Adaptive CQI adjustment with LTE higher-order sectorization

Transcription

1 13 8h Inernaional Conference on Communicaions and Neworking in China (CHINACOM) Adapive usmen wih LTE higher-order secorizaion Xinyu Gu 1, Wenyu Li 2, Lin Zhang 1 1 Beijing Universiy of Poss and Telecommunicaions 2 China Academy of Telecommunicaion Research of MIIT guxinyu.bj@gmail.com Absrac - To cope wih he rapidly increased demands for raffic volumes and end-user daaraes, deploying higher-order secorizaion is an effecive way especially in highly loaded scenario. However, for LTE, he mismach beween he inerference level of he cell-specific reference signal and daa is more severe wih higher-order secorizaion. This resuls in inaccurae channel qualiy esimaion and conservaive link adapaion. This paper proposes an adapive usmen algorihm o compensae he under-esimaed channel qualiy. Simulaion resuls clearly prove ha he proposed algorihm can effecively improve he link adapaion and consequenly improve end-user daaraes. Keywords- Higher-order secorizaion, Link Adapaion,,, iner-cell inerference I. INTRODUCTION The increased populariy of mobile broadband is leading o rapidly increasing raffic volumes. Mobile daa raffic forecass are presened e.g. in [1][2][3]. Yearly growh raes unil vary beween 55% [1][2] and 90% [3]. In parallel, expecaions on user daa-raes [4] are increasing, resuling from more and more capable devices and new applicaions. These demands require ha he neworks should be able o handle an increased raffic volume, e.g. he number of downor uploaded files per monh, and a he same ime provide increased daa raes, corresponding o he speed a which a file is down- or uploaded. There are differen ways o mee hese demands, for example by an improved or densified macro nework as well as adding smaller base saions. Higher-order secorizaion (HOS) is as well one of such ways o mee he increasing demands for raffic and daaraes. HOS refers o he pariioning of cell sies ino more han 3 secors so ha o have denser spaially reuse of he radio resource [5-7]. HOS has been successfully deployed in for example WCDMA (Wideband Code-Division Muliple Access) real nework [7]. In LTE (Long-Term Evoluion) nework, which has been launched by several major operaors in he world, HOS also is expeced o be an effecive way o increase boh sysem capaciy and end-user daaraes. However, o fully ge benefis from HOS, some problems, especially some LTE specific problem, need o be resolved. One of such problems is he even more serious mismach beween he esimaed channel qualiy based on Cell-specific Reference Signals () [5] and he acual channel qualiy ha daa experiences. This problem is no brough abou by higherorder secorizaion, bu in a sysem deploying higher-order secorizaion, his problem would become more serious. This mismach comes from he differen iner-cell inerference level experienced by and daa. Considering a ypical scenario ha he frequency reuse facor of 1, he inercell inerference experienced by daa usually highly depends on he raffic load in he surrounding cells. Wih low raffic loads, here is low probabiliy ha he users in he surrounding cells use he same frequency/ime resource for daa ransmission and consequenly he iner-cell inerference is expeced o be low. Wih high raffic loads, however, here is high probabiliy ha he users in surrounding cells are allocaed wih he same frequency/ime resource for daa ransmission and consequenly are inerfered by high iner-cell inerference. The iner-cell inerference experienced by, however, is usually no dependen on sysem raffic load because of he fixed paern and frequency/ime posiion defined in LTE. No maer how low he daa raffic load is in he sysem, is anyway ransmied in each cell wih he same srucure and using he same frequency/ime resource. Consequenly almos all he cells generae inerference o each oher for. Therefore, here is mismach beween he iner-cell inerference experienced by daa and especially wih low load. This mismach is more serious when HOS is deployed. On he one hand, when HOS is deployed, here are more cells in he same coverage area and consequenly more inerference is experienced by ; on he oher hand, wih HOS, assuming he same sie raffic load, per cell raffic load is lower han a normal 3-secor sysem and he even lower raffic load siuaion makes he mismach more eviden. is ypically used o esimae he Channel Qualiy Indicaor () for link adapaion, herefore, such mismach resuls in underesimaion of he daa channel qualiy and conservaive link adapaion. Conservaive link adapaion consequenly degrades he user daaraes as he channel qualiy is no sufficienly deployed. Even if wih frequency shifs [5], and if considering a ypical downlink 2x2 MIMO configuraion, here is only 3 differen ses of paerns (each se of paerns conains wo paerns corresponding o wo anenna pors). By proper cell planning, each cell could have wo neighboring cells use differen paerns which can miigae he inerference by some exen. However, his is evidenly no sufficien for a HOS sysem. Furhermore, HOS makes he assignmen of frequency shifs more complicaed. Ouerloop usmen based on Hybrid ARQ (Auomaic Repea reques) feedback [11] could as well by some exen cope wih imperfecion. However, ouer-loop usmen is slow and is less efficien when he raffic objec is small. Therefore, when he sysem load is low, he performance of IEEE

2 HOS could suffer from he mismach beween he esimaed and he acual daa channel qualiy and here is no effecive enough way o cope wih his issue. Alhough HOS is usually deployed in areas ha have high raffic demands, considering he dynamic characer of raffic, i is sill desirable o furher invesigae he problem and find proper soluions. In he remainder of he paper, Secion II describes HOS and underesimaion problem in more deails. An adapive usmen algorihm for HOS is proposed in Secion III, followed by sysem evaluaion resuls in Secion IV. Secion V concludes he paper. II. UNDERESTIMATION WITH HIERH ORDER SECTORIZATION HOS brings benefis from wo aspecs. Firsly, he anenna gain in he forward direcion of he HOS secor anenna is higher han he 3-secor anenna. This comes from he narrower half power beam-widh by assuming he same energy. Figure 1 shows he 3GPP SCM anenna paern [9] for 3-secor and 6- secor respecively. I is clear ha he 6-secor anenna paern is narrower corresponding o a narrower secor coverage area. A he same ime he anenna gain in he forward direcion is 3 db higher han a 3-secor anenna. This gives benefis o he served users especially for users in he cell edge. Secondly, HOS provides cell spliing gain or in oher words denser resource reuse. Assuming he same number of users or he same raffic load in a sie, he average number of users or average raffic load in each secor is lower wih HOS and he users can ge more chances o be scheduled or o ge more radio resource. However, denser resource reuse also usually means higher iner-cell inerference because of more inerferers. downlink subframe and in every resource block in he frequency domain, hus covering he enire cell bandwidh. Figure 2 illusraes he srucure of a single [5]. Figure 3 shows he srucure of wo reference signals wihin a cell [5]. In he case of wo reference signals wihin a cell, he second reference signal is frequency muliplexed wih he firs one, wih a frequency-domain offse of hree subcarriers. is ransmied in each cell wih he srucure shown in figure 2 and 3. This makes he iner-cell inerference experienced by no highly depend on he cell load. Even wih very low raffic load in he sysem, in each cell is ransmied anyway and he iner-cell inerference experienced by is jus similar as in a fully loaded sysem. Therefore, in a low load scenario, alhough he iner-cell inerference ha daa experiences is low, he iner-cell inerference ha experiences is sill high. The lower he cell load, he more serious he mismach is. Figure 2: Srucure of cell-specific reference signals Horizonal anenna gain [db] db 3-secor anenna 6-secor anenna Azimuh angel Figure 1: anenna paern for 3-secor and 6-secor As discussed, he iner-cell inerference o daa usually depends on raffic load in he sysem. When he sysem raffic load is low, he iner-cell inerference experienced by daa is usually low because here is probabiliy ha he same frequency/ime resource ha he daa uses is empy in oher cells. However, he iner-cell inerference o does no depend on he raffic load in he sysem because of he fixed paern and frequency/ime posiion defined in LTE., inroduced in he firs release of LTE (release 8), are he mos basic downlink reference signals in LTE. There can be one, wo, or four s in a cell, defining one, wo, or four corresponding anenna pors [5]. is ransmied in every Figure 3: Srucure of cell-specific reference signals in he case of wo reference signals corresponding o 2 anenna pors. In LTE, six possible frequency shifs of he reference symbols are defined wih a frequency-domain offse of 0, 1, 2 5 subcarriers. The frequency shif o use in a cell depends on he physical-layer ideniy of he cell, and by properly assigning physical-layer cell ideniies o differen cells, differen reference-signal frequency shifs may be used in a mos five neighboring cells for cells wih single anenna por. Considering a ypical 2x2 downlink MIMO configuraion, however, he effec of shifs is expeced o be less efficien especially for a HOS sysem as discussed in secion I. In LTE, excep for ransmission mode 9 which is inroduced in Release 10, are used by he erminal o acquire for link adapaion [5, 10]. is esimaed based on and ored o he nework o predic he daa channel qualiy and consequenly decide he downlink modulaion and coding scheme (MCS) for each erminal. In a low load scenario, as discussed above, here is mismach for he iner-cell inerference level of and daa and his will resul in ha he ored underesimaes he acual channel qualiy of daa. Consequenly, a conservaive link adapaion decision is made and a lower MCS is seleced alhough a higher MCS is more desirable for he channel qualiy a he momen. 139

3 Link adapaion aims o selec proper MCS and mainain a arge block error rae (BLER) of daa ransmission. To ge he early erminaion gain from HARQ and save ransmission power, in LTE, he arge BLER in downlink is usually 10%. When such mismach happens, due o he conservaive link adapaion, he BLER would be lower han he arge which is a wase of he channel capaciy. Figure 4 illusraes he average BLER wih differen raffic loads for a 6-secor nework configured wih 2x2 downlink MIMO. frequency shifs is no used in he es as he opimal paern assignmen for a 6-secor nework is very complicaed and he effec is expeced o be less effecive wih wo anenna pors. Ouer-loop usmen is enabled bu he effec seems marginal as he download file size is 1 Mbyes which is fairly small. More deails of he simulaion assumpions can be found in able I. In figure 4, he x-axis is he normalized cell load ranged from 0 o 1. I is clear ha he average BLER is very low compared o he arge BLER when he load is low. When raffic load increases, he difference beween he acual average BLER and arge BLER ges smaller, e.g. when he normalized cell load is larger han 0.8, he average BLER is approaching he arge. In nex secion, some compensaion algorihms are proposed by coordinaion beween cells o cope wih his problem in low load HOS scenario. Figure 4: BLER of 6-secor sysem wih differen cell loads III. ADAPTIVE ADJUSTMENT FOR HIGHER ORDER SECTORIZATION As discussed above, in low load scenario, he HOS sysem suffers from serious under-esimaion and conservaive link adapaion. Alhough HOS is ypically deployed in areas where he raffic loads are usually high, considering ha in he real nework, he raffic load siuaion may vary over he ime, i is herefore valuable o find soluions o improve HOS sysem performance even in low load scenario. The underesimaion resuls from he differen inerference levels of and daa. The experienced inerference of is no dependen on he sysem load while he inerference of daa does depend on he load. Consequenly, he esimaed based on over-esimaes he inerference level of daa. In his secion, an adapive compensaion algorihm is proposed o compensae for he inerference difference beween and daa. Suppose I and daa I are he experienced inerference by daa and respecively. In low load scenario, as discussed, I is usually smaller han daa I. The difference beween he experienced inerference by and daa can be expressed as Δ I = I / I daa (1) Ideally, he esimaed should be used by I o compensae for he inerference difference: = ΔI (2) In (2), denoes he Signal o Inerference and Noise Raio (SINR) convered from ored index and he ored is esimaed from. denoes he used SINR which can be convered back o index. Therefore, he key poin here is o correcly esimae I. To direcly esimae I is difficul. In his secion, a mehod o esimae I according o relaive sysem load is proposed based on he fac ha experienced inerference by daa is proporional o he sysem load: I daa L (3) L is a load facor defined o resens he sysem load siuaion in ime insan. I is a normalized value ranged from 0 o 1. The lower he L, he lower he sysem load is. L equaling 1 resens a fully loaded sysem. For, all he cells ransmi using he same frequency/ime resource (no frequency shifs is assumed in he discussion). The in each cell is inerfered like in a fully loaded sysem and no dependen on he acual sysem load. Define L is he virual sysem load ha experienced, and i should resen he fully loaded sysem: L =1 (4) Based on (1) (3) and (4), he inerference difference should be proporional o he relaive load siuaion of daa and. L 1 Δ I = (5) L L From (5), i is clear ha when he sysem load L is low, I is high and he should be compensaed by a large value. Combining (2) and (5), we can ge he compensaion as: Here k is a scaling facor. 1 = k (6) L To obain sysem load facor L, firsly a meric named Resource Block Uilizaion Raio (RBUR) is defined o measure he load siuaion in each cell. In LTE, each Resource Block (RB) consiss of 12 consecuive subcarriers in he frequency domain and one 0.5 ms slo in he ime domain [5]. RBUR is a cell specific measuremen and can be defined as (7). nrb u u celli RBURi = mean( ) (7) NRB i 140

4 In (7) nrb u is he number of RB allocaed o user u who belongs o cell i a scheduling insan. NRB i is he oal available number of RB in cell i a scheduling insan. The average RBUR in cell i is obained by averaging he measured RB uilizaion over differen scheduling insan. Compared o number of users, RBUR is more accurae o resen he load siuaion in each cell. This is because he load in a cell is no only dependen on number of users in a cell bu also dependen on he raffic volume of each user. RBUR is measured in each cell and resens he cell load siuaion. To ge knowledge of he load siuaion in he sysem, coordinaion beween cells is needed. Generally speaking, o esimae he load siuaion in he sysem more accuraely, i is desirable o ge he load siuaion of as many cells as possible. However, in pracical, such kind of coordinaion could be quie complicaed and requiring he suppor from X2 inerface beween cells no co-sied. This will increase he signaling burden in X2 inerface and inroduce addiional delay. For HOS sysem, forunaely, here are more cells co-sied which can provide fairly sufficien saisics for sysem load esimaion and he coordinaion beween such co-sied cells is much easier. Therefore, in his paper, coordinaion among cosie cells is proposed: RBURi i I L (8) N In (8), RBUR i is he load facor in cell i which belongs o a se of cells co-sied and his se of cells are denoed by se I. N is he number of co-sied cells. For example, in a 6-secor sysem, N is 6. Therefore, he usmen algorihm is proposed as: = k i I N RBUR Considering he underesimaion problem is no longer eviden in high load, i is proposed o se a load hreshold o enable his usmen algorihm o reduce he complexiy. In nex secion, he proposed usmen algorihm is evaluaed by sysem level simulaion wih a 6-secor sysem. To ge a more complee undersanding, besides he performance of (9), he performance of a more ideal case is evaluaed as well. In his more ideal case, i is assumed ha he whole sysem load siuaion can be obained hrough all cells wihou considering he complexiy and delay in realiy. IV. EVALUATION AND RESULTS In his secion he sysem-level evaluaion and resuls are presened and discussed. The simulaion assumpions and parameer seings are illusraed in Table I. Considering ha HOS is usually deployed in urban environmen wih dense users, in he simulaion, ITU UMa scenario is seleced [10]. Furhermore, indoor modeling from ITU UMi is added o UMa o more accuraely model he real urban scenario. More specifically, he indoor modeling relaed parameers are lised in Table II. i (9) TABLE I: Sysem Evaluaion Assumpions and Parameers Parameers Values/descripions Seven 6 secored cells, wih wrap-around Macro base saion deploymen Sie o sie disance: 500m Heigh: 40m above ground Tx power: 60W Carrier frequency 2 GHz Bandwidh MHz Muliplex FDD Channel model ITU UMa + UMi [10] Scheduling Proporional fair in ime and frequency shif No Ouer-loop On usmen Anenna model 3GPP SCM [9] Anenna configuraions 2x2 Transmission mode TM4 Traffic Type Download Mean file size 1MByes Traffic load From low o high BLER arge 0.1 TABLE II: Parameers for Indoor Modeling Parameers Values/descripions Indoor probabiliy 0.8 Average loss for oudoor-o-indoor peneraion sandard deviaion of oudoor-o-indoor 0 peneraion Average loss for in-car peneraion 0 Indoor loss depends on perpendicular 0.5dB per meer disance from wall o UE max perpendicular disance from wall o UE 25m sandard deviaion of he in-car peneraion 0 Figure 5 shows he BLER performance wih and wihou usmens. In he simulaion, is used every 10ms. The usmen enabling hreshold is se o be 0.8. From figure 5, i is clear ha wih usmen, he BLER difference compared o he BLER arge can be reduced in low loads. For he mos low load case (resource block uilizaion raio is abou 0.1), if he load siuaion is esimaed based on all cells in he sysem (shown by he black curve), he average BLER of users in he sysem is improved from around 0.01 o 0.08 and very much closer o he arge BLER (0.1). When he load siuaion is esimaed based on cells in he same sie (as shown in he magena curve), he BLER is also evidenly more approaching he arge BLER compared o no usmen. This indicaes ha he link adapaion is more accurae han wihou usmen. I can also be observed ha when he load siuaion is only esimaed wihin he sie, he performance is worse han he case ha he load siuaion is esimaed in he whole sysem. This is because he load siuaion esimaed wihin a sie is sill no sufficien enough o resen he inerference level experienced by daa especially in very low load scenario where he raffic disribuion is very much ununiform and he esimaion by limied saisics will inroduce more inaccuracy. To compensae for he esimaion inaccuracy in very low load scenario, when is used based on load siuaion esimaed wihin each sie, one way is o apply a scaling facor (k in (9)) o he esimaed usmen value. Noe 141

5 ha he performance shown in Figure 5 is obained wihou applying any scaling facor. BLER BLER Targe 0.03 No usmen 0.02 sumen; load esimaed in all cells sumen; load esimaed wihin sie Resorce Block uilizaion raio Figure 5: BLER performance wih and w/o usmen Figure 6 shows he BLER performance afer applying scaling facor in he mos low load case (he mos lef poins in figure 5). The seleced scaling facor is 2. I can be seen ha by applying a scaling facor, he BLER performance is approaching he more ideal case where he load siuaion is esimaed from all cells. and average cell hroughpu can be improved due o more accurae link adapaion. I can also be observed ha even if he is used only hrough coordinaion wihin each sie, he performance is quie good. Considering he complexiy, signaling overhead and delay issues, he proposed usmen algorihm seems o be a simple and effecive way o cope wih he underesimaion in a low load HOS nework. V. CONCLUSION Higher order secorizaion is an effecive way o increase boh he sysem capaciy and end-user daaraes. However, when he sysem raffic load is low, here is usually severe mismach beween he inerference level of cell reference signal and daa. This resuls in underesimaion of he channel qualiy and conservaive link adapaion. In his paper, an adapive usmen algorihm is proposed for a sysem deploying higher order secorizaion. The proposed algorihm adapively uss he ored by prediced inerference difference beween cell reference signal and daa. The inerference difference is esimaed based on measuremens of he cell load averaged wihin he sie. Sysem level simulaion resuls indicae ha he proposed algorihm is efficien o improve he link adapaion accuracy, and consequenly he sysem hroughpu and user biraes are improved evidenly. Furhermore, his algorihm is simple and does no require furher signaling in X2 inerface. VI. ACKNOWLEDGEMENTS This work was suppored by he Sae Major Science and Technology Special Projecs (Gran No. 13ZX ). Figure 6: BLER wih and w/o scaling facor in low load Mean user birae [Mbps] No usmen sumen; load esimaed in all cells sumen; load esimaed wihin sie Cell hroughpu [Mbps] Figure 7: mean user birae vs. average cell hroughpu Figure 7 shows he performance of he average user birae vs. average cell hroughpu. I is clear ha, by applying usmen (wihou scaling here), boh he average user birae REFERENCES [1] Ericsson, Traffic and marke daa or, accessed [2] Cisco, Cisco Visual Neworking Index: Global Mobile Daa Traffic Forecas Updae, 11-15, [3] UMTS forum, Mobile raffic forecass 10-, [4] Web Size Opimizaion Average Web Page Size Sepuples Since 03, hp:// accessed [5] E. Dahlman, S. Parkvall, J. Sköld, 4G LTE/LTE-Advanced for mobile broadband, Academic Press, 11. [6] Huang, H. ; Alrabadi, O. ec., Increasing hroughpu in cellular neworks wih higher-order secorizaion,signals, Sysems and Compuers (ASILOMAR), 10 [7] Hagerman, B. ; Imbeni, D. ; ec., WCDMA 6-secor Deploymen - Case Sudy of a Real Insalled UMTS-FDD Nework, Vehicular Technology Conference, 06. VTC 06-Spring. [8] Riedel, I. ; Feweis, G., ec., Increasing hroughpu and fairness in he downlink of cellular sysems wih N-fold secorizaion, GLOBECOM Workshops (GC Wkshps), 11 IEEE [9] 3GPP TS , Furher advancemens for E-UTRAN physical layer aspecs, Release 9 [10] ITU-R M.2135, Guidelines for evaluaion of radio inerface echnologies for IMT-Advanced, 08 [11] Klaus I. Pedersen, ec., Frequency Domain Scheduling for OFDMA wih Limied and Noisy Channel Feedback, VTC-07, fall 142