Joint Optimization of Uplink Power Control Parameters in LTE-Advanced Relay Networks

Transcription

1 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 Join Opimizaion of Uplink Power Conrol Parameers in LTE-Advanced Relay Neworks Ömer Bulakci 1,2, Ahmad Awada 1,3, Abdallah Bou Saleh 1,2, Simone Redana 1, Jyri Hämäläinen 2, Bernhard Wegmann 1, Bernhard Raaf 1, and Ingo Viering 4 Absrac Relaying is a promising enhancemen o exising radio access neworks and is currenly being sandardized in 3GPP o be par of he LTE-Advanced Release 10. Relays promise o alleviae he limiaions of convenional macrocellonly neworks such as poor indoor peneraion and coverage holes in a cos-efficien way. Ye, o fully exploi he benefis of relaying, he iner-cell inerference which is increased due o he presence of relays should be miigaed. Besides, a high receiver dynamic range should be avoided o reain he orhogonaliy of he SC-FDMA sysem. In his manner, uplink power conrol is an effecive ool o address such challenges. However, in order o enhance he overall sysem performance, power conrol opimizaion should be joinly done on all links, i.e. on enbrelay node, enb-ue and relay node-ue links. In his paper, we propose a join opimizaion sraegy of power conrol parameers based on Taguchi s mehod ha explois he muual dependencies of he differen links wihou a priori knowledge. The proposed join parameer opimizaion is compared wih a reference sudy ha comprises a four-sep opimizaion. The evaluaion of he opimizaion sraegies wihin he LTE- Advanced uplink framework is carried ou by applying he sandardized LTE Release 8 power conrol scheme boh a enbs and relays. Simulaions show ha he proposed opimizaion yields similar or beer performance relaive o he reference sraegy depending on he considered performance meric. Keywords- LTE-Advanced; relay; uplink; power conrol; join opimizaion; Taguchi s mehod R I. INTRODUCTION ELAYING is considered an inegral par of he Fourh Generaion (4G) radio access neworks, namely IEEE m and 3GPP Long Term Evoluion Release 10 and beyond (LTE-Advanced). The moive behind choosing relaying as an enhancemen echnology o curren radio access neworks has been well elaboraed in lieraure. According o previous echnical sudies, relay nodes (RNs) promise o increase he nework capaciy and o beer disribue resources in he cell, or alernaively, exend he cell coverage area [1], [2]. Relaying is also regarded a cos efficien echnology [3]. Relay deploymens require a more deailed dimensioning 1 Nokia Siemens Neworks, S.-Marin-Sraße 76, 81541, Munich, Germany, {omer.bulakci, abdallah.bousaleh}@ieee.org, {ahmad.awada.ex, simone.redana, bernhard.wegmann, bernhard.raaf}@nsn.com. 2 Aalo Universiy School of Elecrical Engineering, P.O. Box 13000, FIN Aalo, Finland, jyri.hamalainen@aalo.fi. 3 Technische Universiä Darmsad, Communicaions Engineering Lab, Mercksr. 25, 64283, Darmsad, Germany. 4 Nomor Research GmbH, Brecherspizsrasse 8, 81541, Munich, Germany, viering@nomor.de. and planning han convenional single-hop neworks. RNserved user equipmens (relay-ues) can creae severe iner-cell inerference in paricular when a large number of RNs are deployed in he cell wih a reuse facor of one. Consequenly, power conrol (PC) becomes a vial means in he uplink (UL) no only o compensae for channel variaions, bu also o miigae he inerference and o increase he cell edge and sysem capaciies. Furhermore, PC decreases he deviaion beween received power levels of differen nodes in he same cell ensuring ha he receiver dynamic range does no exceed a predeermined level above which inra-cell inerference occurs [4]. Besides, in conras o evolved Node B (enb)-only neworks, PC is necessary for he relay link (enb-rn) because he end-o-end (e2e) hroughpu (TP) of relay-ues depends on he qualiies of boh he access (RN-UE) and relay links. In [5], a four-sep PC parameer opimizaion is suggesed where he parameers are uned in each sep according o he resuls obained in he preceding sep. We ake his work as a reference for performance comparisons and propose a join PC parameer opimizaion based on Taguchi s mehod. The PC parameers on he direc (enb-ue), access, and relay links are uned joinly. Taguchi s mehod uses originally a so-called orhogonal array (OA) [6]. An OA selecs a reduced se of parameer combinaions o be esed from he full search space. The number of seleced combinaions deermines he number of experimens being carried ou and evaluaed agains a performance measure. Using all he experimens' resuls, a candidae soluion is found and he process is repeaed ill a desired crierion is fulfilled. Herein, we use he opimizaion procedure based on Taguchi s mehod applying a nearly OA (NOA) raher han an OA, as an NOA can be consruced for any number of experimens and in urn he complexiy of he algorihm can be reduced. NOAs were used in [7] o opimize anenna parameers in enb-only deploymen wihin he LTE Release 8 (Rel. 8) framework. In his work, evaluaion is carried ou by applying sandardized LTE Rel. 8 PC scheme in simulaions conduced wihin he LTE-Advanced UL conex. The remainder of he paper is organized as follows. Secion II provides he background informaion. In Secion III, he opimizaion problem is oulined along wih he proposed and reference opimizaion sraegies. The sysem model and simulaion assumpions are given in Secion IV. In Secion V, deailed performance evaluaion and analysis are carried ou. Secion VI concludes he paper. II. BACKGROUND AND DEFINITIONS In his secion, we briefly recall he fracional power conrol (FPC) scheme of LTE Rel. 8. The definiions of he /11/$ IEEE 2064

2 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 performance merics are presened aferwards. A. UL Open-Loop Power Conrol The main ask of PC mechanisms is o compensae he longerm channel variaions and o limi he amoun of generaed iner-cell inerference. Ye, he receiver dynamic range 1 of enbs and RNs should also be adjused via PC. Large dynamic range may lead o reduced orhogonaliy beween imefrequency resources wihin a cell and cause inra-cell inerference [4]. To fulfill he aforemenioned objecives, FPC [8] is used for he Physical Uplink Shared Channel (PUSCH) o deermine he UE ransmi power. In his work, FPC is also employed for he relay specific PUSCH (R-PUSCH). Accordingly, he ransmi power of a node u (UE or RN) ha employs open-loop FPC, is given in dbm as: Pu = min{ Pmax, P log10 Mu+ α L}. (1) In his equaion, P max is he maximum allowed ransmi power which has an upper limi of 23 dbm for UE power class 3 and 30 dbm for RN ransmissions for urban scenarios [9], P 0 is he power offse comprising cell-specific and node-specific componens and i is used for conrolling he received signal-o-noise-raio (SNR) arge ha can be se from -126 dbm o P max wih a sep size of 1 db, M is he number of physical resource blocks (PRBs) u allocaed o node u, α is a 3-bi cell-specific pah loss compensaion facor ha can be se o 0.0 and from 0.4 o 1.0 wih a sep size of 0.1, L is he downlink pah loss esimae calculaed a he node. Open-loop PC compensaes slow channel variaions, i.e. pah loss changes including shadowing, while limiing he iner-cell inerference. If α is se o one in (1), he pah loss is fully compensaed and he resuling scheme is called full compensaion power conrol (FCPC). For a given P 0 value, FCPC improves he cell-edge user performance a he cos of increased iner-cell inerference due o higher ransmi power levels. Ye, he iner-cell inerference can be reduced by using values smaller han one, which can increase he cell-cener performance a he cos of penalizing he cell-edge performance [10] [11]. Moreover, one imporan moivaion o sudy he applicabiliy of he exising FPC for he relay enhanced cells is he desired backward compaibiliy beween LTE Rel. 8 and LTE-Advanced erminals. B. Performance Merics The paper uilizes he following key performance merics: Γ 5% : The 5 h %-ile level of he user TP CDF reflecs he celledge bi rae or equivalenly he cell coverage performance. Γ HM : The harmonic mean (HM) of he user TP levels in he cell prioriizes he performance of cell edge UEs and hus an 1 The receiver dynamic range is defined as he difference in db beween he 5 h %-ile and 95 h %-ile of he cumulaive disribuion funcion (CDF) of he oal received power. opimizaion using his meric leads o a more homogeneous user experience in he nework [12]. III. OPTIMIZATION PROBLEM AND METHODS In his secion, we firs inroduce he relay scenario and he opimizaion problem o be addressed. We hen ouline he proposed opimizaion sraegy and give a brief explanaion of he reference opimizaion sraegy. Figure 1. One-ier relay deploymen shown in a secor. 4 RNs are deployed a he cell-edge. The considered relay deploymen along wih he links is depiced in Fig. 1. Cell selecion for he UEs is based on he sronges downlink received signal power, whereas RNs are conneced o he overlaying macrocell. As can be seen in Fig. 1, relay-ues are mosly hose a he macrocell edge, whereas, macro enb-served UEs (macro-ues) are generally locaed in he cell cener. Accordingly, in order o enhance overall sysem performance, he PC parameer opimizaion should be done on all links considering he inerdependencies. Tuning P max, P 0, and α on hese links simulaneously is a challenging ask - given he possible parameer range discussed in Secion II-A. Therefore, a brue-force approach becomes infeasible due o high compuaional complexiy. A reasonable opimizaion approach should ake ino accoun he muual dependencies of relay and access links, where e2e performance is deermined by he qualiies of boh links. A. Four-sep Opimizaion In [5], inerdependencies beween differen links are addressed via a four-sep sraegy. The parameer uning in each sep is done according o he performance resuls of he preceding sep. Namely, he four seps are: 1. Simulaions are carried ou o opimize PC parameers for enb-only scenario. 2. Relay scenario is adoped and parameers resuling from Sep 1 are applied a boh enbs and RNs. Simulaion resuls of his sep are used as a saring poin for Seps 3 and PC parameers are opimized in RN cells. 4. Keeping PC parameers fixed a RNs, he PC parameers a enbs are opimized. In conrary o he four-sep opimizaion, Taguchi s mehod inherenly akes ino accoun he inerdependencies and yields he opimized parameer seing in one simulaion run. Since he main focus of he relay deploymen a he cell edge is o increase coverage, we consider FCPC as described in Secion II-A. The goal is o opimize P 0 values on all links 2065

3 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 joinly, i.e. P 0 direc link, P 0 access link, and P 0 relay link. Moreover, i is known from [5] ha P max of relay-ues (P max relay-ues ) can be also uned o furher enhance he sysem performance. Therefore, P max uning for he relay-ues is considered as well. B. The Opimizaion Procedure Based on Taguchi s Mehod Using NOA We sar by defining he configuraion parameers. Le x be a configuraion parameer where index = 1,2,3,4 refers o as one of he four configuraion parameers, i.e., x = P link link, x2 = P access, x relay 3 = P, x4 = Pmax direc link relay UEs Besides, we denoe he opimizaion arge funcion as y and i can be se eiher o 5%-ile user TP or HM meric, see Secion II-B. Le us nex inroduce Taguchi's mehod [7] [12]. The opimizaion approach is depiced in Fig. 2 and will be discussed in deails in he following. Figure 2. The ieraive opimizaion procedure based on Taguchi's mehod using NOA. 1) Consruc he proper NOA Originally, Taguchi's mehod uses an OA which conains a reduced se of N parameer combinaions from he full search space [6]. Every parameer x has a se of esing values corresponding o a se of levels, i.e., level 1 is mapped o he firs esing value of a parameer, level 2 o he second value and so on (see Secion III-B-2). Each of he N parameer combinaions is esed in a corresponding experimen i where he funcion y is evaluaed resuling in a measured response y i. In an OA, each esing value of a parameer x is esed a leas once wih every oher value of parameer x j. This propery of he OA accouns for he ineracions ha migh exis beween he configuraion parameers. To reduce he number of experimens, an OA is replaced by a nearly orhogonal array (NOA) [7]. Considering an NOA, each esing value of a parameer x is no necessarily esed wih every oher value of parameer x j. Hence, NOA considers only parially he ineracions among he parameers and is easier o consruc. An NOA can be consruced for any number of parameers and number N of experimens a he expense of considering parially he ineracions among he configuraions parameers. The firs sep in he opimizaion procedure is o consruc a proper NOA. For his purpose, he number of configuraion. parameers has o be deermined. In our case, he oal number of configuraion parameers is k = 4. Thus, an NOA having 4 columns should be consruced wih a predefined number of experimens N and levels s. In his work, we consruc an NOA having N = 36 experimens and s = 9 levels, see Table I, using he algorihm described in [13]. Using his NOA, each parameer will be esed wih 9 differen values in each ieraion of he algorihm. Every column of he NOA corresponds o a configuraion parameer. For example, he firs column can be assigned o x 1, he second o x 2 and so on. Having consruced he required NOA, he levels of each parameer should be mapped o parameer values in order o conduc he experimens. In each ieraion, he levels of each parameer are assigned o differen se of values depending on he candidae soluion found in he previous ieraion. This process is repeaed unil he levels of each parameer are mapped o values ha are close enough o each oher. The mapping funcion is explained in he nex subsecion. TABLE I. AN NOA HAVING K=4 CONFIGURATION PARAMETERS, N=36 EXPERIMENTS, S=9 LEVELS WITH THE MEASURED RESPONSES AND THEIR CORRESPONDING SN RATIOS. Experimen # i x 1 x 2 x 3 x 4 Measured Response SN Raio y 1 SN y 2 SN y 3 SN y 4 SN y 5 SN y 6 SN y 7 SN y 8 SN 8 : : : : : : : y 36 SN 36 2) Map each level o a parameer value Le min{x } and max{x } be he minimum and he maximum feasible values for parameer x. In he firs ieraion, m =1, he cener value of he opimizaion range for parameer x is defined as min{ x} + max{ x} V =. (2) 2 In any ieraion m, he level l = s/2 is mapped o V (m). The oher s-1 levels are disribued around V (m) by adding or subracing a muliple ineger of a sep size β (m). For m =1, he sep size is defined as max{ x} min{ x} β =. (3) s+ 1 In ieraion m, he mapping funcion f (m) (l ) for a level lo a dedicaed value of he parameer x can be described as: V ( s /2 l) β 1 l s/2 1 f ( l) = V l= s/2. (4) V + l s/2 β s/2 + 1 l s ( ) For insance, consider he parameer x 4 = P relay-ues max having a minimum value of min{x 4 } = 7 dbm and a maximum max{x 4 } = 23 dbm. If x 4 is esed wih 9 levels, i.e., s = 9, 2066

4 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 level l = 5 is mapped in firs ieraion o (7 + 23) / 2 = 15 dbm, level 4 o 15 - β 4 (1) =13.4 dbm, level 6 o 15 + β 4 (1) = 16.6 dbm and so on. As he power seing x canno be decimal, he mapped value f (m) (l ) of a level l is furher quanized o he neares ineger. The values of V (m) and β (m) are updaed a he end of each ieraion if he erminaion crierion, see Secion III-B-5, is no me. 3) Apply Taguchi s mehod To inerpre he experimenal resuls, Taguchi's mehod convers he measured responses o signal-o-noise (SN) raios which are no o be confused wih he SNRs of he received signals. The SN raio is compued for each experimen i as i 2 10 i SN = 10 log ( y ). (5) Then, he average SN raio is compued for each parameer and level. In he example of Table I, he average SN raio of x 1 a level l = 1 is compued by averaging he SN raios of he experimens where x 1 is esed a level 1, i.e., SN 1, SN 2, SN 3 and SN 4. The bes level of each parameer is he level having he highes average SN raio. According o he mapping funcion f (m) (l ), he bes seing of a parameer x in ieraion m is found and denoed by V (bes,m). 4) Shrink he opimizaion range A he end of each ieraion, he erminaion crierion is checked. If i is no me, he bes values found in ieraion m are used as cener values for he parameers in he nex ieraion m+1: V + = V. (6) ( m 1) (bes, m) I may happen ha he bes value of a parameer x found in ieraion m is close o min{x } or max{x } In his case, here is a need for a procedure o consisenly check if he mapped value of a level is wihin he opimizaion range. Moreover, he opimizaion range is reduced by muliplying he sep size of each parameer x by a reducion facor ξ < 1: β = ξ β ( m+ 1). (7) 5) Check he erminaion crierion Wih every ieraion, he opimizaion range is reduced and he possible values of a parameer ge closer o each oher. Hence, he se used o selec a near-opimal value for a parameer becomes smaller. The opimizaion procedure erminaes when all sep sizes of he parameers are less han a predefined hreshold ε, i.e. β ) ( m <ε. (8) In his work, he algorihm ends when he mapped values of levels 1 and 9 do no differ by more han 1 db for each parameer. To his end, ε is se as a rough approximaion o 1/(s-1) =1/8. A. Resource Sharing IV. SYSTEM MODEL A resource fair round robin (RR) scheduling is applied for all UEs. Besides, he performance of relay-enhanced neworks depends significanly on he resource allocaion sraegy. For relay-ues, he experienced e2e TP depends on he qualiies of boh he access and relay links. Tha is e e ( TP TP ) TP 2 = min,, (9) enb RN RN UE where e2e hroughpu is obained as a minimum over hroughpus on he relay and access links. As depiced in [9], he backhaul subframes are reserved for relay link ransmissions and hus a daa ransmission gap is experienced on he access link. During he ransmission gaps relay-ues are no scheduled. For he resource allocaion on he backhaul subframes and access link, we uilize he scheme in [14], referred o as hop-opimizaion model. In his model, he resource shares of he RNs on he relay link are deermined proporional o he number of aached relay-ues. The available capaciy on he relay link is hen disribued among relay-ues uilizing max-min fairness. Moreover, he number of backhaul subframes o be allocaed o RNs is chosen such ha he overall sysem performance is opimized. B. Simulaion Parameers The simulaed nework is represened by a regular hexagonal cellular layou wih 19 ri-secored sies, i.e. 57 cells. RNs admi regular oudoor deploymen a he secor border and indoor users are assumed, where 25 uniformly disribued UEs are dropped per secor and he full buffer raffic model is applied. In oal, 50 user drops (or snap-shos) are simulaed using a sysem level semi-saic simulaor, where resuls are colleced from he inner mos secor only, o ensure proper modeling of inerference (wo iers of ri-secor sies). Simulaion parameers follow he laes parameer seings agreed in 3GPP [9] and are summarized in Table II. A frequency reuse facor of one (full reuse scheme) is considered in he nework. All available resources in a cell are assumed o be used and hence a raher pessimisic inerference modeling is considered a he access link. TP is compued from signal-o-inerference-plus-noise raio (SINR) using he Shannon approximaion similarly as described in [15]. An overhead of 25% is assumed, which accouns for conrol symbols and pilos. Relay sie planning is assumed as modeled in [9]. Direcional anennas are uilized a he RNs for backhaul ransmission, while omni-direcional anennas are assumed for he access link ransmission. Log-normal shadow fading is as well modeled and applied for NLOS propagaion condiions. V. PERFORMANCE EVALUATION AND ANALYSIS The performance evaluaion is carried ou assuming he 3GPP urban scenario (ISD 500 m) where 4 RNs are deployed per cell. The number of backhaul subframes is se o wo [14]. This seing is found o opimize he overall sysem performance along wih relaively more homogeneous user experience over he whole cell area. The cell capaciy-oriened seing wih P 0 = -55 dbm and α = 0.6 (FPC) [5] is applied for he enb-only deploymen. Recall ha he enb-only deploymen is aken as a benchmark o deermine he relaive gains of differen opimizaion sraegies. Besides, for he four- 2067

5 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 sep opimizaion he cell coverage-oriened seing is used for he relay scenario. Furhermore, in order o maximize he cell edge user performance α is se o one (FCPC). Parameer Carrier Frequency Bandwidh Number of PRBs TABLE II. SIMULATION PARAMETERS Sysem Parameers 2 GHz 10 MHz Value 48 for daa ransmission Highes MCS 64-QAM, R = 9/10 Peneraion Loss Thermal Noise PSD SINR lower bound Transmi Power Elevaion Gain Anenna Configuraion Noise Figure Anenna Paern (Horizonal) Maximum Transmi Power Anenna Configuraion Noise Figure Maximum Transmi Power Anenna Configuraion RN-eNB Elevaion Gain RN-UE Elevaion Gain Relay Link Anenna Paern (Horizonal) Access Link Anenna Paern Noise Figure Disance 20 db on enb-ue and RN-UE links -174 dbm/hz -7 db enb Parameers 46 dbm 14 dbi Tx-2, Rx-2 5 db A(θ) = -min[12 (θ/ θ 3dB) 2, A m] θ 3dB = 70 o and A m = 25 db UE Parameers 23 dbm Tx-1, Rx-2 9 db Relay Node Parameers 30 dbm Tx-2, Rx-2 7 dbi 5 dbi A(θ) = -min[12 (θ/ θ 3dB) 2, A m] θ 3dB = 70 o and A m = 20 db Omni-direcional 5 db Channel Models R [km] Direc Link (enb UE) PL(LOS): log 10(R), PL(NLOS): log 10(R) Prob(LOS) = min(0.018 / R, 1) (1-exp(-R / 0.063))+exp(-R / 0.063) Access Link (RN UE) PL(LOS): log 10(R), PL(NLOS): log 10(R) Prob(LOS) = 0.5 min (0.5, 5exp(-0.156/R)) + min (0.5, 5exp(-R/ 0.03)) Relay Link (enb RN) {a & b accoun for he sie planning gain} PL(LOS): log 10(R), PL(NLOS): log 10(R)-b Prob (LOS) = 1-(1-(min(0.018./R,1)(1-exp(-R/0.072))+exp(-R/0.072))) a a=3 & b=5 owards donor enb, whereas, a=1 & b=0 owards oher enbs. Shadowing Shadow Fading Sandard Deviaion De-correlaion Disance Correlaion Facor Log-normal 8 db (direc link), 10 db (access link) 6 db (relay link) 50 m 0.5 beween sies 1 beween secors The res of his secion is organized as follows. Firs, he numerical resuls for join P 0 opimizaion are analyzed. To perform his join opimizaion, i is enough o drop he column of x 4 = P max relay-ues from he NOA and follow he opimizaion algorihm as described in Secion III. Second, he resuls are presened for join P 0 and P max opimizaion. In addiion, he ranges of [-113, -83] dbm and [7, 23] dbm wih a sep size of 1 db are considered for P 0 and P max relay-ues values, respecively. A. Join P 0 Opimizaion The P 0 values on all links are joinly opimized while fixing P max of UEs and RNs o he upper bounds. Boh Γ 5 % and Γ HM are used as performance merics. In Fig. 3, he convergence of he proposed opimizaion procedure is shown wih respec o he number of ieraions for boh performance merics. Noe ha a complexiy and opimaliy analysis can be found in [12]. Obained parameer seings are abulaed in Table III and he corresponding user TP CDFs are ploed in Fig. 4. I can be observed ha he four-sep opimizaion and he proposed opimizaion using HM meric yield similar performances as well as similar parameer seings. On he oher hand, using he 5%-ile TP meric, he proposed mehod ouperforms he foursep opimizaion on a wide range of CDF levels. Ye, sligh performance degradaion can be seen a he very low perceniles (below 5%-ile). Compared o enb-only deploymen, he proposed mehod wih 5%-ile TP meric achieves 129% and 54% gains a 5%-ile and 50%-ile CDF levels, respecively. Noe ha he 50%-ile user TP gain reads as 42% for he four-sep opimizaion and he proposed mehod wih HM meric. Figure 3. The convergence of he proposed opimizaion; (a) using Γ HM and (b) using Γ 5% as performance merics in join P 0 opimizaion. Parameers TABLE III OPTIMIZED PARAMETER CONFIGURATIONS 4-Sep Opimizaion enbs RNs Relay Link B. Join P 0 and P max Opimizaion Proposed Opimizaion (Meric: 5%-ile user TP) Relay enbs RNs Link Proposed Opimizaion (Meric: HM of user TPs) Relay enbs RNs Link P 0 [dbm] α P max [dbm] In conjuncion wih P 0 values, he parameer P max relay-ues is also joinly uned by he proposed opimizaion mehod. Furhermore, P max of macro-ues and RNs are se o he upper bounds. We adop Γ 5 % performance meric as i has shown a 2068

6 IEEE Inernaional Wireless Communicaions and Mobile Compuing Conference, Isanbul, Turkey, July 2011 Figure 4. UE TP CDFs for differen approaches for join P 0 opimizaion. Figure 6. UE TP CDFs for differen opimizaion approaches in case of join P 0 and P max opimizaion. he cos of decreased performance a higher CDF perceniles. Afer applying he proposed power conrol opimizaion sraegy wih 5%-ile user hroughpu meric in urban scenarios, i was found ha relay deploymen achieves 129% and 54% gains relaive o macrocell-only deploymen a 5%-ile and 50%-ile user hroughpu CDF levels, respecively. REFERENCES Figure 5. The convergence of he proposed opimizaion using Γ 5% as he performance meric for join P 0 and P max opimizaion. superior performance relaive o adoping Γ HM. The convergence of he mehod is depiced in Fig. 5. For he four-sep opimizaion P max relay-ues is found o be 15 dbm. The P 0 and α values are abulaed in Table III. For he proposed mehod he P 0 values -98, -108 and -90 dbm are obained for direc, access and relay links, respecively. Besides, P max relay-ues is uned o 9 dbm. All oher parameers are he same as given in Table III. The resulan user TP CDF plos are shown in Fig. 6. The four-sep opimizaion and he proposed join P 0 and P max opimizaion yield similar resuls. As compared o join P 0 opimizaion, hese seings resul in beer performance a low CDF perceniles. However, he join P 0 opimizaion shows beer performance a high perceniles. VI. CONCLUSION We have proposed a join power conrol parameer opimizaion sraegy for relay enhanced LTE-Advanced neworks based on Taguchi s mehod ha explois he muual dependencies of he differen links. The impac of using differen performance merics as opimizaion funcions has been invesigaed. I is presened ha join P 0 opimizaion using 5%-ile user hroughpu level as he opimizaion arge funcion yields a beer performance on a wide probabiliy range when compared o he case where harmonic mean was used as a meric and o he reference foursep opimizaion. Beween join P 0 opimizaion and join P 0 and P max opimizaion, i has been shown ha join P 0 opimizaion can improve he performance a higher CDF perceniles. On he oher hand, join P 0 and P max opimizaions can improve he performance in he low hroughpu regime a [1] A. Bou Saleh, S. Redana, J. Hämäläinen, and B. Raaf, On he Coverage Exension and Capaciy Enhancemen of Inband Relay Deploymens in LTE-Advanced Neworks, Journal of Elecrical and Compuer Engineering, vol. 2010, Aricle ID ,12 pages, [2] R. Schoenen, W. Zirwas, and B. H. Walke, Capaciy and Coverage Analysis of a 3GPP-LTE Mulihop Deploymen Scenario, IEEE ICC Workshops 2008, pp , May 2008, Beijing, China. [3] E. Lang, S. Redana, and B. Raaf, Business Impac of Relay Deploymen for Coverage Exension in 3GPP LTE-Advanced, LTE Evoluion Workshop, IEEE ICC 2009, June 2009, Dresden, Germany. [4] B.E. Priyano, T.B. Sorensen, O.K. Jensen, In-Band Inerference Effecs on UTRA LTE Uplink Resource Block Allocaion, IEEE VTC Spring 2008, pp , May [5] Ö. Bulakci, S. Redana, B. Raaf, J. Hämäläinen, Sysem Opimizaion in Relay Enhanced LTE-Advanced Neworks via Uplink Power Conrol, IEEE VTC 2010-Spring, Taiwan, May [6] A. S. Hedaya, N. Sloane, and J. Sufken, Orhogonal Arrays: Theory and Applicaions. New York: Springer-Verlag, [7] A. Awada, B. Wegmann, I. Viering, and A. Klein, A join opimizaion of anenna parameers in a cellular nework using Taguchi s mehod, IEEE VTC 2011-Spring, Hungary, May [8] 3GPP TS , Evolved Universal Terresrial Radio Access (E- UTRA) Physical Layer Procedures (Release 10), v10.0.1, Dec [9] 3GPP TR , Furher Advancemens for E-UTRA: Physical Layer Aspecs (Release 9), v9.0.0, March [10] C.U. Casellanos, e al., Performance of Uplink Fracional Power Conrol in UTRAN LTE, IEEE VTC-Spring, Singapore, May [11] A. Simonsson and A. Furuskar, Uplink Power Conrol in LTE - Overview and Performance, IEEE VTC Fall, Canada, Sepember [12] A. Awada, B. Wegmann, I. Viering, and A. Klein, Opimizing radio nework parameers of Long Term Evoluion sysem using Taguchi s mehod, submied o IEEE Transacions on Vehicular Technology, Ocober [13] H. Xu, An algorihm for consrucing orhogonal and nearly-orhogonal arrays wih mixed levels and small runs, American Saisical Associaion and he American Sociey for Qualiy, vol. 44, no. 4, Technomerics, November [14] Ö. Bulakci, e. al., Two-sep Resource Sharing and Uplink Power Conrol Opimizaion in LTE-Advanced Relay Neworks, IEEE Muli- Carrier Sysems & Soluions 2011, Germany, May [15] 3GPP TR , Evolved Universal Terresrial Radio Access (E- UTRA); Radio Frequency (RF) sysem scenarios, v10.2.0, Dec