Traffic Adaptive Formation of mmwave Meshed Backhaul Networks

Transcription

1 Traffc Adaptve Formaton of mmwave Meshed Backhaul Networks Hroak Ogawa Tokyo Insttute of Technology Ga Khanh Tran Tokyo Insttute of Technology Ke Sakaguch Tokyo Insttute of Technology / Fraunhofer HHI ke.sakaguch@hh.fraunhofer.de Thomas Hausten Fraunhofer HHI thomas.hausten@hh.fraunhofer.de Abstract MmWave meshed network s a promsng archtecture for cost-effcent wreless backhaul of mllmeter-wave overlay heterogeneous network (mmwave overlay HetNet). As user dstrbuton n practce s tme-varant and spatally non-unform, mmwave meshed backhaul should be controlled adaptvely. Ths paper proposes a novel method to control mmwave meshed backhaul for effcent operaton of mmwave overlay HetNet. Our algorthm s featured by two functonaltes,.e. backhaulng route multplexng for overloaded mmwave small cell base statons (SC- BSs) and mmwave SC-BSs ON/OFF status swtchng for underloaded spot. Consderng practcal user dstrbuton, rado backhaul resources should be concentrated on overloaded mmwave SC-BSs. Inversely, underloaded mmwave SC-BSs should be deactvated for savng power. The performance of mmwave meshed backhaul controlled by the proposed algorthm s evaluated by system level smulaton. Numercal results show that the proposed algorthm can cope wth the locally ntensve traffc, network scalablty, and can reduce energy consumpton. I. Introducton Nowadays as moble termnals lke smartphones or tablets become commonplace, the amount of moble traffc has been ncreasng exponentally. Accordng to [1], the ncrease speed s about 53% per year, and the ncrease wll surely contnue. In order to keep up wth ths growth, 5G communcaton network s requred to support enhanced moble broadband (embb) servces. One of promsng archtectures to realze embb s mllmeter-wave overlay heterogeneous network (mmwave overlay HetNet) proposed n [2],[3]. In mmwave overlay HetNet, mmwave small-cell Base Statons (SC-BSs) are deployed n the coverage of a macro BS. [4] shows that t can acheve 1 tmes hgher capacty compared wth conventonal systems on the assumpton that all SC-BSs have deal backhaul. However, t s extremely costly to equp deal backhaul such as optcal fbers, because a large number of SC-BSs are requred to be ntroduced n a macro cell [4]. One of possble solutons to reduce CAPEX (captal expendture) s wreless backhaul. The capacty of wreless backhaul should be larger than that of access, thus [5],[6] attempted to use mmwave band for backhaul. As mmwave wreless lnks wth hghly drectonal antennas can be modeled as pseudo-wred lnks, we can operate mmwave meshed backhaul especally n dense deployment of mmwave SC-BSs. As one way to operate mmwave meshed backhaul, [7] ntroduced anchored BSs that have wred backhaul lke fbers as a gateway for other BSs. However, [7] dealt wth only sngle-hop scheme.e. wthout consderng any relay. Accordng to [8], user dstrbuton n practce s tme-varant and spatally non-unform. Therefore, t occurs n practcal scenaros that there wll be some overloaded and underloaded mmwave SC-BSs. The overloaded SC-BSs need much rado backhaul resources to satsfy huge traffc demand. For such case, backhaulng route multplexng,.e. concentraton of rado backhaul resources on the overloaded spots, s effectve through the method of load balancng as conducted n [9]. In another aspect, [1] showed that power consumpton of ICT (nformaton and communcaton technology) ndustry s ncreasng up to 3% of power consumpton of the world. In addton, 9% of the amount are consumed by wreless network BSs, 4G LTE or W-F access ponts [11]. Therefore t s crucal to manage power consumpton of BSs to realze energy-effcent wreless network systems. As there are some underloaded SC-BSs n mmwave overlay HetNet, adaptve swtchng of BSs ON/OFF status are effectve as conducted n [12],[13]. These works consder power consumed by only RAN (Rado Access Network). Another work tred to reduce power consumpton of wred backhaul by solvng flow optmzaton problem n graphcally modeled network [14]. However, [14] does not consder jont power consumpton optmzaton of RAN and backhaul, and also does not consder drecton selectvty as assumed n the case of mmwave backhaul lnks. As we can use drectonal antennas for mmwave meshed backhaul networks, combnatoral optmzaton of the drectons of mmwave lnks should be requred n addton to flow optmzaton. However, there are too many canddates of lnk drecton combnaton, thus t s mpossble to fnd the best combnaton of mmwave backhaul lnks and SC-BSs ON/OFF status for the best lnk drecton combnaton /17/$ IEEE

2 LTE macro BS mmwave SC-BS Mult-Hop scheme Sngle-Hop scheme C-plane U-plane U-plane mmwave Gateway mmwave overlay HetNet Fg. 1. C/U splttng archtecture of the evaluated network. For these reasons, ths paper proposes an algorthm to control mmwave meshed backhaul,.e. fndng a heurstc combnaton of mmwave backhaul lnks, SC- BSs ON/OFF status, and lnk drectons. The assumed scenaro s a large hotspot (an overloaded spot) lke a scramble ntersecton n front of Shbuya staton n Tokyo. The purpose s to satsfy user s traffc demand whle reducng network power consumpton. Our algorthm wll control mmwave backhaul lnks and mmwave BSs ON/OFF status so that the overloaded mmwave SC-BSs are allocated large backhaul rado resources and nversely the underloaded mmwave SC-BSs are deactvated to reduce power consumpton consderng mult- RAT (Rado Access Technology) selectvty of mcrowave LTE and mmwave network. To the best of our knowledge, the proposed method s the frst work on controllng both of wreless backhaul lnks and ON/OFF status swtchng n mmwave meshed network consderng drecton selectvty of mmwave backhaul lnks. In addton, system level smulaton s conducted to evaluate the performance of mmwave meshed backhaul controlled by the proposed algorthm. The nterval of network swtchng control s the order of seconds correspondng to the perod of change of user dstrbuton. Investgaton on specfed transmsson protocols for mmwave backhaul meshed network are out of scope of ths paper. II. Adaptve Backhaul Control A. Network topology We employ mmwave overlay HetNet shown n Fg. 1 as a network topology. In the mmwave overlay HetNet, LTE s assumed to manage the C-plane nformaton,.e. user s locaton, movement, and traffc demand. Ths paper ntroduces one mmwave gateway as a source base staton for mmwave SC-BSs, and focuses only on downlnk communcatons. B. Wreless Bakchaul As ths paper assumes tme-varant and spatally nonunform user dstrbuton, wreless backhaul should be controlled adaptvely n accordance wth traffc dstrbuton. Relay scheme can realze adaptve backhaul control and can conduct backhaulng route multplexng. Ths secton presents two types of wreless backhaul schemes, Receve Backhaul Access Relay Fg. 2. Sngle-Hop scheme and Mult-Hop scheme..e. Sngle-Hop and Mult-Hop schemes summarzed n Fg. 2. 1) Sngle-Hop scheme: In ths scheme, there are only drect connectons between gateway and SC-BSs, thus all backhaul lnks are fxed. If locally ntensve traffc demand exsts, ths scheme wll fal to cope wth such scenaro due to lmted capacty of each wreless backhaul lnk. In addton, ths scheme s napproprate for a large coverage network because the effect of path loss attenuaton s nherent n mmwave lnks. 2) Mult-Hop scheme: In Mult-Hop scheme, gateway s connected to each SC-BS not only drectly but also ndrectly wth relay scheme. Ths scheme can operate adaptve topology backhaul, and also can conduct backhaulng route multplexng on an arbtrary spot. Furthermore, relay scheme can compensate path loss attenuaton by amplfcaton when sgnals are relayed. For stable communcatons and ease of analyss, ths paper allows Mult- Hop scheme to be formed among only lnks whch can acheve maxmum data rate of IEEE 82.11ad standard. C. Proposed Algorthm In order to take advantage of Mult-Hop scheme descrbed n the prevous secton, we have to control relay properly. Ths secton presents an algorthm to determne the approprate backhaul lnks and SC-BSs ON/OFF status n accordance wth traffc dstrbuton. As t s hard to determne backhaul lnks and SC-BSs ON/OFF status all at once, ths algorthm s dvded nto three steps summarzed n Fg. 3. In the descrpton of the proposed algorthm, mmwave gateway and SC-BSs are called smply GW and AP. () Determne tentatve ON/OFF status based on traffc dstrbuton: Step () determnes tentatve ON/OFF status of each AP consderng mult-rat selectvty of mcrowave LTE and mmwave network and the goal s to reduce the total

3 () Determne tentatve ON/OFF based on traffc dstrbuton Receve from 2nd sector x 2 { Constrants x1+ x2 = T x1, x2 CS () Form backhaul lnks through mmwave meshed network T () Reactvate APs for relay (f necessary) Fg. 3. Algorthm flow chart. power consumpton of mmwave network as much as possble. In order to mnmze the total power consumpton, LTE should accommodate as many users as possble wthn ts avalable bandwdth B LTE and underloaded APs should be set OFF. As t s complcated to consder each user ndvdually, all APs are actvated at frst and all users are accommodated by ther nearest APs. Then AP has an aggregated traffc demand T. If T can be nstead accommodated by macro LTE, LTE needs to allocate some bandwdth b gven by Shannon s capacty as follows. b = T / log 2 (1 + γ ) (1) γ s the approxmated SINR (Sgnal to Interference plus Nose power Rato) of sgnals from LTE macro BS to AP consderng only path loss attenuaton. Therefore, n order to determne tentatve ON/OFF status of AP, we have only to determne whch system of LTE or mmwave network should accommodate T. When we defne G k as a state that users around AP are accommodated by the k-th sector of LTE macro BS, the problem to be solved s as follows. for k = 1, 2, 3 fnd: group G k s.t. maxmze: G k subject to: G k b B LTE where G k expresses the number of APs ncluded n G k. As a result, f T s accommodated by LTE, the correspondng users around AP wll be accommodated by LTE, and AP can be set OFF to reduce power consumpton. If AP s set ON, all the 3 sectors for AP s access structure wll be actvated regardless of the number of users n the coverage of AP and the user locaton. In the followng steps, we shall focus on only mmwave meshed network. () Form backhaul lnks through mmwave meshed network: In step (), mmwave backhaul lnks are formed among APs that are set ON n step () to satsfy user s traffc demand. In order to form backhaul lnks, we have to determne approprate backhaulng routes from any sector of GW to AP. If the combnaton of sources and x 1 Receve from 1st sector Locally ntensve traffc load Fg. 4. The concept of backhaulng route multplexng. destnatons s gven, graph-theoretcal expresson of AP and connectvty between APs as node and edge enables us to fnd the shortest path easly. Thus the problem of backhaulng route can be substtuted for another problem to fnd the optmal combnaton of the source sector of GW and destnaton AP. Fgure 4 presents a smple example to derve the combnaton of transmtter sector of GW and recever AP. The flled cell s a hotspot and has an aggregated traffc demand T that s larger than the capacty of one sector of GW. In such a case, we have to conduct backhaulng route multplexng on the hotspot from several sectors of GW to satsfy user s request as the load balancng conducted n [9]. To cope wth the locally ntensve traffc T, we consder a smple soluton that one sector of GW s assgned x 1 of T and the other s assgned x 2 of T. A reasonable way to determne the value of x 1 and x 2 s to mnmze Total Hop defned as Total Hop = Data Hop, where Hop s the dstance of the shortest path to transmt Data from transmtter to recever. And x 1, x 2 should satsfy three constrants below. [A] x 1, x 2 C S [B] x 1 + x 2 T [C] x 1, x 2 where [A] means the capacty of each sector of GW, [B] ensures satsfacton of user s request, [C] assures that the value of traffc s not negatve. If ths example s generalzed to a whole network, a lnear programmng as follows s defned. fnd: x s.t. mnmze: Total Hop=f T x subject to: [A] t S =W S x C S 1 [B] t AP =W AP x a T D [C] x where means Hadamard product. When the number of AP and the number of sectors of GW are denoted by N S and N AP respectvely, total number of flow N V s defned as N V = N S N AP. x R N V means the data amount to be transmtted from any sector of GW to any AP, f R N V weghts the number of relay hop aganst x, t S R N S s the summaton of traffc load accommodated by each sector of GW, t AP R N AP s the summaton of traffc suppled to each AP, a R N AP expresses the ON/OFF state, n other

4 words, ( / (G k ))-th component s 1, and the others are, T D R N AP s the aggregated traffc demand of each AP, W AP R N V N AP s a mappng matrx between t AP and x, W S R N V N S s a mappng matrx between t S and x. The constrant [A] means the capacty of each sector of GW, [B] ensures satsfacton of user s request, [C] assures that the value of traffc s not negatve. We then get the optmal combnaton of transmtter sector of GW and recever AP from solvng for x. As Mult-Hop scheme can only form among lnks assurng maxmum data rate of IEEE 82.11ad standard, the constrant [A] s suffcent to satsfy the capacty constrant of each mmwave backhaul lnk. () Reactvate APs for relayng (f necessary) To confrm whether we should reactvate some APs for relayng or not, we frst fnd parent for each actvated AP n step (). Parent of AP s an upper AP on the backhaulng route from GW to AP, and satsfes the followng requrements. Requrements for the parent of AP 1) Can communcate wth AP at the maxmum data rate 2) Nearest to GW n terms of Hop among APs that satsfy 1) 3) If AP has no parent that satsfes 1) and 2), AP s regarded as solated If a parent succeeds n connecton wth GW, the correspondng subordnate AP wll also succeed. Thus, f there s no solated AP, reactvaton s not necessary. On the other hand, f there are some solated APs, we have only to reactvate approprate APs for them. From the vewpont of power consumpton, t s desrable that a small number of APs are reactvated. In order to mnmze the number of such reactvated APs, we determne the best combnaton of APs to be newly actvated by the followng procedure. Procedure to determne AP to be reactvated 1) Identfy solated APs for each sector of GW 2) Generate the shortest paths from GW to solated APs through all avalable APs 3) Consder all combnatons of the shortest paths and count the number of APs needed to be reactvated for each combnaton 4) Adopt the combnaton that mnmzes the number of reactvated APs By ths procedure, all solated APs that s set ON can be connected wth GW. For APs that are set ON, the number of actvated sectors of AP s backhaul structure depends on whether AP plays a role of relay AP for other APs or not. If AP does not have any relay lnks, the number of actvated sectors of AP wll be one; otherwse t wll be two or three. Hotspot R[m] d[m] Coverage of LTE macro BS A. Base statons 4th sector 3rd sector 5th sector R/5[m] mmwave S-BS 2nd sector 6th sector mmwave Gateway Fg. 5. Deployment of SC-BSs n a macro cell. III. Smulaton Study 1st sector In order to evaluate the performance of mmwave meshed backhaul controlled by the proposed algorthm, numercal analyss s conducted n system level smulaton. Smulaton settngs are as follows. The deployment of mmwave gateway and SC-BSs s presented n Fg. 5. One mmwave gateway and nnety mmwave SC-BSs are deployed n a LTE macro cell. 1) LTE macro BSs: LTE macro cell, whch s hexagonal, has R = 25m coverage and three sectors. Indeed ths paper evaluates only one macro cell, but ntroduces sx more macro cells around the evaluated macro cell n order to consder the effects of nterference by other macro BSs. 2) mmwave small-cell base statons: MmWave SC-BSs have mmwave access structure wth three sectors and mmwave backhaul structure also wth three sectors, and access and backhaul structure are operated ndependently. Note that all mmwave SC-BSs do not have wred backhaul. 3) mmwave Gateway: MmWave gateway s the only base staton wth wred backhaul n our mmwave meshed network, and plays a role of source to mmwave SC-BSs. MmWave gateway s assumed to be equpped wth massve MIMO antenna that has sx sectors, and for smplcty each sector can form one beam. B. Traffc demand In order to reproduce non-unform user dstrbuton, user locaton s assumed to follow 2-dmensonal Gaussan dstrbuton whose form s determned by mean vector µ and covarance matrx Σ. µ corresponds to the locaton of hotspot, and Σ can be nterpreted as non-unformty of user dstrbuton. Here, Σ s assumed to be sotropc,.e. Σ= σ 2 I where σ s standard devaton. Ths paper employs the models for packet generaton nterval and the length of each packet as shown n [8], and assumes that the traffc demand grows about twce every year and then 1 years later t wll become 1 tmes hgher.

5 Locaton [m] Hotspot:(2m,m), σ =1[m] User AP (OFF) AP (ON) Gateway Locaton [m] Fg. 6. Backhaul lnks and ON/OFF status when σ = 1. C. Frequency Ths work assumes 57-66GHz band use and communcaton scheme follows the channel allocaton defned by IEEE 82.11ad standard. Ths standard dvdes 57-66GHz band nto four channels [15]. As there are many wreless lnks n mmwave meshed backhaul, severe nterference between access and backhaul may occur. In ths paper, for stable communcatons two channels are used for backhaul, and the others for access. Note that CSMA/CA s not sutable for mmwave meshed backhaul scenaro, thus deal protocol s assumed nstead. D. Example Fgure 6 presents the result of adaptvely controlled backhaul lnks and ON/OFF status n the case of σ = 1m where the hotspot s centered at (2[m],[m]). From ths result, we can confrm that SC-BSs whch are dstant from the hotspot are deactvated, and rado backhaul resources are concentrated on the area of the hotspot. Followng the above, three types of numercal analyses are conducted at system level smulaton.e. Backhaulng Route Multplexng, Scalablty, and Power Consumpton. Also two types of evaluaton crtera.e. system satsfacton rato (SSR) and power consumpton are defned as follows. [ ] Suppled Backhaul Data SSR = (2) Demanded Backhaul Data N AP Power Consumpton = (N on P on + N off P off ) (3) Here, N on access and backhaul structure, and N off OFF sectors of -th SC-BS, should satsfy N on, the number of ON sectors of -th SC-BS for, the number of + N off = 6, because each SC-BS has three sectors for access structure and also three sectors for backhaul. The second crteron TABLE I system parameter. Parameter LTE Macro mmwave network Bandwdth 1MHz GHz Carrer freq. 2.GHz 6GHz Antenna gan 17dB 26dB Antenna heght 25m 4m/25m(AP/GW) Tx power 46dBm 1dBm Beam pattern [18] [17] Path loss [18] [19] # of BSs 1 9 # of users 5 Nose densty -174dBm/Hz -174dBm/Hz System Satsfacton Rate w/ Relay (Proposal) d = 1.4 d = 15 w/o Relay d = 2.2 d: Dstance from Gateway to Hotspot σ of User Dstrbuton [m] Intensve Fg. 7. SSR aganst σ of user dstrbuton. Unform evaluates only the power consumpton of mmwave network, calculated by the summaton of power consumpton of both access and backhaul structure. E. Backhaulng Route Multplexng Here we evaluate SSR by changng the standard devaton σ of the 2-dmensonal Gaussan user dstrbuton. Then, the smaller σ becomes, the more users are dstrbuted ntensvely near the hotspot,.e. the center of the 2-dmensonal Gaussan dstrbuton. On the other hand, the larger σ becomes, the more unform user dstrbuton becomes. Fgure 7 shows SSR values n changng σ from 1m to 25m. From ths result, we can see that even though users are dstrbuted ntensvely, Mult-Hop scheme wth the proposed algorthm can mantan hgh SSR by backhaulng route multplexng and compensaton of path loss by amplfcaton relay. Conversely, Sngle-Hop scheme cannot deal wth the ntensve traffc, because of lmted capacty of only some gateway sectors able to be drected toward the hotspot. The performance of Sngle-Hop scheme and Mult-Hop scheme depends on the number of routes to receve backhaul data. In Sngle-Hop scheme, SC-BS can receve backhaul data wth only one sector, whereas n Mult-Hop scheme, SC-BS can receve wth three sectors at most. Thus Mult-Hop scheme can cope wth three tmes hgher traffc demand compared wth Sngle-Hop scheme, and ths can be confrmed from SSR when σ goes to 1m. In addton, ths result suggests that

6 System Satsfacton Rate d =.4R d =.6R d =.8R w/o Relay w/ Relay d: Dstance from Gateway to Hotspot Coverage of LTE macro "R" [m] Fg. 8. SSR aganst macro radus. Mult-Hop scheme wth the proposed algorthm can control adaptvely mmwave meshed backhaul and ON/OFF status of SC-BSs followng tme-varant and spatally nonunform user dstrbuton. F. Scalablty Fgure 8 presents SSR by changng the coverage R of LTE macro BS and fxng σ to 1m. Mult-Hop scheme can cope wth the expanson of the network scale, whereas Sngle-Hop scheme cannot due to path loss attenuaton. G. Power Consumpton Here we evaluate power consumpton calculated by (3). Three types of crtera for SC-BS actvaton are compared. The frst one s defned by the proposed algorthm consderng mult-rat selectvty of mcrowave LTE and mmwave network and SC-BSs ON/OFF status swtchng. The second one s User centrc ON consderng only ON/OFF status swtchng.e. wthout mult-rat. In the case of User centrc ON, mmwave AP wll be deactvated only when AP has no assocated user. The last one s Always ON consderng nether of mult- RAT selectvty nor ON/OFF status swtchng. Fgure 9 shows power consumpton by changng σ of user dstrbuton where the hotspot s centered at (2[m],[m]). The proposed algorthm can even reduce power consumpton aganst User centrc ON n most scenaros. Fgure 1 shows power consumpton by changng the number of users n the evaluated area and fxng σ of user dstrbuton. The proposed algorthm and User centrc ON can reduce power consumpton sgnfcantly as compared wth Always ON. As the proposed algorthm ntroduces mult- RAT selectvty as well as ON/OFF status swtchng, t can even reduce more power consumpton than User centrc ON. IV. Concluson and Future Works Dealng wth practcal user dstrbuton, ths paper employed mult-rat selectvty of mcrowave LTE and mmwave network, and SC-BSs ON/OFF status swtchng, to propose an algorthm whch can realze adaptve Energy Consumpton [kw] Energy Consumpton [kw] Hotspot:(2m,m) 1 w/ Assocaton to LTE 5 w/o assocaton to LTE Always ON σ of user dstrbuton [m] Fg. 9. Power consumpton aganst σ of user dstrbuton w/ Assocaton to LTE w/o Assocaton to LTE Always ON Total traffc load [Gbps] Few users Many users Fg. 1. Power consumpton aganst total traffc load. control of mmwave meshed backhaul. To the best of our knowledge, the presented method s the frst work on controllng both of wreless backhaul lnks and ON/OFF status swtchng n mmwave meshed network. Numercal results showed that proposed method can satsfy users traffc demand and reduce power consumpton, and also confrmed the effect of backhaulng route multplexng and ON/OFF status swtchng. In our future works, we wll nvestgate the performance gap between the optmal one and the proposed heurstc approach. Furthermore, transmsson schedulng ncludng pre-fetchng and cachng wll also be ntroduced. In addton, smulaton study and experments on more practcal settngs, e.g. backhaulng delay, uplnk communcaton, should be conducted. V. Acknowledgement Ths work s supported by 5G-MEdge project funded by Horzon22 n EU and MIC n Japan and 5G-Crosshaul project funded by Horzon22 n EU. References [1] Csco VNI Forecast, Csco Vsual Networkng Index: Global Moble Data Traffc Forecast Update, , Csco Publc Informaton, Feb. 216.

7 [2] R.J. Weler, M. Peter, W.Keusgen, E.C. Strnat, A.D. Domenco, I. Flppn, A. Capone, I. Saud, A. Maltsev, T. Hausten, and K. Sakaguch, Enablng 5G Backhaul and Access wth Mllmeter-waves, Proc. EuCNC 214, pp.1-5, Jun [3] K. Sakaguch, S. Sampe, H. Shmodara, R. Rezagah, G.T. Khanh, K. Arak, Cloud Cooperated Heterogeneous Cellular Networks, Intellgent Sgnal Processng and Communcatons Systems (ISPACS) pp , Nov [4] K. Sakaguch, G.K. Tran, H. Shmodara, S. Nanba, T. Sakura, K. Taknam, I. Saud, E.C. Strnat, A. Capone, I. Karls, R. Aref, T. Hausten, Mllmeter-Wave Evoluton for 5G Cellular Networks, IEICE Trans. Commun.,vol.E98-B, no.4, pp , Mar [5] L. Verma, M. Fakharzadeh, S. Cho, Backhaul Need for Speed: 6GHz Is the Soluton, IEEE Wreless Communcatons Vol.22 Issue.6 pp , Dec [6] X. Ge, H. Cheng, M. Guzan, T. Han, 5G Wreless Backhaul Neworks: Challenges and Research Advances, IEEE Network Vol.28 Issue.6 pp6-11, Nov/Dec [7] S. Sngh, M.N. Kulkarn, A. Ghosh, J.G. Andrews, Tractable Model for Rate n Self-backhauled Mllmeter Wave Cellular Networks, IEEE Journal on Selected Areas n Communcatons, Vol.33, No.1 pp , Oct [8] MWEBA Delverable D4.5, Overall system performance evaluaton results, Dec [9] H. Vswanathan, S. Mukherjee, Throughput-Range Trade-Off of Wreless Mesh Backhaul Networks, IEEE Journal on Selected Areas n Communcatons Vol.24, Issue.3 pp593-62, Mar. 26. [1] G. Fettws, E. Zmmermann, ICT Energy Consumpton Trends and Challenges, Proc. 11th Int l Symp. WPMC 28, Wreless Personal Multmeda Communcatons (WPMC 8), Sep. 28. [11] Unversty of Melbourne, The Power of Wreless Cloud, Whte Paper of Centre for Energy-Effcent Telecommuncatons (CEET), Jun [12] G.K. Tran, H. Shmodara, R.E. Rezagah, K. Sakaguch, K. Arak, Practcal evaluaton of on-demand smallcell ON/OFF based on traffc model for 5G cellular networks, IEEE Wreless Communcatons and Networkng Conference 216, Apr [13] E. Chavarra Reyes, I. Akyldz, and E. Fadel, Energy consumpton analyss and mnmzaton n mult-layer heterogeneous wreless systems, IEEE Trans. Moble Computng, 215. [14] L. Charavglo, M. Mella, F. Ner, Reducng Power Consumpton n Backbone Networks, IEEE ICC 29 proceedngs, Jun. 29. [15] Aglent Technologes, Wreless LAN at 6GHz - IEEE 82.11ad Explaned, Aglent Technologes Publc Informaton, May [16] 3GPP TR V9.., Evolved Unversal Terrestral Rado Access (E-UTRA); Further advancements for E-UTRA physcal layer aspects, Mar. 21. [17] I. Toyoda, T.Sek, K.Igusa, H.Sawada, Y.Fujta, A.Mura, N.Orhash, Reference antenna model wth sde lobe for TG3c evaluaton, IEEE /474r, Nov. 26. [18] 3GPP, TR V9.., Further advancements for E-UTRA physcal layer aspects, Mar. 21. [19] MWEBA Delverable D4.1, System Level Smulator Specfcaton, Dec. 214.