Computer Communications

Transcription

1 Computer Communications 32 (2009) Contents lists available at ScienceDirect Computer Communications journal homepage: Multiple requency reuse schemes in the two-hop IEEE j wireless relay networks with asymmetrical topology q,qq Weiwei Wang a, *, Zihua Guo b, Jun Cai c, Xuemin(Sherman) Shen d, Changjia Chen a a School o Electronics and Inormation Engineering, Beijing Jiaotong University, Beijing, China b Lenovo Cooperate Research, Beijing, China c Department o Electrical and Computer Engineering, University o Manitoba, Winnipeg, Man., Canada d Department o Electrical and Computer Engineering, University o Waterloo, Waterloo, Ont., Canada article ino abstract Article history: Available online 12 February 2009 Keywords: Frequency reuse Relay network Isolation band In this paper, throughput perormance o the access links (i.e., base station to mobile station and relay station to mobile station) is analyzed or the two-hop IEEE j wireless relay networks with asymmetrical topology. In speciic, three requency reuse schemes are proposed to improve the spectrum eiciency o the access links: (1) an isolation band based requency reuse scheme (IBF) which introduces an isolation band surrounding each relay station () cluster (i.e, a separate or several adjacent s) so that the throughput o the access link can be improved by allowing requency reuse between s and the base station (); (2) the dynamic requency power partition (DFPP) scheme or reusing the requency among s; (3) the selective reuse (SR) scheme or the s to urther selectively reuse the requency in the isolation band according to the intererence measurement. Comprehensive simulation shows that by applying the proposed IBF+DFPP+SR, the throughput o the access link can be signiicantly improved. Ó 2009 Elsevier B.V. All rights reserved. 1. Introduction With the increasing demand or ubiquitous multimedia data services, uture wireless cellular networks are expected to provide services with wider coverage range and higher data packet throughput. To achieve these goals, wider bandwidth at higher carrier requency above 2 GHz is oreseen to be used. Since the radio propagation in these requency bands is more vulnerable to non- Line-o-Sight conditions, a new network node, called relay station (), is introduced in wireless networks, which could store and orward data packets received rom base stations (s) to mobile stations (MSs), and vice versa [1 3]. There are generally two advantages brought by : irst, instead o increasing the density o s, adding could overcome the coverage hole o and provide ubiquitous wireless services cost-eiciently; second, adding s could improve network throughput due to possible reuse o q This work was supported in part by the Chinese NSFC under Grant and , Chinese Ministry o Education under Grant and Beijing Jiaotong University under Grant 2005SM006. qq This work is partly presented at Qshine 08. * Corresponding author. Tel.: addresses: wwwang@ee.umanitoba.ca (W. Wang), guozh@lenovo.com (Z. Guo), jcai@ee.umanitoba.ca (J. Cai), xshen@bbcr.uwaterloo.ca (X. Shen), changjia chen@sina.com.cn (C. Chen). radio resources. As a result, relay networks have driven both industry and academia interests recently. Relay-related unction is being standardized as the extension to the basic standards, such as IEEE j [4,5]. Another standard, IEEE802.16m, which is deemed as a potential 4G standard, also supports the application o relay stations [6]. In the IST-WINNER project [7], integrating relay unction has been considered as an inevitable part o the system design or cellular deployment. Furthermore, several studies on relay networks have also been reported. One o main ocuses is on the resource reuse and scheduling, which could be divided into two categories based on dierent network topologies under consideration: (1) symmetrical topology where a is surrounded by several s evenly; and (2) asymmetrical topology where the s are established around randomly. For the irst category, as shown in Fig. 1(a), two-hop network is considered, where the data could be transmitted to the destination by at most one. Li et al. [8] proposes a requency partition scheme, which allocates orthogonal requencies to three kinds o links, to MS, to and to MS; in [9], dierent requency reuse actors (FRF), such as 1, 2, 3 and 6, are proposed among s; similarly, in [10], dierent FRFs are used or s and s, respectively. In [11], with FRF = 1, requency hopping scheme among and s is introduced. For the second category, as shown in Fig. 1(b), multi-hop network is considered, where the data may be transmitted to the destination through at least two s. The area /$ - see ront matter Ó 2009 Elsevier B.V. All rights reserved. doi: /j.comcom

2 W. Wang et al. / Computer Communications 32 (2009) (a) symmetrical topology covered by s is adjacent to that covered by. As a result, the serving area by each could be treated as the traditional cell, and the requency reuse method used in the traditional cellular network could be used in the relay network with minor modiications [12], where a sot requency scheme is proposed with adaptive scheduling among and s. However, in the real world, these regular topologies do not always exist. According to the usage model deined in IEEE j [13], in order to overcome the coverage holes and the shadowing areas o the, the typical topology is that several s are deployed in the serving area irregularly, i.e., the areas covered by s are oten surrounded by rather than adjacent to the area covered by. In order to consider all these new characteristics, in this paper, the requency reuse schemes are studied by taking into account the two-hop IEEE j network with an irregular deployment o s. First, the throughput o the access link (i.e., to MS, or to MS) is analyzed. Then, three requency reuse schemes are proposed: (1) isolation band based requency reuse scheme where an isolation band is deined to surround each cluster which includes a separate or several adjacent s in the serving area, and each cluster could reuse the requency out o the isolation band, this scheme is partly presented in [14]; (2) dynamic requency power partition (DFPP) scheme where the requency could be reused among s in one cluster; (3) the selective reuse (SR) scheme where s could selectively reuse the requency in the isolation band according to the intererence measurement. Through the comprehensive simulation, the numerical results veriy the signiicant throughput improvement rom the proposed schemes. The remainder o this paper is organized as ollows. Section 2 provides an overview o IEEE j. Section 3 gives the throughput analysis o the relay network. Section 4 describes IBF and introduces an analytical method to determine the isolation band. Section 5 introduces the DFPP and SR schemes. Section 6 gives the implementation o the proposed requency reuse schemes in IEEE j network. Section 7 presents the simulation results, ollowed by conclusions in Section Overview o IEEE j (b) asymmetrical topology Fig. 1. The network topology under consideration. orwards data to and rom MSs based on the requency allocation inormation obtained rom. In Fig. 2, the rame structures o and are depicted or two modes, respectively [4]. For non-transparent mode, as in Fig. 2(a), each rame is divided into downlink (DL) and uplink (UL) subrames. Both DL and UL sub-rames consist o one access zone and one relay zone. The access zone is used or the communication between (or ) and the corresponding MSs, while the relay zone is used or the communication between and or between and the subordinate s. Both access zones in rame and rame may share the same resource. For transparent mode, similar and rame structures are deined as shown in Fig. 2(b). In DL sub-rame, transmits data to the corresponding MSs and the subordinate s in the access zone. During this period, s are in the receiving state. In the optional transparent zone o the DL sub-rame, transmits data to the corresponding MSs or the subordinate s; while could be in silent state or provide the cooperative diversity or communicating with its subordinate s and MSs. In UL sub-rame, the access zone is used by MSs or transmitting data to the corresponding and s. However, the zones in rame and rame should use dierent requency bands. In the relay zone, s deliver the data to or their superordinate s. Through the comparison between the two modes, the major dierence lies in that, or non-transparent mode, and s may communicate with the corresponding MSs in the access zone using the same requency bands, while requency reuse is not allowed in the transparent mode. In other words, in non-transparent mode, more resource could be used or improving the system perormance in the access zone. Thereore, in this paper, we will ocus on the resource allocation o the non-transparent mode. In [13], rom the perspective o the inrastructure, especially where the coverage is provided, our usage models are deined as the guideline o drating IEEE j. They are ixed inrastructure, in-building coverage, temporary coverage and coverage on mobile vehicle. Thereore, a typical topology can be ormed, as shown in Fig. 3, where several s are established asymmetrically in each cell to overcome the coverage holes or the shadowing areas where the could not serve. Obviously, with non-transparent mode, when s can use the same resource as the in the access zone, they could obtain the whole system bandwidth to serve the users in their coverage area with little intererence rom the. However, they may cause severe intererence to the users currently rame rame rame DL Access Zone ( to MS) DL Access Zone ( to MS) DL Access Zone ( to MS or ) DL Relay Zone ( to ) UL Access Zone (MS to ) DL Relay Zone UL Access Zone ( to (MS to ) or receiving ) UL Relay Zone ( receiving) UL Relay Zone ( to or to ) DL sub-rame UL sub-rame (a) non-transparent mode Optional transparent zone (silent or transmitting) UL Access Zone (MS to ) UL Relay Zone ( receiving) t t t As the extension o the current standards (IEEE802.16d and IEEE e), IEEE j aims at deining the multi-hop relay speciication including the MAC and the physical (PHY) layers. According to the newest baseline document [4], two modes, nontransparent mode and transparent mode, are speciied to support those application scenarios. The ormer one indicates that the has the scheduling unction; while or the latter one, the just rame DL Access Zone ( receiving) Optional UL Access Zone transparent zone (MS to ) ( to MS or ) DL sub-rame (b) transparent mode UL Relay Zone ( to or to ) UL sub-rame Fig. 2. IEEE j rame structure. t

3 1300 W. Wang et al. / Computer Communications 32 (2009) served by, especially or the users located close to their coverage area. In order to mitigate such intererence rom the s and improve the perormance o the access zone, in this paper, an isolation band based requency reuse scheme (IBF) between and s is introduced. Then, the requency reuse scheme among s and the selective requency reuse scheme between the isolation band and the cluster are proposed to urther improve the system throughput. 3. Throughput analysis In this section, throughput analysis o the relay network is carried out or the access zone, as shown in Fig. 2(a). There are two kinds o links, the access link including to MS and to MS, and the relay link including to. Let their throughputs be T BM ; T RM and T BR, respectively. Then T RM ¼ XN i¼1 T ðiþ RM ; where T ðiþ RM denotes the throughput achieved between i and its corresponding MSs, and N is the number o s in the network. In order to utilize the requency eiciently, the ollowing condition should be satisied T RM t A ¼ T BR t R ; Coverage hole where t A and t R are the durations o the access zone and relay zone, respectively. Otherwise, the buer in could be overlowed i T RM t A < T BR t R, or some resource between and MS could be wasted i T RM t A > T BR t R. Thus, by (2), the eective system throughput could be denoted as Shadowing area Fig. 3. The typical topology o relay network. ð1þ ð2þ T BR ðt BM DT BM Þ T A þ DT þ T BR T BM < 0 T A A ) T ADT BM DTþ A ðt BR T BM Þ T A ðt A þ DT þ A Þ < 0: ð5þ Thereore, DT BM DT þ A < T BR T BM T A : ð6þ The above inequality indicates the conditions o decreasing. The right-hand side o (6), as the original, could be regarded as a ixed value; thereore, the let-side o (6) should be as small as possible. In other words, some schemes should be designed to make sure that less decrease on T BM and more increase on T A would be achieved. From these two situations, we could conidently conclude that increasing T A has positive eects on improving the eective system throughput. Thereore, the ollowing discussion will ocus on improving the throughput o the access link, T A. 4. Isolation band based requency reuse scheme 4.1. Introduction o IBF For simplicity, hexagonal cells are used to denote the areas covered by and. Deine the cluster which denotes a separate or several adjacent s. Because our ocus is on the requency reuse between and s, we assume all s in a cluster serve same MS simultaneously to provide macro-diversity. The IBF is illustrated by considering the DL. However, the similar idea can be applied or UL. As shown in Fig. 4, the IBF ollows three rules: each cluster is surrounded by an isolation band; the users in the isolation band are served by ; the s in the cluster can reuse the resource, which is not used by the users in the isolation band. In Fig. 4, the whole coverage o the is separated into three subareas, which are the area covered by the cluster (-area), an isolation band, and the rest area called reuse-area. The -area is surrounded by the isolation band. Users locating in both isolation band and the reuse-area can access the whole system Isolation band T sys ¼ T BMt A þ T BR t R t A þ t R ¼ T BMt A þ T RM t A t A þ t R ¼ T BR T BM þ T RM T BR þ T RM : Let T A ¼ T BM þ T RM, which means the throughput o the access link, (3) can be rewritten as T sys ¼ T BR 1 þ T BR T BM T A : ð4þ ð3þ cluster Ater the deployment o s, the relay link is determined; thus, T BR in (4) could be regarded as a constant. Thereore, decreasing T BR T BM T A is the best way to improve the eective system throughput. Assume ¼ T BR T BM T A, the ollowing situations should be noticed. 1. T BM is not decreased (ixed or increased). Apparently, i T A could be increased by some schemes, is deinitely decreased. 2. T BM is decreased. Assume T BM is decreased by DT BM ð> 0Þ and T A is increased by DT þ A ð> 0Þ. Then, the ollowing condition should be satisied or decreasing = = Fig. 4. Illustration o IBF. System Bandwidth

4 W. Wang et al. / Computer Communications 32 (2009) bandwidth or transmission, while only the requency band used in the reuse-area can be exploited in the -area. As a result, the users served by the and close to the edge o the cluster would not be interered by the s therein, and the intererence is mitigated or the users out o the isolation band even s reuse their requency due to the large distance rom the s Determination o isolation band Determining the isolation band o the cluster is the key or the perormance o the proposed IBF. In this subsection, an analytical method is introduced or isolation band determination based on the system model deined in Fig. 4. (1) Deinition o variables S: the total acreage o the cell; A : the area served by the with the acreage o S ; A c : the area served by the cluster with the acreage o S c ; A nr : the isolation band with the acreage o Snr ; A r : the area served by the but out o the isolation band, which has an acreage o S r ; B: system bandwidth; C (bit/s/hz): the spectrum eiciency o A when no requency reuse between the and the cluster; C c (bit/s/hz): the spectrum eiciency o the cluster. From the deinitions, we have S þ S c ¼ S; S r þ Snr ¼ S : (2) Optimal isolation band ð7þ ð8þ Assume the users are uniormly distributed in A. Let h 2½0; 1Š be the decrease o the spectrum eiciency o A r ater the s reuse the requency rom the reuse-area. The throughput o A, denoted as T,is T P Snr ðhþ B C þ Sr ðhþ B C ð1 hþ; ð9þ S S where Snr ðhþ S B is the requency used in the isolation band and Sr ðhþ S B is the potential requency band, which could be reused by the cluster. For the worse-case scenario, where all potential reusable requency bands are applied by the cluster, the throughput o the cluster ater the requency reuse is given as T c ¼ Sr ðhþ S B C c : ð10þ Thereore, the system throughput o the DL access zone is T P T c þ T ¼ Sr ðhþ B C c þ Snr ðhþ B C þ Sr ðhþ B C ð1 hþ S S S ¼ C B þ Sr ðhþ BðC c C hþ: ð11þ S Obviously, maximizing the right-hand side o (11) could improve the throughput o the access link. Thereore, the optimal value o h should satisy h opt ¼ argmax h2½0;1š C B þ Sr ðhþ BðC c C hþ S : ð12þ To derive C ; C c and S r ðhþ in (12), we consider a cellular system with 19 cells and let the central one is the home cell. First, we calculate C in (12). Assume a location, x,ina o the home cell. The signal to intererence plus noise ratio (SINR) at x is given as c x ¼ P L v Pm2I P ðmþ LðmÞ v ðmþ þ g ðx 2 A Þ; ð13þ where I is the set o the intererence sources to x in two-tier cells. In both numerator and denominator o (13), P ðdþ is the transmitting power o in cell d; L ðdþ and vðdþ are the path loss and shadowing rom in cell d to x, respectively. In general, the latter one in db is modeled as a lognormal variable. g means the noise which is ignored in the ollowing analysis due to its smaller value compared to the intererence. According to [15], c x in (13) can be approximated by a lognormal variable with mean lðx Þ and standard variance rðx Þ in db. Thereore, the cumulative density unction (CDF) o c x is given as! F x ðcþ ¼Pðc x < cþ ¼1 Q 10log 10ðcÞ lðx Þ rðx Þ ; ð14þ where QðxÞ ¼ R 1 p1iiii exp x 2p Z2 dz. Given CDF, the probability density unction (PDF) o SINR, x 2 ðcþ, can be obtained by dierentiating F x ðcþ. Let the relationship between the SINR and the throughput in unit bandwidth be m ¼ gðcþ; ðm 2 VÞ, where V is the set o possible values o m. Then, the PDF o the throughput in unit bandwidth at x is T x ðmþ ¼ Z c2u x ðcþdc ðm 2 V; U ¼cjm ¼ gðcþgþ: ð15þ From (15), the average throughput in unit bandwidth at x T x ¼ X m2v mt x ðmþ: Finally, C is derived as I C ¼ Pðx Þdx ; A T x is ð16þ ð17þ where Pðx Þ is the probability that the user is located at x. For C c in (12), let a location o x c in A c o the home cell. Then, its SINR could be written as c x c ¼ Pn2X PðnÞ c LðnÞ cv ðnþ c Pm2I c P ðmþ c LðmÞ cv ðmþ c þ g c x c 2 A c ; ð18þ where X is the set o s in the cluster, and I c is the set o the intererence sources to x c. Due to the requency reuse between the and the cluster, I c should include the which covers the cluster and s in two-tier cells. Here, the intererence rom the cluster in other cells is ignored due to the relatively small transmitting power o the. By the similar method used or C, we could derive C c as I C c ¼ Pðx c Þdx c ; ð19þ where T x Pðx c A c T x c c P m2i r is the average throughput in unit bandwidth at x c and Þ is the probability that the user is located at x c. Similarly, or S r ðhþ in (12), consider a location o x;r in A r o the home cell. Since the resource used by users located at x ;r is reused by the cluster, its SINR is P c ;r L ;r v ðnþ;r ;r x ¼ x ;r P ðmþ;r L ðmþ;r v ðmþ;r þ g ;r 2 A r ; ð20þ where I r is the set o intererence sources to x;r, which includes s in two tiers and the s in the cluster which is within the same coverage. The intererence rom the s in the other cells is ignored. Likewise, the average throughput in unit bandwidth at x ;r is given as

5 1302 W. Wang et al. / Computer Communications 32 (2009) T ;r x ¼ X m2v mt ;r x ðmþ; ð21þ where T ;r x is the PDF o the throughput in unit bandwidth at x ;r ater the resource is reused by cluster. Thereore, A r is given by 8 9 < T A r ¼ x;r jx ;r 2 A ; x ;r ¼ x ; x T ;r x = < h : ; : ð22þ T x Eq. (22) indicates that at any location in A r, the average throughput in unit bandwidth is decreased by no more than h ater the requency reuse between the and the cluster. Combining (12), (17), (19) and (22), the optimal h and the corresponding isolation band can be obtained. 5. Further discussion In Section 4.1, we introduce macro-diversity among s in one cluster. To urther improve the throughput o the access link, other two requency reuse methods are discussed as ollows: in each cluster, requency could be reused among s; there is the possibility o reusing the requency in the isolation band. Thus, in this section, the above two aspects would be urther discussed Dynamic requency partition scheme among s Due to the similarity between the s in the traditional cellular network and the s in one cluster, the requency reuse scheme used or the s may be applied to the s. As a new requency reuse scheme, sot requency reuse (SFR) [16,17] has attracted lots o interest, which divides the requency o each cell into two power levels, and the high-power requency o the adjacent s should be orthogonal. The main advantage o SFR is that FRF = 1 could be applied among s and the co-channel intererence o the edge area o each cell could be mitigated. Further discussion could be ound in [18]. However, due to the possible uneven service distribution, the coordination among s is needed to change the ratio o the high-power requency to the low-power one, which brings high signal overhead. Thus, the application o the dynamic SFR in practical scenarios is still under study. On the contrary, in the relay network, since the service requirement o each serving area is known to the, the SFR could be implemented among s dynamically. Based on this act, we propose a dynamic requency power partition scheme (DFPP) or the requency reuse among s in the cluster, which is elaborated as ollows: the requency o each serving area is divided into two parts: primary requency and secondary requency; the primary requency, which is orthogonal among adjacent s, is transmitted by the high power level; while the secondary requency is transmitted by the low power level; the ratio o the primary requency to the secondary requency could be adjusted dynamically according to the service requirement o each serving area Determination o the primary requency Let G ¼ðV; EÞ denote an intererence graph, where the node set V denotes s, and the edge set E represents geographical proximity o serving area and thereore the possibility o co-channel intererence. Here, due to the small transmission power o s, only intererence among adjacent s is considered. Thereore, the s in one cluster orm a weighted graph ðg; xþ, where G is an intererence graph and x is a weight vector indexed by the nodes o G, and xðtþ represents the bandwidth requirement o node t. Assume C subchannels are included in the system. Since the orthogonality o the primary requency o the adjacent s is required, a proper multicoloring o G is needed to allocate the primary requency. In this paper, the method in [19] could be used. Since a hexagonal cell is used to denote the serving area, the G is a 3-colorable graph. Thereore, serving areas in one cluster could be represented by three colors: red, blue and green. Deine the nominal subchannel o each colored area as one third o the system bandwidth, then three steps should be ollowed to allocate the primary requency o each serving area: 1. allocate the nominal subchannel to the corresponding serving area; 2. i there are serving areas whose bandwidth requirements are not satisied, the red (blue/green) serving areas borrow the nominal subchannels which are not used by all the adjacent blue (green/red) serving areas; 3. i there are still serving areas whose bandwidth requirements are not satisied, the red (blue/green) serving areas borrow the nominal subchannels which are not used by all the adjacent blue and green (green and red/ red and blue) serving areas Determination o the secondary requency I the bandwidth requirement o the serving area is not satisied, the secondary requency should be allocated. Obviously, the secondary requency o the serving area could interere with the primary requency o the adjacent serving areas; thereore, it should be the requency causing ewer intererence. Taking a serving area as the reerence, denoted by k, the requirements o its primary and secondary requencies are supposed to be jpf k j and m k subchannels, respectively; thus jpf k jþm k 6 C; ð23þ where C is the number o the system subchannels. Then, the ollowing two situations should be discussed: 1. I m k ¼ C jpf k j, there is no possibility to reduce the intererence to the adjacent serving areas since all o the rest subchannels should be used as the secondary requency in k. 2. I m k < C jpf k j, there is the extra requency ater the requency requirement o k is satisied, which means there are several choices when allocating the secondary requency. Thereore, in order to reduce the intererence to the adjacent serving areas as much as possible, the secondary requency o k should be chosen rom the subchannels, which are used as the primary requency by the adjacent serving areas, but with the descending order o the number o the adjacent serving areas which use them Allocation o the requency reused by IBF Ater determining the primary requency and the secondary requency o each serving area, the s in the cluster could be divided into several reuse groups, each o which includes the s reusing the same subchannel. Then, the subchannels reused by IBF could be allocated to the reuse groups one by one according to the descending order o the number o the s in each reuse group.

6 W. Wang et al. / Computer Communications 32 (2009) Selective reuse scheme between the isolation band and the cluster Obviously, ater the allocation o the requency reused by IBF, there are still some serving areas under bandwidth requirement. An interesting problem is whether there is any space to reuse more requency? The answer is positive. As mentioned in Section 5.1.3, some reuse group may include ew s; thereore, it could cause small intererence to the users in the isolation band. For instance, as in Fig. 5, due to the long distance, the subchannel used by user A may be reused by the in the reuse group 2. Thereore, based on the intererence measurement, the reuse groups could be sorted by the ascending order o the intererence to each user in the isolation band. Then, in order to guarantee the airness o the users served by, the requency used by the user in the isolation band could be reused by the reuse group which causes the throughput decreasing less than h opt. 6. Implementation o the proposed requency reuse schemes In each cell, when a cluster is established, the corresponding isolation band can be determined through the method deined in Section 4.2 and could be maintained or a long period unless the cluster is changed. Ater that, the requency schemes mentioned above could be implemented rame by rame. Through the rame structure in Fig. 2, since and work at the same time in the DL access zone, in each rame, the should schedule the requency o the next rame and inorm the in the relay zone o the current rame what the reusable resource is in the next rame. In summary, the implementation at each rame ollows three steps: 1. schedules the resource o the next rame and pre-allocates the requency o each according to DFPP; 2. inds out the users in the isolation band o the cluster by the location technology, such as GPS (Global Position System), and then allocates the reusable resource o the cluster in the next rame, which includes the requency used in and out o the isolation band; 3. inorms the s in the cluster the reusable resource o the next rame through the message in the relay zone. Compared to the existed schemes, the additional implementations o our scheme are ixing the isolation band by a long period Isolation band and pre-scheduling the resource o the next rame. Thus, the complexity o the proposed scheme is acceptable. 7. Simulation and discussions An OFDM cellular network with 19 cells is considered in both numerical analysis and simulation. Sot requency reuse scheme is used among cells. The main simulation parameters are listed in Table 1. According to (22), we irst study the isolation band o the cluster. s are placed in the cell randomly and at most 10 s are included in a cluster. Fig. 6 shows an example o the isolation band (A nr ) o a cluster and its corresponding h. The area enclosed by the contour except the coverage area o the cluster is the isolation band and the number on the contour denotes the corresponding h. Obviously, with the decrease o the area o the isolation band, h is increased since the cluster will interere with the users out o the isolation band more severely. Fig. 7 shows the average ratio o S r to S with respect to the number o s in the cluster. It can be seen that the ratio is decreased with the increase o the number o s. That is because the intererence to the users served by the increases as the number o s increases. Nevertheless, the ratio is still above 80% which means most requency could be reused by the cluster. We compare the proposed scheme with a traditional scheme where the and the s in the cluster use dierent requency bands to serve the respective users [8] by extensive simulations with Matlab. For simplicity, each user is allocated one subchannel, so that both the and the cluster could serve 30 users at most. During the simulation, 20 kinds o topology are emulated. For each topology, the clusters are deployed randomly. The number o clusters in each cell and the number o s in each cluster are uniormly distributed in [1, 3] and [1, 5], respectively. One hundred samples are simulated or each topology, and the steps or user generation in each sample are as ollows: 1. Generating 30 users distributed randomly in the cell (including serving areas) or the traditional scheme. 2. Based on the users in step 1, or IBF, increase the number o users in the serving area (except serving areas) to 30, while the number o users in each cluster is uniormly distributed in [1, 30]. 3. Based on the users in step 2, or all the schemes in this paper, the number o users in each serving area is increased to be uniormly distributed in [1, 30]. In step 1, due to no requency reuse in the traditional scheme, 30 users means the system is ully loaded. However, actually, the reason o the deployment o the is that more users can be sup- Reuse group 1 Reuse group 2 Table 1 Simulation parameters. User A Fig. 5. Example o reusing the requency in the isolation band. Carrier requency 2:5 GHz System bandwidth 10 MHz Number o subchannels 30 Number o sub-carriers in a 24 subchannel cell radius 1 km TX power 38dBm (high); 33dBm (low) path loss 138:6 þ 34:79log 10 ðdþ, d is the distance in km cell radius 0:1km TX power 5dBm (high); 1dBm (low) path loss 143:69 þ 37:2log 10 ðdþ, d is the distance in km Shadowing Lognormal variable with mean 0 db and standard variance 8 db Modulation and coding scheme See IEEE e [20]

7 1304 W. Wang et al. / Computer Communications 32 (2009) The average area ratio Y(Km) x (Km) Fig. 6. An example o the isolation band Number o s in cluster Fig. 7. The average area ratio o S r to S. ported in the serving areas; thus, in step 2, by considering the requency reuse in IBF, 30 users could be served by ; while 30 users could be served by each cluster at most due to the assumption o the macro-diversity. Finally, in step 3, due to the urther requency reuse among s, more users could be generated in each -area. Fig. 8(a) (d) gives the average throughput o the system, serving area, serving area and the isolation band, respectively, in the access zone. In the igures, IBF+DFPP+SR means the schemes including IBF, DFPP and selective reuse scheme between the isolation band and cluster. For the traditional scheme, since only 30 users could be supported in the whole cell, there is only one point in Fig. 8(a), and in Fig. 8(b) and (c), the number o users in the and serving areas would not exceed 30, respectively. From Fig. 8(a) and (c), due to requency reuse between and clusters in IBF, more users could be served by s. Further, in IBF+DFPP+SR, requency reuse among s and the selective reuse between the isolation band and the cluster are permitted; thus, the throughput is improved signiicantly with the increase o users. In Fig. 8(b), only one point corresponding to 30 users is illustrated or IBF and IBF+DFPP+SR because the number o users served by the has reached the maximum o the system. Obviously, the throughput o serving area in IBF and IBF+DFPP+SR decreases a little because o the requency reuse by the users in the cluster, and since the selective reuse scheme causes the intererence to the users in the isolation band, the value in IBF+DFPP+SR is smaller than that in IBF. In Fig. 8(d), the throughputs o the isolation band in IBF and IBF+DFPP+SR are also decreased a little. As mentioned beore, in the simulation, each cell may include several clusters. Since the requency used by one isolation band may be reused by other clusters, the throughput o the isolation band is lower than that in the traditional scheme; moreover, due to the selective reuse scheme, the throughput o the isolation band in IBF+DFPP+SR is the smallest. In summary, through IBF and IBF+DFPP+SR, the throughput o the access link is improved largely with little negative inluence to the users served by the link o to MS, and IBF+DFPP+SR brings the largest improvement. 8. Conclusions In this paper, three requency reuse schemes or the two-hop relay network based on IEEE j has been proposed. By introducing an isolation band or each cluster, the proposed isolation band a Average system throughput(mbps) Tradition IBF IBF+DFPP+SR Number o users in the system b Average throughput o serving area(mbps) Tradition IBF IBF+DFPP+SR Number o users served by c Average throughput o serving area(mbps) Tradition IBF IBF+DFPP+SR d Average throughput o the isolation band(mbps) Tradition IBF IBF+DFPP+SR Number o users served by Number o users in the isolation band Fig. 8. Throughput comparison.

8 W. Wang et al. / Computer Communications 32 (2009) based requency reuse (IBF) scheme allows each cluster reuses all requency resources out o the isolation band. The DFPP scheme deines how to use the reused requency eectively among s, and the selective reuse scheme utilizes the possibility o reusing the requency in the isolation band. The simulation results indicate that the proposed IBF+DFPP+SR scheme can signiicantly improve the throughput o the access links with little negative inluence to other users served by the. Our uture work will ocus on more complex scenarios, such as networks with directional antennas. Reerences [1] R. Pabst et al., Relay-based deployment concepts or wireless and mobile broadband radio, IEEE Commun. Mag. 42 (9) (2004) [2] J. Cai, X. Shen, J.W. Mark, A.S. Ala, Semi-distributed user relaying algorithm or ampliy-and-orward wireless relay networks, IEEE Trans. Wireless Commun. 7 (4) (2008) [3] J. Cai, A.S. Ala, P. Ren, X. Shen, J.W. Mark, Packet level perormance analysis in wireless user-relaying networks, IEEE Trans. Wireless Commun., in press. [4] IEEE j-06/026r4, Baseline document or drat standard or local and metropolitan area networks, [5] < [6] IEEE m-07/002r4, IEEE m system requirements, [7] < [8] P. Li, M. Rong, Y. Xue, E. Schulz, Reuse one requency planning or two-hop cellular system with ixed relay nodes, in: Proc. IEEE WCNC 07, Hong Kong, China, March [9] P.W. Hyoung, B. Saewoong, Resource management policies or ixed relays in cellular networks, in: Proc. IEEE GLOBECOM 06, San Francisco, USA, November [10] P. Li, M. Rong, T. Liu, Y. Xue, D. Yu, E. Schulz, Reuse partitioning based requency planning or relay enhanced cellular system with nlos bs-relay links, in: Proc. VTC-2006 Fall, Montreal, QC, Canada, September [11] O. Mubarek, H. Yanikomeroglu, Dynamic requency hopping in cellular ixed relay networks, in: Proc. IEEE VTC 2005-Spring, Stockholm, Sweden, May [12] K. Doppler, X. He. C. Wijting, A. Sorri, Adaptive sot reuse or relay enhanced cells, in: Proc. IEEE VTC 07-Spring, Dublin, Ireland, April [13] IEEE j-06/015, Harmonized contribution on j (mobile multihop relay) usage models, [14] W. Wang, Z. Guo, J. Cai, X. Shen, C. Chen, Isolation band based requency reuse scheme or IEEE j wireless relay networks, in: Proc. Qshine 08, Hongkong, China, July 28 31, [15] W. Choi, J.G. Andrews, Downlink perormance and capacity o distributed antenna systems in a multicell environment, IEEE Trans. Wireless Commun. 6 (1) (2007) [16] 3GPP R , Sot requency reuse scheme or UTRAN LTE, Huawei, May [17] 3GPP, R , Further analysis o sot requency reuse scheme, Huawei, August [18] W. Wang, Z. Guo, X. Shen, C. Chen, J. Cai, Flow-satisaction-degree based scheduling algorithm in static reuse partitioning system, in: Proc. IEEE Globecom 07, Washington, DC, USA, November 26 30, [19] Lata Narayanan, Channel Assignment and Graph Multicoloring, Handbook o Wireless Networks and Mobile Computing, John Wiley and Sons, [20] IEEE Standard , IEEE standard or local and metropolitan area networks Part 16: Air interace or ixed and mobile broadband wireless access systems, 2005.