main Fan et al. 2011
description
Transcript of main Fan et al. 2011
-
MCS Selection for Throughput Improvement inDownlink LTE Systems
Jiancun Fan, Qinye Yin, Geoffrey Ye Li, Bingguang Peng, and Xiaolong Zhu MOE Key Lab for INNS, School of Electronics and Information Engineering, Xian Jiaotong University, Xian, China
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA Huawei Shanghai Research Institute, Shanghai, China
AbstractIn this paper, we investigate resource block (RB)assignment and modulation-and-coding scheme (MCS) selectionto maximize downlink throughput of long-term evolution (LTE)systems, where all RBs assigned to the same user in anygiven transmission time interval (TTI) must use the same MCS.We develop several effective MCS selection schemes by usingthe effective packet-level SINR based on exponential effectiveSINR mapping (EESM), arithmetic mean, geometric mean, andharmonic mean. From both analysis and simulation results, weshow that the system throughput of all the proposed schemesare better than that of the scheme in [7]. Furthermore, theMCS selection scheme using harmonic mean based effectivepacket-level SINR almost reaches the optimal performance andsignificantly outperforms the other proposed schemes.
I. INTRODUCTIONKey features of wireless communication channels are with
frequency-selective fading and time varying. In order to matchthese features, adaptive modulation and coding (AMC) hasbeen proposed to enhance system throughput in [1]-[3]. Itsbasic idea is to select an optimal combination of modulationand coding scheme (MCS) according to channel quality sothat high rate data can be transmitted when channels are ingood condition and low rate data is transmitted to guaranteereliability when channel quality is poor. AMC has beenwidely adopted for many wireless communication systems orstandards, such as 3GPP long-term evolution (LTE) [4], IEEE802.16 WiMAX [5], and IEEE 802.11n [6]. In this paper, wewill focus on MCS selection in the downlink transmission ofLTE systems.
In the downlink transmission of LTE systems, only channelquality indicator (CQI) corresponding to a resource block(RB) or multiple RBs in the form of MCS index is fed backto the base station (BS). RB in LTE systems is the smallestresource unit and consists of 12 adjacent sub-carriers and 6 or7 consecutive OFDM symbols [4]. Based on the MCS indexfed back from mobile terminals, BS will allocate RBs to usersand select an appropriate MCS. Different from the generalAMC schemes [1]-[2], LTE specification requires that all RBsallocated to the same user in any given transmission time
This work was supported in part by the Research Gift from HuaweiTechnologies Co., the National Natural Science Foundation of China undergrant No. 60971113 & 61071125 & 61071216, the Research Foundationfor Doctoral Program of Higher Education of China under Grant No.200806980020, and the Science Foundation for Innovation Research Groupof China under Grant No. 60921003.
interval (TTl) must use the same MCS. A key issue is whichMCS should be selected for each user. In order to guaranteeblock-error rate (BLER) performance of the RB with the poorchannel condition, it has been proposed in [7] to select a MCSfor the RBs belonging to the same user in any TTI accordingto the poor channel quality. The scheme in [7] is furtherextended to proportional fair multiuser scheduling in [8]. Suchschemes are too conservative and result in a throughput lossfor those RBs with better channel condition. However, if tooaggressive, the BLER is too large, which may not have a highthroughput either.
In this paper, we will investigate MCS selection to maxi-mize the system throughput. We will develop effective MCSselection schemes that not only ensure the BLER performanceof the RB with poor channel condition but also make full useof the RBs with good channel condition.
The rest of this paper is organized as follows. SectionII introduces the MCS feedback model and presents theoptimization objective function for the downlink transmissionof LTE systems. In Section III, suboptimal RB assignment andMCS selection are proposed. Simulation results are presentedin Section IV to demonstrate performance improvement of theproposed schemes and conclusions are drawn in Section V.
II. SYSTEM MODEL
In this section, we introduce the MCS feedback model inLTE dowlink and then formulate the optimization problem forRB assignment and MCS selection.
A. MCS Feedback
In LTE downlink, CQI in form of MCS index is veryimportant feedback information, which can be used to assignRBs and determine modulation order and coding rate [4].Each RB in LTE systems includes multiple subcarriers andtheir channel gains are different. To determine MCS index,an effective block-level signal-to-interference-plus-noise ratio(SINR) mapping method has been proposed in [9]-[13] tocollect all channel information and select an appropriate MCSfor each RB, which is also called exponential effective SINRmapping (EESM). It has been shown that the effective block-
978-1-4577-0638-7 /11/$26.00 2011 IEEE
AbdullahRectangle
AbdullahHighlight
AbdullahHighlight
-
TABLE IMCS FEEDBACK TABLE IN LTE DOWNLINK
CQI Modulation Code Rate Rate r [14] m SINR threshold Index Order 1024 (bits/symbol) [12] w/ 10% BLER (dB)
0 out of range
1 QPSK 78 0.1523 1.00 -9.4782 QPSK 120 0.2344 1.40 -6.6583 QPSK 193 0.3770 1.40 -4.0984 QPSK 308 0.6010 1.48 -1.7985 QPSK 449 0.8770 1.50 0.3996 QPSK 602 1.1758 1.62 2.4247 16QAM 378 1.4766 3.10 4.4898 16QAM 490 1.9141 4.32 6.3679 16QAM 616 2.4063 5.37 8.456
10 64QAM 466 2.7305 7.71 10.26611 64QAM 567 3.3223 15.5 12.21812 64QAM 666 3.9023 19.6 14.12213 64QAM 772 4.5234 24.7 15.84914 64QAM 873 5.1152 27.6 17.78615 64QAM 948 5.5547 28.0 19.809
level for a given MCS can be defined as
(m) = m ln(
1Nsb
iNsb
exp( i
m
)), (1)
where Nsb is the set of subcarriers in a RB, Nsb = |Nsb| isthe number of subcarriers in the RB, i is the SINR at the ithsubcarrier, and m, is an adjusting factor and depends on aspecific MCS. For LTE systems, the relationship between theMCS index, m, and the adjusting factor, m can be found inTable I. In fact, the effective block-level SINR changes slightlywith the adjusting factor and with the MCS index since itessentially reflects the average of SINRs on all subcarriersfor each RB no matter which adjusting factor is used.
Let switching threshold, m, be the SINR with 10% BLERfor the MCS m. For a given set of SINRs corresponding tosubcarriers of a RB, {i}iNsb , the MCS index can be obtainedby
m = max{i| i (i), for i = 1, ...,M. }, (2)where M is the total number of MCSs and is 15 in the currentLTE specification. The MCS index of each RB for each user isthen fed back to the BS, where resource allocation, includingRB assignment and MCS selection, is performed.
Once the MCS index information is obtained at the BS byfeedback, the effective SINR corresponding to each RB ofeach user can be well approximated by
m + , for m = 1, ...,M, (3)
where is a positive offset value over the switchingSINR, m, and is determined by the two adjacent switchingthresholds. Based on such approximate block-level SINR, wewill perform RB assignment and MCS selection to maximizesystem throughput.
B. Problem FormulationWe focus on the downlink transmission of LTE systems with
Nrb RBs and K active users, and the CQI information in theform of MCS index for all RBs of each user is assumedto accurately feed back to the BS without delay. With theMCS index, the corresponding effective block-level SINR andtransmission rate will be well estimated. We assume that NkRBs are allocated to user k in a given TTI and all transmitteddata blocks of each user in the given TTI are combined intoa data packet. Then, the packet throughput corresponding touser k can be expressed as
Tmk(k,mk) = Rmk(1 Pmk(k,mk)), (4)where mk is the MCS index of user k, Rmk is the transmissionrate determined by the modulation order and code rate and canbe found in Table I for LTE systems, and Pmk(k,mk) is thepacket-error rate of the data packet corresponding to user k,which can be obtained according to [11], [12]. In addition,k,mk is an effective EESM based packet-level SINR for userk and is determined by the SINRs at all subcarriers of thepacket and the MCS index, which can be expressed as
k,mk = mk ln(
1NkNsb
nNk
iNsb
exp(k,n,i
mk
)),
(5)where Nk is the set of RBs allocated to user k and k,n,i isthe SINR at the ith subcarrier of RB n for user k.
With the SINRs on the subcarriers of all users, we canassign RBs to users, obtain the corresponding packet-levelSINR of each user, and select the MCS. To achieve the highestsystem throughput, the BS will optimize the RB assignmentand select the most appropriate MCS for each user basedon the effective packet-level SINR. Mathematically, we canformulate the joint optimization problem of RB assignmentand MCS selection as
argmax{Nk, mk}
Kk=1
Tmk(k,mk), (6a)
subject toNi Nj = , for i = j, (6b)N1 N2 NK = Nrb, (6c)mk = 0, 1, ...,M, (6d)
where Nrb is the set of all RBs. The constraint in (6b)guarantees that each RB can be allocated to only one user,the constraint in (6c) ensures that all available RBs can beallocated to the users, and the constraint in (6d) means thatMCS of each user must be selected from the given MCS setin Table I.
AbdullahHighlight
AbdullahHighlight
AbdullahHighlight
AbdullahHighlight
AbdullahTextboxRmk = rate r
AbdullahHighlight
AbdullahRectangle
AbdullahRectangle
AbdullahHighlight
AbdullahHighlight
-
The above optimization problem uses the effective packet-level SINR for each user, which is determined by SINRs ofall subcarriers in the packet of each user and the MCS index.However, the only feedback information in LTE systems isthe MCS index for each RB of each user, which can be usedto evaluate the effective block-level SINR for each RB. Inthe following sections, we will investigate how to use theapproximate effective block-level SINRs belonging to eachuser to estimate the effective packet-level SINR for resourceallocation and develop the MCS selection schemes.
III. HEURISTIC SCHEMES
To simplify the optimization problem, we separate RBassignment and MCS selection. We first assign appropriateRBs to users and then select the MCS based on the effectivepacket-level SINR estimated by different ways.
A. Principle
We will first relax the MCS constraint for each user andthen develop a heuristic solution to maximize the systemthroughput. We know that Tmk in (4) always monotonicallyincreases with SINR. Therefore, if the MCS constraint foreach user is ignored, then each RB should be allocated tothe user with the highest SINR so that highest MCS can beused to maximize the system throughput, which can be directlyexpressed as
Nk = {n : n,k > n,l for k = l and 1 n Nrb}, (7)
where n,k is the effective block-level SINR of RB n for userk at the BS based on the MCS index feedback. If severalusers have the same channel quality or the same MCS indexat the same RB, then such a RB will be randomly assigned toarbitrary one of these users.
This RB assignment scheme is very simple since it onlyneeds to compare the SINR of all different users on each RB.However, the RB assignment scheme is suboptimal since theconstraint on MCS selection is not considered here. We willfurther optimize the system throughput by the MCS selectionin the following.
Once RB assignment is completed, we will find the best-estimated packet-level SINR to obtain the best MCS for eachuser. The optimization problem in this case can be expressedas
argmax0mkM
Tmk(k,mk). (8)
For such an optimization problem, we can obtain the optimalsolution of MCS selection by an exhaustive search algorithmif the accurate effective packet-level SINR is known. However,only MCS index of each RB is available at the BS in practicalLTE systems. Therefore, we will propose some heuristicschemes of MCS selection to deal with this issue.
B. Packet-Level SINR EstimationA very natural ideal is to use the EESM method to esti-
mate the effective packet-level SINR from the approximationeffective block-level SINRs for each user. In this case, theeffective packet-level SINR can be estimated as
EESMk,m = m ln(
1Nk
nNk
exp(n,k
m
)). (9)
If accurate block-level SINRs at all RBs belonging to user kare known and they all are calculated by the same adjustingfactor, m, then we have EESMk,m = EESMk,m . Otherwise, theyare different.
Besides EESM based estimation, we can exploit arithmeticmean (AM), geometric mean (GM), and harmonic mean (HM)of the approximate block-level SINRs on the RBs for eachuser to obtain estimated packet-level SINRs, which can beexpressed as
AMk =1
Nk
nNk
n,k, (10)
GMk =
( nNk
n,k
)1/Nk, (11)
and
HMk =Nk
nNk1
n,k
, (12)
respectively.With these estimation, the corresponding MCS for each user
to transmit the data packet can be easily determined by
mk =
0, k < 1, i = 0,i, i k < i+1, i = 1, ...,M 1,M, M k, i = M.
(13)
Compared with the exhaustive search scheme, they can signifi-cantly reduce the complexity. For different ways of estimatingthe effective packet-level SINR, the impact of each RB isdifferent. For the AM based scheme, the estimated effectivepacket-level SINR is mainly dominated by the RB with themaximum SINR. For the GM based scheme, it fairly considersthe impact of all the RBs belonging to the same user. Whilefor the HM based scheme, it is mainly decided by the RBswith the minimum SINR. The difference will result in differentthroughputs.
MCS selection could be also based on the minimum ormaximum effective SINR among the set of effective block-level SINRs belonging to the same user, which are calledminimum MCS selection scheme [7] and maximum MCSselection scheme, respectively. Obviously, the former one istoo conservative while the later one is too aggressive. Theywill result in a throughput loss.
AbdullahRectangle
AbdullahHighlight
AbdullahNoteHeuristic-refers to experience-based techniques for problem solving, learning, and discovery that find a solution which is not guaranteed to be optimal, but good enough for a given set of goals. Where the exhaustive search is impractical, heuristic methods are used to speed up the process of finding a satisfactory solution via mental shortcuts to ease the cognitive load of making a decision.
AbdullahLine
AbdullahHighlight
-
According to the relationship among AM, GM, and HM,we have the following inequalities
mink HMk GMk AMk maxk , (14)where mink and maxk is the minimum and maximum block-level SINR among all block-level SINRs belonging to userk. The selection schemes based on the effective packet-level SINR in (14) from left to right gradually change fromthe conservative to the aggressive. The conservative schemeensures the reliability at the cost of low transmission ratewhile the aggressive one improves transmission rate, however,results in a high packet-error rate. The system throughput isnot high in either case. In practice, the system packet-errorrate is mainly affected by the RBs with the poor channelcondition. The scheme based on HMk exactly considers thispoint so it can determine a more appropriate MCS and obtainhigher throughput compared with the other schemes besidesthe optimal one.
From a simple mathematical derivation, we also have
mink EESMk AMk maxk , (15)which means that the EESM based scheme is more con-servative than the AM based scheme. However, there is noanalytical relationship between the EESM, the GM, and theHM based the schemes which can be only compared bycomputer simulation in the next section.
C. ImplementationIn the above, we present RB assignment and MCS selection
schemes and compare the propose MCS selection schemeswith the existing MCS selection scheme in [7]. In this section,the implementation of the proposed schemes can be summa-rized into the following steps:
First assign RB to the user with the best channel qualityaccording to (7). If several users have the same channelquality at the same RB, the RB will be randomly assignedto arbitrary one of these users.
Then obtain the effective packet-level SINR for each useraccording to (9), (10), (11), or (12).
Finally determine the most appropriate MCS accordingto the SINR thresholds using (13).
IV. SIMULATION RESULTSIn this section, we compare performance of the proposed
schemes with the optimal scheme based on exhaustive searchand the existing scheme based on minimum MCS in [7] bycomputer simulation. For the scheme in [7], after RBs areassigned to all users, all RBs belonging to the same userselect the minimum MCS for transmission. For the optimalscheme, the effective packet-level SINR is first determined bythe EESM using SINRs of all subcarriers allocated to eachuser and then the MCS index is selected by exhaustive search.The optimal scheme can only be used as a performance boundand is impractical in LTE systems. Besides its high complexity,only the MCS indices of the RBs are available in LTEspecification and the SINRs of all subcarriers are unknown
TABLE IISIMULATION PARAMETERS
Carrier frequency: 2GHz Sampling frequency: 7.68MHzTTI duration: 1ms OFDM symbols per frame: 14FFT size: 512 Subcarrier spacing: 15kHzSystem bandwidth: 5MHz Antenna configure: 1 1
HARQ Type: Chase combiningMaximum retransmission number: 2
Channel model [15], [16] Type: ITU Ped-BSpeed: 3km/h
at the BS. In our simulations, the corresponding simulationparameters are listed in Table II and hybrid automatic repeatrequest (HARQ) is exploited. In addition, all simulation curvesare obtained by averaging over 1000 channel realizations.
Figures 1 and 2 demonstrate the average throughput andthe average MCS index versus the average SINR for differentschemes, respectively. From Figure 1, the proposed schemewith HM based MCS selection obtains almost the sameaverage MCS index as the optimal one while others are eithertoo conservative or too aggressive compared with the optimalone. Therefore, the proposed scheme with HM based MCSselection will obtain almost the same throughput as the optimalone and outperform the rest, which can be verified in Figure2.
In Figures 3 and 4, we give the average MCS index andthe average throughput versus the number of users for variousresource allocation schemes, respectively. From these figures,we find that the proposed scheme with HM based MCSselection also has almost the same average MCS as the optimalscheme and achieves the higher throughput compared withthe proposed schemes with AM, GM, and EESM based MCSselection as well as the existing scheme in [7]. Furthermore,we have also found that the throughput of the proposed schemewith HM based MCS selection is improved by 18% comparedthat in [7] when the number of users is 4 and SINR = 10 dB.
V. CONCLUSIONS
In this paper, we present novel adaptive block-level RBassignment and MCS selection schemes for the downlinktransmission of LTE systems. In these schemes, we firstassign RBs to the user with the best channel quality andthen select an appropriate MCS according to several effectivepacket-level SINR estimations. In particular, the scheme withHM based MCS selection reaches a good tradeoff betweendata transmission rate and packet-error rate. Consequently, itachieves almost the same performance as the optimal one andoutperforms the rest. The proposed schemes are very easy toimplement in the downlink transmission of LTE systems toimprove system throughput.
REFERENCES
[1] A. J. Goldsmith and S.-G. Chua, Adaptive coded modulation for fadingchannels, IEEE Trans. Commun., vol. 46, pp. 595-602, May 1998.
AbdullahRectangle
AbdullahHighlight
AbdullahHighlight
-
4 6 8 10 12 147
8
9
10
11
12
13
SNR(dB)
Ave
rage
MC
S In
dex
AMGMHMMinimumEESMOptimal
Fig. 1. Average MCS index versus SINR for the proposed schemes, theoptimal scheme, and an existing scheme with minimum MCS selection in[7].
4 6 8 10 12 141
1.5
2
2.5
3
3.5
4
SNR(dB)
Thro
ughp
ut (b
its/s
ymbo
l)
AMGMHMMinimumEESMOptimal
Fig. 2. Average throughput versus SINR for the proposed schemes, theoptimal scheme, and an existing scheme with minimum MCS selection in[7].
[2] D. Kim, B. C. Jung, H. Lee, D. K. Sung, and H. Yoon, Optimalmodulation and coding scheme selection in cellular networks with hybrid-ARQ error control, IEEE Trans. Wireless Commun., vol. 7, no. 12, pp:5195-5201, Dec. 2008.
[3] R. Fantacci, D. Marabissi, D. Tarchi, and I. Habib, Adaptive modulationand coding techniques for OFDMA systems, IEEE Trans. WirelessCommun., vol. 8, no. 9, pp: 4876-4883, Sep. 2009.
[4] S. Sesia, I. Toufik, and M. Baker, The UMTS Long Term Evolution: FromTheory to Practice. Wiley, 2009.
[5] J. G. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of WiMAX.Prentice Hall, 2007.
[6] IEEE 802.11n, Wireless lAN medium access control (MAC) and physicallayer (PHY) specifications: amendment 5 - enhancements for higherthroughput, Sep. 2007.
[7] R. Kwan, C. Leung, and J. Zhang, Resource allocation in an LTE cellularcommunication system, in Proc. IEEE Int. Conf. Commun., Dresden,German, pp. 1-5, June 2009
[8] R. Kwan, C. Leung, and J. Zhang, Proportional fair multiuser schedulingin LTE, IEEE Signal Processing Letters, vol. 16, no. 6, pp. 461-464, June2009
[9] Ericsson, System-level evaluation of OFDM - further considerations,
2 3 4 5 6 7 86
7
8
9
10
11
12
Number of users
Ave
rage
MC
S In
dex
AMGMHMMinimumEESMOptimal
Fig. 3. Average MCS index versus number of users for the proposed schemes,the optimal scheme, and an existing scheme with minimum MCS selection in[7].
2 3 4 5 6 7 81.5
2
2.5
3
3.5
Number of users
Thro
ughp
ut (b
its/s
ymbo
l)
SNR=10(dB)
AMGMHMMinimumEESMOptimal
Fig. 4. Average throughput versus number of users for the proposed schemes,the optimal scheme, and an existing scheme with minimum MCS selection in[7].
TSG-RAN WG1 #35.[10] Y. W. Blankenship, P. J. Sartori, B. K. Classon, V. Desai, and K. L.
Baum, Link error prediction methods for multicarrier systems, in Proc.IEEE Int. Vehicular Tech. Conf., Los Angeles, CA, vol.6, pp. 4175-4179,Sep. 2004.
[11] K. Sayana, J. Zhang, and K. Stewart, Link performance abstractionbased on mean mutual information per bit (MMIB) of the LLR channel,IEEE 802.16 Broadband Wireless Access Working Group.
[12] J. C. Ikuno, M. Wrulich, and M. Rupp, System level simulation of LTEnetworks, in Proc. IEEE Int. Vehicular Tech. Conf., Taipei, Taiwan, pp.1-5, May 2010.
[13] S. Kant and T. L. Jensen, Fast link adaptation for IEEE 802.11n,M.S. thesis, Institute of Electronic Systems, Aalborg University, Aalborg,Denmark, Aug. 2007.
[14] 3GPP TS 36.213 V9.0.0, E-UTRA Physical Layer Procedures, Dec.2009.
[15] Recommendation ITU-R M.1225, Guidelines for evaluation of radiotransmission technologies for IMT-2000, Int. Telecommun. Union, 1997.
[16] 3GPP TS 05.05 V8.15.0, Technical specification group GSM/EDGEradio access network; radio transmission and reception (release 1999),Aug. 2001.
/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 200 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 2.00333 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 400 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.00167 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False
/CreateJDFFile false /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles true /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /NA /PreserveEditing false /UntaggedCMYKHandling /UseDocumentProfile /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ]>> setdistillerparams> setpagedevice