Interference Avoidance through Dynamic Downlink OFDMA Subchannel Allocation using Intercell Coordination 1

Transcription

1 Iterferece Avoidace through Dyaic Dowlik OFDMA Subchael Allocatio usig Itercell Coordiatio 1 Mahudur Raha ad Hali Yaikoeroglu Broadbad Couicatios ad Wireless Systes (BCWS) Ceter Departet of Systes ad Coputer Egieerig Carleto Uiversity, Ottawa, Otario K1S 5B6, Caada {raha,hali}@sce.carleto.ca. Abstract- OFDMA subchael allocatio has bee wellstudied. However, ost of the available literature cosiders a sigle cell syste without cochael iterferece. We preset a ovel itercell iterferece avoidace schee, i which the dowlik OFDMA subchaels are allocated through itercell coordiatio i a ulticell eviroet. The proposed schee ot oly ais to achieve axiized etwork throughput but also focuses o providig iproved throughput for cell or sector edge users that are ost affected by itercell iterferece. Ehaced cell edge perforace ay result i fewer base statios eeded to cover a regio; this, i tur, ay yield substatial savigs i deployet cost. The schee is coprised of two differet algoriths; oe at the base statio level, ad the other at a cetral cotroller to which a group of eighborig base statios are coected. The perforace of the proposed schee is copared with that of a referece schee i which coordiatio is ot eployed. It is observed fro siulatio results that the proposed schee outperfors the referece schee i ters of etwork ad cell edge throughput. Idex Ters- OFDM resource allocatio, iterferece avoidace, radio resource optiizatio. I. INTRODUCTION As a proisig odulatio techique for the ext geeratio wireless systes, such as IEEE , 3GPP logter evolutio (LTE), ad 4G cellular (see, for istace, WINNER [1]), orthogoal frequecy divisio ultiplexig (OFDM) has attracted sigificat attetio recetly. Besides OFDM s iheret ability to cobat iter-sybol iterferece (ISI) resultig fro frequecy selective fadig, orthogoal frequecy divisio ultiple access (OFDMA) offers flexibility i radio resource allocatio as each subcarrier ca be allocated itelligetly ad odulated adaptively to exploit ultiuser ad frequecy diversities, respectively. As a result, the achievable data rates o the available frequecy spectru ca be iproved sigificatly. I this paper, we propose a viable iterferece avoidace schee that uses dyaic OFDMA subchael allocatio with aggressive frequecy reuse i a ulticell eviroet. Nuerous studies o subchael allocatio i OFDMA systes i a sigle cell cotext are available i the literature. I ost of these studies, the ultiuser diversity gai is explored usig various optiizatio techiques [2-4]. I desigig practical etworks, however, optiizatio should be perfored i a ulticell eviroet cosiderig oe of the ajor challeges associated with future wireless systes, i.e., itercell iterferece. To the authors kowledge, resource optiizatio i a ulticell eviroet has ot yet bee ivestigated well ad oly a few research works o ulticell allocatio ca be foud i the literature [5,6]. I [5], a liear prograig (LP) forulatio is proposed, where subchaels are partitioed ad assiged fixed reuse factors such that user terials (UTs) at the cell edge ca oly use subchaels with higher reuse factors. Cotrary to the schee i [5], our proposed schee does ot require ay prior resource partitioig or cell plaig. While such partitioig of resources reduces resource utilizatio proactively, our proposed schee results i a dyaic ad efficiet reuse factor for each subchael depedig o utual iterferece situatios ad the UTs data rate requireets. The schee preseted i [6] is iterestig fro a optiizatio poit, where algoriths residig at the radio etwork cotroller (RNC) ad base statios (BSs) ai to optiize the total etwork throughput. However, sice the UTs rate requireets are ot cosidered i the forulatio, the schee is expected to favor UTs that are closer to the BS leadig to starvatio ad ufairess for the cell edge UTs ost affected by iterferece. I this paper, a chuk is regarded as the iiu graularity of the radio resource allocatio uit, which is defied as a collectio of cosecutive subcarriers over a defied tie period. By usig itercell coordiatio, our proposed schee thrives to axiize throughput o allocated chuks cosiderig the rate requireets of the cell edge UTs. I particular, ay optial or sub-optial allocatio schee, such as i [6], ca exploit ultiuser diversity to achieve axiized sector throughput by optially or sub-optially assigig best chuks to UTs. However, as the cell edge UTs experiece higher path-losses o the desired liks ad higher iterferece fro earby cells, these schedulig schees ted to overlook such disadvataged UTs. Therefore, iterferece avoidace is crucial i order to offer iiu required data rate to the cell edge UTs. The proposed schee is coprised of two separate algoriths residig at the BS ad at a cetral etity. Based o the iterferece received by its UTs ad their data rate require- 1 This work was supported by the Natural Scieces ad Egieerig Research Coucil (NSERC) of Caada uder participatio i the Wireless World Iitiative New Radio (WINNER) project Cocepts preseted i this paper were cotributed, i part, to the WINNER deliverable D4.7.2.

2 -ets, each sector (via its BS) seds a request to the cetral cotroller; this request icorporates a tetative list of chuks i the requestig sector. The cetral cotroller gathers all such requests ad processes to prepare a refied list of chuk restrictios to be applied i all ivolved sectors i differet cells. Oce the restrictio list is available to a sector, its scheduler allows trasissios oly o the eligible chuks. The perforace of the proposed schee is copared with that of a referece schee i which itercell coordiatio is ot used ad resource allocatio is perfored i the presece of total iterferece. The reaider of this paper is orgaized as follows. Detailed descriptios of the schee alog with a brief overview of the syste uder study are preseted i Sectio II. Sectio III describes siulatio odels ad syste paraeters. Siulatio results are discussed i Sectio IV. The ipleetatio coplexity issues of the proposed schee are addressed i Sectio V followed by the coclusios i Sectio VI. II. DESCRIPTIONS OF E SCHEME The etwork layout uder study is illustrated i Fig. 1. A total of 19 cell sites, each with 3 hexagoal sectors, are cosidered. Available frequecies are assued to be reused i each sector as per WINNER baselie syste cofiguratio [7]. Sectors are equipped with directioal trasit ateas, while the UTs receive ateas are cosidered to be oidirectioal. The atea gai patter for the trasit atea is discussed i the followig sectio. arked by the saller circle), its first tier sectors 2 as eclosed by the wider circle would be the ost doiat iterferers due to their relative locatios ad atea directivities. A cell edge UT, as show i Fig. 1, experieces higher path-loss ad receives sigificat iterferece fro the sectors of earby cells. Also, UTs closer to the servig sector ay experiece severe iterferece fro eighborig sectors of ow cell. As a cosequece, these UTs are expected to see ore poor-quality chuks havig low sigal-to-iterferece-plus-oise ratios (SINRs). The optial or sub-optial allocatio schees that axiize etwork throughput ay overlook such disadvataged UTs, which are less attractive to cotribute to the total throughput copared to the oes closer to the BS. Therefore, it is very iportat to avoid iterferece o these UTs i order to guaratee their iiu required rates. As etioed earlier, the proposed schee is coposed of two differet algoriths residig at each sector (i.e., BS) ad at a cetral cotroller. The workig priciple of the schee is described by a block diagra i Fig. 2. The algorith at the sector prepares chuk restrictio requests based o the desired ad doiat iterferer chaels as well as the UTs resource deads derived fro the required ad achieved rates. I particular, this is a utility axiizatio proble where the utility easure cosiders the above etioed factors as well as the fairess across the sectors. The obvious questio is that what happes whe a pair of utually iterferig sectors wats to restrict the sae chuk fro each other. I this case, a cetral etity is required to resolve the coflict. The algorith at the cetral cotroller solves request coflicts i a optial aer, where the proble is forulated as a liear costraied proble with biary variables. The details of these algoriths are discussed below. UTs sed chael states icludig doiat iterferece to sector i Sector i prepares utility atrix Sector i solves utility atrix usig iterative Hugaria algorith, prepares ad seds tetative chuk restrictio list to the cetral cotroller Cetral cotroller processes requests fro all sectors i a cluster ad resolves coflicts i a optial way Fig. 1: Network layout uder ivestigatio For the give layout, it is apparet that for a dowlik trasissio to a UT i ay sector (for istace, the sector Algorith at the sector Fig. 2: Block diagra of the schee Algorith at the cetral cotroller 2 Doiat iterferece is received fro sector trasit ateas of differet cells as well as fro its ow cell. Itercell coordiatio iplies that these cells coordiate resource usage restrictios at correspodig sector ateas.

3 A. Sector Algorith Let us deote j ad k as the idices of the first ad secod tier iterferers, respectively, for the UTs i sector i; ad are the idices for UTs ad chuks, respectively. For ay chuk, the SINR see at UT i sector i ca be give by, (), i () Pt H i, ρ, = ()( i, j) ()( i, k ), (1) P + P H + P H N t, t j= 1:6 k= 1: 12 where P t is the trasit power applied o each chuk derived fro equal power distributio; that is, Pt = PT N, where P T is the total trasit power per sector ad N is the uber of available chuks. P N is the theral oise over the chuk ()( i j) badwidth. H, ()( i k ), ad H,, are the first ad secod tier iterferer chaels o chuk at UT i sector i. H reflects the large scale path-loss, atea gai ad directivity, fadig, ad shadowig. The first suatio i the deoiator has the ost doiat effect o the quality of the chuk. The iiu ad axiu rates o chuk, deoted as r, ad r ( i ),, ca be achieved fro (1) whe oe ad all ( ax ) first tier iterferers are restricted, respectively. However, ovig fro r, ( i ) to r, iplies icreasig the pealty to ( ax ) the iterferer sectors, as ore ad ore chuks are to be restricted there. The dead factor, D, is defied as the ratio of the target rate to the average rate achieved by UT i sector i. At the sector algorith, a threshold-based strategy is used to deterie which iterferers are requested to be restricted as described below. Based o its dead factor ad chael coditios, a UT ca restrict two ost doiat iterferers at ost. This liits resultig chuk restrictios i the eighborig sectors ad the coplexity of the cetral algorith. First, a utility atrix U is prepared at each sector. The, the utility atrix is solved usig iterative Hugaria (also kow as Kuh-Mukres) algorith [8] i order to reserve chuks for its UTs. I order to build the utility atrix, the followig steps are repeated for each UT ad chuk. 1) Doiat iterferers are sorted i descedig order ito a doiat iterferer set. 2) r, is calculated fro the SINR expressio i (1) ( i ) cosiderig the presece of all doiat iterferers. a. If r, r ( i ) 1, o iterferers are requested to be restricted if chuk is assiged to UT, irrespective of its dead factor. UT is the either havig a strog desired lik fro servig BS or is experiecig weak iterferece fro all doiat iterferers o chuk. I siulatios, we have used r 3 bps/hz. 1 =, b. Else, calculate the ew rate r, ( ew with the ost ) doiat iterferer beig restricted. i. If D 1 (UT has bee rate-satisfied i the past) ad r, r ( ) r ew, ( i ) 2, UT will request this doiat iterferer to be restricted (here, r2 < r1 ). The above coditios iply that for a rate-satisfied UT, iterferer restrictio is justified oly whe a cosiderable gai ca be achieved. r 2 = 1 bps/hz has bee used i siulatios. () ii. Else if D > 1 ad r, i r 0 ( ), >, ew ( i ) the ost doiat iterferer will have to be restricted. I soe istaces, r, ( ew ight ) be zero. I such a case, the secod ost doiat iterferer would also be restricted if doig so provides ay achievable rate o chuk for UT. After fidig the doiat iterferer(s) to be restricted o each chuk ad each UT, achievable rates r i (), are calculated. Now, the utility of chuk for UT ca be expressed as: = r NC D, (2) U,,,, where NC, is the uber of first tier iterferers that UT allows, yet achieves a rate of r,. I (2), the paraeter NC, is used i order for the sector algorith ot to be greedy (i.e., to esure fairess across sectors). The utility atrix is give by = U. Each etry of U is associated with [ ] U, correspodig iterferer(s) to be restricted as well as achieved rate whe the chuk is used by the respective UT. U is solved usig the iterative Hugaria algorith. This is a efficiet ad low coplexity optial algorith for oe-to-oe assiget proble. However, a UT eeds ultiple chuks i order for it to achieve the required rate, ad the iterative Hugaria ethod becoes a rather good suboptial solutio. Chuks are tetatively allocated (as the cetral cotroller ight override the restrictio requests) i order to reserve chuks for each UT with the followig steps. 1) Iitialize the rate achieved for each UT: R = 0. Iitialize the iterferer restrictio request list to be epty. 2) Apply the Hugaria algorith to U. Update the achieved rates for the assiged chuks, R = R + r,*, where r,* is the rate of optially assiged chuk to UT. If the utility etry has a correspodig iterferer restrictio, the restrictio list will be updated with the etered iterferer.

4 3) Delete the colus of the utility atrix correspodig to assiged chuks fro the utility atrix. Also, if req R R, delete the correspodig row(s) of the req utility atrix. Here, R is the required rate for UT. Now, apply the Hugaria algorith to the updated utility atrix. 4) Repeat steps 2) ad 3) util all UTs rates are satisfied or the resource is exhausted. 5) The restrictio request list is ow ready to be forwarded to the cetral cotroller. B. Cetral Cotroller Algorith The cetral cotroller receives requests fro a cluster of BSs ad resolves coflictig requests i a optial way. For a particular chuk, Fig. 3 shows a exaple proble to be solved at the cetral cotroller usig its algorith. Here, the gree (solid lie) ad red (dashed lie) arrows idicate that iterferece received at the arrow-origiatig-sector fro the arrowhead-sector is acceptable ad uacceptable (requested to be restricted), respectively. For exaple, for a chuk of iterest, sector B ca tolerate iterferece fro sector A, but the opposite is ot true as there is a red (dashed) arrow fro sector A toward B. I this case, either sector A or B has to be restricted for this chuk. Fig. 3: Chuk restrictio request for a particular chuk The followig iteger liear optiizatio proble ca be forulated to resolve request coflicts optially: Maxiize ( j ) U, xi, + U, x j, ;, (3) i, j; j i i B A C j where sector i (sector of iterest) ad sector j (first-tier of doiat iterferer) are havig coflictig chuk restrictio requests for chuk, ad i ad j are cadidate UTs for this chuk i sector i ad j, respectively. subject to xi, + x j, 1;, (4) where ay two sectors i ad j have coflictig restrictio requests. The variables x i ad x j are biary. x i or x j takes a value of 1 if the coflict is resolved i favor of sector i or j, otherwise, it is zero. The cetral cotroller resolves request coflicts ad seds refied restrictio lists to all ivolved BSs. The proposed restrictio processig ca be doe fro tie to tie as log as the chael coherece tie perits. This period is usually uch loger tha the schedulig iterval ad shorter tha chael coherece tie. Oce the chuk restrictio list is available at a sector, the scheduler ca perfor chuk schedulig based o its ow criteria. I that sese, the above chuk restrictio techique ca be cosidered idepedet of schedulig. I this study, we assue the iterative Hugaria algorith for schedulig i both the proposed ad the referece schees. Chuk restrictio processig is perfored o 10 chuks tie iterval ad schedulig is doe at every chuk tie duratio. III. SIMULATION MODELS AND PARAMETERS A total of 19 cell sites (i.e., 57 hexagoal sectors) are cosidered, as show i Fig. 1. The iter-site distace is 1 k. UTs are radoly distributed [9] i the cetre 19 shaded sectors withi a iiu ad axiu radius i each sector. While the sector algorith is executed i these 19 sectors, other sectors reai as iterferece cotributors oly. Perforace statistics are collected fro 7 cetre sectors (iside the wider circle). 45 MHz spectru i the 3.95 GHz frequecy bad is available to each sector givig a total of 1152 subcarriers each of which has a badwidth of khz. A chuk (subchael) cosists of 8 cosecutive subcarriers. It is a tiefrequecy resource uit occupyig s ad KHz, which traslates ito 8 12 OFDM sybols [7]. All UTs are assued to fall i the sae service class havig a dowlik rate requireet of 5.6 Mbps (i.e., 1920 bits per chuk tie duratio). For the siplicity of siulatios, a full queue odel has bee cosidered for the dowlik traffic. Rayleigh chael saples, correlated i tie ad frequecy, are geerated fro the power delay profile for WINNER wide area sceario [9]. UT obility is assued to be 20 k/hr. The followig path-loss (L) odel assuig pathloss expoet of 3.5 has bee used as defied i [9], L log10 ( d ) = [db], (5) where d is distace i eters betwee trasitter ad receiver. While UTs are equipped with oidirectioal receive ateas, the gai patter for directioal sector trasit ateas is cosidered to be as follows [7]: 2 θ A( θ ) = i 12, 20 [db], (6) θ 3dB

5 where the value of θ 3dB is 70 0 ad θ varies fro to We have used a sigle-i-sigle-out (SISO) atea cofiguratio. A oise figure of 7 db is used to calculate average theral oise power. Idepedet logoral rado variables with a stadard deviatio of 8 db are cosidered for shadowig. Trasit power is assued to be Watts per sector ad chuks are assiged fixed equal powers. Retrasissio for erroeous data is ot cosidered. Adaptive odulatio with a block low-desity paritycheck (B-LDPC) code is used. Thresholds for trasissio schees are deteried assuig a block legth of 2304 bits ad 10% block error rate (BLER) as show i Table I [7]. The correspodig water-fall curves were produced fro lik level siulatios. Here, a chuk usig QPSK with a codig rate of 1/2 ca carry 96 iforatio bits. Probability(SINR<=absissa) With coordiatio Without coordiatio CDF of chuk SINR SINR(dB) TABLE I LOOKUP TABLE FOR AMC MODES: B-LDPC WI BLOCK LENG OF 2304 BITS AND BLER OF 10% SINR Rage (db) AMC Mode Efficiecy, η (Bits/Sec/Hz) γ < 0.14 BPSK rate 1/ γ < 1.15 BPSK rate 2/ γ < 3.14 QPSK rate 1/ γ < 4.15 QPSK rate 2/ γ < 6.55 QPSK rate 3/ γ < QAM rate 1/ γ < QAM rate 2/ γ < QAM rate 3/ γ < QAM rate 2/3 4.0 γ QAM rate 3/4 4.5 I order to solve the biary iteger optiizatio proble at the cetral cotroller, YALMIP [10] ad LPSOLVE (a iteger liear prograig (ILP) solver) [11] have bee used alog with MATLAB. Achieved rate (Mbps) Fig. 4: CDF of chuk SINRs o scheduled chuks With coordiatio Without coordiatio Fitted (coordiatio) Fitted (o coordiatio) UT Throughput UT distace () Fig. 5: Scatter plots ad fitted curves of UT throughput IV. SIMULATION RESULTS Perforace results have bee observed for a user desity of 10 UTs per sector. Perforace of the proposed schee is copared with that of a referece schee where itercell coordiatio is ot used. The referece schee copes with itercell iterferece ad axiizes throughput cosiderig UTs rates i the presece of all iterferers. While both schees achieve ultiuser diversity, the gai i the proposed schee is solely due to iterferece avoidace. SINRs are predicted o each chuk fro the pilot sigals. As the referece schee does ot use itercell coordiatio, the predicted SINRs are deteried coservatively based o the assuptio that all other sectors are usig ay particular chuk cocurretly. The cuulative distributio fuctios of the SINRs see o scheduled chuks for both schees are show i Fig. 4. The proposed schee shows a iproveet of 2.5 db i the 50 th percetile of chuk SINRs. However, this SINR iproveet does ot have a coplete eaig without takig throughput ito accout. Probability (Throughput <= absissa) With coordiatio Without coordiatio CDF of UT throughput UT throughput (Mbps) Fig. 6: CDF of UT throughput

6 Fig. 5 shows achieved UT throughput for both schees i scatter plots alog with quadratic-fitted curves. It is observed fro the figure that up to 20% throughput gai ca be achieved at the cell edge UTs i the proposed schee with coordiatio. This iproveet is crucial fro the user satisfactio poit. For exaple, as show i Fig. 6, aroud 90% of the UTs ejoy a data rate of 4 Mbps or ore with the proposed avoidace schee, while oly 77% do whe coordiatio is ot used. V. IMPLEMENTATION COMPLEXITY ISSUES The coputatioal coplexity of the algoriths ad the sigalig overhead coplexity are two differet kids of coplexities associated with the proposed schee. The overall coplexity of the sector algorith is doiated by the coplexity of the Hugaria assiget algorith. This cobiatorial optiizatio algorith solves the assiget proble i polyoial tie with worst-case 3 coplexity of O ( N ), where N is the uber of chuks. For each chuk, the coplexity of the algorith at the cetral cotroller depeds o the uber of sectors that have coflictig restrictio requests. This deteries the uber of variables ad costraits ivolved i the biary iteger proble. For exaple, if there are 3 pair-wise coflictig requests for a particular chuk, the the uber of biary variables to solve is 6 ad the uber of costraits is 9. Moreover, there are efficiet ways to solve such iteger probles. For istace, LPSOLVE first fids a relaxed liear feasible solutio efficietly ad the applies brach ad boud iteratios for the optial solutio i order to reduce the overall coplexity. The aout of sigalig overhead i a certai tie iterval is related to the frequecy of the chael reportig ad resource allocatio operatios, which essetially depeds o the obility of the UTs ad hece the resultig chael coherece tie. For a UT ovig at 20 k/hr, the chael coherece tie is aroud 6 s. I this case, chael reportig ad resource allocatios have to be perfored withi every 18 chuks tie duratio. However, for a speed of 70 k/hr, the tie iterval of this chael reportig would be aroud 5 chuks tie duratio. Therefore, a higher sigalig overhead would have to be supported for higher obility UTs. The sigalig overhead bottleeck would be over-the-air chael reportig fro UTs to BS; the sigalig betwee BS to the cetral cotroller ca be perfored usig high data rate backboe coectios such as fiber liks ad thus it is less of a issue. Furtherore, i order to reduce both the coputatioal ad overhead coplexities further, the proposed schee could potetially be applied to the cell edge UTs oly, for which it is eeded ost. Siulatios with this reduced uber of UTs reai for future study. VI. CONCLUSIONS A ovel viable iterferece avoidace schee usig dowlik ulticell chuk allocatio with dyaic itercell coordiatio is preseted i this paper. Accordig to siulatio results, the schee outperfors the referece schee ot oly i ters of higher total throughput, but also i achievig better satisfactio ad fairess for the cell/sector edge UTs. It should be oted that the referece schee exploits siilar ultiuser diversity as the proposed schee, sice both use the Hugaria algorith for schedulig. The referece schee axiizes throughput cosiderig the utual iterferece ad UTs data rate requireets. Therefore, the observed perforace gai i the proposed schee is solely due to iterferece avoidace. Ehaced cell edge throughput i the proposed schee ca potetially allow a saller uber of BSs to cover a regio yieldig substatial savigs i the deployet cost. I the sector algorith, a siple utility fuctio has bee used, where the cosidered utility is proportioal to the achieved rate o the chuk of iterest, UT resource dead at the allocatio istat, ad the uber of allowed cocurret trasissios i the first tier sectors. However, it ay be possible to devise a better utility fuctio that ay cosider other critical factors for better etwork perforace. Also, i order to reduce coplexities, schee ay be odified to allow oly UTs with poor rate to restrict chuks i the doiat iterferers. These issues reai uder ivestigatio at the tie of writig this paper. ACKNOWLEDGEMENTS The authors would like to express their gratitude to Dr. Abdulkaree B. Adioyi for his review ad useful coets o the auscript. REFERENCES [1] Wireless World Iitiative New Radio, [2] C.Y. Wog, R.S. Cheg, K.B. Letaief, ad R.D. Murch, Multiuser OFDM with adaptive subcarrier, bit, ad power allocatio, IEEE Joural o Selected Areas i Couicatios, vol. 17, o. 10, pp , October [3] W. Rhee ad J.M. Cioffi, Icrease i capacity of ultiuser OFDM syste usig dyaic subchael allocatio, i Proc. IEEE Vehicular Techology Cof., vol. 2, pp , May [4] D. Niyato ad E. Hossai, Adaptive fair subcarrier/rate allocatio i ultirate OFDMA etworks: radio lik level queuig perforace aalysis, IEEE Trasactios o Vehicular Techology, vol. 55, o. 6, pp , Noveber [5] H. Ki, Y. Ha, ad J. Koo, Optial subchael schee i ulticell OFDMA systes, i Proc. IEEE Vehicular Techology Cof., vol. 3, pp , May [6] G. Li ad H. Liu, Dowlik dyaic resource allocatio for ulti-cell OFDMA systes, i Proc. IEEE Vehicular Techology Cof., vol. 3, pp , October [7] Test Scearios ad Calibratio Cases Issue 2, WINNER Deliverable D6.13.7, Deceber 2006, [8] H.W. Kuh, The Hugaria ethod for the assiget proble, Naval Research Logistic Quarterly, vol. 2, o. 1, pp , [9] Fial report o lik level ad syste level chael odels, WINNER Deliverable D5.4, Noveber 2005, [10] J. Löfberg, YALMIP: a toolbox for odelig ad optiizatio i MATLAB, I Proceedigs of the CACSD Coferece, Taipei, Taiwa, [11] M. Berkelaar, K. Eiklad, ad P. Notebaert, lp_solve versio 5.5,