ofdma vs sc fdma performance comparison on local area imt-a scenarios

8/2/2019 OFDMA vs SC FDMA Performance Comparison on Local Area IMT-A Scenarios

1/9IEEE Wireless Communications October 200864 1536-1284/08/$25.00 2008 IEEE

PAPR (db)

7 8 9 10

2.2 dB

The authors discuss

the suitability of

using OFDMA or

SC-FDMA in the

uplink for local area

high-data-rate

scenarios by

considering as target

performance metrics

the PAPR and

the multi-user

diversity gain.

RECEN T AD VA NC ES A ND EV OL UT IO N O F

WLAN AND WMAN STANDARDS

INTRODUCTION

The International Telecommunication Union(ITU) is currently specifying the requirementsfor the next generation of mobile communica-tion systems, the so-called International MobileTelecommunications Advanced (IMT-A).The deployment of the latter at the mass marketlevel is expected to take place around 2015 andfacilitate what has already been a buzzword forthe last decade: the fourth generation (4G).IMT-A systems are expected to provide peakdata rates on the order of 1 Gb/s in local areasand 100 Mb/s in wide areas [1]. In order to copewith such targets, the employment of multiple-input multiple-output (MIMO) antenna technol-ogy and very high spectrum allocations (in the

range of 100 MHz) is foreseen.The standardization of the Long Term Evolu-

tion (LTE) of Universal Terrestrial Radio Accessbegan in the first half of 2005 and, practically,will set the main reference for next-generationsystems assumptions, LTE- Advanced (LTE-A).As a consequence, in order to achieve maximumcompatibility, LTE and LTE-A should bealigned. LTE has selected orthogonal frequency-division multiple access (OFDMA) in the down-link and single-carrier frequency-divisionmultiple access (SC-FDMA) in the uplink [2].However, other standards, such as WorldwideInteroperability for Microwave Access(WiMAX), use OFDMA in both links, given thebenefits of having the same access scheme interms of reciprocity, allocation flexibility, andbandwidth efficiency [3]. While OFDMA is astrong candidate for IMT-A systems in the down-link, the same cannot be ensured for SC-FDMAin the uplink. Note that the influence of the sce-nario in which future technologies are expectedto be deployed, and backwards compatibilitywith earlier releases, will definitely play a deter-minant role in the selection of the proper solu-tion. In this article we provide a performanceevaluation of both uplink candidates, OFDMAand SC-FDMA, for IMT-A systems in local areascenarios. Furthermore, in order to cope with

the 100 MHz requirements, we propose new

GILBERTO BERARDINELLI, LUIS NGEL MAESTRO RUIZ DE TEMIO, SIMONE FRATTASI,MUHAMMAD IMADUR RAHMAN, AND PREBEN MOGENSEN, AALBORG UNIVERSITY

ABSTRACT

The system requirements for IMT-A arecurrently being specified by the ITU. Targetpeak data rates of 1 Gb/s in local areas and100 Mb/s in wide areas are expected to be pro-vided by means of advanced MIMO antennaconfigurations and very high spectrum alloca-tions (on the order of 100 MHz). For thedownlink, OFDMA is unanimously consideredthe most appropriate technique for achievinghigh spectral efficiency. For the uplink, theLTE of the 3GPP, for example, employs SC-FDMA due to its low PAPR properties com-pared to OFDMA. For future IMT-A systems,the decision on the most appropriate uplinkaccess scheme is still an open issue, as many

benefits can be obtained by exploiting the flex-ible frequency granularity of OFDMA. In thisart ic le we discuss the suitabi l i ty of usingOFDMA or SC-FDMA in the uplink for localarea high-data-rate scenarios by considering astarget performance metrics the PAPR andmulti-user diversity gain. Also, new bandwidthconfigurations have been proposed to copewith the 100 MHz spectrum allocation. In par-ticular, the PAPR analysis shows that a local-ized (not distr ibuted) al locat ion of theresource blocks (RBs) in the frequency domainshall be employed for SC-FDMA in order tokeep its advantages over OFDMA in terms ofPAPR reduction. Furthermore, from the multi-user diversity gain evaluation emerges the factthat the impact of different RB sizes and band-width configurations is low, given the propaga-tion characteristics of the assumed local areaenvironment . For ful l bandwidth usage,OFDMA only outperforms SC-FDMA whenthe number of frequency multiplexed users islow. As the spectrum load decreases, instead,OFDMA outperforms SC-FDMA also for ahigh number of frequency multiplexed users,due to its more flexible resource allocation. Inthis contex different channel-aware schedulingalgorithms have been proposed due to theresource allocation differences between the

two schemes.

OFDMA VS. SC-FDMA: PERFORMANCE

COMPARISON IN LOCAL AREA IMT-A SCENARIOS

Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 15, 2010 at 07:38 from IEEE Xplore. Restrictions apply.


2/9IEEE Wireless Communications October 2008 65

bandwidth configurations where the selection ofthe spectrum parameters will again depend onthe scenario conditions.

The rest of the article is organized as follows.The next section presents the basic principles ofOFDMA and SC-FDMA. The following sectionintroduces the proposed 100 MHz bandwidthconfigurations, and their pros and cons are dis-cussed in detail. We then show the peak-to-aver-age power ratio (PAPR) analysis for thepreviously defined configurations, and the fol-

lowing section focuses on the multi-user diversitygain evaluation of both access schemes. In par-ticular, the latter describes the channel-awarescheduling algorithms employed for OFDMAand SC-FDMA, and the semi-analytical frame-work used to derive the simulation results. Thefinal section summarizes the main conclusions ofthis work and outlines the future directions ofinvestigation.

OFDMA AND SC-FDMA PRINCIPLES

OFDMA is a multiple access scheme based onthe well-known orthogonal frequency-division

multiplexing (OFDM) modulation technique. Itsmain principle is to split the data stream to betransmitted onto a high number of narrowbandorthogonal subcarriers by means of an inversefast Fourier transform (IFFT) operation, whichallows for an increased symbol period. The lat-ter, together with the use of a guard intervalappended at the beginning of each OFDM sym-bol, provides this technology great robustnessagainst multipath transmission [4]. A realizationof this guard interval is the so-called cyclic prefix(CP), which consists of a repetition of the lastpart of an OFDM symbol. As long as the CP islonger than the maximum excess delay of thechannel, degradations due to intersymbol inter-

ference (ISI) and intercarrier interference (ICI)are avoided. Furthermore, the goal of employingnarrowband subcarriers is to obtain a channelthat is roughly constant over each given sub-band, which makes equalization much simpler atthe receiver. Finally, since these subcarriers aremutually orthogonal, overlapping between themis allowed, yielding a highly spectral efficient sys-tem. Despite all these benefits, OFDM also pre-sents some drawbacks: sensitivity to Dopplershift, synchronization problems, and inefficientpower consumption due to high PAPR [4].

SC-FDMA is a multiple access scheme basedon the single-carrier frequency-division multi-plexing (SC-FDM) modulation technique, some-times also referred to as discrete Fouriertransform (DFT)-spread OFDM. Its main princi-ple is the same as for OFDM; thus, the samebenefits in terms of multipath mitigation andlow-complexity equalization are achievable [5].The difference though is that a DFT is per-formed prior to the IFFT operation, whichspreads the data symbols over all the subcarrierscarrying information and produces a virtual sin-gle-carrier structure. As a consequence, SC-FDMpresents a lower PAPR than OFDM [6]. Thisproperty makes SC-FDM attractive for uplinktransmissions, as the user equipment (UE) bene-fits in terms of transmitted power efficiency. On

one hand, DFT spreading allows the frequency

selectivity of the channel to be exploited, since allsymbols are present in all subcarriers. Therefore,if some subcarriers are in deep fade, the informa-tion can still be recovered from other subcarriersexperiencing better channel conditions. On theother hand, when DFT despreading is performedat the receiver, the noise is spread over all thesubcarriers and generates an effect called noiseenhancement, which degrades the SC-FDM per-formance and requires the use of a more com-plex equalization based on a minimum mean

square error (MMSE) receiver [5].

100 MHZ BANDWIDTH CONFIGURATIONS

In order to satisfy the 100 MHz spectrumrequirements for IMT-A systems, we proposenew definitions of bandwidth allocation (Table1), where the 20 MHz LTE case is taken as areference [7] so that the new system solutionsappear as its natural evolution. As there arestrong dependencies among configurationparameters, let us briefly introduce some of thelatter as follows: Subcarrier spacing (f): spectrum offset

between adjacent subcarriers.FFT size (NFFT): total number of orthogonal

subcarriers in the system; for computationalefficiency reasons, this value must be power of2 [8].

Number of used subcarriers (Nc): in a real sys-tem, due to some spectral constraints such asbandwidth allocation and spectral mask, notall the subcarriers are employed. This reducednumber of tones together with the subcarrierspacing defines the useful transmission band-width.

Sampling frequency (F0): minimum time resolu-tion the receiver must adopt in sampling thereceived signal in order to correctly recon-

struct it; practically,F0 =NFFT f.

I Table 1. 100 MHz bandwidth configurations.

20 MHz 100 MHz

f(kHz) 15 15 30 60 120

NFFT 2048 8192 4096 2048 1024

Nc 1200 6000 3000 1500 750

F0 (MHz) 30.72 122.88 122.88 122.88 122.88

TS (s) 66.67 66.67 33.33 16.67 8.33

Tslot (ms) 0.5 0.5 0.5 0.5 0.5

Symbols perslot

7 7 14 28 56

CP(s/samples)

5.2/1601

4.68/14425.2/6404.68/576

2.6/3202.34/288

1.3/1601.17/144

0.65/800.58/72

1 First OFDM/SC-FDM symbol in a slot.2 2nd7th OFDM/SC-FDM symbol in a slot.



3/9IEEE Wireless Communications October 200866

Useful symbol duration (Ts):inverse of the sub-carrier spacing.

Cyclic prefix (CP): guard interval appended atthe beginning of each OFDM/SC-FDM sym-bol to avoid degradations due to ISI and ICI.As the subcarrier spacing increases, the use-

ful symbol duration is reduced; therefore, inorder to keep low spectral efficiency loss, the CPshould be proportionally decreased. However,this could lead to ISI and ICI for long channelimpulse response environments. On the other

hand, reducing the symbol duration provideshigher robustness against Doppler shift [9] andyields a loss in frequency-domain granularity,which is instead important for scheduling pur-poses. In order to keep contiguity with LTE, itsslot duration (Tslot = 0.5 ms) is preserved. Thisleads to different numbers of transmittedOFDM/SC-FDM symbols within a slot accordingto the chosen configuration. Moreover, the longCP defined in Table 2 shall be employed each 7symbols in order to keep the desired slot dura-tion. Regarding complexity, it can be seen thatwhen the subcarrier spacing decreases, the FFTsize increases, thus bringing higher computation-

al complexity and thereby higher power con-sumption at the UE. Finally, it has to be pointedout that in all cases the sampling frequency is

increased four times with respect to the LTEvalue.

If we now consider the local area scenario,which is within the scope of this article, someof the above-mentioned constraints can berelaxed. For example, the s ubcarrier spacingcan be increased up to 120 kHz without ISIand ICI degradations, as the CP is long enoughto cover for the arrival of all the multipathcomponents.1 Therefore, lower values of FFTsize can be also employed, thus reducing the

complexity of the system. However, the impactof this large subcarrier spacing on the frequen-cy granularity should be further analyzed interms of scheduling.

PAPR EVALUATION

Despite its wide acceptance, OFDM exhibitslarge envelope variations of the transmitted sig-nal. This is due to the transmission of data overparallel subcarriers, which could constructivelyadd in phase and yield high instantaneous peakpower compared to its average. Signals with highPAPR require highly linear power amplifiers in

order to avoid excessive intermodulation distor-tion. Therefore, the amplifier must operate witha large backoff from its peak value. This leads tolow power efficiency, which is measured as theratio of the transmitted power over the dc dissi-pated power. High PAPR is particularly criticalfor uplink transmissions, given the power con-straints at the UE. As pointed out previously,the low PAPR properties of the SC-FDM signalmake it a very attractive solution for the uplink.In this section we carry out a PAPR evaluationfor OFDM and SC-FDM with the proposed 100MHz bandwidth configurations, where resultsare obtained by means of Monte Carlo simula-tions, assuming 16-quadrature amplitude modu-

lation (QAM).From the cumulative complementary distribu-

tion functions (CCDFs) in Fig. 1a, we observethat the PAPR of the transmitted user signal isstrictly dependent on the number of subcarriers.In particular, the configuration with the lowestnumber of subcarriers (750/1024) provides thebest results (i.e., the lowest PAPR values) forboth OFDM and SC-FDM. Note that for all theconfigurations, the PAPR gain of SC-FDM overOFDM is approximately 2.2 dB.

In a multi-user scenario the available band-width is divided in basic units called resourceblocks (RBs), which are shared among users. Asa consequence, it is also important to considerthe case of partial bandwidth usage for each userin our evaluation. In particular, we decided toadopt two resource allocation strategies: Localized: All the RBs are allocated to the

user in a contiguous manner.

I Table 2. Simulation parameters and models.

Parameter Value

Cell radius 30 m

Minimum BS-UE distance 3 m

Carrier frequency 3.5 GHz

Antenna scheme SISO, SIMO (1x2)

Number of users 1-20

UE speed 3 km/h

Simulated slots 2000

Maximum transmitted power per user 24 dBm

Power control no

Noise power 160 dBm/Hz

Channel model A1NLOS [10]

PathLoss and shadowing models from Winner II project [10]

Access schemes OFDMA, SC-FDMA

Power backoff for OFDM 3.5 dB

Power backoff for SC-FDM 2 dB

Scheduling metric PF

Scheduling algorithm for OFDMA Greedy

Scheduling algorithm for SC-FDMA RME

1 The maximum delay spread of the local area indoor

channel that we consider is 175 ns [10].

2 We refer to distributed allocation at the RB level rather

than at the subcarrier level [6]. The latter has not been

considered because of its high sensitivity to the frequency

offset between different types of UE [11].




Distributed: All the RBs allocated to the userare randomly distributed over the whole band-width.2

In Fig. 1b performance results are obtained for the(6000/8192) configuration, considering 10 percentbandwidth usage (i.e., 600 assigned subcarriers).The RB size is fixed to 15 subcarriers, which meansthat a total of 40 RBs are assigned to the user ofthe 400 available. Despite its low PAPR benefit,we observe that SC-FDM is much more sensitiveto resource allocation than OFDM. For localizedallocation, the PAPR gain of SC-FDM overOFDM is almost 2 dB. Nevertheless, we found out

that this gain is dramatically reduced for distribut-ed allocation (0.3 dB). Therefore, if the RBs arerandomly distributed along the bandwidth, themain selling point of SC-FDM with respect toOFDM falls. Hence, in order to keep its low PAPRproperties, localized allocation of the RBs shall beconsidered for SC-FDM. This result is indeedremarkable, although it imposes a constraint onthe scheduling design. Finally, note that forOFDMA the increase of PAPR when passing fromlocalized to distributed allocation is approximately1dB, while for SC-FDMA it is 2.7 dB.

MULTI-USER DIVERSITY GAIN

In a multi-user scenario the available bandwidthmust be shared among several users. Each usermay experience different conditions in terms ofvelocity, path loss, and shadowing. Furthermore,each may have different requirements in termsof quality of service. A smart design of the net-work should therefore take into account the dif-ferent user conditions while providing fairness,without a drastic reduction in the overall cellthroughput. A higher spectral efficiency is actu-ally the main goal of all radio interface design.In an adaptive OFDMA-based system the cellspectral efficiency can be increased as the num-ber of users is. This effect is called multi-user

diversity gain [12], and it is mainly due to:

The increase of the total received power at thebase station (BS) with the number of users.

The possibility of assigning orthogonal time-frequency resources to the user who can uti-lize them best. This is quite different from thepresent systems based on code-division multi-ple access, in which the nature of imperfectresource orthogonality leads to a reduction ofthe cell throughput as the number of usersincreases [13].The multi-user diversity gain of OFDMA-

based systems can be exploited by adopting chan-nel-aware schedulers, where the main idea is to

allocate the RBs to users experiencing betterchannel conditions in those frequency slots. Inthis way the overall cell throughput is increasedwith the number of users, given the availability ofmany different channel conditions that can beconsidered for the best bandwidth allocationsearch. Therefore, an analysis of the multi-userdiversity gain should be focused on channel-awarescheduling algorithms as well as the behavior ofthe selected multiple access schemes in a multi-user environment. Different scheduling algo-rithms have therefore been considered forOFDMA and SC-FDMA, as the latter presentsthe constraint of localized allocation (see the pre-vious section). Our effort has been particularlyfocused on the development of a novel channel-aware scheduling algorithm for SC-FDMA (dis-cussed below). Note that the performanceevaluation of OFDMA and SC-FDMA in a multi-user scenario has been investigated adopting asemi-analytical approach, which is described later.

CHANNEL-AWARE

SCHEDULING ALGORITHMS

OFDMA This allows high freedom in resourceallocation because no constraint on the contigu-ousness of the RBs is generally assumed. The

optimal scheduling is based on a combinatorial

I Figure 1. CCDF of PAPR for: a) 100 MHz bandwidth configurations; b) localized vs. distributed allocation.

PAPR (db)

16-QAM

(a)

650

0.1

CCDF

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

7 8 9 10 11 12 13 14PAPR (db)

6000/8192, 16-QAM, 10% BW usage

(b)

650

0.1

CCDF

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

7 8 9 10 11 12 13 14

OFDM, localizedOFDM, distributedSC-FDM, localizedSC-FDM, distributed

OFDM 750/1024OFDM 1500/2048OFDM 3000/4096OFDM 6000/8192SC-FDM 750/1024SC-FDM 1500/2048SC-FDM 3000/4096SC-FDM 6000/8192

2.2 dB2.7 dB

0.3 dB

1 dB

1.9 dB




I Figure 2.Example of resource allocation by the RME algorithm.

Resource allocation

UE2

UE0

UE1

RB0

(a)

Frequency

Metrics

Resource allocation

UE2

UE0

UE1

(b)

Frequency

Metrics

Resource allocation

UE2

UE0

UE1

(c)

Frequency

Metrics

Resource allocation

UE2

UE0

UE1

(d)

Frequency

Metrics

Resource allocation

UE2

UE0

UE1

(e)

Frequency

Metrics

Fourth max

First max

Second max

Third max

Fifth max




comparison of all the possible allocations ofRBs. Nevertheless, this solution is intractablebecause of its high computational expense. Inour investigation the well-known greedy algo-rithm has been considered [14]. Practically, thelatter allocates the resources to the user thatmaximizes the marginal scheduling metric gainin each RB, and it is expected to represent asuboptimal solution.

SC-FDMA It requires localized allocation of the

RBs in order to exploit its PAPR benefits overOFDMA; this, on the other hand, reduces theresource allocation flexibility and thus rises achallenge in the scheduling design. As in theOFDMA case, the optimal solution is inapplica-ble. In this section we propose a relatively low-complexity channel-aware scheduling algorithmfor SC-FDMA, which is based on a recursiveexpansion of the bandwidth to allocate to eachuser by starting from the maximum value of thescheduling metric. Considering as an input thematrixM, whose dimension is [number_of_users,number_of_RBs] and whose values represent allthe user metrics for each RB, the steps of the

proposed recursive maximum expansion (RME)algorithm are the following:Step 1: Select the combination UE-RB with the

highest metric value (UE0RB0 in Fig. 2a).Step 2: AssignRB0 to UE0.Step 3: Expand the allocation in step 2 for UE0

on both the right and left sides ofMuntilanother UE with a better metric is found(UE1in Fig. 2a).

Step 4: Put UE0 in idle mode.Step 5: Repeat steps 14 by searching for the

maximum among the non-idled UEs (Fig.2b2c). Stop when all the UEs are idled or allRBs have been allocated.

Step 6: If not all RBs have been allocated, search

for the UE with the maximum value of themetric among the remaining RBs.

Step 7: Check if one of the adjacent alreadyassigned RBs belongs to the same UE foundin step 6.

Step 8: If the UE is not the same, delete thismaximum fromMand repeat step 6.Otherwise, expand its allocation on both theright and left sides ofMuntil contiguousnesswith the previous allocation is achieved onone side. Stop to expand on the other sidewhenever another (idled) UE with a highermetric value is found (Fig. 2d).

Step 9: Repeat steps 68 until all RBs are allo-cated (Fig. 2e).

SEMI-ANALYTICAL APPROACHGiven the different nature of their signal genera-tion as well as the different constraints in termsof resource allocation, OFDMA and SC-FDMAare expected to behave distinctly in a multi-userscenario. An evaluation of both access schemesperformance can be done by focusing on theiranalytical behaviors or by means of system-levelsimulations. A fully analytical approach becomesintractable, though, if rigorously applied to arealistic scenario. As a consequence, in order togive our investigation a certain degree of gener-ality and, at the same time, keep it easily man-

ageable, we opted for a trade-off between the

two approaches. The evaluation has thereforebeen carried out following a semi-analyticalapproach. First, we simulate the scenario (i.e.,generate user locations, fast fading channels,shadowing, path loss, etc.). Then, based on thelatter, we employ the scheduling algorithmsdescribed previously. Once the resource alloca-tion is performed, the signal-to-noise ratio expe-rienced by each user in the assigned resources isused in the analytical expressions below in orderto calculate the optimal spectral efficiency. In

this way we retrieve an upper bound that cangive useful insights on the performance of thetwo schemes. All pieces of UE are assumed totransmit at the same power Pmax, where thetransmitted power per subcarrier of user (Pk

(sub))depends on the number of frequency resourcesallocated to him. For OFDMA, the data symbolsare directly k mapped over the subcarriers.Therefore, the upper spectral efficiency of userkis simply obtained by summing up the upperspectral efficiency values over the subcarrierswithin each RB assigned to that user, which isgiven by

(1)

whereNsub,RB is the number of subcarriers perRB, IRB,k is the set of RBs assigned to user k,NRB is the total number of RBs, and the signal-to-noise ratio per subcarrier is defined as

(2)

where |Hi,k|2 is the channel gain of subcarrier ifor user k;Lloss,k is the path loss and shadowingterm of user k; n2 is the noise power per Hertz;and fis the subcarrier spacing in Hertz. In SC-FDMA, the data detection is performed after aninverse DFT operation. Therefore, the upperdata rate of a certain UE cannot be expressed asa linear sum of the upper data rates over all theallocated RBs. The upper spectral efficiency ofuserk can be then written as [15]

(3)

S P II

NSC FDMSA k RB k

RB k

RB =, max ,,

( , ) log2

111

1

1

1+

+ ( )

Isub k

i Ii k

i ksub k

,

,

,,

1

,

i k

ksub

i k

loss k n

P H

L f,

( ),

,

,=( )

2

2

S P IN N

OFDMA k RB k RB sub RBi Isub k

, max ,,

( , )

,

=

1 1

log ,( ) ,,,

2 11+ +

i N j k

j Nsub RB

sub RB

Given the different

nature of their signal

generation as well

as the different

constraints in terms

of resource

allocation, OFDMAand SC-FDMA are

expected to behave

distinctly in a

multi-user scenario.

3Equations 1 and 3 have been obtained hypothesizing

that the length of the CP is greater than the length of the

discrete-time baseband channel impulse response so that

ISI and ICI are avoided. In the local area scenario this is

consistent with all the proposed bandwidth configurations.




where |IRB,k| is the number of RBs assigned touserk and |Isub,k| is the total number of subcar-riers of userk3 each RB. By means of these val-ues, the scheduling metrics are calculatedaccording to the proportional fair (PF) criterion[16], which are then exploited by the algorithmsin the previous section, Finally, Eqs. 1 and 3 areused in the final upper spectral efficiency com-putation for each UE given its assigned set ofRBs.

SIMULATION RESULTSTable 2 shows the main parameters of the MAT-LAB simulator, which has been developed

according to the previous semi-analyticalapproach. The scenario we have investigated isan isolated cell with no surrounding interferercells, where propagation and channel models aretaken from [10]. Both single and multiple receiv-ing antenna configurations have been consid-ered. In particular, the simulator generates apredefined number of users and uniformly dis-tributes them within the cell. The average spec-tral efficiency is obtained after a simulation timeequal to 2000 slots.

In Fig. 3a the maximum achievable spectralefficiency of OFDMA is shown as a function ofthe number of users in the cell and for differentRB sizes (number of subcarriers within an RB),where the (750/1024) configuration and onereceiving antenna have been considered. As ageneral trend, we observe that the spectral effi-ciency increases with the number of users, due tothe multi-user diversity gain property of OFDMA-based systems. For a few users in the system (upto 10), no relevant difference between the RBsizes is appreciable. As the number of usersincreases, the scheme with lower RB size leads tobetter performance; this is due to the finer granu-larity in resource allocation. Nevertheless, the gapbetween the configurations is still small. In Fig. 3bthe comparison between different bandwidth con-figurations is presented. The spectral efficiency

curves appear almost overlapped; therefore, the

throughput performance in a local area scenariocan be considered independent of the selectedbandwidth configuration. This consequentlyallows the choice of the proper bandwidth config-uration to be relaxed, making it only dependenton hardware/computational complexity considera-tions. Similar trends have been observed for SC-FDMA; therefore, the same conclusions can bedrawn for it.

In Fig. 4 the comparison between OFDMAand SC-FDMA is shown for the (750/1024) con-figuration and two-receiving-antennas system. Inorder to guarantee amplifier linearity and there-by avoid intermodulation distortion, different

power backoff values, derived from cubic metricvalues adopted by the 3GPP for power deratingcomputation [17], are considered in the trans-mitter for both OFDMA and SC-FDMA. Notethat the power backoff value for OFDM is high-er than that for SC-FDM because of its inherenthigher PAPR. Results for different traffic loadconditions are presented. In a fully loaded sce-nario (100 percent of bandwidth usage), SC-FDMA performs worse than OFDMA for fewusers in the system. The higher transmittedpower does not allow the lower scheduling flexi-bility to be fully compensated for. For a numberof users higher than 12, SC-FDMA achieveshigher upper spectral efficiency. A high numberof users in the system leads to few resourcesallocated to each of them; therefore, the lowerscheduling flexibility is not so critical. Note thatthese results are dependent on the chosenscheduling algorithms, which are suboptimal, asoutlined previously. Let us examine the case offew users in the system (up to 5). As the band-width usage decreases to 75 percent, the perfor-mance of SC-FDMA and OFDMA get closer.This gap is further reduced for 25 percent band-width usage. Low traffic load in the systemenhances the freedom in SC-FDMA resourceallocation, making it similar to OFDMA. On theother hand, for a high number of users (e.g., 20),

the gap between SC-FDMA and OFDMA is

I Figure 3. OFDMA spectral efficiency results for: a) different RB sizes; b) different subcarrier spacing.

Number of users

(a)

100 MHz, SISO, A1NLOS, OFDMA, 750/1024, PF

56

7

Spectralefficiency(b/s/Hz)

8

9

10

11

12

10 15 20Number of users

(b)

100 MHz, SISO, A1NLOS, OFDMA, RB size = 15, PF

56

7

Spectralefficiency(b/s/Hz)

8

9

10

11

12

10 15 20

750/1024 (f = 120 kHz)6000/8192 (f = 15 kHz)

RB size = 15RB size = 30RB size = 45




slightly reduced when passing to 75 percentbandwidth usage. For very low bandwidth usage(25 percent), OFDMA always performs betterthan SC-FDMA, and their gap tends to increasewith the number of users.

CONCLUSIONS AND FUTURE WORK

The definition of the system requirements forthe upcoming IMT-A is currently under discus-sion in ITU Radiocommunication Standard-

ization Sector (ITU-R) Working Party 5D.Solutions should properly weigh flexibility andefficiency in order to realistically cope with thedata rate targets of 1 Gb/s in local areas and 100Mb/s in wide areas. Such targets can be reachedby a combination of very wide spectrum alloca-tion (i.e., on the order of 100 MHz) combinedwith high peak spectral efficiency by using multi-ple antennas. In this article we have analyzed100 MHz bandwidth configurations for theuplink of local area IMT-A. The study has beencarried out by varying the subcarrier spacing andkeeping the slot duration as in the 3GPP LTERelease 8 specifications (i.e., varying the number

of symbols within a slot). Within this framework,the performance of OFDMA and SC-FDMA, asstrong access scheme candidates for the uplink,have been evaluated and compared in terms ofPAPR and multi-user diversity gain.

The PAPR analysis has shown a dependenceon the number of subcarriers. In particular, inorder to preserve a low PAPR, bandwidth con-figurations with a low number of subcarriers arepreferable. Furthermore, we found out that onlylocalized allocation of RBs shall be consideredin order to keep the PAPR benefits of the SC-FDM signal. However, this leads to a constrainton the flexibility of resource allocation in amulti-user scenario. As a consequence, we have

proposed a new channel-aware scheduling algo-rithm for SC-FDMA, which combines subopti-mal performance with low implementation andcomputational complexity. However, note thatthe focus of this article was not to compare sev-eral scheduling algorithms for SC-FDMA, butjust to propose and utilize a solution that wouldcope with the aforementioned issue.

The cell spectral efficiency performance hasbeen evaluated for a local area scenario, withlow user mobility (3 km/h) and a PF metric forscheduling. Despite its higher transmitted powerdue to lower power backoff, SC-FDMA per-forms slightly worse than OFDMA for the casewith a low number of frequency multiplexedusers. In a partial loaded scenario, OFDMAalways performs slightly better than SC-FDMAeven for a high number of users. Furthermore,no significant differences have been observed fordifferent bandwidth configurations or differentRB sizes. This is a consequence of the selectedscenario in which the wide coherence bandwidthreduces the sensitivity to different configurationparameters. To sum up, the following main con-clusions can be derived:

Bandwidth configurations with a low num-ber of subcarriers (e.g., 750/1024) can be consid-ered in a local area IMT-A scenario if nobackward compatibility issues are required.

Their short CP length still allows ISI and ICI to

be avoided due to the low delay spread of theindoor office channel. Furthermore, using a lownumber of subcarriers is advantageous in termsof PAPR. Finally, the granularity reduction dueto the larger subcarrier spacing does not lead to worse performance in terms of aggregatethroughput in a multi-user scenario with a PFscheduling metric.

The sensitivity of the aggregate throughputto RB size in the range of 1545 subcarriers is

negligible, especially for few users. Consideringthe benefits in terms of signaling overheadreduction of a large resource unit, an RB lengthof 45 subcarriers could therefore be consideredwithout degrading the performance up to 10users.

The cell throughput difference betweenOFDMA and SC-FDMA is related to the trafficload in the system. In a fully loaded scenarioOFDMA performs slightly better than SC-FDMA for a low number of frequency- multi-plexed users and slightly worse for a largenumber of multiplexed users. Reducing the traf-fic load in the cell, and therefore the bandwidthusage, OFDMA tends to slightly outperform SC-FDM regardless of the number of users, evenconsidering the increased power backoff penaltyof OFDMA. Future work will be focused on thedevelopment of smarter scheduling algorithmsfor SC-FDMA, allowing its performance to beleveraged without dramatically increasing thecomputational complexity, as well as an evalua-tion of the behavior of both access schemes in amulti-cell scenario, where issues related to inter-cell interference and intercell synchronizationlosses will be investigated.

ACKNOWLEDGMENTSThis work has been supported by Nokia Siemens

Network (NSN).

I Figure 4.Performance comparison between OFDMA and SC-FDMA for atwo-receiving-antenna system.

Number of users

100 MHz, SIMO, A1NLOS, 750/1024

2

2

4

Spectralefficiency

(b/s/Hz)

6

8

10

12

4 5 6 8 10

2% BW

75 % BW

12 14 16 18 20

OFDMA PFSC-FDMA PF

100% BW




REFERENCES[1] F. Ivanek, Convergence and Competition on the Way

Toward 4G: Where Are We Going?, IEEE Radio andWireless Symp., Long Beach, CA, Jan. 2007, pp. 26558.

[2] 3GPP TS 36.201V 8.1.0, LTE Physical Layer GeneralDescription (Release 8), Nov. 2007.

[3] IEEE . 802.16e-2005, Standard for Local and MetropolitanArea Networks. Part 16: Air Interface for Fixed and MobileBroadband Wireless Access Systems, Feb. 2006.

[4] L. Hanzo et al., Eds., OFDM and MC-CDMA for Broad-band Multi-User Communications, WLANs and Broad-casting, Wiley/IEEE Press, 2003.

[5] S. L. Falconer et al., Frequency Domain Equalization

for Single-Carrier Broadband Wireless Systems, IEEECommun. Mag., vol. 40, no. 4, Apr. 2002, pp. 5866.

[6] Myung, Lim, and Goodmand, Single Carrier FDMA forUplink Wireless Transmission, IEEE Vehic. Tech. Mag.,vol. 1, Sept. 2006, pp. 30-38.

[7] 3GPP TS 36.211V 8.1.0, Physical Channels and Modu-lation (Release 8), Nov. 2007.

[8] J. W. Cooley and J. W. Tukey, An Algorithm for theMachine Calculation of Complex Fourier Series, Math.Comp., vol. 19, 1965, pp. 297-301.

[9] J. G. Proakis, Digital Communications, 4th ed., McGrawHill, 2001.

[10] Winner II Deliverable 1.1.2 V 1.0 IST-4-027756, Win-ner II Channel Models, Sept. 2007; https://www.ist-winner.org

[11] 3GPP TSG RAN WG1 Mtg. 44 R1-060540, Performanceof Localized and Distributed SC-FDMA, Feb. 2006.

[12] W. Wang et al., Impact of Multi-User Diversity and

Channel Variability on Adaptive OFDM, IEEE Proc. 58thVTC., vol. 1, Oct. 2003, pp. 54751.[13] F. Swarts et al., Eds., CDMA Techniques for Third Gen-

eration Mobile Systems, Springer, 1999.[14] K. Kim, Y. Han, and S. Kim, Joint Subcarrier and

Power Allocation in Uplink OFDMA Systems, IEEE Com-mun. Lett., vol. 9, June 2005, pp. 52628.

[15] J. Lim et al., Channel-Dependent Scheduling of UplinkSingle Carrier FDMA Systems, IEEE Proc. VTC, Sept.2006, pp. 15.

[16] C. Wengerter, J. Ohlhorst, and A. G. E. von Elbwart,Fairness and Throughput Analysis for Generalized Pro-portional Fair Frequency Scheduling in OFDMA, IEEEProc. 61th VTC, vol. 3, June 2005, pp. 190307.

[17] 3GPP TSG RAN4 Mtg. 38 R4-060179, LTE Cubic Met-ric, Feb. 2006.

BIOGRAPHIES

GILBERTO BERARDINELLI ([email protected]) received his first andsecond level degrees in telecommunication engineering,cum laude, from the University of LAquila, Italy, in 2003and 2005, respectively. He also received a second levelmaster in techniques and economics of telecommunica-tions in 2006 from the University of Padova, Italy. In 2006he worked with the Radio Frequency Engineering Depart-ment in Vodafone NV, Padova, Italy, where he studied theissues related to coverage of HSDPA services, and alsoradio propagation in urban and suburban environments.Since 2007 he has been employed as a research assistantin the Radio Access Technology Section of Aalborg Univer-sity, Denmark, and successively appointed as a Ph.D. stu-dent in the same section. His research interests are mostlyfocused on physical layer design for 4G systems, thusincluding multiple access schemes, multi-antenna systems,channel sounding, and iterative turbo receivers.

LUIS NGEL MAESTRO RUIZ DE TEMIO (lames.aau.dk) graduat-ed in mobile communications from Miguel Hernndez Uni-versity, Elche, Spain, in July 2007. Since September 2007he has been working as a research assistant in the Elec-tronic Systems Department of Aalborg University, where hecollaborates with Nokia-Siemens Networks in the develop-ment of solutions for upcoming LTE-A systems. His maininterests are in OFDM, MIMO, and LTE-related physicallayer studies.

SIMONE FRATTASI ([email protected]) received his Ph.D. in elec-trical and electronic engineering from Aalborg Universityin 2007, and his M.Sc. degree cum laude and B.Sc.degree in telecommunications engineering from Tor Ver-gata University, Rome, Italy, in 2002 and 2001, respec-tively. From 2002 to 2005 he was employed as a researchassistant at Aalborg University, where he worked on twoEuropean projects (STRIKE and VeRT) and one industrialproject (JADE) in collaboration with the Global Standards& Research Team, Samsung Electronics Co. Ltd., Korea.Since 2005 he has been an assistant professor, where,besides still contributing to the JADE project, he has beenleading a Danish-funded project (COMET). In November2007 he joined the Radio Access Technology Section(RATE), where he is currently project leader of an industri-al project (LA-TDD) in collaboration with Nokia SiemensNetworks, Aalborg, Denmark. He is author/co-author ofaround 50 papers published in journals, magazines, andproceedings of international conferences; book chapters;encyclopedia papers; technical reports; and patent appli-cations. He has served as a reviewer for several journals,magazines and international conferences, and as a GuestEditor for Springer, Wireless Personal Communications

Journal; IEEE, Technology & Society Magazine; Wiley,Wireless Communications and Mobile Computing Journal;and Academy Publishers,Journal of Communications. Hewas the main instructor for two half-day tutorials onwireless location at IEEE PIMRC 07 and IADIS WAC07.He was General Chairman of the First International Work-shop on Cognitive Radio and Advanced Spectrum Man-agement (CogART 08) and the First InternationalSymposium on Applied Sciences on Biomedical and Com-

munication Technologies (ISABEL 08). In January 2008 hewas appointed Chairman of the IEEE GOLD AG Denmark.His research interests mainly include cooperation in wire-less networks, link layer techniques, wireless location,quality of service mechanisms, next-generation wirelessservices and architectures, user perspectives, and socio-logical dimensions related to the evolution of technologyand society.

MUHAMMAD IMADUR RAHMAN ([email protected]) He received hisPh.D., M.Sc., and B.Eng. degrees in 2007, 2003, and 2000respectively from Aalborg University, Helsinki University ofTechnology, Finland, and Multimedia University, Malaysia,respectively. Since January 2008 he has been working as aresearch engineer in the Access Technologies and SignalProcessing Department of Ericsson Research, Kista, Swe-den. He is primarily involved in 3GPP standardizationresearch on PHY and MAC layer issues. Prior to that he was

an assistant professor at Aalborg University in 2007, whenhe also managed a project, Local Area TDD Solutions forFuture Gigabps Wireless Systems, funded by Nokia SiemensNetwork and Danish Research Council. At Aalborg Universi-ty he taught a number of courses related to wireless com-munications, and supervised many M.Sc. and two Ph.D.students. He has published a number of papers in interna-tional journals and conferences. He is an active IEEE volun-teer, and has led many projects in student and recentgraduates activities for the IEEE Denmark section. Currentlyhe is involved in THE IEEE Sweden section. He was one ofthe initiators and first General Chair of the Aalborg Univer-sity IEEE Student Paper Conference in 2007. His mainresearch interests are radio access techniques, signal pro-cessing, radio resource management, and future cognitivenetworks.

PREBEN MOGENSEN ([email protected]) received M.Sc.E.E. and

Ph.D. degrees from Aalborg University, Denmark, in 1988and 1996, respectively. He is currently a professor at Aal-borg University, where he is head of the RATE section. Heis internationally recognized for his research in mobilecommunication systems and has more than 130 interna-tional publications; he has supervised 14 successfully com-pleted Ph.D. projects. Since 1995 he has also worked parttime with Nokia and later on with Nokia Siemens Net-works, working on LTE and IMT-Advanced related projects.

ofdma vs sc fdma performance comparison on local area imt-a scenarios

Documents