Mobile Backhaul in Heterogeneous Network Deployents: Technology Options and Power Consuption Paolo Monti*, Meber, IEEE, Sibel Tobaz, Meber, IEEE, Lena Wosinska, Senior Meber, IEEE, Jens Zander, Meber, IEEE 1 KTH Royal Institute of Technology, Counication Systes Departent, Electru 229, 164 40 Kista, Sweden * Tel: +468) 790 4676, Fax: +468) 790 4090, e-ail: ponti@kth.se ABSTRACT Mobile counication networks account for 0.5% of the global energy consuption, a value that is expected to double within the next five years. For this reason, eans of reducing the energy consuption in cellular obile radio networks has recently gained great interest within the research counity. In obile networks the backhaul contribution to the total power consuption is usually neglected because of its liited ipact copared to that of the radio base stations. However, eeting the alost exponential increase in obile data traffic requires a large nuber of ainly sall) base stations. This eans that backhaul networks will take a signific share of the cost and the energy consuption in future systes. Their actual contribution to the energy consuption will depend on the radio base station deployent scenario as well as on the technology and topology choices for the backhaul itself. This paper presents an initial assessent of the power consuption of two established backhaul technologies, i.e., fiber and icrowave. For the icrowave case, three backhaul topologies are considered, i.e., tree, ring and star, while for the fiber case only one topology is analysed, i.e., a dedicated point-to-point star. The presented results, assuing off-the-shelf products and based on todays network capacity levels, confir the iportance of considering the backhaul when iniizing the total power consuption in heterogeneous network scenarios. They also show the ipact of the basic technology and topology choices of the backhaul for iniizing total power consuption. Keywords: Green networks, obile backhaul, heterogeneous wireless networks, pico base stations, icrowavebased backhaul, fiber-based backhaul. 1. INTRODUCTION Despite the large potential of ICT for reducing carbon eissions in other sectors of society, there is a growing concern for the power consuption of ICT itself. Proections show that by the year 2015 capacity requireents in obile networks will be a up to 30 ties larger than today [1]. How to eet this deand without leading to an unacceptable increase in network power consuption is a topic that currently is gaining a lot of interest not only in acadeia. One way of catering for the increasing capacity requireents in obile networks is the use of heterogeneous network deployent strategies. Heterogeneous networks present an alternative to acro densification where the acro layer is copleented by the deployent of sall, low power, Pico/Micro base stations. The key rationale behind a heterogeneous network deployent is to tailor the network deployent to the expected traffic levels, i.e., to provide high capacity only where it is needed. Thereby we reduce the nuber base stations needed and the energy used. The perforance of different deployent schees in ters of energy efficiency were also studied by eans of siulation in [2] and [3]. However, ost studies of power consuption in obile networks oit contributions fro the backhaul to the total network power and only consider the aggregated power consuption of the base stations [4][5]. In ters of total network power consuption the backhaul contribution is not necessarily the sae for all obile network deployent scenarios. The authors in [6] have shown that the relative effect of the backhaul power consuption is non-negligible for scenarios with an increasing nuber of sall low power) base stations. This is due to the large difference in power consuption between base stations of different size whereas the backhaul contribution per base station is siilar. This eans that the total power consuption of a heterogeneous network deployent to a larger degree is affected by the backhaul architecture. This calls for an understanding of how different architectural options for the backhaul affect the total backhaul power consuption. Thus, in order to obtain consistent and realistic results on the overall network power consuption, so to define strategies for network optiization, an assessent of the various backhaul technologies and their energy consuption should be odelled and incorporated into the analysis. Currently backhaul for conventional Macro base stations is to a large extent based on icrowave, copper and fiber. These technologies are copleentary in a sense that they offer different advages and disadvages for different deployent scenarios. A fiber-based backhaul coes at a relatively high CAPEX and costly deployent but offers long-ter support with respect to increasing capacity requireents. Microwave-based 978-1-4673-2229-4/12/$31.00 2012 IEEE 1
backhaul is an attractive choice for eeting backhaul requireents [7] with short tie-to-arket, low investent in infrastructure and siple deployent [8]. Digital Subscriber Line DSL) is attractive in the presence of an existing copper infrastructure. Continued iproveents in DSL-technology are also enabling DSL to play an iport role for obile backhaul with increasing capacity requireents [9]. In an eerging heterogeneous network scenario, with a large nuber of sall cells that need to be backhauled, an increasing variety of backhaul technologies and solutions is expected, including also new wireless options. Heterogeneous network backhauling leads to new challenges copared to Macro backhauling, which is expected to result in a different technology ix for the backhaul solution. Deployent will depend on several factors, such as, existing infrastructure, spectru and licence costs, availability of equipent, operator business situation, etc. This paper presents an analysis of the power consuption for todays heterogeneous network deployent scenarios, including the effect of backhaul for two established backhauling technologies, i.e., fiber, and icrowave. The obective is to ake an initial assessent of what is the ipact on the total network power of the various technology and architectural options for the backhaul, where off-the-shelf products and current traffic levels i.e., capacity requireents) are used for the assessent. The ain differences, potentials and challenges of these technologies are discussed in [7] with a specific focus on cost, however to the best of the authors knowledge, the power consuption perspective has not been well odelled and investigated. In order to do so first the ain characteristics of each backhaul technology will be introduced, and their ost coon topology options will be analysed. Then two power odels, one for the icrowave case and one for the fiber case, will be presented and used in a case study to assess the power perforance of each technology. The obtained results will then be leveraged to see how different architectural options for a given backhaul technology affect both the total backhaul and the total obile network power consuption. Siulation results show that already in todays heterogeneous network deployents, fro a ere energy consuption perspective, fiber-based backhaul solutions ight be preferable due to their relatively low ipact in ters of extra power consuption. 2. MOBILE BACKHAUL ARCHITECTURES When it coes to the choice of a particular backhaul technology and topology, there are a nuber of factors that need to be considered, such as, Capital Expenditures CAPEX), Operating Expenditures OPEX), propagation and spectru conditions, Quality of Service QoS) requireents i.e., ostly delay), and requested resiliency levels. This section presents the assuption ade for each technology and architectural choice. In this work it is assued that only two types of base stations are available for a heterogeneous network deployent scenario, i.e., Macro and Pico base stations. Fig. 1 shows the basic architectural backhaul options that are considered and later used for coparison fro a power consuption perspective. a) b) c) Figure 1: Different backhaul topology solutions[10]: a) ring topology, b) star topology, an c) tree topology. Circles represent Radio Access Network RAN) sites, while rectangles represent hub nodes. As can be seen fro Fig. 1, each base station, i.e., RAN, site is connected to one or ore sites, depending on the topology under consideration. All traffic fro the base stations is assued to be backhauled through a hub node. In this work it is assued that any base station in the network, regardless of its type i.e., Macro and/or Pico), can be a hub node. Depending on the total nuber of base stations in the studied area, and also depending on the backhaul architectural choice i.e., axiu nuber of base stations in each topology), ore than one level of aggregation can be considered, i.e., ultiple hub nodes ight be present in an area. Each hub node is connected to the area aggregation point i.e., a node, not shown in the figure) that in turn connects to an aggregation network i.e., either a etro or core segent). If ore than one backhaul link originates or terinates at any base station site, that particular site is assued to be equipped with a. For the icrowave-based backhaul we consider tree, ring and star topologies and these are copared to a star-like fiber topology. The backhaul links are diensioned for peak rate fro the base stations to the hub node. This is ore aggressive copared to diensioning guidelines in [11] since it iplies signific overbooking already at the first aggregation stage. Each backhaul link is diensioned to cater for the capacity requireents of 2
the connected base station and of any additional base station backhauled via that base station site this applies only for the ring and tree case). It should however be noted that despite the effort to diension the backhaul networks in a siilar way for the different topologies there are inherent differences between the topologies with respect to aspects such as latency and resilience. For star topologies there are separate icrowave links fro the hub to each site. The drawback with a starlike topology is the liited area that can be covered by a single star due to the reach liitations of the links. To provide coverage via icrowave, star topologies require longer LOS line of sight) links. Star topologies also lead to poor frequency reuse. All links terinate at the sae point i.e., the hub node), and links using the sae frequency are ore likely to interfere with each other. Rings are advageous in ters of protection but latency ay be an issue. The axiu capacity of a ring will also liit the nuber of sites that can be served by each ring. A tree topology, on the other hand, is sensitive to faults on the feeder links. In ters of power consuption the star topology can be seen to be the ost efficient assuing that a single star can serve the considered area and assuing that the link power is independent of the reach. This is because a star topology requires a iniu of one link per base station site and enables axiu statistical ultiplexing at the central hub. Since fiber is able to provide sufficient reach for the considered area, for the fiber-based backhaul case a single star scenario is assued. In the considered fiber-based star all base stations are connected to a single central hub node, that works also as the towards the aggregation network ore details about this backhaul architecture can be found in [6]). The hub node is equipped with one or ore aggregation es if the nuber of the traffic streas to be aggregated so requires). 3. BACKHAUL POWER CONSUMPTION MODELS This section presents power odels for obile networks including the contribution fro the backhaul. A characterization of the power consuption of the individual Macro and Pico base stations can for exaple be found in [3] where the odel paraeters are derived based on data found in the literature. In the icrowave case the total power consuption of a heterogeneous obile radio network including the backhaul part can be written as: Ptot = NiPi + P, 1) i= 1 where is the nuber of base station types used in the network, N i is the total nuber of base stations of a specific type i e.g., Macro or Pico base stations), P i is the power consuption of a base station of type i, and P is the entire backhaul power consuption. P i is odelled as a linear function of the radiated power: Pi = aiptx + bi. 2) In equation 2) the paraeter a i accounts for the scaling of the base station power consuption with the transitted power, i.e., the power associated with radio frequency RF) aplifiers and the feeder losses. The paraeter b i is associated with the baseline power that is independent of the transitted power, i.e. signal processing and site cooling. The backhaul power contribution P N BS = 1 can be odelled as: P = P + P, 3) where P represents the power consuption of the area node, N BS the total nuber of base stations regardless of their type) in the entire area, and P represents the power associated with icrowave backhaul operations at the base station site. Assuing that C represents the total aggregated capacity that base station has to backhaul, and that N represents the nuber of icrowave ennas i.e., nuber of icrowave links) used at base station, P can be expressed as: P = P, agg C ) + P N, Ci ), 4) where 0, if N = 1 Plow c, if C Thlow c, agg ) = ;,, ) = Phigh c PS MAX C P C P N C C. 5), otherwise *, otherwise According to equation 5) the power consuption for transitting and receiving the aggregate backhaul traffic via icrowave P,agg ) is a step function with respect to the total backhauled capacity C ). Two power regions P low-c and P high-c ) for each base station were defined, one for low capacity traffic conditions and the other for 3
high traffic capacity conditions. For the sake of siplicity only two capacity regions are considered in this work. The extension beyond two regions is straightforward. Equation 5) is also used to account for the power consuption of the P s ) that needs to be used at any base station site aggregating traffic fro ore than one base station. The nuber of es is a function of the total backhauled capacity and the axiu MAX capacity of a C ). Identical es are assued to be used at all base station sites, regardless of the type. The sae reasoning can be applied to copute the value of power consuption of the node. P can be expressed as follows: where P = P C ) + P N, C ), 6), agg, 0, if N = 1, agg ) Plow c, if C Thlow c ;,, ) P high c PS MAX C P C = P N C = C. 7), otherwise *, otherwise For the fiber-based case the total power consuption of a heterogeneous obile radio network including the obile backhaul can be written as: FIB FIB Ptot = NiPi + P, 8) i= 1 where according to the paraeters defined in [6]: and Pi = aiptx + bi + ci, 9) FIB 1 P = Ni Ps Ni Pdl Nul Pul ax + +. 10) dl i= 1 i= 1 The paraeters, N i, a i, b i, and P s are the sae as in the icrowave case. On the other hand c i represents the power consuption of a sall-for factor pluggable SFP) optical interface used to connect a base station to the aggregation at the hub node), and ax dl is the axiu nuber of downlink interfaces available at one aggregation. This paraeter is used to copute the total nuber of es that are needed to collect the backhauled traffic of the obile network. On the other hand, P dl is the power consued by one interface in the aggregation one for each backhauled base station), while N ul, and P ul are the total nuber of uplink interfaces, and the power consuption of one uplink interface, respectively. These last two paraeters account for the power consuption of the interfaces used to aggregate the backhauled traffic towards a etro/core segent. A ore detailed explanation of these paraeters can be found in [6]. 4. POWER CONSUMPTION COMPARISONS As a basis for coparing the various backhaul architectures we assue two different base station deployent scenarios, each one providing the sae area throughput and coverage. The first scenario considers a Macro deployent where capacity is increased by a densification of the Macro base station sites. The second scenario considers a heterogeneous network deployent where capacity is increased by the deployent of Pico cells. The underlying syste odel and the assuptions used for the network deployent are ade in accordance with the work presented in [12]. In each scenario the nuber of base stations for the Dense Macro and the Macro + Pico cases are deterined sequentially until the target area throughput is satisfied. Table 1. Power consuption paraeters, icrowave backhaul case [3][13]. Base station type a i b i P tx [db] P low-c [W] P high-c [W] Th low-c [Mbps] P s [W] Macro 21.45 354.44 43 37 92.5 500 53 36 Pico 5.5 38 30 37 92.5 500 53 36 C MAX [Gbps] In this work the base stations are assued to be of 3G Universal Mobile Telecounications Syste UMTS) type each with a peak user downstrea data rate of 100Mbps in total. Macro base stations are assued to have 3-sectors with ennas above rooftop for longer range. They have a axiu of three carriers per sector. On the other hand, Pico base stations have single carrier and oni-directional ennas. The area considered is a square region of four by four kiloeters. For the icrowave case the power consuption paraeters of each type of base station used in the coparison are suarized in Table 1, while Table 2 suarizes the power consuption paraeters for the fiber backhaul case. Results presented in this section are for the ring and star topologies shown in Fig. 1, where the nuber of nodes is varied as part of the assessent. For siplicity, in the 4
tree topology case, a coplete binary tree is assued copared to the vari shown in Fig. 1). The exact structure of the tree is not expected to affect the general validity of the conclusions that will be drawn fro the results. Table 2. Power consuption paraeters, fiber backhaul case [3][6]. Base station type a i b i c i [W] P dl [W] P ul [W] ax dl P s [W] C MAX Macro 21.45 354.44 1 1 2 24 300 24 Pico 5.5 38 1 1 2 24 300 24 [Gbps] Tables 3 and 4 show, for the icrowave case, the power consuption associated with backhaul operations, i.e., P defined in section 3, as a function of the area throughput requireents. For all three cases i.e., tree, star and ring) the value P is noralized using the value of the total area covered, i.e., P /16 k 2. Table 3. Value of Area Throughput [Mbps/k2] P [W] noralized by the total area covered: Dense Macro case. Tree Star Ring Max # hops Max # nodes Max # nodes 2 3 4 5 6 8 12 16 4 6 8 10 40 103.3 122.9 126.4 126.4 87.1 80.0 76.4 76.4 145.6 208.5 218.0 222 60 194.9 218.1 221.6 242.6 157.0 146.3 135.7 132.1 270.3 416.0 422.9 423 80 277.0 301.8 312.4 312.4 224.5 210.3 192.5 185.4 389.0 607.6 607.6 611.1 Based on the results it is possible to draw soe conclusions on the topology ipact on power consuption. As expected the value of P increases with the capacity requireents which affects the nubers of base station deployed in the network. Aong the three cases, the ring-based backhaul solution is the ost costly in ters of power. This is due ainly to two factors. Firstly, each base station needs to be equipped with a to support the ring structure. Secondly, in the considered scenarios the backhaul links always work in the high traffic capacity region due to a large aount of pass-through traffic. The ring structure, on the other hand is able to handle single link failures, copared to the star topology case, which is the least power consuing aong the three topology cases. This highlights the iport trade off between power consuption and resiliency. In the star case, the only base station equipped with a and working in the high traffic capacity region is the hub node. The tree topology case, can in ters of power consuption, be seen as an interediate case where a few nodes are required to operate in the high traffic region. Going towards a heterogeneous network deployent scenario, we see that the backhaul has a non-negligible ipact on the power consuption. This is due to the increased nuber of base stations that need to be backhauled. All three topology cases experience a power consuption increase of at least 100% for the considered backhaul segent) copared to the Dense Macro scenario. Table 4. Value of Area Throughput [Mbps/k2] P [W] noralized by the total area covered: Macro + Pico case. Tree Star Ring Max # hops Max # nodes Max # nodes 2 3 4 5 6 8 12 16 4 6 8 10 40 264.2 290.9 301.5 308.6 208.2 197.5 183.3 176.2 365.3 559.6 573.5 556.1 60 582.4 641.0 662.1 672.6 457.3 425.3 393.5 375.7 804.7 1226.1 1269.5 1269.5 80 852.5 925.1 956.8 958.5 658.6 616.1 566.4 545.1 1166.9 1843.3 1843.3 1843.3 Fig. 2 provides a coparison of the noralized power consuptions of the backhaul segent for both a icrowave-based and a fiber-based backhaul, assuing a heterogeneous network deployent scenario, i.e., the Macro + Pico case. Two extree cases in ters of topology size, for each icrowave topology option, are depicted to deonstrate the relative difference in ters of power consuption perforance between fiber and icrowave technologies. For sall size topologies 2, 6, and 4 nodes are considered for the tree, star, and ring topologies, respectively. On the other hand, large size topologies for trees, stars, and rings are characterized by 5, 16 and 10 nodes, respectively. The figure shows that a fiber-based backhaul consues less power regardless of the considered topology, and topology paraeters. Hence, fro a pure energy consuption perspective the single star fiber-based solution where each base station in the area has a dedicated point-to-point connection with the node is preferable. Obviously, such considerations need to be weighted against other iport factors such a deployent costs. 5
a) b) Figure 2. Area power consuption for the backhaul: icrowave vs. fiber in a Macro + Pico case. Two scenarios are considered: sall size a), and b) large size icrowave topologies. Fig. 3 shows the ipact of the power consuption of the backhaul segent on the overall network power consuption. The value of the total network power consuption is shown as a function of the area throughput requireents. Both technology options, i.e., fiber and icrowave with all three topologies options) are considered. The figure confirs the intuition that the ipact of the backhaul power consuption is larger for a heterogeneous network scenario copared to dense Macro. This is true regardless of the technology choice for the backhaul. This eans that technology and topology considerations for the backhaul will be increasingly iport for optiizing the total network power in a heterogeneous network scenario. On the other hand, even with the inclusion of the extra power consued by the backhaul segent, the use of low power Pico base station still reains the best option to have a network with an overall reduced power consuption, i.e., the power benefits of using a heterogeneous deployent still hold. Figure 3. Total area power consuption. Three scenarios are considered, no backhaul, icrowave-based backhaul, and fiber-based backhaul. 5. CONCLUSIONS This paper presented a study on the ipact of various backhaul options on the overall power consuption of a heterogeneous obile network. Two backhaul technologies are considered, icrowave- and fiber-based. Various topological choices were analysed assuing off-the-shelf products and current network capacity requireents) and their respective power consuption odels were derived. The presented results confir that the power consuption of the backhaul segent is an iport part of the total network power consuption. For this reason the backhaul needs to be carefully included in any deployent strategy with obective of iniizing total network power consuption. Furtherore, results also show that fro a pure power consuption perspective a 6
star shaped fiber-based topology is preferable. This finding ust be weighted against other iport factors such a deployent costs, which would be interesting to address in a continuation of this work. ACKNOWLEDGEMENTS The authors would like to thanks to Klas Johansson for sharing his siulation results and Börn Skubic for his valuable technical conversations and input. The work presented in this paper was carried out with the support of the Energy-efficient Wireless Networking ewin) and Optical Networking Syste ONS) focus proects, part of ICT The Next Generation TNG) Strategic Research Area SRA) initiative at the Royal Institute of Technology. REFERENCES [1] Cisco visual networking index: Global obile data traffic forecast update, 2010-2015, Technical Report, Cisco, 2010. [2] S. Tobaz, M.Usan, and J. Zander, Energy efficiency iproveents through heterogeneous networks in diverse traffic distribution scenarios, in Proc. CHINACOM, Harbin, China, August 2011. [3] F. Richter, et al., Energy efficiency aspects of base station deployent strategies for cellular networks, in Proc. IEEE VTC, Anchorage, USA, Sep. 2009. [4] F. Richter, et al., Traffic deand and energy efficiency in heterogeneous cellular obile radio networks, in Proc. IEEE VTC Spring, Taipei, Taiwan, May 2010. [5] K. Dufkova, et al, Energy savings for cellular network with evaluation of ipact on data traffic perforance, in Proc. European Wireless Conference, Lucca, Italy, April 2010. [6] S. Tobaz, et al., Ipact of Backhauling Power Consuption on the Deployent of Heterogeneous Mobile Networks, in Proc. IEEE GLOBECOM, 2011. [7] O. Tipongkolsilp, S. Zaghloul, and A. Jukan, The evolution of cellular backhaul technologies: Current issues and future trends," IEEE Counications Surveys Tutorials, vol. 13, no. 1, pp. 97-113, 2011. [8] High speed technologies for obile backhaul, White Paper, Ericsson, Sep. 2008. [9] Leveraging VDSL2 for obile backhaul, White Paper, Alcatel Lucent, Sep. 2010. [10] M. Paoloni, Crucial econoics for obile data backhaul, White paper, Senza Fili Consulting, 2011. [11] Guidelines for LTE Backhaul Traffic Estiation, White Paper, NGMN Alliance, 2011. [12] K. Johansson, Cost effective deployent strategies for heterogeneous wireless networks, Ph.D. dissertation, Sweden, Noveber, 2007, KTH Inforation and Counication Technology. [13] Ericsson Microwave Networks portfolio, available at http://www.ericsson.co/ourportfolio/products/icrowave-networks 7