. http://www.alcatel-lucent.com/wps/portal/NewsReleases/, Alcatel-Lucent, March 2008.

19. Alcatel-Lucent. Alcatel-Lucent unveils 3G CDMA/EV-DO dis-tributed base station that offers greater deployment flexi-bility while lowering power requirements. http://www.alcatel-lucent.com/wps/portal/newsreleases/, Alcatel-Lucent, April 2008.

20. Yaacoub E, Dawy Z. A Game Theoretical Formulation forProportional Fairness in LTE Uplink Scheduling. Proceedings of theIEEE Wireless Communications and Networking Conference 2009,Budapest, Hungary.

21. Yaacoub E, Dawy Z. Low Complexity Scheduling Algorithms for theLTE Uplink. Proceedings of the IEEE International Symposium onComputers and Communications 2009, Sousse, Tunisia.

22. Song G, Li Y. Cross-Layer Optimization for OFDM WirelessNetworks-Part I: Theoretical Framework. IEEE Transactions onWireless Communications 2005; 4(2):614–624.

23. Song G, Li Y. Cross-Layer Optimization for OFDM WirelessNetworks-Part II: Algorithm Development. IEEE Transactions onWireless Communications 2005; 4(2):625–634.



24. Wong CY, Cheng RS, Letaief KB, Murch RD. Multiuser OFDMwith Adaptive Subcarrier, Bit and Power Allocation. IEEE Journalon Selected Areas in Communications 1999; 17(10):1747–1757.

25. Wang X, Giannakis GB. Ergodic Capacity and Average Rate-Guaranteed Scheduling for Wireless Multiuser OFDM Systems.Proceedings of the IEEE International Symposium on InformationTheory 2008, Toronto, Canada.

26. Wong IC, Evans BL. Optimal Downlink OFDMA ResourceAllocation with Linear Complexity to Maximize Ergodic Rates. IEEETransactions on Wireless Communications 2008; 7(3):962–971.

27. Kim K, Han Y, Kim S-L. Joint Subcarrier and Power Allocationin Uplink OFDMA Systems. IEEE Communications Letters 2005;9(6):526–528.

28. Gao L, Cui S. Efficient Subcarrier, Power, and Rate Allocation withFairness Consideration for OFDMA Uplink. IEEE Transactions onWireless Communications 2008; 7(5):1507–1511.

29. Ng CY, Sung CW. Low Complexity Subcarrier and Power Allocationfor Utility Maximization in Uplink OFDMA Systems. IEEETransactions on Wireless Communications 2008; 7(5):1667–1675.

30. Huang J, Subramanian VG, Agrawal R, Berry R. Joint Schedulingand Resource Allocation in Uplink OFDM Systems for BroadbandWireless Access Networks. IEEE Journal on Selected Areas inCommunications 2009; 27(2):226–234.

31. Myung HG, Lim J, Goodman DJ. Single Carrier FDMA for UplinkWireless Transmission. IEEE Vehicular Technology Magazine 2006;48(1):30–38.

32. 3rd Generation Partnership Project (3GPP). 3GPP TR 25.814. 3GPPTSG RAN Physical Layer Aspects For Evolved UTRA, v7.1.0, 2006.

33. Lunttila T, Lindholm J, Pajukoski K, Tiirola E, Toskala A. EUTRANUplink Performance. Proceedings of the International Symposium onWireless Pervasive Computing (ISWPC) 2007, San Juan, Puerto Rico.

34. 3rd Generation Partnership Project (3GPP). 3GPP TS 36.213 3GPPTSG RAN Evolved Universal Terrestrial Radio Access (E-UTRA)Physical layer procedures, version 8.3.0, Release 8, 2008.

35. Wu H, Haustein T. Energy and Spectrum Efficient TransmissionModes for the 3GPP-LTE Uplink. Proceedings of the IEEEInternational Symposium on Personal Indoor and Mobile RadioCommunication 2007, Athens, Greece.

36. Calabrese FD, Michaelsen PH, Rosa C, Anas M, Ubeda CastellanosC, Lopez Villa D. Search-Tree Based Uplink Channel Aware PacketScheduling for UTRAN LTE. Proceedings of the IEEE VehicularTechnology Conference-Spring 2008, Marina Bay, Singapore.

37. Wong IC, Oteri O, McCoy W. Optimal Resource Allocation in UplinkSC-FDMA Systems. IEEE Transactions on Wireless Communications2009; 8(5):2161–2165.

38. Qiu X, Chawla K. On the Performance of Adaptive Modulationin Cellular Systems. IEEE Transactions on Communications 1999;47(6):884–895.

39. Lim J, Myung HG, Oh K, Goodman DJ. Channel-DependentScheduling of Uplink Single Carrier FDMA Systems. Proceedingsof the IEEE Vehicular Technology Conference-Fall 2006, Montreal,Canada.

AUTHORS’ BIOGRAPHIES

Elias Yaacoub received the B.E. degree in Electrical Engineering from the Lebanese University in 2002, and the M.E. degree inComputer and Communications Engineering from the American University of Beirut in 2005. He worked as a Research Assistant inthe American University of Beirut from 2004 to 2005 and in the Munich University of Technology in Spring 2005. From 2005 to 2007,he was a Telecommunications Engineer with Dar Al-Handasah, Shair and Partners. He is currently a PhD student at the AmericanUniversity of Beirut. His research interests include wireless communications and antenna theory.

Zaher Dawy received the B.E. degree in Computer and Communications Engineering from the American University of Beirutin 1998. He received his M.Sc. and Dr.-Ing. degrees in Electrical Engineering from Munich University of Technology (TUM)in 2000 and 2004, respectively. Between 1999 and 2000, he worked as a part-time Communications Engineer at Siemens AGresearch labs in Munich focusing on the development of enhancement techniques for UMTS. At TUM, between 2000 and 2003he managed and developed a research project with Siemens AG where he designed advanced multiuser receiver structures for UMTSbase stations. Since September 2004, he is an Assistant Professor at the Electrical and Computer Engineering Department at theAmerican University of Beirut. His research interests include Cooperative Communications, Cellular Technologies (WCDMA, HSPA,LTE), Radio Network Planning and Optimization, Multiuser Information Theory, Multimedia over IP Networks, and ComputationalBiology.


uplink scheduling in lte systems using distributed base stations

Documents