final-das-conference_edit_by_mbi.pdf

8/14/2019 Final-DAS-Conference_edit_by_MBI.pdf

1/5

Performance of Limited Feedback MU-MIMO inDistributed Antenna Systems with Different

Scheduler

Mehran Behjati, Mahamod IsmailDept. of Electrical, Electronic and System Engineering

Universiti Kebangsaan Malaysia43600 UKM Bangi, Selangor Malaysia{mbehjati, mahamod}@eng.ukm.my

Muhammad@Yusoff IbrahimFaculty of Electrical Engineering

Universiti Teknologi MARA40450 Shah Alam, Selangor Malaysia

[email protected]

Abstract Distributed antenna system (DAS) acts as an

effective solution to mitigating interference and path-loss by

decreasing the access distance for users, and increases the cell

coverage and system capacity as well. This study evaluatesperformance of scheduling methods on the DAS where practical

system constraints are considered, such as path-loss, out-of-cell

interference and limited feedback. Zero-forcing multiuser MIMO

precoding is utilized as downlink transmission strategy, when

imperfect channel state information is available at the

transmitter. System performance is evaluated by empirical

cumulative density functions of the cell throughput, where a long-

term evolution advanced (LTE-A) standard compliant simulator

is utilized for simulation. It is demonstrated that by spreading the

transmit antennas throughout the cell more cell throughput is

achievable. Moreover, by utilizing an appropriate scheduling

algorithm, more potential of DAS can be extracted and leads to

substantial cell throughput.

Keywordsdistributed antenna system; limited feedback; zero-

forcing precoding; scheduling

I. INTRODUCTIONIn cellular networks, interference is known as a major

destructive factor to provide uniform high data-rate throughoutthe coverage area. A partial solution to this problem is reducingthe overall transmit power by utilizing distributed antennasystem (DAS). Moreover, DAS enhances the coverage area,capacity, and throughput, especially in shadowed and blindenvironments [1], [2]. To do so, remote radio units (RRUs) aredistributed over the cell area and are connected to the central

base station (BS) via low-latency and high-bandwidth

dedicated connections acting as distributed antenna arrays(DAAs).

DAS reduces the access distance for users, wherebyreduces the required uplink transmission power and achievessignificant transmission power gain, moreover increases thesum-rate capacity versus conventional cellular system [3], [1].Results in [2] show that DAS reduces inter-cell interferenceand significantly improves capacity, specially for users arelocated close to the cell edges. By applying multiuser MIMO(MU-MIMO) transmission to the DAS, more spatial multiuser

diversity can be exploited. Authors of [4] compared thethroughput of MU-MIMO zero-forcing (ZF) beamforming withand without DASs, and demonstrated advantages of MU-MIMO in DAS, and expressed that by utilizing full MU-MIMO to all RRUs best performance in term of area spectralefficiency is achievable.

Accordingly, this study applies full MU-MIMOtransmission at the central BS and each RRU, as well,according to the LTE-A specifications [5]. Furthermore,system performance evaluated under practical systemlimitations, such as limited feedback and out-of-cellinterference. Recently performance of DASs under perfect andimperfect CSI with different quantization methods investigatedin [6]. Authors of [3] showed that DAS can properly mitigateinterference, if multiple users scheduled simultaneously.Furthermore, authors of [7] shows that with scheduling

multiple best users simultaneously, the system capacitysurpasses over scheduling the best user, because inter-userinterference mitigated effectively and the spatial degree offreedom can be fully exploited. The results in [7] are onlyconsidered for three single antenna RRUs. Therefore, this studyinvestigates the cell throughput in terms of empiricalcumulative density functions (ECDF) when differentscheduling methods and different configuration of DAS areapplied to the MU-MIMO system.

In the case of frequency division duplex (FDD) system,channel state information (CSI) should be fed back via userequipments (UEs) to the BS, in order to compute the precoderfor inter-user interference cancelation and schedule users and

set a suitable modulation and coding scheme (MCS). To saveuplink resource, at the receiver side, the achieved CSI isquantized by random channel direction quantization (RCDQ)method [8] and conveyed to the BS by limited bits. Afterwards,at the transmitter side, ZF beamforming [9] as a promisingtransmission strategy is utilized to exploit multiplexing gainand mitigate interference. The simulation results demonstratethat, by utilizing proper scheduler method the potential of DAScan be extracted and with increasing the number of RRUs andtransmit antenna per RRU more cell throughput can beachieved.


2/5

Fig. 1. MU-MIMO DAS architecture

The rest of the paper is organized as follows. Section IIoutlines the system model of limited feedback MU-MIMO,which contains downlink transmission, CSI feedback,

precoding, and scheduling. Simulation methodology isillustrated in Section III. Section IV presents the simulationresults. Finally Section V summarizes and concludes the paper.

II. LIMITED FEEDBACK MU-MIMOSYSTEM MODELA.MU-MIMO Downlink Channel

Figure 1 depicts the architecture of considered MU-MIMODAS with I cells which each cell i = {1, , I} contains onecentral BS with Mc,i transmit antennas and Ri RRUs whicheach of them equipped with Mr,i, where r{1 , , Ri}. The totalnumber of users which served in cell i is Ki and each userk{1,,Ki}is equipped with Nk,ireceive antennas. The input-output relationship on kth user in cell iis given by

yk,i= Hk,ixk,i+ Hk,i xu,iKu=1,uk + Hk,i(j)XjIj=0,ji + nk,i (1)

where, , is the perturbed received signal vector to user incell , ,(), is the channel matrix between user incell and all transmit antennas of cell (=,+ ,=1 ), for simplicity whenever =, the superscriptomitted, ,=,(). ,~(, 2. ) is additive whiteGaussian noise (AWGN) with variance 2, and,~( , )is the scheduled and precoded symbol vectoras follows

xk,i= fk,isk,i (2)

Xj= fk,jsk,j= FjSjk

,

jI\

i

where {1, , K i} is the set of selected users that areserved in parallel over a given time-frequency resource in celli, sk,i (O, PTx D ) is the selected users symbols whichindependently generated by channel encoders with statistical

power sk,i2= 1 , and fk,i is the precoder vector whichmaps the transmit symbol vector onto the Mitransmit antennas,and allocates the available transmit power P among users.

In the receiver side users apply a linear receiver filters,wk,i N,1, to quantize their respective channels.Consequently the symbol of kth user can be estimated byapplying w

kto the perturbed receive vector y

k,

i, as

sk,i= wk,iHyk,i= wk,iHHk,ifk,isk,i+ wk,iHHk,i fk,isk,iKu=1,uk +wk,iH Hk,i(j)FjSjIj=0,ji + nk, (3)

B. Channel State Information FeedbackIn the MU-MIMO FDD systems, transmitter requires CSI,

to compute the precoding matrix, schedule users and select anappropriate MCS for downlink transmission. As channelmatrix contain multi-dimensional variable, UEs should toexploit beneficial characteristics of their channels and quantize

them in order to reduce the feedback overhead. In cellularnetworks, CSI is divided into two categories:

1) Channel quality indicator (CQI): There is an unusualproblem to calculate the CQI in the receiver side, whencalculation of CQI depends to the scheduler ahead of

transmission, and scheduler computation depends to the CQIfeedback. Therefore, to solve this problem, scheduler canmake decision based on the expected SINR value (not on theexact SINR value). As a preliminary solution, it can beassumed that there is no quantization error in the CDIfeedback, therefore

SINRk,ipk,i||hk,i||2cos2k,i=k,i(1) (4)

As a better approximation, authors of [10] proposed a lower-

bound of SINR as CQI, as

SINRk,iPM|h,|

cos,1+PM|h,|

sin,=k,i(2) (5)

where ,is the angle between effective channel direction andquantized channel direction. Therefore CQI feedback provided

by

CQIk=(k) (6)where

is quantization function.

To select an appropriate MCS, a look-up table is used asdefined in [11]. Moreover, to decrease the quantization error,and reduce the feedback overhead, as well, a combination ofquantized CQI feedback with hybrid automatic repeat request(HARQ) protocol [12] is utilized at the simulator.

2) Channel direction indicator (CDI): The effectivechannel vector,which is a concatenated of channel matrixand the receive filter (,=,,) contains essentialchannel statistics for interference mitigation. Therefore, userscomputes their effective channel directions as,=


3/5

TABLE I. SIMULATION PARAMETERS

Parameters Value

Terminal power 43 dBmNoise power spectral density -174 dBm/HzShadowing standard deviation 8 dBCarrier frequency 2.1 GHzChannel bandwidth 1.4 MHzSubcarrier bandwidth 15 kHzChannel model Correlated Flat RayleighNumber of transmit antennas 8Number of UEs receive antennas 1Cell radius 500 mUE speed 3 Km/h

Available MCSs

QPSK (code rate=0.076, 0.117,0.188, 0.301, 0.439, 0.588)16QAM (code rate=0.369, 0.479,0.602)64QAM (code rate=0.455, 0.553,0.650, 0.754, 0.853, 0.926)

equalizer filter MMSE

, ||,|| and in order to save the uplink bandwidth,quantize the, to a unit norm vector with utilizingquantization codebook. The simplest way to design acodebook is randomly generate the2quantization codewordsfrom an isotropic distribution on the M-dimensional unitsphere, which results to a lower complexity. Therefore, theCDI is given by

CDI: hk=hk (7)where is the effective channel quantization function. In thisstudy, the RCDQ method defined in [8] is used for quantization

purpose.

C.Zero Forcing PrecodingIn this study ZF beamforming is utilized as multi-user

transmission strategy, when BS selects a large number of users( ) and sends one data stream to each of them.Precoding matrix is computed based on the quantized effectivechannel direction , to exploit multiplexing gain and mitigateinterference. Therefore, precoding vector

,

should be

selected in a way to be orthogonal to the quantized effectivechannel vectors of other users , (\ {}), thus ,,=0. Therefore, the ZF beamforming matrix can be computed as

F = H jH(H jH jH)1diag(pj)1 2 (8)

D. SchedulingCQI feedback provides information such as channel

amplitude and quantization error for transmitter, whereby BSselects a subset of users and assigns the available resource tothem for downlink transmission. In ZF MU-MIMO precodingtransmission when the total number of receive antennas

exceeds the total number of transmit antennas, a schedulingalgorithm is vital to select a subset of orthogonal users to servein parallel over a given time-frequency resource.

Authors of [10] proposed a greedy user selection algorithm,where transmitter uses the available SINR value and searchesthrough the unscheduled users to select users which canmaximize sum-rate, where sum-rate defined as

()= log2(1 +k)k (9)

Therefore, set of selected users is according to

= arg maxK TkK (10)

where k is the current achievable sum-rate and Tk is theaverage past sum-rate of this user. (for original definition seealgorithm 2 in [10]).

As an alternative method, per user unitary and rate control(PU2RC) scheme has been proposed to 3GPP [13], where userscheduling and beamforming are jointly and practicallydesigned. PU2RC supports limited feedback MU-MIMOsystem and it capable to exploit multiuser diversity. [14]showed that in presence of large enough number of users,

PU2RC outperforms the zero-forcing beamforming scheme,and it is more robust against CSI quantization errors.

III. METHODOLOGYIn this study, to simulate the results the Vienna LTE-A

link-level simulator [15] is used which is compliant to theLTE-A specifications. In order to simulate the results of DAS,all users are randomly distributed in one central cell which issurrounded with two interfering base stations tiers. Central cellis equipped with one central BS and some uniformlydistributed antenna arrays which located equiangularly on aring with radius of (2 3 ) . The simulation

parameters are listed in Table I.

System-level simulation of MU-MIMO requires detailedknowledge of physical layer and results to massivecomputational complexity. Therefore, Vienna simulator usesthe hybrid link/system level simulations to consider the

physical details of one cell and out-of-cell interference as well(see [6] for out-of-cell interference model). The system

performance evaluated under limited feedback methods whichdescribed in Section II, where eight bits assigned for CSIfeedback. The performance of system evaluated under differentscheduling method as illustrated in table II. Finally, simulationresults are presented as empirical CDFs of the average cellthroughput.

IV. RESULTS AND DISCUSSIONThe simulation results obtained for DASs with different

DAS configurations and different scheduling methods. DASconfigurations are denoted as Mc,i Ri/Mr,i. To evaluate theperformance of scheduling methods in DASs, simulations areperformed in two scenarios: with and without DAAs, meansthat Ri{2,4,6}, and Ri= 0, respectively. Figure 2 presents theempirical CDFs of cell throughput under considerations. Theresults of PU2RC scheduler are showed in Figure 2 (a), andillustrate that by increasing the number of RRU, the cellthroughput increases as well. Figure 2 (b) presents the results


4/5

(a)

(b)

(c)

1 1.5 2 2.5 3 3.50

0.2

0.4

0.6

0.8

1

Cell throughput [Mbit/s]

ECDFofcellthroughput

8-0/0

4-2/2

4-4/1

2-6/1

0 2 4 6 8 100

0.2

0.4

0.6

0.8

1



8-0/0

4-2/2

4-4/1

2-6/1

0 2 4 6 8 10 12 140

0.2

0.4

0.6

0.8

1



8-0/0

4-2/2

4-4/1

2-6/1

TABLE II. SCHEDULING METHODS

Model NotesPU2RC Proposed by [13] to the 3GPP-LTE standards

ZF-MUMIMO Proposed by [10],k,i(1)placements in (9)ZF-MUMIMO-flat Proposed by [10],k,i(2)placements in (9)

Fig. 3. Comparision of performance of schedulers in different DASconfigurations, in term of 0.95 of ECDF

0

24

6

8

10

12

14

8-0/0 4-2/2 4-4/1 2-6/1Cellthrou

ghputwith0.9

5ECDF

DAS configuration

PU2RC

ZF-MUMIMO

ZF-MUMIMO-flat

of ZF-MUMIMO scheduler, and shows when DAS utilized, thecell throughput considerably improved. Moreover, result of 4-2/2 configuration shows that by equipping RRU with multipletransmit antennas more throughput is achievable, wheremultiplexing gain of MU-MIMO can be exploited. Results of

Figure 2(c) belong to ZF-MUMIMO-flat scheduler and reveal aremarkable dependency to the DASs. The performance ofschedulers under different DAS configurations is compared in

Figure 3. As results show, when all transmit antennas areallocated on the central BS (8-0/0 configuration), the lowestcell throughput is achieved. It that case, ZF-MUMIMO-flat

provides better performance, because more accurate andsufficient knowledge of SINR fed back to the BS compared toZF-MIMO scheme, and under small number of users,

performance of PU2RC is not sufficient (see [14]). PU2RCperforms a weak dependency to the DAS configuration, wherecell throughput slowly improves with RRU increment. While,ZF-MUMIMO-flat scheduler goodly extracts the potential ofDASs, where for 2-6/1 configuration approximately achievesfour folds more cell throughput compared to 8-0/0configuration.

V. CONCLUSIONThis study investigates the performance of different

scheduling methods in distributed antenna systems with limitedfeedback zero-forcing multiuser MIMO transmission.Simulation results show that by spreading the transmit antennasthroughout the cell more cell throughput is achievable,moreover by assigning more transmit antenna per RRU, moremultiplexing gain can be extracted. Furthermore, it isdemonstrated that PU2RC scheme as a well-known precodingand scheduling method has a poor dependency to the DASs.The comparison of results shows that by utilizing anappropriate user selection strategy substantial cell throughputcan be achieved.

ACKNOWLEDGMENT

The study was funded by Universiti Kebangsaan Malaysia andand Universiti Teknologi MARA under grants number DPP-2013-006 and 600-RMI/Dana 5/3/RIF(285/2012) respectively.

REFERENCES[1] B. Song, R. L. Cruz, and B. D. Rao, "Downlink optimization of indoor

wireless networks using multiple antenna systems," in INFOCOM 2004.Twenty-third AnnualJoint Conference of the IEEE Computer andCommunications Societies, 2004, pp. 2778-2789.

Fig. 2. Cell throughput performance in therm of ECDF, with differentDAS configuration and scheduling methods: (a) PU2RC, (b) ZF-MUMIMO, and (c) ZF-MUMIMO-flat


5/5

[2] W. Choi and J. G. Andrews, "Downlink performance and capacity ofdistributed antenna systems in a multicell environment," WirelessCommunications, IEEE Transactions on, vol. 6, pp. 69-73, 2007.

[3] L. Dai, "Distributed antenna system: Performance analysis in multi-userscenario," inInformation Sciences and Systems, 2008. CISS 2008. 42ndAnnual Conference on, 2008, pp. 85-89.

[4] R. Heath, T. Wu, Y. H. Kwon, and A. C. Soong, "Multiuser MIMO indistributed antenna systems," in Signals, Systems and Computers(ASILOMAR), 2010 Conference Record of the Forty Fourth Asilomar

Conference on, 2010, pp. 1202-1206.[5] "3GPP, Techn. Spec. Group Radio Access Network; E-UTRA; Phy.Chan. and Mod," (Release 10) 2010.

[6] S. Schwarz, R. W. Heath, and M. Rupp, "Multiuser MIMO in distributedantenna systems with limited feedback," in Globecom Workshops (GCWkshps), 2012 IEEE, 2012, pp. 546-551.

[7] X. Li, M. Luo, M. Zhao, L. Huang, and Y. Yao, "Downlink performanceand capacity of distributed antenna system in multi-user scenario," inWireless Communications, Networking and Mobile Computing, 2009.WiCom'09. 5th International Conference on, 2009, pp. 1-4.

[8] S. Schwarz, R. W. Heath, and M. Rupp, "Single-user MIMO versusmulti-user MIMO in distributed antenna systems with limited feedback,"EURASIP Journal on Advances in Signal Processing, vol. 2013, p. 54,2013.

[9] N. Jindal, "MIMO broadcast channels with finite-rate feedback,"Information Theory, IEEE Transactions on, vol. 52, pp. 5045-5060,2006.

[10] M. Trivellato, F. Boccardi, and F. Tosato, "User selection schemes forMIMO broadcast channels with limited feedback," in VehicularTechnology Conference, 2007. VTC2007-Spring. IEEE 65th, 2007, pp.2089-2093.

[11] J. C. Ikuno, M. Wrulich, and M. Rupp, "System level simulation of LTEnetworks," in Vehicular Technology Conference (VTC 2010-Spring),

2010 IEEE 71st, 2010, pp. 1-5.[12] J. C. Ikuno, M. Wrulich, and M. Rupp, "Performance and modeling ofLTE H-ARQ," in Proc. International ITG Workshop on Smart Antennas(WSA 2009), Berlin, Germany, 2009.

[13] Downlink MIMO for EUTRA, 3GPP TSG RAN WG1 44/R1-060335,Feb. 2006.

[14] K. Huang, J. G. Andrews, and R. W. Heath, "Performance of orthogonalbeamforming for SDMA with limited feedback," Vehicular Technology,IEEE Transactions on, vol. 58, pp. 152-164, 2009.

[15] C. Mehlfhrer, J. C. C. Ikuno, M. imko, S. Schwarz, M. Wrulich, andM. Rupp, "The Vienna LTE simulators-Enabling reproducibility inwireless communications research," EURASIP Journal on Advances inSignal Processing, vol. 2011, pp. 1-14, 2011.

final-das-conference_edit_by_mbi.pdf

Documents