1

5G Backhaul Challenges and Emerging ResearchDirections: A Survey

Mona Jaber1, Muhammad Imran 1, Rahim Tafazolli1, and Anvar Tukmanov2

1 Institute for Communication Systems, Home of 5G InnovationCentre, University of Surrey, Guildford, GU2 7XH, UKEmail: {m.jaber, m.imran, r.tafazolli}@surrey.ac.uk

2 BT Research and Innovation, Adastral Park, Ipswich, IP5 3RE, UKEmail: anvar.tukmanov@bt.com

Abstract— 5G is the next cellular generation and is expectedto quench the growing thirst for taxing data rates and to enablethe Internet of Things. Focused research and standardizationwork have been addressing the corresponding challenges fromthe radio perspective whilst employing advanced features suchas network densification, massive multiple-input-multiple-outputantennae, coordinated multi-point processing, inter-cell interfer-ence mitigation techniques, carrier aggregation, and new spec-trum exploration. Nevertheless, a new bottleneck has emerged:the backhaul. The ultra-dense and heavy traffic cells shouldbeconnected to the core network through the backhaul, often withextreme requirements in terms of capacity, latency, availability,energy, and cost efficiency. This pioneering survey explains the 5Gbackhaul paradigm, presents a critical analysis of legacy,cutting-edge solutions, and new trends in backhauling, and proposesanovel consolidated 5G backhaul framework. A new joint radioaccess and backhaul perspective is proposed for the evaluation ofbackhaul technologies that reinforces that no single solution cansolve the holistic 5G backhaul problem. The study also revealshidden advantages and shortcomings of backhaul solutions whichare not evident when backhaul technologies are inspected asan independent part of the 5G network. This survey is keyin identifying essential catalysts that are believed to jointlypave the way to solving the beyond-2020 backhauling challenge.Lessons learned, unsolved challenges, and a new consolidated 5Gbackhaul vision are thus presented.

Index Terms— 5G, backhaul, fronthaul, small cells, heteroge-neous network, C-RAN, SDN, SON, backhaul as a service.

I. I NTRODUCTION

Societal changes, witnessed since the explosion of dataservices, and the growing appetite for wireless broadband haveincentivised the speedy development of the fifth generationof cellular systems (5G), envisioned for year 2020 [1]. Inorder to cater for the anticipated 1000x capacity, key playersforesee the need for a “revolution” in some aspects of legacysystems accompanied by enhancements in existing technolo-gies [2]. Based on early consortiums in the development of 5G,promising enablers have been identified: ultra-dense networks(UDN), advanced inter-cell interference coordination (ICIC)schemes, massive multiple-input-multiple-output(MIMO)andcoordinated multi-point processing (CoMP), centralised/cloudprocessing, and user/control plane decoupling. UDNs areoften heterogeneous networks (HetNets), i.e., multi-layeredincluding legacy high power macro-cells and very densecells with lower power (small cells). Small cells are multiradio access technologies (multi-RAT) capable and repre-sent an essential part of UDNs, which are considered an

imperative 5G solution [3]–[8]. Sharing the spectrum in aUDN requires intelligent inter-cell interference coordination,cancellation or exploitation. Accordingly, key radio UDNfacilitators have been developed: CoMP and enhanced ICIC(eICIC). The increasing need in processing power coupledwith the emerging small cells diversity in traffic patterns,bothspatial and temporal, render the concept of centralized radioaccess network (C-RAN) very attractive. C-RAN consistsof splitting the functions of the traditional evolved NodeB (eNB, i.e., the cellular radio station) and migrating themtowards a distant shared pool of baseband resources, referredto as baseband unit (BBU). Basic radio functions remainat the radio site, hence the terminology remote radio unit(RRU). The C-RAN architecture capitalises on the diversityof traffic peaks hence improves the utilisation efficiency ofthe infrastructure. A the same time, it promotes the greenaspect of 5G, owing to close proximity of cells and usersand corresponding lower transmission power requirements [9].Another challenge resulting from UDNs is the user mobilitymanagement; traditionally, users moving from one cell toanother require a handover procedure managed by the mobilitymanagement entity in the RAN. If this model were applied toUDNs, it would generate a crippling signalling overhead dueto the limited footprints of small cells, hence frequent cellborder crossing. Accordingly, splitting data and control planesis another essential 5G technology: small cells are used as dataoffloading points whereas mobility handover is triggered whenusers move between clusters as opposed to small cells [6],[10]. According to NTT DoCoMo, most of the presentedenablers are either not new or not “intelligent” as stand-alonetechniques, but consolidating them into a complete coordi-nated solution results in innovation, such as the advanced C-RAN [11]. Although such technologies can potentially addressthe greedy 5G capacity requirements and reduced RAN-relatedcapital and operational expenditures (CapEX and OpEX), anew challenge has nonetheless emerged: the 5G backhaul.

The backhaul (otherwise referred to as back-net or backboneor transport network), in cellular networks, is the networkthatconnects the eNBs to the core network and consists mostly ofdedicated fibre, copper, microwave, and occasionally satellitelinks. In pre-LTE (Long term evolution) cellular generations,the radio controller node often acts as a backhaul aggregationpoint, thus, concentrating backhaul connections from all radiostations within its reach, towards the core. LTE’s architecturedoes not employ a radio controller node, however, backhaul

2

Fig. 1. The GSM base station controller (BSC) and the UMTS radionetwork controller (RNC) are often co-located, and are usedas backhaulaggregation points for BTS, Node B and eNB backhaul links. TheeNB connects directly to the service gateway (SGW) for user datatransmission over the S1-u and to the mobility management entity(MME) for control data transmission over the S1-c interface. In addition,inter-eNB interfaces, referred to as X2, are often routed through theaggregation point.

Fig. 2. Example 5G network mobile backhaul network consisting offronthaul, midhaul, and traditional backhaul. The fronthaul refers tothe last mile transport links connecting the RRH to the network. Themidhaul is the link between the fronthaul aggregation pointand thebackhaul network. The backhaul links are those that connectthe BBUsto the network; the group of all transport links is also referred to asbackhaul network.

aggregation remains desirable for both wired and wirelessconnections, as shown in Figure 1. With the rise of C-RANarchitecture, the 5G backhaul has evolved to a more complexnetwork composed offronthaul, midhaul, andbackhaul. Thebackhaul section connecting the remote radio head (RRH)to the baseband unit (BBU) directly, or to an intermediateaggregation point, is labelled fronthaul. The basic fronthaul isassumed to run over a common public radio interface (CPRI)separating the RRH from the BBU. Due to the stringentrequirements of the CPRI-based fronthaul, novel interfaces arebeing explored such as the fronthaul-lite [12], next-generationfronthaul interface (NGFI) [13], or xHaul [14]. In this paper,all forms of fronthaul are referred to asfronthaul. Based on the3GPP terminology, the inter-eNB X2-based interface is calledthe midhaul [15]; the term has recently been used to refer tothe group of links connecting the fronthaul aggregator to thebackhaul aggregation point (see Figure 2). While the networkconnections between aggregation points and the core, basedonthe S1-interface [15], have retained the term backhaul. In thispaper, we use the term backhaul to refer to the entire transportnetwork including midhaul and fronthaul.

Inhibitive bandwidths greater than 10 Gbps and maximumallowed latency in the orders of hundreds of microseconds,render fibre optics, perhaps the only fronthaul viable solu-tion [6], [16]. However, laying fibre to connect all envisagedRRH to the core may be impossible in some cases andcertainly very costly otherwise. In view of the immensechallenge facing 5G deployment, the 5G backhaul researchhas been triggered, aiming at bridging the gap between therequirements stipulated by the 5G RAN and the realisticbackhaul capabilities from two different perspectives. The firstconsists of evolving the current backhaul (microwave, opticalfibre, copper, etc.) to meet 5G expectations and encompassingnew wireless technologies such as in-band links (reuse theradio access spectrum), millimetre wave (mmWave), free space

optical communications (FSO), and sub-6GHz (e.g. World-wide Interoperability for Microwave Access-WiMAX, WiFi).The other backhaul research perspective looks at adapting the5G RAN to the available backhaul with realistic performance,such as investigating intermediate RAN architectures betweenthe C-RAN and the distributed RAN (D-RAN) to fit thefronthaul capabilities [12], [14], [17].

The evolutions of backhaul solutions from 2G to 3G andfrom 3G to 4G are well surveyed in [18] and [19], respectively.Also, [20] provides a comprehensive study of circuit switchedand packet switched backhaul technologies as perceived in2011, before the launch of LTE. However, to the best of ourknowledge, there has been no comprehensive survey on 5Gbackhaul challenges and evolution in the literature. iJOIN1

group, provides a representative review of backhaul evolutionfor UDNs and cloud-RAN within a broader context of RANevolution. The backhaul review targets the physical level,medium access control and resource management level, andnetwork level in three deliverables [21]–[23], respectively.Next Generation Mobile Networks (NGMN) provide a con-sensus around operators’ views on UDN backhaul requirementwith a focus on “the last mile” link to small cells [8].The Small Cell Forum (SCF) builds on findings reported byNGMN, and presents accordingly a technical review on diverseUDN backhaul solutions and explores their suitability foridentified use-cases [24]. These works, [8], [21]–[24], jointlyprovide an insightful starting point to grasping the problemof next generation’s backhaul network and essential reviewonsolutions portfolio.

In this work, we present the first comprehensive surveythat explains the 5G backhaul problem, examines proposedsolutions, and puts forward a set of tangible guidelines foradapting the backhaul solution to the various 5G scenarios

1Interworking and Joint Design of an Open Access and BackhaulNetworkArchitecture for Small Cells based on Cloud Networks

3

whilst offering a consolidated vision of a dynamic, flexibleand adaptive 5G backhaul framework. We first identify the5G features with high impact on the backhaul and quantifytheir impact on backhaul performance. Then the availablebackhaul solutions are categorised in terms of their respectivenominal performance and limitations. The first contributionis a matching exercise between the required and availableperformance metric from the compiled data, shedding lighton the impact of various RAN features on the restriction ofbackhaul choices, and vice-versa. Another contribution isthepresented joint RAN/backhaul perspective to different combi-nations of possible RAN architectures and backhaul solutions,offering a tangible tradeoff analysis between gains and lossesincurred from each. The results from these studies help indelimiting the solution space and higher level requirements ofthe 5G backhaul solution: a heterogeneous network composedof various wired and wireless links with the ability to dy-namically adjust and adapt to the changes in the network in aflexible, efficient, and timely manner. A comprehensive surveyof backhaul state-of-the-art research is conducted highlightingkey trends and how these can collaborate to form a holistic 5Gbackhaul solution. Thus, another contribution in this workisthe consolidated 5G backhaul vision that we propose, referredto as backhaul as a service (BHaaS), which builds on keyresearch findings such as software defined networks, heteroge-neous backhaul technologies, joint RAN/backhaul operation,self-optimisation techniques and proactive caching to deliverthe required backhaul flexibility and adaptability.

Any attempt to frame and survey the 5G backhaul problemshould start with a pertinent definition of 5G networks. Ac-cordingly, in Section II we first identify major 5G aspectsthat impact the backhaul network and proceed towards in-vestigating related challenges and state-of-the-art solutions.Consequently, Section III presents legacy backhaul networksand discusses how 5G characteristics impact the backhaul andtheir corresponding requirement. Main research directions inbackhaul technologies are classified in Section IV under sixmajor categories: fibre-based, wireless, SDN-enabled, cacheenabled, green efforts, and joint RAN-backhaul intelligence.Section V concludes the article with an outlook on fore-seen challenges and research directions. Abbreviations andacronyms used are first introduced in the their first occurrencein the text and are also tabulated Table IX for ease of reference.

II. A BACKHAUL 5G PERSPECTIVE

The 5G backhaul research topic exists as a consequence ofthe holistic 5G network ambitious challenge. To this end, 5Gfeatures that impact the backhaul ought to be first identifiedand studied, from a backhaul perspective, in order to delineatethe 5G backhaul research topic. There is currently no completestandardised definition of 5G networks, however, a general di-rection of main goals is emerging through diverse group effortsand can possibly be summarised in this paragraph. The main5G initiatives globally are: in the United States such as 4GAmerica, in China e.g., IMT-2020 (5G) promotion group, inJapan e.g., 2020 and beyond, in Korea with the 5G forum, andin Europe, mainly, the 5G Private Public Partnership (5G PPP)

funded by the European Union and the 5G Innovation Centre(5GIC) at the University of Surrey in the UK. Key examples ofEuropean projects researching technology beyond 4G can befound in [25]. The International Mobile Telecommunicationssystem (IMT) has initiated research and technology trials in2013 and plans to start the standardisation phase in 2016.The 2015 World Radiocommunication Conference identifiedchunks of mmWave spectrum bands, between 20 and 80GHz,for testing, but postponed the allocation of 5G spectrum tillthe next conference in 2019. It was also decided that the thirdgeneration partnership project (3GPP) will have a technicalspecification group (TSG) that will start the work on the 5GRAN in 2016 [26]. In addition, the International Telecommuni-cation Union-Radio Communication Sector (ITU-R) workingparty 5D is responsible for the definition and evaluation of 5G(IMT2020) networks; the plan is to deliver 5G specificationsby October 2020 [27].

Ericsson, in a presentation on IMT-2020, state that 5G isabout “making the extremes possible” [28]. Indeed, the IMTvision as summarised by METIS2, targets 1000x capacityincrease, 10-100 more connected devices, compared to today’scellular performance, data rates in the order of Gbps andsub-millisecond latency, at the cost and energy consumptionof current networks [29] [30]. These requirements impactthe backhaul directly which should bear a data explosionat minimum delay and cost with high resilience and greenconsiderations. Such ambitious expectations are perhaps unre-alistic and impossible to realise holistically; nonetheless, thedominant 5G feature that the backhaul needs to capitalise onis the diversity of requirements. For example, latency in smartmeter applications could stretch to tens of minutes, or more,whereas tactile internet requires end-to-end values less than1 msec. Similarly, real-time video applications would requireextremely high data rates, whereas fleet management transmitsand receives at low data rates. In addition, 5G should providethe platform to connect a massive number of objects to theinternet, thus, supporting the Internet of Things. Smartphonesand devices (e.g., smart meters), however, differ greatly intheir processing power, radio components, and battery capa-bilities. Smart meters need to be low-cost devices with verylong battery life (up to 10 years), thus, energy efficiency incommunication is crucial for this application; while, videoconferencing, for instance, would have latency and throughput,instead of energy efficiency, as a priority target. A detailedlisting of typical 5G user experience and system performancerequirements is available in [31], showing data rates varyingfrom 1 kbps to 1 Gbps and latency from microseconds tohours. Such heterogeneity in service expectations reflectsonthe backhaul requirements. Accordingly, a 5G backhaul shouldnot necessarily be holistically compliant with the most strin-gent specifications; instead, it should be flexible and adaptivein such a way that all services are catered for in an efficientmanner while conforming with their attributes.

In order to provision for the increase in capacity, devices,and data rates, efforts are invested in three axes jointly:

2Mobile and Wireless Communications Enablers for the Twenty-TwentyInformation Society

4

network densification, improved spectrum efficiency, and spec-trum extension [16]. The features employed towards these endsare partly an evolution of existing technologies (e.g., HetNets,CoMP, massive MIMO) and partly disruptive to state-of-the-artcellular systems (e.g., C-RAN and control/user plane split). InSections II-A and II-B, we discuss key evolved and disruptive5G features, respectively, from a backhaul perspective. For arecent comprehensive review on 5G technology and converg-ing trends, please refer to [25].

A. Evolved 5G features

5G technology trends include the evolution of existingfeatures such as carrier aggregation, MIMO, CoMP, andHetNets, which have already been standardised for LTE/LTE-A and have shown promising gains in boosting the numberof connected devices and corresponding data rates. Carrieraggregation is introduced in release 10 of LTE and furtherextended in release 11. It basically consists of equipping acell with more than one carrier components (total maximumbandwidth up to 100 MHz) with joint scheduling, hence,reaching users with higher data rates. Release 13 is expectedto support aggregation of 32 carrier components, hence, largeraggregate bandwidth [32]. The aggregate radio throughput ofa cell scales with the available radio bandwidth, thus, maynecessitate larger capacity on the connecting backhaul link.

MIMO is based on spatial multiplexing, in which datastreams from several branches are multiplexed and transmittedover several spatially separated channels. MIMO is an essentialfeature in LTE-Advanced; R10 transmission modes allow4x4 MIMO for the uplink and 8x8 MIMO for the downlink.NTT DoCoMo demonstrated, in December 2012 10 Gbpsusing 8x16 MIMO with 400 MHz bandwidth, later showedsimulated data rate of 30 Gbps using 24x24 MIMO [33] [34].Massive MIMO is, thus, a prime enabler of 5G due to itsdata rate boosting capability. Network MIMO is a class oftransceiver techniques, where the transmission and receptionof signals, among multiple spatially distributedbase stations,are coordinated so that interference is mitigated [35]. NetworkMIMO is often referred to as coordinated multi-point process-ing or CoMP. NGMN foresee a pivot role for massive MIMOand CoMP in ameliorating quality, fairness, and overall systemefficiency [31]. Considering the variety of different CoMPmethods already proposed for LTE, ranging from coordinatedscheduling to joint transmission, 5G is expected to nativelysupport the most effective techniques. However, both massiveMIMO and CoMP transmission rely on the availability oftimely channel state information, hence, would necessitatevery low backhaul latency to realize their full potential.Moreover, these features often require that the user data bepresent (transmitted and/or received) in all cells in the CoMPcluster, consequently, the connecting backhaul links of thosecells would have to be equipped with higher bandwidth tocater for the data of all users in the cluster (as opposed tousers served by the cell alone).

HetNets also stem from the evolution of existing technology,which consists of various cell layers (e.g., macro-cell, micro-cell, pico-cell, etc.) and various radio access technologies

(e.g., GSM, 3G, LTE, WiFi, etc.). They are considered anindisputable part of future cellular networks and have receiveda focused attention from 3GPP standardisation work [36].Small cells may have different sizes, may be indoor or outdoor,and may be operator-planned or not (e.g., femtocells, lowpower cellular access points connected to the internet), buttheir common characteristic is that they are low power nodesat low heights used for data offloading [37]. Foreseeable UDNsmall cell density in highly populated areas may well reach1500 cells perkm

2, including femtocells. HetNets maximisearea spectral efficiency, due to the tight reuse of the preciousspectrum, and economise on transmit power requirements,owing to the proximity of transmitters and receivers, thus,endorsing the “green” aspect of 5G. Advantages of small cellsspring from these characteristics which allow high area spec-tral efficiency, and low uplink and downlink power, thus, longUE battery life and greener communications. When small cellsare part of a HetNet, they may share the spectrum with themacro-layer (radio network layer formed by a group of macro-cells), a desired feature to maximise the usage of the spectrum.In this case, however, a cell association (or handover) problemarises since the cell selection is traditionally based on thestrength of signals received from the candidate cells. Themacro-cell is a high power station, thus, often reaches the userswith higher signal strengths than small cells. The cell rangeextension feature was recently defined for HetNets to biasthe small cells signals for attracting more users (e.g., [38]).With this feature, enhanced inter-cell-interference coordination(ICIC) schemes are used to limit the interference caused bythe macro-cell to the small cells’ UEs. In LTE release nine,ICIC was first introduced to exchange load and interferenceinformation over X2 (LTE interface between base stations).The base station would then consider received information tooptimise scheduling, mostly targeting edge users [39]. With theemergence of HetNets, enhanced ICIC (eICIC) is defined inrelease 10, in which almost blank sub-frames are used on themacro-layer to reduce downlink interference on UEs associ-ated with small cells [40] [41]. LTE release 11 includes furtherenhanced ICIC (feICIC), which aims at handling interferenceby the UE through inter-cell interference cancellation forcontrol signals, enabling even further cell range extension [42].With these features in place, small cells in ultra-dense HetNetsare able to absorb the anticipated 5G massive traffic and aninvasive number of devices. The capillaries of the backhaulnetwork need to expand at the same pace and breadth as thesmall cell growth while providing higher throughput and lowerlatency, a staggering target on its own.

B. Disruptive 5G features

UDNs imply an invasive spread of small cells at a high costof RAN equipment, even if the exorbitant cost of backhaullinks is excluded. In addition, the limited coverage of smallcells results in a large peak-to-average ratio of traffic demand,which would necessitate large allocation of baseband resourcesper cell if the traditional, or distributed, RAN (D-RAN)approach were adopted. The C-RAN aims at addressing bothof these problems by reducing the complexity (hence cost) of

5

small cells and pooling baseband resource, thus, improvingtheir utilisation efficiency, irrelevant of the individualcelltraffic patterns.

The “C” in C-RAN often stands for centralised or cloudbut also clean and cooperative RAN. The focal concept is toredistribute functions, which are traditionally found in basestations, towards a cloud-operated central processor. Suchcentralised intelligence would consequently enable cooperativeoperation among cells for greener and cleaner (i.e., less carbonemissions) communication. A fully centralised RAN consistsof taking most of the base stations functionalities away fromthe eNB and leaving only the radio functions at the remoteradio unit (RRU) or RRH. Consequently, BBU, which istraditionally located in the base station cabinet, is relocated tothe cloud or central processor, hence, forming a shared pooltoall connected RRHs. With the C-RAN architecture, the LTEeNB is migrating towards the virtual eNB (VeNB), referringto the joint functionalities of the BBU and RRH that are indifferent locations [9].

In fact, a similar architecture has been deployed, as early asthe second generation (2G) of cellular networks (e.g., GSM),for indoor coverage such as airports, shopping malls, and cor-porate building, called the distributed antenna systems (DAS).It consists of breaking the traditional 2G radio site, called basetransceiver station (BTS), into two parts: the BBU and a setof RRHs. These two parts are normally connected with opticfibre links inside the radio cabinet, thus, the solution requiresremoving and distributing the RRH by prolonging the fibreconnection using radio over fibre transmission. Consequently,spreading diligently these RRHs around a building providesacontinuous and close-to-uniform indoor coverage, irrespectiveof the traffic distribution. DAS may be considered as animplementation option of C-RAN in which quantised signalsare exchanged among the RRHs to enable centralised or de-centralised joint decoding, as described in [43] and [44]. C-RAN is thus an evolution of DAS which introduces the novelconcept of cloud BBU, whereby various BBUs may be locatedin different geographical areas while forming a cloud andconnecting to more than on DAS. However, with the C-RANarchitecture, the covered distances between RRH and BBUare larger than indoor solutions, and fibre is a luxury that isoften unavailable. The fully centralised C-RAN, also referredto as baseline C-RAN configuration, consists of migrating theprocessing functions of layers one, two and three to the centralprocessor, and leaving the basic function of analogue/digitalconversion to the RRH. However, the resulting CPRI-basedfronthaul requirements, in such a configuration, become over-whelming and, worse, independent of the actual traffic load inthe RRH. Indeed, the CPRI-based C-RAN migrates both celland user functions to the BBU, thus burdening the fronthaulwith full-load even when no users are served by the BBH.Moreover, the MIMO-related functions are also migrated tothe BBU resulting in a fronthaul traffic that scales with thenumber of MIMO antennae [14], [45].

Consequently, the level at which the traditional base stationfunctions should be split, has become a prime research topic,termedfunctional split, which aims at finding an ideal split,by analysing the impact of different options on possible gains

and fronthaul exigence. Most of the work in this domain hasbeen conducted through iJOIN and has resulted in essentialquantification of latency and capacity requirements imposedon the fronthaul, for different eNB breaking points [17].Moreover, the impact of the functional split of CoMP andxICIC gains was also analysed by the same group (e.g., [9])with the identification of implementation key challenges. Theultimate joint message from these works promotes a flexibleand dynamic functional split orchestrated by the cloud, sinceall research show that there is no one-solution-fits-all inthis area [45]. Another interesting work, [46], agrees on theneed of various splits for different scenarios and categorisesdifferent split levels with respect to achievable gains andcost imposed on the backhaul network (as a function of thebandwidth and latency required). iJOIN endorse the conceptofflexible cooperative processing through their RAN as a service,RANaaS. The RANaaS enables coordination between cells,thus, interference mitigation, intelligent spectrum utilisation,and energy efficiency in cellular communication [21]. C-RANarchitecture is a leading solution towards economising on thecapital expenditure by using low-cost RRH and pool sharingexpensive BBU. Moreover, data rate boosting radio featuressuch as xICIC, massive MIMO, and CoMP require tight andfast coordination between various cells, hence, would benefitfrom centralised processing.

However, the C-RAN has also disrupted the backhaul net-work architecture and created a new type of links, that isa hybrid between RAN and backhaul: the fronthaul. Theselinks connect essential parts of the virtual eNB, thus canbe considered as RAN parts, but they can also be seenas extensions to the backhaul. Moreover, the C-RAN andcorresponding fronthaul can only be designed jointly to en-sure coordinated performance over the virtual eNB, hence, adisruption to the traditional network design is also incurred.In other words, the functional split can only be decided basedon the available fronthaul solutions, and the required fronthaulperformance can only be stipulated by determining the levelofRAN centralisation. Diverse efforts in the industry are leadingtowards the convergence of using the common Ethernet packetbackhaul for the fronthaul, motivated by the advantages of easeof deployment, inter-operability, and cost, e.g., [13]. Delay andloss of synchronization remain challenges for a the adoptionof Ethernet in the fronthaul and are currently being addressedby the iCIRRUS3 project [12].

Another network architectural revolution is the decouplingof the user data and control plane; a need fuelled by theintrinsically restricted footprint of small cells in a UDN.Mobile users would be crossing cell borders very often ina UDN, thus, generating a debilitating signalling load fromhandovers and cell reselections. By separating the user dataand control planes, handovers and reselections are requiredwhen the user moves between anchor cells only; these aremacro-cells that cater for the control plane while the data planeis tunnelled through various small cells within the macro-cell coverage. The concept of control/user plane split is often

3iCIRRUS (intelligent Converged network consolidating Radio and opticalaccess aRound USer equipment) is an EU Horizon 2020 project.

6

referred to as soft cell or phantom cell. Such a split wasenvisioned for release 12 of LTE but has recently been movedto release 13 [32], [47]. System information is broadcast overthe anchor cell which also manages most of the radio resourcecontrol and signalling, while the small cells play an assistantrole, having data offload as a main task [48]. Furthermore,a decoupling of uplink and downlink connection points isalso suggested to enable greener and cleaner communicationby selecting the network layer (i.e., small cell or macro-cell) that requires least transmit power [49]. The soft cellconcept is strongly endorsed by key 5G pioneers such as NTTDoCoMo [10], iJOIN [6], and MiWEBA4 [50].

C-RAN, control/plane split, and decoupling of uplink anddownlink all necessitate very low latency on the backhaulto ensure coordination and timely synchronisation amongpertinent parallel channels. In addition, the C-RAN architec-ture inflates the effective backhaul throughput such that linkssuitable for an eNB deployment act as a funnel for a VeNB,under the same user traffic load.

III. E VOLUTION TO 5G CELLULAR BACKHAUL NETWORK

5G targets are evidently ambitious and intensify the designchallenges of the backhaul. This section presents a technicalappraisal of backhaul technologies followed by quantified 5Grequirements in order to assess the pertinent performancegaps and shed light on the possible solutions. To this end,a summary of existing and emerging backhaul technologies’performance is first compiled, taken from various key sources.Different 5G deployment use-cases are then considered em-ploying C-RAN and CoMP, since both features affect thestipulated backhaul performance. SCF states that “the backhaulis NOT a barrier to small cell deployment” given the largeavailable backhaul toolbox and the diversity of small celluse-cases [24]. Accordingly, we present a quantitative andrepresentative set of tailored 5G backhaul solutions for variousdeployment scenarios.

A. Backhaul technologies

There are many documents that describe the growing portfo-lio of backhaul/fronthaul solutions including legacy and noveltechnologies such as [18], [19], [20], [21], and [24]. Keyinformation pertinent to this survey is summarised in Tables Iand II for wireless and wired solutions, respectively. For moredetails on a specific topic, readers are invited to refer thecorresponding cited documents in the tables.

Current backhaul networks are mostly built with microwavelinks (often operator owned) and fibre/copper-based links(often leased) with different proportions per operator andcountry [51]. A study conducted in 2014 brings attention tothe fact that optic fibre backhaul is not available nationwide inEurope and that current microwave replacements cannot sus-tain the traffic growth of LTE/LTE-A beyond 2017-2018 [52].Indeed, fibre to the home (FTTH) is scarce worldwide withonly 16 countries exceeding 15% FTTH penetration [53].The need for innovation in backhaul provisioning is thus

4MiWEBA - Millimetre-Wave Evolution for Backhaul and Access

evident and becomes more vital in the dawn of 5G; to thisend, a compilation of available and potential technologiesispresented here.

Any of the listed backhaul technologies could be deployedin different topologies that would, in turn, impact the nominalperformance in Tables I and II. Point-to-point (PtP) linkscould be mounted in chain, tree, ring or mesh networks, forinstance, but the incurred delay will increase with link length,the number of hops, and delay in aggregation/demultiplexingpoints. Point-to-multi-point (PtmP) architectures, on the otherhand, curtail the dependency of the backhaul network perfor-mance on the number of aggregation nodes while enablingeasy addition/deletion/modification of nodes. In the presenceof C-RAN and pooled BBU, PtmP becomes a favourablearchitecture for the mobile fronthaul.

B. 5G backhaul requirements

Evidently, the mobile backhaul/midhaul/fronthaul networkis an essential milestone towards realising a 5G network. Itisclear that more capacity, less latency, synchronisation, security,and resilience are needed. However, it is a pre-requisite toquantify the required improvements in an attempt to identifythe optimum solution (or set of solutions) from the listedoptions in Tables I and II. SCF defines four main use-cases forsmall cell deployments: capacity hotspot, peppered capacityfor quality of experience (QoE) boost, outdoor not-spot, andindoor not-spot as listed in Table III.

Each of the use-cases has different requirements in terms ofcapacity, latency, resilience, and others. For each of these use-cases, different RAN architectures could be deployed, startingwith the traditional D-RAN and different levels of functioncentralisation. We consider five different functional splits, asdepicted in Figure 3; corresponding backhaul requirementsfor a simple deployment of LTE-FDD assuming 20 MHzbandwidth, two receive antennae, and 50% cell load are listedin Table IV. A key aspect is that the backhaul bandwidthrequirement scales with larger radio access bandwidth andbecomes crippling with≥100 MHz possible 5G allocationsper small cell. It should be highlighted that Split A depictsthebaseline C-RAN architecture whereby the RRH and the BBUare connected over a CPRI interface. The round trip time overthe CRPI interface is 5µsec and the effective admissive delayover the fronthaul link, including propagation delay is in theorder of 100µsec. Moreover, Split-A has the most exigentthroughput requirement over the fronthaul, even when theactual user traffic is minimum (or null). All split options aboveSplit-A scale with the actual traffic, hence, allow exploitingthe statistical multiplexing gain based on occupied physicalresources. Moreover, the backhaul throughput requirementbecomes flexible and more relaxed since it depends on theactual user throughput related to the user channel quality.On the other hand, latency requirements remain critical andare determined by the channel coherence time, hence, theuser speed of movement. In these functional splits, latencyrequirements are governed by the hybrid automatic repeatrequest (HARQ), link adaptation, and scheduling processes.

7

TABLE I

WIRELESS BACKHAUL TECHNICAL SOLUTIONS.

Technology Options Upstreamthroughput

Downstreamthroughput

Latency/ Jitter Distance Note

Microwave PtP† PtP 1 Gbps 1 Gbps < 1 msec/ hop 2-4 km 6-60 GHz remote not-spot

Microwave PtmP† PtmP 1 Gbps 1 Gbps < 1 msec/hop 2-4 km 6-60 GHz peppered ca-pacity

Satellite† LOS 15 Mbps 50 Mbps 300 msec one-way latency 5-30 msec jitter

∼ubiquitous due to cost per Mbpsrealistic Tput 2-10 Mbps DL/1-2 MbpsUL

TVWS † NLOS 18 Mbps/ch 18 Mbps/ch 10 msec 1-5 km up to 4 channels up to10 km at 10 Mbps us-ing 2 ch with LOS

mmWave 60 GHz† LOS 1G bps 1 Gbps 200µ sec 1 km scalablemmWave 70-80 GHz† LOS 10 Gbps 10 Gbps 65-350µ sec 3 km scalableSub-6GHz 800 MHz-6 GHz† NLOS 170 Mbps 170 Mbps 5 msec single hop

one way1.5-2.5 kmurban10 kmrural

licensed (20 MHzTDD) expected toincrease to 400 Mbps

Sub-6 GHz 2.4, 3.5, 5 GHz† NLOS 150-450 Mbps

150-450 Mbps

2-20 msec 250 m unlicensed data rate de-pends on MIMO

FSO‡ LOS 10 Gbps 10 Gbps low 1-3 km† Refer to [24].‡ Refer to [54].

Fig. 3. Functional split points considered (UL/DL) [17]. AD: Analogue to Digital; DA: Digital to analogue, CP: Cyclic prefix, FFT: Fast fourier transform;IFFT: inverse FFT, RE: Resource element, FEC: Forward errorcorrection.

Opportunistic HARQ is proposed in [59], which divides theHARQ process into a time-critical part conducted at the RRHand computationally intense part that takes place at the BBU;such an approach relaxes the latency requirements over thebackhaul, as highlighted in Table IV.

In addition, for each use-case coupled with an RAN archi-tecture, different CoMP techniques may be employed. Threelevels of downlink coordination and two levels of uplinkcoordination are considered, as described in Table V. Adeployment scenario is defined by selecting an option fromeach of Tables III and IV, and choosing an UL and DLCOMP option from Table V. Consequently, The combinationset is 4 · 5 · 3 · 2 = 120 deployment options with varyingbackhaul requirements, hence, different tailored solution. Itshould be highlighted, that the mentioned 120 possible de-ployments are only a subset of all actual possibilities sincethey do not include other optional features such as aggregateradio bandwidth, MIMO size, cluster size, data compression,etc. Accordingly, 5G deployment alternatives are numerous;and with each, come different backhaul requirements and acorresponding tailored set of suitable backhaul solutions.

C. Tailored 5G backhaul solutions

By considering the limited options listed in Tables III, IV,and V, it is possible to combine 120 different deploymentoptions. In reality, the required backhaul bandwidth, in aC-RAN architecture, depends also on many factors such asaggregate carrier bandwidth, cell load, the number of sectors,modulation and coding scheme, the number of antennae, andothers. In this paper, we consider a basic set of parameters,consisting of 20 MHz bandwidth, LTE FDD technology, 64quadrature amplitude modulation, two receive antennae, and acell load of 50% (i.e., half of resource blocks are occupied).Moreover, a CoMP cluster size of seven cells is considered.Accordingly, we select the peppered capacity use-case fromTable III, and corresponding backhaul requirements of all5 · 3 · 2 = 30 (five splits, three DL CoMP, and two UL CoMP)possible RAN/CoMP combinations, to all the listed backhaulsolutions in I and II. The results are shown in Table VI.

We would like to emphasise, at this stage, that the datapresented in Table VI is a compilation from diverse sourcesas mentioned in the previous section, and the reader is referredto these sources as listed in Tables I, II, III-A, III, IV, andV

8

TABLE II

WIRED BACKHAUL TECHNICAL SOLUTIONS.

Technology Options Upstreamthroughput

Downstreamthroughput

Latency/ Jitter Distance Note Ref

Fibre PtP PtP ≥10 Gbps ≥10 Gbps <1 msec ∼20 km [24]PON/VDSL2 FTTC 10-

50 Mbps80-100 Mbps

3.8-11.8 msecone way

<1 km cop-per length

[24]

GPON FTTP 1.24 Gbps 2.5 Gbps 1-7 msec up to29 km

[24]

NGPON FTTP 2.5 Gbps 5 Gbps 1-7 msec up to29 km

[24]

EPON FTTP 1 Gbps 1 Gbps 1-7 msec N/A [24]10G-EPON FTTP 1 Gbps 10 Gbps 1-7 msec N/A [24]VDSL2 15 Mbps 75 Mbps 5-15 msec one

way1 km capacity is up

to 100/20 Mbps(DS/US)§at<600 m

[24]

VDSL2 ph2 20 Mbps 100 Mbps 3 msec† 1.5 km phantom mode; 2 pairbonding

[55]

VDSL2 ph4 40 Mbps 230 Mbps 3 msec† 1.5 km phantom mode; 4 pairbonding

[55]

VDSL2 ph8 150 Mbps 750 Mbps 3 msec† 1.5 km phantom mode; 8 pairbonding

[55]

G.FAST 50 m 50 m 1 Gbps‡ 1 Gbps‡ <1 msec‡ 50 m the data rate shown isUS/DS aggregate

[56] [57]

G.FAST 100 m 100 m 500 Mbps‡ 500 Mbps‡ <1 msec‡ 100 m the data rate shown isUS/DS aggregate

[56] [57]

G.FAST 200 m 200 m 200 Mbps‡ 200 Mbps‡ <1 msec‡ 200 m the data rate shown isUS/DS aggregate

[56] [57]

DOCSIS 3.0 108 Mbps 304 Mbps 10-20 msec 1.5 km 8 channels for down-stream; 4 channels forupstream

[58]

DOCSIS 3.1 1 Gbps 5-10 Gbps N/A N/A [58]EuroDOCSIS 3.0 108 Mbps 400 Mbps 10-20 msec 1.5 km 8 channels for down-

stream; 4 channels forupstream

[58]

§ US= Upstream and DS= Downstream.† Realistic latency in the order of 10 msec.‡ These are theoretical values; however, realistic throughput may be sub-1 Gbps and latency closer to 3 msec. Moreover, G.fast is

a TDD technology thus the throughput quoted assumes full occupation of either upstream or downstream.

TABLE III

SMALL CELL FORUM USE-CASES[24] .

Use-case Capacity hotspot Peppered capacity Outdoor not spot Indoor not spotInter cell coordination yes yes no noCoMP possibly possibly unlikely unlikelyFrequency synch yes yes yes yesPhase synch yes yes no noAvailability 99-99.9% 99-99.9% 99.9-99.99% 99.9-99.99%BH Capacity should not be limiting should not be limiting relaxed relaxedEstimated distance† <1 km <1 km 1-5 km <1 kmDistance to hotspot flexible to not spot to buildingGNSS available likely likely likely unlikely

† Distances are practical estimates, not defined in [24].

for more details. However, the information provided is addedvalue, because it consolidates key findings from differentresearch into a practical guide that could help in creating atangible vision of possible 5G deployment options, and iden-tifying areas that require further research and improvement.

Looking at results in Table VI, it is clear that moving thefunctional split towards the medium access control (MAC)layer relaxes both latency and bandwidth requirements, thus,

allows a larger backhaul toolbox and vice-versa. Notably, darkfibre is the only wired backhaul solution that conforms withbaseline C-RAN configuration requirements, while mmWaveand FSO are the only wireless possibilities. Another interestingpoint, inferred from Table VI, is that the functional split ismore dominant in limiting the choice of backhaul solutionsthan the CoMP features. Actually, the only apparent impactof CoMP on backhaul selection is found with the MAC

9

TABLE IV

BACKHAUL REQUIREMENTS FORC-RAN AND FUNCTIONAL SPLIT OPTIONS(SEEFIGURE 3) [46].

Option Throughput LatencySplit A- CPRI - I/Q forwarding 2457 Mbps 150 µ secSplit B- Subframe forwarding 720 Mbps 150 µ sec†Split C- Rx Data forwarding 360 Mbps 150 µ sec†Split D- Soft bit forwarding 180 Mbps 150 µ sec‡Split E- MAC forwarding 27 Mbps ∼10 msec

Scenario considered assumes LTE-FDD 20 MHz bandwidth, 2 receive antennae, and 50% cell load (refer to [17] for full details).† For slow-moving users and with opportunistic HARQ the latency requirement can be relaxed to 4 msec [59].

‡ For slow-moving users the latency requirement can be relaxed to 1-4 msec at the risk of higher probability of block errors.TABLE V

BACKHAUL EXTRA CAPACITY REQUIREMENTS FOR DIFFERENT FORMS OFCOMP AND C-RAN [46].

Option Split A§ Split B§ Split C§ Split D§ Split E§ D-RANDL Joint transmission† 0 200 Mbps 200 Mbps 200 Mbps 200 Mbps 200 MbpsDL Coordinated beamforming† 0 0 0 0 0 N*1DL Coordinated scheduling† 0 0 0 0 0 N*1UL Coordinated scheduling‡ 0 250 Mbps 250 Mbps 250 Mbps 0 N*1UL Joint reception/ MUD‡ 0 250 Mbps 250 Mbps 250 Mbps N*100 N*100§ Centralised functional splits from Table IV.† = Downlink; [‡]=Uplink.

N is the number of cells in the CoMP cluster.

Scenario considered assumes LTE-FDD DL, 2 receive antennae, 20 MHz 10 users in 64 QAM (peak), and 50% cell load.

TABLE VI

POSSIBLE BACKHAUL SOLUTIONS FOR THE PEPPERED CAPACITY USE-CASE.

Functional split UL CoMP DL CoMP Possible backhaul/fronthaul solutions

Split A - CPRI-I/Q forwarding JR or CS‡

JT or CB or CS‡ Fibre point to point†

N-GPONmmWave 70-80 MHz†FSO†

Split B- Subframe forwarding JR or CS JT or CB or CS

Fibre point to pointxPON†G-fast (up to 50 m)†DOCSIS 3.1mmWave 60 MHz and 70-80 MHz†FSO†

Split C- RX Data forwarding JR or CS JT or CB or CS

Fibre point to pointxPON†G-fast (up to 50 m)†DOCSIS 3.1microwave (PtP, PtmP)†mmWave 60 MHz and 70-80 MHz†FSO†

Split D- Soft bit forwarding JR or CS JT or CB or CS

Fibre point to pointxPON†G-fast (up to 100m)†DOCSIS 3.1microwave (PtP, PtmP)†mmWave 60 MHz and 70-80 MHz†Sub-6GHz 2.4, 3.5, 5 GHzFSO†

Split E- MAC data JR JT

Fibre point to pointxPON†VDSL2 with 8 pair bonding†G-fast (up to 200 m)†DOCSIS 3.0, 3.1 and EuroDOCSIS 3.0microwave (PtP, PtmP)†mmWave 60 MHz and 70-80 MHz†Sub-6GHz 2.4, 3.5, 5 GHzFSO†

CS CB or CSsame as JR and JTVDSL2 with 4 pair bonding†Sub-6GHz 800 MHz-6 GHzTVWS (with 2 channels)

‡ JR is joint reception, JT is joint transmission, CS is coordinated scheduling, CB is coordinated beamforming.

† Backhaul technologies marked with†indicate that they conform with both throughput and latencyrequirements. Otherwise, the technology only

accommodates the throughput requirements but is under-performing in terms of latency.

10

data functional split (Split E), in which joint reception andtransmission limits the backhaul toolbox further comparedtocoordinated scheduling and beamforming.

The backhaul throughput requirements of all functionalsplits above Split-A scale with the traffic load; consequently,one can capitalise on tailored usage of the rich backhaultechnology toolbox available. On the other hand, latencyrequirements are more crippling, inasmuch that RAN cen-tralisation (below Split-E) is only possible in areas withrestricted user movement or where dark fibre is available,due to stringent sub-msec latency needs. A recent work byBartlet et al. advocates the importance of converged fronthauland backhaul and a flexible functional split that adapts tothe available backhaul links [60]. The authors also providea comparative study between common (but not exhaustive)toolbox of backhaul technologies and C-RAN requirements,which complies with our findings in Table VI.

D. Joint backhaul/RAN perspective on the tradeoff betweenC-RAN gain and fronthaul cost

C-RAN is presented as a key disruptive technology, vital tothe realisation of 5G networks. However, based on Table VI,C-RAN-Split A is only feasible with direct optical fibre asa wired solution. In this case, how would 5G evolve in theabsence of fibre, knowing that only five countries in Europehave more than 15% coverage of fibre to the home [53]?Wireless CPRI-fronthaul (e.g., mmWave) is another promisingway forward, however, propagation challenges and incurredresilience issues are still partially unsolved and requiremoreresearch to provide a mature reliable fronthaul solution. On theother hand, D-RAN is less demanding on the backhaul but isbelieved to lack in performance in terms of resource usage andefficiency of RAN deployment. Under such constraints, it iscrucial to identify the most suitable architecture from a jointbackhaul/RAN perspective.

The C-RAN versus D-RAN comparison has been addressedqualitatively in the literature, whereas it requires a quantitativeanalysis to enable tangible guidelines for this dilemma. Studiesthat advocate C-RAN for its superior RAN functionality andsignificant RAN cost reduction emphasise that it is onlyfeasible with a fibre-based fronthaul; nonetheless, the latteris often unavailable and very expensive and impractical todeploy. On the other hand, there are studies that promote D-RAN because it operates over a realistic backhaul, but warnsagainst losing the centralisation benefits (cost reductionandease of deploying RAN features). Various functional splitsare also analysed from a fronthaul perspective and resultingreduction in overhead, while highlighting the increase in RRHcomplexity and the incurred limitation in RAN features. Ta-ble VII summarises the general messages from the C-RAN/D-RAN comparison. The gap in these studies is a quantitativecomparison of how much is lost and how much is gained withthe various RAN architectures when looking at the problemfrom a joint backhaul-RAN perspective.

In this section, we present a cost-versus-benefit analysis ofdifferent functional splits, considering three types of back-haul technologies: copper-based G.fast, point-to-multi-point

microwave, and optical fibre based GPON (see Figures 4, 5, 6).The study takes on a joint backhaul-RAN perspective andis based on a holistic network dimensioning method usingbackhaul-aware dynamic cell range extension approach; moredetails can be found in [61]. The difficulty in this analysisstems from tagging a realistic relative cost to each of theadvantages and disadvantages listed in Table VII. The studyis thus based on published cost-related material mostly, andindustry-based estimates where information is not readilyavailable. The total cost of ownership (TCO) of each scenariois computed by adding the CapEX to the five years OpEX.

Figure 7 displays the gains/losses of each of various de-ployment scenarios featuring variable levels of centralisation,compared to the D-RAN. The benchmark deployment scenariois D-RAN with G.fast; the capacity gains/losses of all otherscenarios are derived by comparing their respective cumula-tive effective throughput to the benchmark. In parallel, theincrease/decrease in TCO of each scenario is also defined withrespect to the same benchmark scenario. The diagonal lineseparates the region of advantageous from the unprofitablescenarios; those that fall on the line incur comparable costincrease and capacity gain, those below the line require highercost than capacity gain achieved. The promising deploymentscenarios are those that fall above the diagonal line since thecapacity gains exceed the respective cost increase.

Scenarios deploying PtmP microwave are seen as the leastinteresting solutions since the resulting increase in capacityis comparable, if not less, than the incurred cost. The costof PtmP microwave fronthaul may be reduced if more smallcells fall within its coverage, however, that would increase thecontention of cells to the shared bandwidth and may result inthroughput degradation. The microwave solution consideredhere utilises a licenced spectrum; hence, different licencecost assumptions may alter the results and render the PtmPmicrowave solution more attractive. Interestingly, centralisa-tion of MAC and FEC coding/decoding (Split D) resultsin considerable reduction in cost (16%), while maintainingcomparable throughput as the benchmark scenario. This maybe a efficient migration strategy from D-RAN to C-RAN thatrelies on existing backhaul infrastructure. Contrary to commonbelief, the case study shows that the C-RAN architecture withfibre-based fronthaul is profitable when considered from ajoint backhaul/RAN perspective. The capacity gain is almostdouble the incurred increase in TCO owing to the RAN costreductions due to centralisation and the throughput boost onaccount of the unlimited fronthaul capacity.

Although these results cannot be conclusive because theydepend on delicate cost assumptions; nonetheless, some usefulinsights can be drawn. In the presented case study, the highestgain reached with centralisation is 37% increase in effectivethroughput; on the other hand, the highest increase in cost is30%. Thus, the gain from centralisation dominates the increasein TCO, even when fibre to the cell is assumed. But perhaps amore critical factor than cost is the practicality of layingfibre,which is difficult to capture in the analysis.

11

TABLE VII

COMPARISON BETWEEND-RAN AND C-RAN ARCHITECTURES[61].

Factor D-RAN C-RANCost of RRH / small cell High LowPlanning, deployment, maintenance of RRH / small cell High LowEnergy efficiency of RRH / small cell Low† HighCost of BBU N/A HighPlanning, deployment, maintenance of BBU N/A LowEnergy efficiency of BBU N/A High†Potential for resource pooling Limited HighFronthaul requirements Relaxed ExigentCost of backhaul/fronthaul‡ High HigherLevel of inter-cell coordination Limited Maximum

†On/Off switching in a D-RAN architecture may be used to economise on energy consumption, but due to the complexity of the

small cell, each would require additional energy for cooling and environment control and would still consume more energy when it

is ON. The C-RAN RRH has low complexity, hence, is more robustand requires less energy to operate.

‡The cost of the backhaul in a D-RAN architecture is high in view of the additional capillaries needed to connect all cells to the

backbone. For the same scenario the cost of the fronthaul is higher as a result of higher exigence, thus the need for more bandwidth,

less latency, resilience, etc.

Fig. 4. Last mile of the small cell backhaul employscopper-based G.fast.

Fig. 5. Last mile of the small cell backhaul isprovisioned using is PtmP microwave coverage.

Fig. 6. Last mile of the small cell backhaul isprovisioned using fibre to the home (FTTH).

−20 −10 0 10 20 300

5

10

15

20

25

30

35

40

Increase in TCO (%)

Incr

ease

in T

hrou

ghpu

t (%

)

Fibre−Split AFibre−Split BMicrowave−Split EMicrowave−Split DMicrowave−Split CG.fast−Split D

Fig. 7. The increase/decrease in effective throughput of each scenario, relativeto the benchmark scenario (D-RAN and G.fast), is compared tothe corre-sponding increase/decrease in TCO. The diagonal line separates the profitableand non-profitable regions; scenarios that fall below the line indicate higherincrease in TCO than in throughput.

IV. STATE-OF-THE-ART RESEARCH FOR5G BACKHAUL

Recent years have witnessed a plethora of diverse efforts,from different research bodies, related to the future of back-hauling, including network operators, equipment manufactur-ers, and academia. The topics of research cover a very broadarea and are often interrelated. Hence, it is challenging tolistall related research, in view of the profusion of publications,and even more to categorise them, because they often overlap.Nonetheless, we identify an essential and representative groupof topics that could jointly pave the way to defining the

5G backhaul. Each of these research directions could bediscussed in a dedicated survey with more technical depth;however, the scope of our survey is to explain the entire5G backhaul problem and compile a representative list ofkey and state-of-the-art references. Recent advances in op-tical networks will first be covered, since fibre is often thepreferred technical choice for backhaul (Section IV-A). Next,we introduce recent work on mmWave, among other wirelesstechnologies; a promising alternative for the last mile, wherefibre is not possible (technically or economically) (Section IV-B). This will be followed by a section detailing the arrivaland integration of SDN in the transport network, coveringboth optical and mmWave technologies among other topics(Section IV-C). Energy efficiency in backhauling is anotherimportant subject that has gained serious attention, and toucheson different technologies while often involving self-optimisednetworks (SON) capabilities (Section IV-D). Caching achievedgreat gains in the internet technology and is now being inte-grated into the mobile network at different levels, nonethelessaffecting the backhaul (Section IV-E). Examples of joint RANand backhaul design and optimisation will finally be presentedto demonstrate the potential of this partnership in unlockingnetwork bottlenecks (Section IV-F).

A. Advances in optical networks

Fibre optical connections are an ideal solution for connect-ing the fronthaul in view of their intrinsic low-latency-high-capacity characteristics, that match stringent requirements ofthe C-RAN architecture. Moreover, many cities in our days

12

enjoy a reliable fibre network that could be further exploited toprovide the mobile fronthaul. Ranaweenra et al. have exploredthe benefits of PtmP architecture, reusing existing laid fibre tothe node, as opposed to PtP deployment in [62] and [63].They show cost reductions reaching60% with the usageof passive optical network (PON) with an aggregation nodemultiplexing/filtering 18 fronthaul connections to lamp-postsmall cells.

NTT DoCoMo provide a comparative analysis of timedivision multiplexing (TDM) and wavelength division mul-tiplexing (WDM) for a PtmP architectures in [64]. Currently,TDM-PON systems, such as GPON and GE-PON are deployedas FTTH, offering rates in the order of one Gigabit persecond and are being upgraded to XG-PON and 10G-EPON,thus, taking TDM-PON to the 10 Gbps class. Accordingly,TDM-PON is a cost effective solution that could meet thecapacity demand of 5G fronthaul; however, the incurredlatency, especially on the upstream, is not compliant withrequirements. Moreover, TDM-PON is a rigid architecturethat hinders easy and dynamic adaptability and scalability,thus, limits resource pooling. On the other hand, WDM-PONallows physical sharing of fibre medium by several opticalnetwork units (ONU) while providing a virtual PtP architecturewith a PtP wavelength realisation. WDM-PON solves all theissues facing TDM-PON, i.e., enables scalable, dynamic, andadaptive resource allocation at low latency. Of course, theseadvantages translate directly to a potential reduction in fibrelinks and lower capital expenditure due to resource sharingandpooling. Nevertheless, WDM devices, such as transceivers andmultiplexers/demultiplexers, are still very costly to deploy andmaintain, hence, the advantage on reduced CapEX is not fullyreaped at this stage. Due to the inherent cost of WDM-PONs,NTT explore further the TDM-PON limitations and providea solution for reducing the latency, which is mainly due tothe dynamic bandwidth allocation (DBA) algorithm. In [16],they present a novel DBA which uses radio access controlinformation to allocate the bandwidth on the fronthaul insteadof waiting for uplink data transmission to trigger a report-gatedialogue between the ONU and the optical network terminal(ONT), hence, reducing latency to better than 40µsec.

In another work, a hybrid TWDM-PON solution is pro-posed, which introduces fronthaul aggregation points managedwith WDM-PON, and the ONUs within each fronthaul aggre-gation node are managed with a TDM-PON architecture [65].This proposal offers a balanced solution between the simplicityand reduced cost of TDM while gaining a level of dynamicflexibility and adaptability from the WDM architecture. Thestandardization of the next-generation passive optical networkstage 2 (NG-PON2) has been recently confirmed and relies onTWDM-PON system to provide 40 Gbps bandwidth [66].

An optical technology research group, introduces an or-thogonal frequency division multiplexing (OFDM) schemefor the downstream combined with a TDM scheme for theupstream [67]. The use of OFDMA is motivated by thecapability to establish virtual PtP links in the frequency-domain, thus, serving a high density of cells (200 cells inthe simulations provided with more than100 Mbps per cell).OFDMA cannot be used for the upstream since transmis-

sions are uncoordinated, instead, a judicious hybrid of DSP-enhanced digital radio-over-fibre and TDMA is used, resultingin reduced latency (< 1 msec). The solution proposed uses off-the-shelf devices such as avalanche photodiodes, consequentlyprovides the required flexibility, with high capacity and lowlatency at a moderate cost.

SODALES5 promotes the introduction of an active remotenode (ARN) between the central office (CO) and end user,hence, creating anactive (not passive) optical network [68].The ARN architecture exploits the existing PON where avail-able and employs mmWave technology otherwise to deliverhigh bandwidth wireless final-drop. Moreover, the ARN couldalso act as a central CPRI switch interfacing BBU locatedin the CO to many RRHs, thus, enabling the VeNB/C-RANconcept. Connecting fixed users through fibre and mmWaveand mobile users by feeding eNBs or RRHs from withinthe same node is indeed a novel approach that promises5G essential features such as scalability, adaptability, andefficiency in resource usage. An example is given of two10 Gbps incoming wavelengths feeding the ARN, originatingfrom an arrayed-waveguide grating passive node away fromthe central office. The ingress capacity is distributed in theARN to provide 10 Gbps to each of three small business en-terprises and one eNB and 1 Gbps to 96 residential users, thus,maximising the usage of backhaul resources. Although addingactive components in the backhaul is normally undesirable inview of carbon footprint and added complexity, ARNs couldshare power supply with existing/planned eNBs, or could userenewable energy for the required 1.5 kW.

Fibre-optic-based backhaul is a leading attractive 5G so-lution owing to its superior performance relative to othertechnologies. Advances in this topic such as latency reductionand efficient multiplexing and aggregation are promising,however, they all require extensions in fibre links to connectthe proliferation of small cells. On the other hand, FTTHor FTTB (building) coverage is still limited worldwide andthe task of laying new fibre links to improve it is dauntingin view of the cumbersome trenching and related exorbitantcost, discouraging telecommunication regulators and networkoperators from pushing for such an endeavour. To this end, analternative solution, that is easy and cost-effective to deploy, iscrucial in order to bridge the existing fibre-based network andthe pervasive small cells in an ultra-dense network; mmWavemay be such a solution as examined int he next section.

B. Advances in mmWave

The anticipated overwhelming capacity and data rates arefaced with a legacy spectrum that is overloaded. Advancedfeatures such as UDN, xICIC, CoMP, massive MIMO, andcarrier aggregation are all essential features to reach thetargetcapacity, but spectrum remains a barrier that needs to beunlocked. A. Goldsmith challenged the common belief ofspectrum shortage in a recent talk asking “Do we have a short-age of bandwidth or imagination?” [69]. Indeed, a large chunkin the 60 − 100 GHz spectrum remains untapped and seemsto be an inevitable option for 5G. This bandwidth is largely

5Software-Defined Access Using Low-energy Subsystems

13

unused, globally, but propagation characteristics withinthesaid range differ greatly; accordingly it is often partitioned intotwo bands with the following popular terminology:60 GHzband (or V-band) and the E-band (> 60 GHz).

A pioneering EU FP7 project, BuNGee6, promotes theadoption of mmWave to enable broadband radio access net-works, using mmWave self-backhauling [70]. T. Rappaportand his team have carried intensive work on validating andadvocating the usage of60 GHz, through field measurementsfor both scenarios: wireless mobile access and wireless back-haul/fronthaul [71]–[75]. Their work has set up the base forfurther research in this field by providing tangible propagationmeasurements and link outage results as well as propagationmodelling. By virtue of its inherent high absorption andlimited coverage, mmWave communication is immune to othercell interference, consequently allows tight frequency reuseand maximisation of spectrum efficiency. Line of sight is astrict requirement for mmWave connectivity, however, in abackhaul/fronthaul application, it would be less challengingto ensure a reliable connection since the endpoints of the linkare fixed, thus, enabling high gain antennae located diligently.

Authors in [76] present mmWave as the prominent solutionfor UDNs in 5G, used for boosting data rates to∼10 Gbpsat lower delays (∼1 ms). Moreover, mmWave-based self-backhauling and interference-aware routing are proposed toavoid cumbersome and costly wired fronthaul connections.A work by the Commonwealth Scientific and Industrial Re-search Organization (CSIRO) offers a novel two-tier small-cell backhaul architecture that employs aggregation nodesand integrates sub-6GHz PtmP and PtP E-band links [77].Local small cells are connected to an aggregation node bysub-6GHz PtmP and low-cost medium capacity PtP E-bandlinks. In the top tier of the architecture, various aggregationnodes are interconnected by PtP LOS high capacity E-bandlinks. The proposed architecture pledges a flexible and scalableheterogeneous solution with easy upgrades and additions.

In-band backhauling is another direction gaining momentumin 5G research, which consists of reusing the radio access forwireless backhaul links. The advantages of in-band solutionstem mostly from the reuse of hardware and spectrum, thus,maximising resource utilisation and reducing CapEX. Authorsin [78] provide a solution framework for supporting an in-bandPtmP mmWave backhaul complemented with tradeoff analysisof gains and incurred reduction in radio access capacity. Thejoint in-band backhaul scheduling and interference mitigationin 5G HetNet is addressed in [79] as an optimisation problemwith promising user throughput gains, especially for densenetworks. mmWave deployment in HetNets is also the topicof [80], and is used for radio access and/or backhaul in TDDmode. A novel frame structure is presented that allows multi-plexing and is backward compatible with LTE, in view of timeslot dimensioning. Simulation results show that an aggregatecell throughput of nearly13 Gbps is possible with 10 smallcells per macro sector whilst using mmWave for both radio

6BuNGee: Beyond Next Generation Mobile Broadband. A projectpart ofthe 7th Framework Programme funded European Research and TechnologicalDevelopment.

and multi-hop backhaul. The joint European-Japanese researchproject, MiWEBA, has adopted mmWave as the main enablerfor 5G network, employing the technology for radio accessand backhaul/fronthaul, empowered with control/user planesplits, cognitive radio, and C-RAN [50]. A recent article [81]examines the challenges of incorporating massive MIMO andmmWave technologies in 5G networks to “provide vital meansto resolve many technical challenges of the future 5G HetNet”.

Wireless connections are prominent contenders to fillingthe shortage gaps of optical fibre links in the 5G backhaulnetwork. In-band backhauling is attractive since it does notrequire additional investments or spectrum license, but maynot satisfy the bandwidth needs in many scenarios. PtmPmicrowave, requires additional spectrum license but benefitsfrom high spectrum efficiency since it is shared by multiplesmall cells; nevertheless, may lead to shortage in bandwidthwhen simultaneous traffic peaks occur, as seen in Section III-D. mmWave entails minimum license cost, if any, and hasample bandwidth but suffers from vulnerability to shadowingwhich becomes crippling in a street-to-roof or street-to-streetscenario. Moreover, mmWave propagation is limited and sen-sitive to weather conditions; however, advances in massiveMIMO may be able to address this shortcoming. In brief,wireless backhauling is a promising alternative to fibre links;each of the solutions in this portfolio has distinctive advantagesand shortcomings, but they are all easier and potentiallycheaper to deploy than fibre optic links.

C. SDN in the backhaul

Projections indicate that the market of SDN and networkfunction virtualisation (NFV) market will reach$11 billionin 2018 with 68% share from new segments [82]. These aremostly the virtualised network functions (VNF), but also ports,routers, switches, and optical gear that have become SDN-capable. Open Networking Foundation (ONF) is the enginebehind the promotion and adoption of SDN through openstandards development (e.g., OpenFlow). SDN essentiallydecouples control from the data forwarding function, in aprogrammable manner, thus, creating “a dynamic, manageable,cost-effective, and adaptable architecture that gives adminis-trators unprecedented automation, and control” [83].

SDN is certainly taking cellular networking by storm, and isseen as a crucial facilitator to 5G networks by many key play-ers. SODALES’s vision is that SDN will allow multi-operators,with multi-RAT technology, to share the same heterogeneousphysical network, thus, exploiting resource utilisation andreducing CapEX and OpEX [84]. SDN-enabled fronthaul isproposed, by the same group, using CPRI over Ethernet andthe ARN, as detailed in [68]. Furthermore, distributed securityis implemented, using SDN, with direct links that are confinedinside the access domain, hence, achieving low latency.

NEC’s research group introduces, in [85], a novel softwaredefined networking tool: the backhaul resource manager, toprovision a flexible high-capacity hybrid mmWave/optical mo-bile backhaul network. In the proposed architecture,60 GHzand E-band mmWave technologies are employed for high-capacity last mile and pre-aggregation backhaul, comple-mented with OFDMA-PON as previously introduced in [67].

14

The backhaul resource manager performs automated dynamicresource provisioning and capacity-aware path computation,consequently improving fairness, network utilization andend-to-end user QoE. The same group adopts OpenFlow to enablesoftware-definedλ-flow architecture for flex-grid OFDMAmobile backhaul over PON in [86] and [87]. Furthermore, theypropose an SDN-controlled optical topology for reconfigurablefronthaul for bidirectional CoMP and low latency inter-celldevice-to-device (D2D) connectivity in [88].

Authors in [89] and [90] propose a virtualised architecturefor next generation systems in which the control plane con-sists of a group of SDN applications starting from the basestations i.e., VeNB, backhaul transport, mobility management,radio access, caching, monitoring, and service delivery. Thebackhaul is realised through the usage of OpenFlow andcarrier grade Ethernet switches, where the I-SID (Instanceservice ID in 802.1ah) is used to mark the path between thefirst aggregation point and the internet gateway, the B-VLAN(backhaul virtual local area network) tag to separate traffic ofdifferent virtual operators and the C-VLAN (customer VLAN)to identify a user in one eNB.

SK-Telecom endorse SDN in the transport network, andpropose an SDN-based unified Converged Transport Net-work in [91]. They also report on successful SDN transportprojects, such as Google, who used SDN to improve theirresource utilisation, and CORONET who employed SDN toautomatically and efficiently reconfigure the network and datadistribution in case of natural disaster. A recent work in thiscontext demonstrates the strength of SDN in optimising theperformance of the mobile backhaul network by dynamicallyfinding the optimum backhaul route (based on latency andavailable capacity), allocating required wavelength, andinstan-tiating the location of the local controller (or BBU) [92]. Theauthors assume a fibre-based meshed network and show greatimprovement in throughput and reduction in packet loss withthe proposed SDN-based algorithm.

The strengths of developing an SDN-based backhaul aremanifold. Firstly, the inferred separation of control and dataforwarding facilitates the co-existence of heterogeneousback-haul links. In addition, such an architecture expedites thepos-sibilities of adding, extending, and dynamically reallocatingresources in the backhaul network. On the other hand, SDNavails the backhaul network for multi-operators and multi-technology sharing, dampening the cost per bit to the enduser and maximising the resource usage efficiency. Similarendeavours to engage competing network operators into shar-ing network infrastructure have often been faced by reluctantconcerned parties. However, the fact that SDN allows networkoperators to have virtual control over their backhaul maysucceed in convincing them, especially when affronted withthe deterring cost of building, operating, and maintainingthe5G backhaul. On the other hand, the separation of control anddata forwarding exposes the network to security challenges,especially when used with cloud computing, due to malicioususage or malfunctioning in the system. Authors in [93] exposesecurity breaches that may result from masquerading as adata plane and overwhelming the control plane with denialof service (DoS) attacks, or errors in the system or malicious

software that compromises the security of the control plane.Some solutions are proposed, nevertheless impacting on thesystem response latency and, in some cases, limiting thescalability of the network.

D. Energy efficiency in the backhaul

Energy efficiency, in next cellular generations, is a globaland paramount requirement driven by the desire to reducecommunication carbon footprint, as well as energy bills, andextend terminal battery life. 5G systems are expected to caterfor the explosive rise in devices and capacity without causinga dramatic increase of energy consumption. Until recently,studies on energy consumption of wireless communicationsystems have diverted from the backhaul contribution, due toits trivial role in macro-cell networks (5% according to [94]).However, with the invasion of small cells, blossoming ofHetNets, and creation of UDNs, the backhaul mark of energyconsumption is expected to grow to50% [94]. Consequently,solving the backhaul bottleneck entails looking at the energyaspect which is as important as capacity and latency.

Tombaz et al. consider three deployment scenarios for thefuture backhaul: fibre to the node (FTTN) with VDSL2 tothe cell, microwave, and a hybrid solution of fibre to thebuilding (FTTB) and microwave [94]. Through simulations,they show that the first deployment scenario is more energyefficient than the one that employs microwave only, whereasthe hybrid scenario outperforms both in a UDN, capitalisingon existing fibre infrastructure. In another work, mmWave forbackhaul is investigated from an energy consumption pointof view comparing different spectrum bands and deploymentscenarios [95]. As expected, provisioning wireless backhaulfrequencies at lower frequencies results in higher energyefficiency. Moreover, in a small cells network, the energyconsumption difference resulting from the wireless frequencybecomes negligible in view of the high gain realised with thebackhaul architecture.

An earlier EU FP7 project, BuNGee, proposes a joint designof backhaul and access networks, using heterogeneous radioelements and a cognitive radio backhaul approach enabled bySON capabilities [4]. In one of their publications they offer agreen-oriented implementation of the cognitive backhaul [96]in which user association is geared towards prioritising RRHswith higher load, when possible, to allow a higher number ofRRHs to be in sleep mode, thus, economising energy.

Motivated to design green networks, authors in [97]and [98], propose an ICIC resource allocation scheme thatis energy-aware, thus, improving energy efficiency (by upto 50%) at the expense of reduction in spectrum efficiency.Backhaul energy consumption is incorporated in the totalenergy budget, and fibre is shown to consume less energythan microwave in a heterogeneous backhaul. Authors in [99]first study the energy impact of various backhaul technologiesunder two scenarios: uniform UE distribution and hotspot.They show that mmWave is the most efficient solution andthat the backhaul could consume up to78% of total energy ifprovisioned using sub-6GHz technology in a hotspot scenario.Next, they elaborate an energy-aware cell-association scheme

15

based on cognitive heuristic algorithm with two objectives.It first exploits the available context-aware information tofind the path with the least number of hops, in order tominimize the backhaul energy consumption. Then it selectsthe less loaded backhaul route, in case more than one optionis available, to achieve load balancing. The proposed algorithmis shown to consistently improve energy efficiency, especiallyin a hotspot scenario, in which42% amelioration is seencompared to the normal UE association criteria being thereference signal received power (RSRP).

Green 5G network operation is no more optional and therapidly rising role of the backhaul in energy consumptionmakes energy efficiency an imperative 5G backhaul objec-tive. The major challenge, however, is to achieve this goalwithout compromising other performance indicators such asuser throughput or network mean packet delay. This recentdimension renders the 5G network (both RAN and backhaul)optimisation highly complex with multi-objectives and multi-constraints. To this end, SON becomes an essential tool inthis endeavour. 5G network elements such as radio cells,aggregation points, routers, etc., need to be equipped withSON while the holistic optimisation is orchestrated jointly bythe RANaaS and a similar backhaul entity.

E. Caching for the backhaul

A promising way, that is rapidly gathering momentum, forsolving the backhaul bottleneck, is to cache the content at theedge of network, namely at the small cells and UEs. Caching,thus, transforms the network intelligence from being “reac-tive” to “proactive”, and leverages the latest developments instorage, context-awareness, and social networking [100].Thus,if user data was predicted and cached in advance during lowtraffic periods, it can be transmitted during peak hours withoutburdening the backhaul while still achieving good QoE.

Bastug et al., in [101] elaborate on the role of cachingin a small cell network, with respect to backhaul alleviationand D2D communication, when the context is pre-stored ina UE within reach. Authors in [102] propose a distributedalgorithm, based on alternating direction method of multipli-ers, to optimise the choice of files from a fixed cataloguefor every storage-capable small cell. Another paper proposesa user association scheme that aims at improving user QoEby exploiting small cell caching capabilities to overcomebackhaul constraints [103]. In the proposed approach, smallcells individually look at content availability, realisable datarates (with respect to interference and backhaul capacity)anddecide which UEs to serve accordingly.

A novel solution framework of cache-induced opportunisticCoMP, enabled by caching a portion of media files at the smallcells, is proposed in [104]. The challenge is to decide on whichfiles to pre-code, how to generate constructive MIMO pre-coding, and in which small cells to store the pre-coded file.The mixed-timescale (short term for MIMO pre-coding andlong term for cache control) optimisation problem is solvedby exploiting the timescale separations.

A novel cache-aware user association algorithm is proposedin [105] which minimises the backhaul usage of each small

cell while respecting the quality of service (QoS) requirementsof users. A survey on recent progress in this field is providedin [100] with a historic on usage of caching, benefits, andintegration in cellular networks.

Nonetheless, the success of caching remains conditionalupon many challenges ahead, such as the storage capacityof cells, very large catalogue size of users’ files, and theneed for fast and dynamic learning of cells while makingthe caching decision. These challenges have been exposedin [106], in which the authors exploit big data and applymachine learning for the purpose of proactive caching. Al-though major complications remain unsolved, the potentialofcaching is nevertheless promising, rendering it a prominent 5Gbackhaul research direction.

F. RAN/Backhaul joint design and optimisation

Although 3GPP considers that the backhaul is a separate en-tity from the RAN, nonetheless, most research paths discussedso far require some level of coordination between the backhaulnetwork and the radio access network. Indeed, advances inoptical technologies are geared towards dynamic wavelengthallocation, coordinated RAN and backhaul resource allocation,and the addition of the active remote node, which all benefitfrom having access to RAN information (see Section IV-A). A major part of research, related to mmWave, considersin-band backhauling or rely on cognitive radio, equippedwith intelligence solicited from RAN, thus, joint RAN andbackhaul operation is crucial (see Section IV-B). Energyefficiency in backhauling tackles wired and wireless backhaulaccess technologies and architectures as part of the globalenergy consumption model, thus, requires close collaborationbetween RAN and backhaul to yield constructive results (seeSection IV-D). Research on SDN-enabled transport networkintersects with all other listed study groups and acts as anenabler to dynamic, flexible, and adaptive green backhaul-ing. Accordingly, SDN benefits and builds on coordinationand information exchange between RAN and backhaul (seeSection IV-C). Caching is another feature that capitaliseson context-awareness and instantaneous network information(e.g., system interference, QoS requirements, and backhaulcapacity) to achieve gains in alleviating the backhaul trafficduring peak hours, consequently, coordination between RANand backhaul is essential (see Section IV-E).

An aggressive joint radio access and backhaul design wasintroduced by the BuNGee project earlier in 2010-2012, pro-moting the benefits of such joint operation for the purposeof optimised performance and efficiency [107]. The outcomesof the BuNGee project were used in the ETSI (EuropeanTelecommunications Standards Institute) technical reports onBroadband Radio Access Networks (BRAN [108], [109]).iJOIN’s view of network evolution towards 5G is that “...blur-ring borders between access and the backhaul networks requirea joint design of both...” [6]. iJOIN have identified jointbackhaul/RAN design as a prime enabler to next generationnetworks and have pinned the terminology RAN as a serviceRANaaS to define a flexible RAN architecture that is neitherfully distributed nor fully centralised [6]. Furthermore,the

16

backhaul and RAN cooperation is classified in two distinctcategories: backhaul/RAN awareness and joint RAN/backhaulfunctional design [110].

Examples of backhaul/RAN awareness are many, such asbackhaul aware resource allocation (e.g., [97], [98]) and cellassociation (e.g., [111]). A recent work looks at adjustingthe radio coverage of a cell in view of backhaul availabilityand capacity, exploiting reinforcement learning to adjustthecell range extension offset [112]. The RAN uses backhaulinformation to redistribute users in a way that maximisesuser QoE and that adapts to temporal backhaul constraints.Thus, the proposed scheme is a typical joint backhaul/RANawareness, realised using SON capabilities. Authors in [113]use a centralised optimisation mechanism to also adjust thecell range extension offset, in order to minimise the meannetwork packet delay. Another new article addresses the issuesof backhaul latency and resilience through a backhaul-awareuser association that aims at improving QoS while balancingthe network load [114]. A recent paper by N. Wang et al. offersa radio resource management perspective to the 5G backhaulproblem [115]. The authors discuss the potentials of backhaul-aware resource allocation in a multi-RAT environment andpropose the usage of a unified wireless backhaul bandwidthallocation in a small cell case study employing in-band back-hauling and massive MIMO.

On the other hand, joint functional design consists ofnetwork-wide functionality such as global energy efficiencyoptimisation (e.g., [94], [96]) or spectral efficiency max-imisation (e.g., [78], [80], [116]). iJOIN in [9] present anovel architecture for next generation systems in which dataand control planes are decoupled. RRHs are used for dataoffloading, but an anchor point (that ideally overlooks severalRRHs) is dynamically configured for the control plane. Anetwork controller node, that communicates with the VeNBcontroller and the SDN-enabled backhaul, finds an adequateanchor point for each incoming UE, based on QoS. It alsodetermines the backhaul optimal route based on the anchorpoint, QoS, as well as the current network status, takinginto account energy consumption in the RAN and backhaul,congestion, and requirements of the VeNB.

Another advantage of joint design reported in [6] is theflexible control of CoMP modes depending on application re-quirements, network status, and backhaul constraints. Anotherperspective is offered in [116] which proposes in-band TDDbackhauling and optimised resource allocation scheme thatmaximises user throughput. A recent work by Wang et al., il-lustrates the potential of SDN-based framework for joint RANand backhaul operation through logically centralized manage-ment of IP-based mobility and energy consumption [117].

5G network operators are concerned with one prime ob-jective: maximise their revenue. To this end they need tomaximise the users’ QoE to increase their market share whileminimising the network expenditure. In previous networks,the radio access was the main bottleneck and the networkoptimisation consisted largely of reducing the number of cellsand maximising their spectral efficiency. On the other hand,5G comes with broader challenges and new opportunities, asdetailed in Section II, and network optimisation has become

an end-to-end endeavour in which joint RAN/backhaul designplays a central role.

V. CONCLUSIONS ANDOUTLOOK

5G backhaul research is probably at its peak, witnessinga profusion of published papers and focused research fromkey 5G players. Unlike incumbent cellular generations, 5Gis partly an evolution of existing technologies but is alsobased on disruptive technologies affecting all parts of thenetwork, nonetheless the backhaul, and revolutionising thetraditional approaches to network design. As a result, the5G backhaul challenges are manifold:>10 Gbps capacity,<1 msec end-to-end latency, high security and resilience, timeand frequency synchronisation, low energy consumption andlow cost. None of the current backhaul solutions can deliverall of the above as a stand-alone solution; perhaps fibre-optic-based backhauls rank the best in all aspects, except cost. Tothis end, the backhaul portfolio has broadened to include newtechnologies, such as mmWave and sub-6GHz spectrum, in-band backhauling, in addition to innovations in wired backhaultechnologies. Besides, heterogeneity prevails in 5G networksand describes all network levels including users, services,RAN, and backhaul. Consequently, a realistic 5G backhaul isone that is comprised of many backhaul technologies. Besides,it is flexible, adaptive, dynamic to allow catering for the5G stringent performance needs where possible (and needed)whilst adapting the RAN network to its hard limitations andconstraints in case of more relaxed requirements.

The survey identifies six key research directions that wouldjointly pave the way to 5G backhaul, as presented in Sec-tion IV. Key surveyed sources are tabulated in Table VIII,highlighting, in each case, the sub-topics covered such as back-haul technologies, optimisation objective, RAN architecture,and RAN options, among others.

A. Lessons learnt

Key lessons drawn from the inspection of the 5G backhaulproblem and state-of-the-art related literature are summarisedbelow:

• Lesson 1: In summary, there is no-one-solution-fits-all in5G backhaul and, more importantly, there is no uniqueset of 5G backhaul requirements. The main lesson learnt,towards designing the future backhaul, is that we needto make the best of existing transports networks, evolveincumbents solution (e.g., xPON), and explore new tech-nologies such as mmWave, sub-6GHz, FSO, etc.

• Lesson 2: Dynamic, adaptive and flexible operation of the5G backhaul is the most stringent requirement, stemmingfrom the heterogeneity of the backhaul and networkrequirements and the need to efficiently adapt networkresources in a timely manner.

• Lessons 3: Moreover, the fusion of the RAN and backhaulis a dominant shift, rendering the joint design, operation,and optimisation of both traditional parts of the networkcrucial to the success of 5G. Thus, a joint RAN/backhaulperspective is critical in assessing backhauling solutions,leading to different outcomes when compared to the

17

RAN-unaware backhaul ranking. We have shown thatsome levels of RAN centralisation are always beneficial,even when deployed with copper-based backhaul and thatthe gain attained from C-RAN with fibre-based fronthaulprevails the incurred cost, rendering the solution moreadvantageous than previously believed.

• Lesson 4: The overwhelming growing size of the networkand relevant parameters, and the mesh-like growth ofthe 5G backhaul dictate employing SON to automateorganisation and optimisation of the network in a dis-tributed manner. The main advantage of SON is its fastadaptability to the dynamic network, relative to a centrallyoptimised solution. The challenge is to design efficientSON algorithms with low complexity to avoid an increasein cost and energy consumption of network elements thatemploy these algorithms.

• Lesson 5: Another critical lesson, drawn from this re-search, is that technology adoption from informationtechnology (IT), such as SDN, will play a key role in the5G backhaul evolution. These technologies are facilitatorsto backhaul management in the presence of heterogeneitybut also render infrastructure sharing more attractive tovarious concerned parties, leading to reduced cost andenergy consumption. However, an SDN architecture mayexpose network security; an open problem that should beaddressed without compromising the network adaptabilityand flexibility.

B. Consolidated 5G backhaul solution

Based on the lessons learned, a promising 5G backhaulvision would be part of a whole network restructuring in whichthere are no boundaries between radio access and transportnetwork. Indeed, the RAN functionality would be presented asa cloud-service, RANaaS, and joint design and optimisationofRAN and backhaul as a solution to coordinated evolution (referto [6]). Interworking of access and backhaul networks wouldenable dynamic functional split in the C-RAN in view ofconstraints and requirements from both domains, and dynamiclink setup in the heterogeneous backhaul network based onUDN status.

A new SDN-enablednetwork layer functionthat has globalnetwork visibility, would complement the joint RAN/backhaularchitecture, thus, allowing fully coordinated RAN and back-haul operation. Such a solution would provide the crucialflexibility, scalability and adaptability needed, and would allowopening the physical network to multi-operators with multi-RAT and multi-vendors while enforcing required securityand virtual individuality (mobile virtual network operatorMVNO). This forms the utmost level of resource sharing andpooling, hence, meets cost constraints in both capital andoperational expenditures. Indeed, the 5G PPP present their5G vision as one that“... will integrate networking, computingand storage resources into one programmable and unifiedinfrastructure” [124]. They foresee that telecom and IT will beintegrated towards a common, very high capacity, ubiquitousinfrastructure in 2025. Hence, the evolution and integration ofSDN/NFV is seen to play an essential role.

Based on this survey and key players’ visions of 5G, theconsolidated 5G backhaul solution can be provisioned as aservice (BHaaS), that is part of a software defined network,with common RAN intelligence, SON, and caching capabil-ities, that operates on a heterogeneous physical network ofwired and wireless connections as shown in Figure 8. The jointBHaaS/RANaaS collaboration ensures a holistic visibilitytothe end-to-end network and enables coordinated optimisationand operation. The BHaaS, based on gathered backhaul andRAN dynamic network data, performs the first level of opti-misation which entails adjusting the prioritisation of networkgoals and disseminating network faults, additions, load, etc.The second level of optimisation is SON-based and distributedover the network elements (e.g., routers, gateways, aggregationpoints, multiplexers, radio cells, etc.), but is guided by theinformation stemming from the BHaaS. Consequently, thebackhaul network is inherently RAN-aware and dynamicallyadapts to network changes and conditions.

In order to demonstrate further of the BHaaS, as presentedin Figure 8, we present typical examples of information flow.As proposed in [110], the RANaaS dynamically controlsthe “flexible” Cloud-RAN based on backhaul changing ca-pabilities and constraints; the BHaaS, in this case, has thecomplete visibility of the backhaul network conditions andisable to report required information to the RANaaS, allowingbackhaul-aware RAN service provisioning. Backhaul networkstatus information is permanently changing due to varyingtraffic load, link outages, router faults or overloading, etc.In this case, the role of the BHaaS becomes even moreimportant when multiple operators including cellular and fixedservice provisioning share the same backhaul network. In suchsituations the RANaaS would not have access to full back-haul information whereas the BHaaS, owing to its networkvirtualisation capabilities would be able to liaise the neededinformation without breaching operators confidentiality.Onthe other hand, the BHaaS collects timely information relatedto the RAN status from the RANaaS and dynamically adjuststhe routeing tables, ingress/egress bandwidth, and link optimi-sation based on various needs and traffic conditions from theRAN. Another example pertains to the backhaul-aware user-cell association, such as presented in [112], in which the radioaccess cells need to be continuously informed of the link statusinformation of the connecting fronthaul/backhaul to adjust thevirtual cell ranges accordingly. In such a scenario, the BHaaSmanages the exchange of this information, which could includecapacity, delay, energy efficiency, resilience etc.

C. Unsolved challenges

Progress from the research community is reducing thegap between 5G backhaul requirements and backhaul capa-bilities, however, major challenges remain along the way.The dominant disparities are capacity, synchronisation, andlatency. We have demonstrated in this survey that advances intechnologies such as fibre, copper, mmWave, sub-6GHz, andFSO have scaled down the capacity challenge considerably.However, these solutions are not fully developed yet and sufferfrom high cost, unreliability, or shortage of bandwidth. The

18

Direction Fib

reop

tics

mm

Wav

e

Oth

erB

Hte

chno

logi

es

SD

N

Ene

rgy

effic

ienc

y

Cac

hing

Join

tR

AN

/Bac

khau

l

Bac

khau

lth

roug

hput

Bac

khau

lla

tenc

y

Bac

khau

lsy

nchr

onis

atio

n

C-R

AN

C-p

lane

/U-p

lane

Spl

it

UL

/DL

deco

uplin

g

Cos

t

SO

N

[4] X X X X X X

[6] X X X X X X X X

[8] X X X X X X

[9] X X X X

[10] X X X X X X

[12]–[14] X X X X X

[16] X X X X X

[17] X X X X X X X

[18]–[20] X X X X X

[21] X X X X X X X X X

[24] X X X X X X

[31] X X X X X X X X X X

[33], [34] X X X

[35] X

[43] X X

[45] X X X X X X X

[46] X X X X X X X X

[49] X X X

[50] X X X X X X X X

[51], [52], [55] X X X

[53] X

[56] X X X

[57] X X X X

[58] X X X X

[59] X X X

[60] X X X X X X

[61] X X X X X X X

[62], [63] X X X

[64], [65] X X X X X X

[67] X X X

[68] X X X X X X

[71]–[75] X

[76] X X

[77] X X X X X

[78], [79], [81] X X X

[80] X X X X

[84] X X X X X X X X

[85] X X X X

[86]–[88] X X X X

[89]–[91], [93] X

[92] X X X X X

[94] X X X

[95] X X X

[70], [96], [107] X X X X X X X

[97], [98] X X X X X

[99] X X X X X X

[100], [101], [104] X X X

[102] X X X X

[103] [105] X X X X

[110] X X X X X X X X X

[113], [115] X X X X X

[114] X X X X X

[116] X X X

[117] X X X X X

[118] X X X

[119] X X X

[107]–[109] X X X X X X X X X

[112] X X X X X X

[120], [121] X X X

[122], [123] X X X X

TABLE VIII

STATE-OF-THE-ART CATEGORISED RESEARCH AND SOURCES.

19

Fig. 8. Heterogeneous radio access and backhaul networks with diverse user devices and applications. The joint BHaaS/RANaaS collaboration allows network-wide visibility and enables dynamic optimised network operation. The BHaaS overlooks the backhaul network operation and adapts it based on informationfrom the RANaaS such as adjusting routeing tables, optimising bandwidth allocation for egress/ingress traffic flows, traffic balancing, self-healing, etc.

remaining challenge is developing and intelligent, adaptive,and dynamic adoption and allocation scheme of these solutionsin an optimised manner that capitalises on the heterogeneityof the backhaul network while catering for the diversity ofusers’ requirements.

There are two types of synchronisation: frequency andtime/phase. Frequency synchronisation is needed in all celldeployment use-cases but is also possible with most backhaulsolutions such as xDSL, xPON, and PtP connections. Phaseand time synchronisation is required with features such asCoMP but is not available in all backhaul techniques. Globalnavigation satellite system (GNSS) assistance is a possiblesolution in some cases, but increases the complexity and costof the small cells and does not operate in indoor small cellsolutions. Arguably, indoor small cells are unlikely to employCoMP schemes, nevertheless, the need and incurred cost ofGNSS on all outdoor small cells motivate more research to-wards alternative solutions for phase and time synchronisation.

However, the prevailing difficulty resides in bringing thebackhaul latency to the required levels of C-RAN and CoMP,i.e. down to 150µsec (see Table IV). Currently, direct fibreand mmWave are the only technologies capable of such lowdelay, but future research is expected to make more optionsavailable. However, direct fibre is often not available andwould be too cumbersome to lay, and mmWave is, relatively,a low-cost emerging technique facing major challenges relatedto propagation. This limitation has many implications and ispredicted to be a leading research motivation.

Another key challenge is to capture the diverse performanceaspects of the 5G heterogeneous transport network in ananalytical model, to enable evaluation and assessment of inno-vative 5G backhaul solutions. Different works have addressed

modelling of various backhaul performance indicators. Forinstance, the cost of the backhaul network has been modelledin view of the technology deployed and the network topologyin different works, such as [120]–[122]. Authors in [122],[123] propose analytical models to capture the delay of thebackhaul network assuming it is wireless or heterogeneous(i.e., a combination of wired and wireless technologies),respectively. Authors in [123] model the delay in networksusing heterogeneous backhaul solutions, composed of fibrelinks, xDSL, mmWave, and sub-6 GHz, and derive the meanpacket delay over both the radio and backhaul networks.Given that energy consumption has pivotal importance infuture networks, recent works have addressed modelling thisaspect based on carried traffic and topology, such as [126]and [127]. Reliability and security of the backhaul are alsocritical and are captured in the proposed analytical modelin [128]. This is certainly a key research direction that stillrequires development to represent fully the performance andconstraints of a heterogeneous 5G backhaul network composedof different backhaul technologies.

ACKNOWLEDGMENT

The views expressed here are those of the authors and donot necessarily reflect those of the affiliated organisations.The authors would like to thank the UK Engineering andPhysical Science Research Council (EPSRC) and BT Researchand Innovation for funding this research through an IndustrialCooperative Awards in Science & Technology (iCASE) stu-dentship. We would also like to acknowledge the support ofthe University of Surrey 5GIC (http://www.surrey.ac.uk/5gic)members for this work.

20

REFERENCES

[1] D. Soldani and A. Manzalini, “Horizon 2020 and beyond: Onthe 5Goperating system for a true digital society,”IEEE Vehicular TechnologyMagazine, vol. 10, pp. 32–42, Mar 2015.

[2] SK-Telecom, “SK Telecom’s view of 5G vision, architecture, technol-ogy, and spectrum,” white paper, 2014.

[3] G. Wunder, P. Jung, M. Kasparick, T. Wild, F. Schaich, Y. Chen,S. Brink, I. Gaspar, N. Michailow, A. Festag, L. Mendes, N. Cas-siau, D. Ktenas, M. Dryjanski, S. Pietrzyk, B. Eged, P. Vago,andF. Wiedmann, “5GNOW: non-orthogonal, asynchronous waveforms forfuture mobile applications,”IEEE Communications Magazine, vol. 52,pp. 97–105, Feb 2014.

[4] Z. Roth, M. Goldhamer, N. Chayat, A. Burr, M. Dohler, N. Bartzoudis,C. Walker, Y. Leibe, C. Oestges, M. Brzozowy, and I. Bucaille, “Visionand architecture supporting wireless Gbit/sec/km2 capacity densitydeployments,”Future Network and Mobile Summit, Jun 2010.

[5] V. Jungnickel, K. Manolakis, W. Zirwas, B. Panzner, V. Braun,M. Lossow, M. Sternad, R. Apelfrojd, and T. Svensson, “The role ofsmall cells, coordinated multipoint, and massive MIMO in 5G,” IEEECommunications MagazineÂa, vol. 52, pp. 44–51, May 2014.

[6] C. Bernardos, A. De Domenico, J. Ortin, P. Rost, and D. Wubben,“Challenges of designing jointly the backhaul and radio access networkin a cloud-based mobile network,”Future Network and Mobile Summit(FutureNetworkSummit), Jul. 2013.

[7] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch,M. Mater-nia, O. Queseth, M. Schellmann, H. Schotten, H. Taoka, H. Tullberg,M. Uusitalo, B. Timus, and M. Fallgren, “Scenarios for 5G mobileand wireless communications: the vision of the METIS project,” IEEECommunications Magazine, vol. 52, pp. 26–35, May 2014.

[8] NGMN-Alliance, “Small cell backhaul requirements,” white paper, Jun2012.

[9] P. Rost, C. Bernardos, A. Domenico, M. Girolamo, M. Lalam,A. Maeder, D. Sabella, and D. Wubben, “Cloud technologies forflexible 5G radio access networks,”IEEE Communications Magazine,vol. 52, pp. 68–76, May 2014.

[10] H. Ishii, Y. Kishiyama, and H. Takahashi, “A novel architecturefor LTE-B: C-plane/U-plane split and phantom cell concept,” IEEEGlobecom Workshops (GC Wkshps), Dec. 2012.

[11] S. Onoe, “The dawn of 5G.” http://www.ieeevtc.org/conf-admin/vtc2013fall/7.pdf, Sep. 2013.

[12] N. J. Gomes, P. Chanclou, P. Turnbull, A. Magee, and V. Jungnickel,“Fronthaul evolution: From CPRI to Ethernet,”Optical Fiber Technol-ogy, vol. 26, pp. 50–58, 2015.

[13] C.-L. I, Y. Yuan, J. Huang, S. Ma, C. Cui, and R. Duan, “Rethinkfronthaul for soft RAN,” IEEE Communications Magazine, vol. 53,no. 9, pp. 82–88, 2015.

[14] C. L. I, J. Huang, Y. Yuan, S. Ma, and R. Duan, “NGFI, the xHaul,”in 2015 IEEE Globecom Workshops (GC Wkshps), Dec 2015.

[15] 3GPP, “An interview with Philippe Reininger - RAN3 chairman,” 2015.[16] T. Tashiro, S. Kuwano, J. Terada, T. Kawamura, N. Tanaka, S. Shige-

matsu, and N. Yoshimoto, “A novel DBA scheme for TDM-PONbased mobile fronthaul,”Optical Fiber Communications Conferenceand Exhibition (OFC), Mar. 2014.

[17] A. Maeder, M. Lalam, A. de Domenico, E. Pateromichelakis,D. Wubben, J. Bartelt, R. Fritzsche, and P. Rost, “Towards a flexiblefunctional split for cloud-RAN networks,”European Conference onNetworks and Communications (EuCNC)Âa, Jun 2014.

[18] H. Raza, “A brief survey of radio access network backhaul evolution:Part i,” IEEE Communications Magazine, vol. 49, pp. 164 – 171, Jun2011.

[19] H. Raza, “A brief survey of radio access network backhaul evolution:Part ii,” IEEE Communications Magazine, vol. 51, pp. 170 – 177, May2013.

[20] O. Tipmongkolsilp, S. Zaghloul, and A. Jukan, “The evolution ofcellular backhaul technologies: Current issues and futuretrends,”Communications Surveys & Tutorials, IEEE, vol. 13, pp. 97 – 113,First Quarter 2011.

[21] iJOIN, “D2.1 State-of-the-art of and promising candidates for PHYlayer approaches on access and backhaul network,” Tech. Rep. D2.1,INFSO-ICT-317941 iJOIN, Nov. 2013.

[22] iJOIN, “D3.1: Final report on MAC/RRM state-of-the-art, require-ments, scenarios and interfaces in the iJOIN arch,” Tech. Rep. D3.1,INFSO-ICT-317941 iJOIN, Nov. 2013.

[23] iJOIN, “D4.1: Report on SotA and requirements for network-layeralgorithms and network operation and management,” Tech. Rep. D4.1,INFSO-ICT-317941 iJOIN, Nov. 2013.

[24] Small Cell Forum, “Backhaul technologies for small cells: Use cases,requirements and solution,” Tech. Rep. 049.01.01, Feb. 2013.

[25] R. Chávez-Santiago, M. Szydełko, A. Kliks, F. Foukalas, Y. Haddad,K. E. Nolan, M. Y. Kelly, M. T. Masonta, and I. Balasingham,“5G: The convergence of wireless communications,”Wireless PersonalCommunications, pp. 1–26, Mar. 2015.

[26] 3GPP TSG-RAN, “5G: From Myth to Reality.” [Online], Available:http://www.3gpp.org/specifications-groups/ran-plenary.

[27] Working Party 5D (WP 5D) - IMT Systems, “ITU towards âAIJIMT for2020 and beyond”.” [Online], Available: http://www.itu.int/en/ITU-R/.

[28] J. Skold, “Research trends and IMT be-yond 2020.” https://www.itu.int/en/ITU-D/Regional-Presence/AsiaPacific/Documents/, Feb. 2014.

[29] ITU- International Telecommunication Union, “Recommendation ITU-R M.2083-0: IMT Vision - Framework and overall objectives ofthefuture development of IMT for 2020 and beyond,” Tech. Rep. ITU-RM.2083-0, November 2015.

[30] J. F. Monserrat, G. Mange, V. Braun, H. Tullberg, G. Zimmermann,and O. Bulakci, “METIS research advances towards the 5G mobile andwireless system definition,”EURASIP journal on Wireless Communi-cations and Networking 2015, vol. 2015, Mar. 2015.

[31] NGMN Alliance, “5G white paper.”http://www.3gpp.org/technologies/presentations-white-papers, Feb.2015.

[32] D. Flore, “LTE evolution and road to 5G.” https://eu-ems.com/, Oct2015.

[33] S. Suyama, J. Shen, Y. Oda, H. Suzuki, and K. Fukawa, “11 GHzBand 8ÃU16 MIMO-OFDM outdoor transmission experiment for 10Gbps super high bit rate mobile communications,” pp. 1008–1012, Sept2013.

[34] S. Suyama, J. Shen, H. Suzuki, K. Fukawa, and Y. Okumura,“Eval-uation of 30 Gbps super high bit rate mobile communications usingchannel data in 11 GHz band 24x24 MIMO experiment,” pp. 5203–5208, Jun. 2014.

[35] D. Samardzija and H. Huang, “Determining backhaul bandwidth re-quirements for network MIMO,” in17th European Signal ProcessingConference, pp. 1494–1498, Aug 2009.

[36] 3GPP, “Technical Specification Group Radio Access Network andScenarios and requirements for small cell enhancements forE-UTRAand E-UTRAN,” Tech. Rep. 36.932 (release 12), 3rd GenerationPartnership Project, Mar. 2013.

[37] A. Aijaz, H. Aghvami, and M. Amani, “A survey on mobile dataoffloading: Technical and business perspectives,”IEEE Wireless Com-munications, vol. 20, pp. 1536–1284, Apr. 2013.

[38] Motorola, “R1-103181: On range extension in open-access heteroge-neous networks,” in3GPP TSG RAN WG1 Meeting, vol. 61, (Montreal,Canada), May 2010.

[39] C. G. Gerlach, I. Karla, A. Weber, L. Ewe, H. Bakker, E. Kuehn,and A. Rao, “ICIC in DL and UL with network distributed and self-organized resource assignment algorithms in LTE,”Bell Labs Technicaljournal, vol. 15, pp. 43 – 62, Dec. 2010.

[40] NEC Group, “R1-113871: Study on pico coverage with CRE offset,”in 3GPP TSG RAN WG1 Meeting, vol. 67, (San Fransisco, U.S.A.),Nov. 2011.

[41] NTT DoCoMo, “R1-103264: Performance of eICIC with control chan-nel coverage limitation,” in3GPP TSG RAN WG1 Meeting, vol. 61,(Montreal, Canada), May 2010.

[42] Motorola Mobility, “R1-124951: Maintenance of up-to-date SIB1 inCRE zone,” in3GPP TSG RAN WG1 Meeting, vol. 79bis, (Bratislava,Slovakia), Oct. 2012.

[43] P. Marsch and G. Fettweis, “Uplink CoMP under a constrained back-haul and imperfect channel knowledge,”IEEE Transactions on WirelessCommunications, vol. 10, pp. 1730– 1742, Jun. 2011.

[44] Y. Beyene, R. Jantti, and K. Ruttik, “Cloud-RAN architecture for indoorDAS,” Access, IEEE, vol. 2, pp. 1205–1212, 2014.

[45] D. Wubben, P. Rost, J. Bartelt, M. Lalam, V. Savin, M. Gorgoglione,A. Dekorsy, and G. Fettweis, “Benefits and impact of cloud computingon 5G signal processing,”IEEE Signal Processing Magazine, vol. 31,pp. 35 – 44, Nov. 2014.

[46] U. Dotsch, M. Doll, H.-P. Mayer, F. Schaich, J. Segel, , and P. Sehier,“Quantitative analysis of split base station processing and determinationof advantageous architectures for LTE,”Bell Labs Technical journal,vol. 18, pp. 105 – 128, Jun. 2014.

[47] D. Astely, E. Dahlman, G. Fodor, S. Parkvall, and J. Sachs, “LTErelease 12 and beyond,”IEEE Communications Magazine, vol. 51,pp. 154 – 160, Jul. 2013.

21

[48] A. Mohamed, O. Onireti, M. Imran, A. Imran, and R. Tafazolli,“Control-data separation architecture for cellular radioaccess networks:A survey and outlook,” IEEE Communications Surveys Tutorials,vol. 18, pp. 446–465, Firstquarter 2016.

[49] H. Elshaer, F. Boccardi, M. Dohler, and R. Irmer, “Load &backhaulaware decoupled downlink/uplink access in 5G systems,” inIEEEInternational Conference on Communications (ICC), pp. 5380–5385,Jun 2015.

[50] R. Weiler, M. Peter, W. Keusgen, E. Calvanese-Strinati,A. de Domenico, I. Filippini, A. Capone, I. Siaud, A.-M. Ulmer-Moll,A. Maltsev, T. Haustein, and K. Sakaguchi, “Enabling 5G backhaul andaccess with millimeter-waves,”European Conference onÂaNetworksand Communications (EuCNC), Jun. 2014.

[51] J. Allen and F. Chevalier, “Report for Vodafone - Mobilebackhaulmarket: Phase 2 report,” 2014.

[52] J. Allen and F. Chevalier, “Report for Vodafone - Mobilebackhaulmarket: Phase 1 report,” 2014.

[53] F. Barros, K. Ahl, and V. Chaillou, “Creating a connectedcontinent: Press conference Warsaw.” [Online], Available:http://www.ftthcouncil.eu/, 2015.

[54] M.-C. Jeong, J.-S. Lee, S.-Y. Kim, S.-W. Namgung, J.-H.Lee, M.-Y.Cho, S.-W. Huh, Y.-S. Ahn, J.-W. Cho, and J.-S. Lee, “8 x 10-Gb/sterrestrial optical free-space transmission over 3.4 km using an opticalrepeater,”IEEE Photonics Technology Letters, vol. 15, pp. 171–173,Jan 2003.

[55] J. Segel and M. Weldon, “Light radio portfolio: Technical overview,”technology white paper, 2011.

[56] S. Lins, P. Figueiredo, and A. Klautau, “Requirements and evaluationof copper-based mobile backhaul for small cells LTE networks,”SBMO/IEEE MTT-S International Microwave OptoelectronicsConfer-ence (IMOC), pp. 1–5, Aug 2013.

[57] D. Bladsjo, M. Hogan, and S. Ruffini, “Synchronization aspects in LTEsmall cells,”IEEE Communications Magazine, vol. 51, pp. 70–77, Sept.2013.

[58] R. Parker, A. Slinger, M. Taylor, and M. Yardley, “Finalreport forOfcom: Future capability of cable networks for superfast broadband.”http://stakeholders.ofcom.org.uk/binaries/research/technology-research/2014/cable-sfbb.pdf, Apr. 2014.

[59] P. Rost and A. Prasad, “Opportunistic Hybrid ARQ-Enabler ofcentralized-RAN over nonideal backhaul,”IEEE Wireless Communi-cations Letters, vol. 3, pp. 481 – 484, Oct. 2014.

[60] J. Bartelt, P. Rost, D. Wubben, J. Lessmann, B. Melis, and G. Fettweis,“Fronthaul and backhaul requirements of flexibly centralized radioaccess networks,”Wireless Communications, IEEE, vol. 22, pp. 105–111, Oct. 2015.

[61] M. Jaber, D. Owens, M. Imran, R. Tafazolli, and A. Tukmanov, “Ajoint backhaul and RAN perspective on the benefits of centralisedRAN functions,” inIEEE International Conference on Communications(ICC), Workshops, (Kuala Lumpur, Malaysia), May 2016.

[62] C. Ranaweera, M. Resende, K. Reichmann, P. Iannone, P. Henry, B.-J.Kim, P. Magill, K. Oikonomou, R. Sinha, and S. Woodward, “Designand optimization of fiber optic small-cell backhaul based onan existingfiber-to-the-node residential access network,”IEEE CommunicationsMagazine, vol. 51, pp. 62–69, Sept. 2013.

[63] C. Ranaweera, P. Iannone, K. Oikonomou, K. Reichmann, andR. Sinha, “Design of cost-optimal passive optical networksfor smallcell backhaul using installed fibers,”IEEE/OSA Journal of OpticalCommunications and Networking, vol. 5, pp. A230 – A239, Oct. 2013.

[64] S. Kuwano, J. Terada, and N. Yoshimoto, “Operator perspective onnext-generation optical access for future radio access,”IEEE Interna-tional Conference on Communications Workshops (ICC), Jun. 2014.

[65] D. Iida, S. Kuwano, J. Kani, and J. Terada, “Dynamic TWDM-PON formobile radio access networks,”OPTICS EXPRESS, vol. 21, pp. 26209–26218, Nov. 2013.

[66] ITU-T, “40-Gigabit-capable passive optical networks(NG-PON2):General requirements.” http://www.itu.int/rec/T-REC-G.989.1-201303-I/en, Mar. 2013.

[67] N. Cvijetic, A. Tanaka, M. Cvijetic, Y. K. Huang, E. Ip, Y. Shao, andT. Wang, “Novel optical access and digital processing architectures forfuture mobile backhaul,”Journal of Lightwave Technology, vol. 31,pp. 621– 627, Feb. 2013.

[68] N. Cvijetic, A. Tanaka, M. Cvijetic, Y. K. Huang, E. Ip, Y. Shao,and T. Wang, “Convergent radio and fibre architectures for high-speed access,”15th International Conference on Transparent OpticalNetworks (ICTON), Jun. 2013.

[69] A. Goldsmith, “Do we have a shortage of bandwidth or imagination?.”http://technical.ly/brooklyn/2014/04/30/andrea-goldsmith-bandwidth/,Apr. 2014.

[70] Z. Roth, M. Goldhamer, N. Chayat, A. Burr, M. Dohler, N. Bartzoudis,C. Walker, Y. Leibe, C. Oestges, M. Brzozowy, and I. Bucaille, “Visionand architecture supporting wireless GBit/sec/km2 capacity densitydeployments,” inFuture Network and Mobile Summit, pp. 1–7, Jun2010.

[71] G. MacCartney, J. Zhang, S. Nie, and T. Rappaport, “Pathloss modelsfor 5G millimeter wave propagation channels in urban microcells,”IEEE Global Communications Conference (GLOBECOM), 2013.

[72] T. Rappaport, J. Murdock, and F. Gutierrez, “State of the art in 60-GHz integrated circuits and systems for wireless communications,”Proceedings of IEEE, vol. 99, pp. 1390 – 1436, Aug. 2011.

[73] T. S. Rappaport, “Millimeter wave wireless communications: Therenaissance of computing and communications.” http://icc2014.ieee-icc.org/speakers-28-4101586138.pdf, Jun. 2014.

[74] S. Nie, G. MacCartney, S. Sun, and T. Rappaport, “28 GHz and 73GHz signal outage study for millimeter wave cellular and backhaulcommunications,”IEEE International Conference on Communications(ICC), Jun. 2014.

[75] T. S. Rappaport, R. W. Heath, R. C. Daniels, and J. N. Murdock,Millimeter wave wireless communications. Upper Saddle River, NJ:Prentice Hall, 1 ed., Sept. 2014.

[76] R. Baldemair, T. Irnich, K. Balachandran, E. Dahlman, G. Mildh, Y. Se-len, S. Parkvall, M. Meyer, and A. Osseiran, “Ultra-dense networksin millimeter-wave frequencies,”IEEE Communications Magazine,vol. 53, pp. 202–208, Jan. 2015.

[77] J. Zhang, W. Ni, J. Matthews, C.-K. Sung, X. Huang, H. Suzuki, andI. Collings, “Low latency integrated point-to-multipointand E-bandpoint-to-point backhaul for mobile small cells,”IEEE InternationalConference on Communications Workshops (ICC), Jun. 2014.

[78] R. Taori and A. Sridharan, “In-band, point to multi-point, mm-Wavebackhaul for 5G networks,”IEEE International Conference on Com-munications Workshops (ICC), 2014, Jun. 2014.

[79] T. K. Vu, M. Bennis, S. Samarakoon, M. Debbah, and M. Latva-aho, “Joint In-Band backhauling and interference mitigation in 5Gheterogeneous networks,” inEuropean Wireless, (Oulu, Finland), May2016.

[80] K. Zheng, L. Zhao, J. Mei, M. Dohler, W. Xiang, and Y. Peng,“10 Gb/s HetSNets with millimeter-wave communications: Accessand networking âAS challenges and protocols,”IEEE CommunicationsMagazine, vol. 53, pp. 222 – 231, Jan. 2014.

[81] T. E. Bogale and L. B. Le, “Massive MIMO and mmWave for 5Gwireless HetNet: Potential benefits and challenges,”IEEE VehicularTechnology Magazine, vol. 11, pp. 64–75, March 2016.

[82] Infonetics Research, “Carrier SDN and NFV hardwareand software: Aannual market size and forecasts.”http://www.infonetics.com/pr/2014/Carrier-SDN-NFV-Market-Highlights.asp, Nov. 2014.

[83] “Open networking foundation.” https://www.opennetworking.org/. Ac-cessed: 2015-03-10.

[84] V. Jungnickel, K. Habel, M. Parker, S. Walker, C. Bock, J. Ferrer Riera,V. Marques, and D. Levi, “Software-defined open architecture for front-and backhaul in 5G mobile networks,”Transparent Optical Networks(ICTON), 2014 16th International Conference onÂa, Jul. 2014.

[85] D. Bojic, E. Sasaki, N. Cvijetic, T. Wang, J. Kuno, J. Lessmann,S. Schmid, H. Ishii, and S. Nakamura, “Advanced wireless andopticaltechnologies for small-cell mobile backhaul with dynamic software-defined management,”IEEE Communications Magazine, vol. 51,pp. 86– 93, Sept. 2013.

[86] N. Cvijetic, “Optical network evolution for 5G mobile applications andSDN-based control,”16th International Telecommunications NetworkStrategy and Planning Symposium (Networks), Sept. 2014.

[87] N. Cvijetic, A. Tanaka, P. Ji, K. Sethuraman, S. Murakami, andT. Wang, “SDN and OpenFlow for dynamic Flex-Grid optical accessand aggregation networks,”Journal of Lightwave Technology,, vol. 32,pp. 864 – 870, Feb. 2014.

[88] N. Cvijetic, A. Tanaka, P. Ji, K. Kanonakis, and T. Wang,“SDN-controlled topology-reconfigurable optical mobile fronthaul architec-ture for bidirectional CoMP and low latency inter-cell D2D in the 5Gmobile era,”OSA Opt. Express, vol. 22, Aug. 2014.

[89] J. Costa-Requena, R. Kantola, J. Llorente, V. Ferrer, J. Manner, A. Y.Ding, Y. Liu, and S. Tarkoma, “Software defined 5G mobile backhaul,”1st International Conference on 5G for Ubiquitous Connectivity (5GU),Nov. 2014.

22

[90] J. Costa-Requena, J. Llorente Santos, and V. Guasch, “Mobile backhaultransport streamlined through SDN,” in17th International Conferenceon Transparent Optical Networks (ICTON, pp. 1–4, July 2015.

[91] J. Shin, R. Hwang, Y. Shim, S. Cho, J. Park, and A. Choi, “TransportSDN : Trends, standardization and architecture,”International Con-ference on Information and Communication Technology Convergence(ICTC), 2014, Oct. 2014.

[92] M. Draxler and H. Karl, “Dynamic backhaul network configuration inSDN-based Cloud RANs,”arXiv, vol. 1, Mar. 2015.

[93] M. Liyanage, I. Ahmed, M. Ylianttila, J. Santos, R. Kantola,O. Lopez Perez, M. Itzazelaia, E. Montes De Oca, A. Valtierra, andC. Jimenez, “Security for future software defined mobile networks,” in9th International Conference on Next Generation Mobile Applications,Services and Technologies, pp. 256–264, Sept 2015.

[94] S. Tombaz, P. Monti, F. Farias, M. Fiorani, L. Wosinska,and J. Zander,“Is backhaul becoming a bottleneck for green wireless access net-works?,” IEEE International Conference on Communications (ICC),Jun. 2014.

[95] X. Ge, H. Cheng, M. Guizani, and T. Han, “5G wireless backhaulnetworks: challenges and research advances,”IEEE Network, vol. 28,pp. 6 – 11, Nov.-Dec. 2014.

[96] J. Lun and D. Grace, “Cognitive green backhaul deployments forfuture 5G networks,”1st International Workshop on Cognitive CellularSystems (CCS), Sept. 2014.

[97] Saidul Huq, K.M. and Mumtaz, S. and Bachmatiuk, J. and Rodriguez,J. and Wang, X. and Aguiar, R., “Green HetNet CoMP: Energyefficiency analysis and optimization,”IEEE Transactions on VehicularTechnology, vol. 64, no. 10, pp. 4670 – 4683, 2015.

[98] K. Saidul Huq, S. Mumtaz, J. Rodriguez, and C. Verikoukis, “In-vestigation on energy efficiency in HetNet CoMP architecture,” IEEEInternational Conference on Communications (ICC), Jun. 2014.

[99] A. Mesodiakaki, F. Adelantado, L. Alonso, and C. Verikoukis, “Energyimpact of outdoor small cell backhaul in green heterogeneous net-works,” IEEE 19th International Workshop on Computer Aided Mod-eling and Design of Communication Links and Networks (CAMAD),2014.

[100] E. Bastug, M. Bennis, and M. Debbah, “Think before reacting: proac-tive caching in 5G small cell networks,”http://www.laneas.com/, Sept.2014.

[101] E. Bastug, M. Bennis, and M. Debbah, “Living on the edge: The roleof proactive caching in 5G wireless networks,”IEEE CommunicationsMagazine, vol. 52, pp. 82–89, Aug. 2014.

[102] A. Abboud, E. Bastug, K. Hamidouche, and M. Debbah, “Distributedcaching in 5G networks: An alternating direction method of multipliersapproach,”http://laneas.com/, 2015.

[103] F. Pantisano, M. Bennis, W. Saad, and M. Debbah, “Cache-awareuser association in backhaul-constrained small cell networks,” 12thInternational Symposium on Modeling and Optimization in Mobile,Ad Hoc, and Wireless Networks (WiOpt), May 2014.

[104] A. Liu and V. Lau, “Cache-induced opportunistic MIMO cooperation:A new paradigm for future wireless content access networks,” IEEEInternational Symposium on Information Theory (ISIT), Jun. 2014.

[105] F. Pantisano, M. Bennis, W. Saad, and M. Debbah, “Matchto cache:Joint user association and backhaul allocation in cache-aware smallcell networks,” inIEEE International Conference on Communications(ICC), pp. 3082–3087, Jun 2015.

[106] E. Bastug, M. Bennis, E. Zeydan, M. Kader, I. Karatepe,A. Er, andM. Debbah, “Big data meets telcos: A proactive caching perspective,”Communications and Networks, Journal of, vol. 17, pp. 549–557, Dec2015.

[107] P. Blasco and M. Dohler, “The ETSI BRAN 1 Gbps/km2 cost-efficientcognitive communications architecture,” in3rd International Workshopon Cognitive Information Processing (CIP), pp. 1–6, May 2012.

[108] Alvarion, CTTC, Polska Telefonia Cyfrowa, Siklu, Thales, “BroadbandRadio Access Networks (BRAN); Very high capacity density BWAnetworks; Protocols,” tech. rep., ETSI TC BRAN, TR 101 589, 2012.

[109] Alvarion, CTTC, Polska Telefonia Cyfrowa, Siklu, Thales, “Very highcapacity density BWA networks; System architecture, economic modeland technical requirements,” tech. rep., ETSI TC BRAN, TR 101 534,2012.

[110] E. Pateromichelakis, A. Maeder, A. De Domenico, R. Fritzsche,P. de Kerret, and J. Bartelt, “Joint RAN/backhaul optimization incentralized 5G RAN,” inEuropean Conference on Networks andCommunications (EuCNC), pp. 386–390, Jun. 2015.

[111] A. Mesodiakaki, F. Adelantado, L. Alonso, and C. Verikoukis, “Energy-efficient context-aware user association for outdoor smallcell heteroge-

neous networks,”IEEE International Conference on Communications(ICC), Jun. 2014.

[112] M. Jaber, M. Imran, R. Tafazolli, and A. Tukmanov, “An adaptivebackhaul-aware cell range extension approach,” inIEEE InternationalConference on Communications (ICC), Workshops, (London, UK), Jun.2015.

[113] G. Zhang, T. Quek, A. Huang, M. Kountouris, and H. Shan,“Backhaul-aware base station association in two-tier heterogeneous cellular net-works,” in IEEE 16th International Workshop on Signal ProcessingAdvances in Wireless Communications (SPAWC), pp. 390–394, Jun2015.

[114] H. Beyranvand, W. Lim, M. Maier, C. Verikoukis, and J. Salehi,“Backhaul-aware user association in FiWi enhanced LTE-A hetero-geneous networks,”IEEE Transactions on Wireless Communications,vol. 14, pp. 2992–3003, Jun 2015.

[115] N. Wang, E. Hossain, and V. Bhargava, “Backhauling 5G small cells:A radio resource management perspective,”IEEE Wireless Communi-cations, vol. 22, pp. 41–49, Oct 2015.

[116] M. Shariat, E. Pateromichelakis, A. Quddus, and R. Tafazolli, “JointTDD backhaul and access optimization in dense small cell networks,”IEEE Transactions on Vehicular Technology, vol. 64, no. 11, pp. 5288– 5299, 2015.

[117] D. Wang, E. Katranaras, A. Quddus, F.-C. Kuo, P. Rost, N. Sapountzis,C. Bernardos, L. Cominardi, and I. Berberana, “SDN-based jointbackhaul and access design for efficient network layer operations,”in European Conference on Networks and Communications (EuCNC),pp. 214–218, Jun 2015.

[118] M. Rahman, C. Machuca, K. Grobe, and W. Kellerer, “Advantages ofjoint access network planning in dense populated areas,”19th EuropeanConference on Networks and Optical Communications - (NOC), Jun.2014.

[119] R. Ford, C. Kim, and S. Rangan, “Opportunistic third-party backhaulfor cellular wireless networks,”Asilomar Conference on Signals, Sys-tems and Computers, Nov. 2013.

[120] M. Mahloo, P. Monti, J. Chen, and L. Wosinska, “Cost modeling ofbackhaul for mobile networks,” inIEEE International Conference onCommunications Workshops (ICC), pp. 397–402, Jun 2014.

[121] V. Suryaprakash and G. Fettweis, “Modeling backhaul deploymentcosts in heterogeneous radio access networks using spatialpointprocesses,” in12th International Symposium on Modeling and Opti-mization in Mobile, Ad Hoc, and Wireless Networks (WiOpt), pp. 725–732, May 2014.

[122] D. Chen, T. Quek, and M. Kountouris, “Wireless backhaul in small cellnetworks: Modelling and analysis,” inIEEE 79th Vehicular TechnologyConference (VTC Spring), pp. 1–6, May 2014.

[123] G. Zhang, T. Quek, M. Kountouris, A. Huang, and H. Shan,“Funda-mentals of heterogeneous backhaul design - analysis and optimization,”IEEE Transactions on Communications, vol. PP, no. 99, pp. 1–1, 2016.

[124] 5G PPP, “5G Vision.” http://www.3gpp.org/technologies/presentations-white-papers, Mar. 2015.

[125] G. Zhang, T. Quek, M. Kountouris, A. Huang, and H. Shan,“Funda-mentals of heterogeneous backhaul design - analysis and optimization,”IEEE Transactions on Communications, vol. PP, no. 99, pp. 1–1, 2016.

[126] X. Ge, S. Tu, T. Han, Q. Li, and G. Mao, “Energy efficiencyofsmall cell backhaul networks based on Gauss-Markov mobile models,”Networks, IET, vol. 4, no. 2, pp. 158–167, 2015.

[127] W. Van Heddeghem, M. Parker, S. Lambert, W. Vereecken,B. Lannoo,D. Colle, M. Pickavet, and P. Demeester, “Using an analytical powermodel to survey power saving approaches in backbone networks,” in17th European Conference on Networks and Optical Communications(NOC), pp. 1–6, June 2012.

[128] C. K. Chan and H. Pant, “Reliability and security modeling in upgrad-ing wireless backbone networks,”Bell Labs Technical Journal, vol. 8,no. 4, pp. 39–53, 2004.

23

TABLE IX

ABBREVIATIONS

10G-EPON 10 Gigabit EPONARN Active Remote NodeBBU Baseband UnitBHaaS Backhaul as a ServiceCapEX Capital EXpenditureCB Coordinated BeamformingCO Central OfficeCoMP Coordinated Multi-point ProcessingCPRI Common Public Radio InterfaceC-RAN Cloud/centralised RANCS Coordinated schedulingCSI Channel State InformationD2D Device to DeviceDAS Distributed Antenna SystemDBA Dynamic Bandwidth AllocationDL DownlinkDOCSIS Data Over Cable Service Interface SpecificationeICIC enhanced ICICEPON Ethernet PONFDD Frequency Division DuplexFEC Forward Error CorrectionfeICIC further enhanced ICICFFT Fast Fourier TransformFSO Free Space Optical communicationFTTx Fibre To The Node/Building

G.FASTFast Access to Subscriber Terminal; the letter Gstands for ITU-T G series of recommendations

GNSS Global Navigation Satellite SystemGPON Gigabit capable PONHARQ Hybrid Automatic Repeat RequestHetNet(s) Heterogeneous Network(s)ICIC Inter-Cell Interference CoordinationIFFT Inverse Fast Fourier TransformISP Internet Service ProviderIT Information TechnologyJR Joint ReceptionJT Joint TransmissionLOS Line Of SightLTE Long Term EvolutionMAC Medium Access ControlMIMO Multiple Input Multiple OutputmmWave Millimetre waveMUD Multi-User DetectionNFV Network Function VisualisationNGPON Next Generation PONNLOS Non LOSOFDM orthogonal frequency division multiplexingONT Optical Network TerminalONU Optical Network UnitOpEX Operational EXpenditurePON Passive Optical NetworkPtmP Point to multi-PointPtP Point to PointQAM Quadrature amplitude modulationQoE Quality of ExperienceQoS Quality of ServiceRAN Radio Access NetworkRANaaS RAN as a ServiceRAT Radio Access TechnologyRF Radio FrequencyRRH Remote Radio (RF) HeadRRU Remote Radio UnitSDN Software Defined NetworkSON Self Organising/Optimising NetworkTCO Total Cost of OwnershipTDD Time Division DuplexTDM Time Division MultiplexingTVWS TeleVision White SpaceUDN Ultra Dense NetworkUE User EquipmentUL UplinkVDSL Very high bit rate Digital Subscriber LineVeNB Virtual enhanced Node BWDM Wavelength Division MultiplexingXGPON 10 Gigabit capable POND