Xu et al. EURASIP Journal onWireless Communications andNetworking (2016) 2016:264 DOI 10.1186/s13638-016-0762-6

RESEARCH Open Access

Towards next generation software-defined radio access network–architecture,deployment, and use caseFangmin Xu*† , Haipeng Yao†, Chenglin Zhao and Chao Qiu

Abstract

As one of the key enabling technologies of the fifth generation wireless network (5G), software defined network(SDN) offers a logically centralized control model, flexible programmability, and a flow-based paradigm that is ideallysuited for highly scalable wireless networks, from access to core part. Following this paradigm, a novel software-definedradio access network (SDRAN) architecture and the function modules have been proposed in this paper. In particular,the motivation, challenge, and deployment roadmap of SDRAN framework are discussed. The relationships betweenalternative solutions (Cloud RAN, network function virtualization) and complementary technologies (cognitive radio,self-organizing network, big data analysis) are analyzed in detail. Taking interference management of heterogeneousmobile network as the example use case, scheme design and preliminary system evaluations are given to show thebenefit of SDRAN architecture.

Keywords: Software-defined network (SDN), radio access network (RAN), network function virtualization (NFV),the fifth generation wireless network (5G)

1 IntroductionWith the maturing of the fourth generation wireless (4G)technologies and the ongoing widely deployment of 4Gnetworks, research activities on 5G communication tech-nologies have emerged in both the academic and indus-trial communities. To support ultra-dense network trafficand massive connections with low latency, rethinking andrestructuring of the current wireless network architectureis required [1].As a networking paradigm that separates the control

and data forwarding planes, software-defined networking(SDN) is currently being considered as an alternative totraditional distributed approaches based on highly spe-cialized hardware. With SDN, network operators can con-figure the behavior of both the traffic and the network ina centralized way. Until now, most of the use cases forSDN paradigm have been limited to wired environmentsor core part of the wireless network [2]. By separating the

*Correspondence: xufm@bupt.edu.cn†Equal contributorsSchool of Information and Communication Engineering, Beijing University ofPosts and Telecommunications, Xitucheng Road, No 10, Beijing, People’sRepublic of China

control plane and data plane, building software via pro-gramming interfaces and virtualization technology, it ispossible to achieve lower cost and higher efficiency usingSDN and network function virtualization (NFV) [3].There have also been some efforts looking at the use

of SDN in wireless networks. OpenRoad [4] project atStanford University introduced OpenFlow to wireless net-works to enhance the control plane. SoftRAN [5] fromAlcatel-Lucent considered a logically centralized controlplane and scalable distributed enforcement of quality ofservice (QoS) policies in the data plane. The control planein SoftRAN programs and manages the abstracted radioresources in multiple dimensions. A SDN-like mobile net-work architecture was proposed in [6] with the examplecase on mobility management. Both RAN side and CNside are enhanced with programmability. Mobile networkSDN controller manages the shared radio access and corenetwork among multiple (virtual) operators. MobileFlow[7] from Huawei was another software-defined futuremobile network architecture that enables operators tocapitalize on a flow-based forwarding model. The virtualMobileFlow controller programs the forwarding of theuser plane in a software-defined fashion. SoftMobile [8]

© The Author(s). 2016 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to theCreative Commons license, and indicate if changes were made.

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 2 of 12

provided a SDN-based control framework to coordinateradio access functions(including mobility, load balancing,spectrum allocation, interference management, etc.) inheterogeneous networks.In summary, above literatures discussed the architec-

tures by the introduction of SDN to cellular network.However, few works have addressed the practical feasibil-ity aspects, such as challenges and roadmap in the evo-lution, whether the control functions could be virtualizedand programmed in centralized controller, the potentialgain of SDN-based RAN and the practical design.To address above issues, the paper represents deep

rethinking on the architecture of SDN-based wirelessaccess network (SDRAN), including the driving force,challenge, and the evolution roadmaps from mobile oper-ators’ view. Based on the virtualization and orchestrationof network components, virtual centralized or distributedcontrollers in SDRAN separates the complicated controlfunction from data plane. In addition, the relationshipsbetween SDN-based RAN and C-RAN, NFV are clarified.And then a typical example case (interference manage-ment in heterogeneous network) is provided to show thepossible advantage of the design.The rest of this article is organized as follows. We

first present the architecture SDRAN and its key ben-efits and challenges in Section 2. Section 3 focuses onthe joint consideration of SDRAN and C-RAN, NFV, andSON. Section 4 presents a special use case developed onSDRAN, and simulation are illustrated to show the poten-tial benefits. A brief conclusion with some discussion on

the current implementations and further developments ofSDRAN is presented in Section 5.

2 SDN-based radio access network architectureand evolution roadmap

In this section, we will discuss the architecture of SDN-like radio access network and the motivation to deployit from operators’ view. Moreover, we analyze the imple-mentation roadmaps of SDN-based radio access network.

2.1 Why use SDN here?Unlike the evolution from 2G to 3G or from 3G to 4G, themotivation to enhance the current cellular network archi-tecture from 4G to 5G is driven by the mobile networkoperator in addition to the requirements from upcominguse cases. For instance, machine-to-machine communi-cation scenario requires exchange of huge amounts ofsmall data among massive devices, and autonomous vehi-cle control requires real-time and reliable data exchanges.All of these diverse quality of service(QoS) requirementsdetermine the transformation of next generation mobilenetwork architecture in addition to the evolution of airinterface technologies.As illustrated in Fig. 1, in LTE networks, the user equip-

ment (UE) connects to an eNodeB where data packetsare transferred through a serving gateway (S-GW). Thetraffic tunneling origins from S-GW to the packet datanetwork gateway (P-GW). The P-GW enforces QoS poli-cies and monitors traffic to perform charging. The P-GWalso connects to the Internet and other cellular networks.

Uu

eNodeB eNodeBX2

S-GW MME

P-GW HSS

PCRF

IP Backbone

E-UTRAN

CN

Control Signaling

Data Traffic

S1u S1MME

S11

S5

S7

SGi

S6a

S1u

Data Path

Control PathMobility Management Function

Idle Mode Mobility

Connected Mode Mobility

Admission Control

Inter-RATInterference Management

Resource Schedule

Load Balance

Multi-RAT Scheduling

GERAN

BTS BSC

UTRAN

NodeB RNC

SGSN

GGSN

CN

Gn

Iu

HLR

Gr

Gb

S4

Abis

Um

UE

Uu

Iub

Gx

Gi

Power Control

RRM Function

Inter-RAT Function

Fig. 1Mobile network architecture defined in 3GPP

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 3 of 12

Besides above data plane functionalities, the eNodeB, S-GW, and P-GW also participate in several control planefunctions. Cooperating with the Mobility ManagementEntity (MME), they handle functions such as sessionsetup, reconfiguration, as well as mobility management.The Policy and Charging Rules Function (PCRF) managesflow-based charging and QoS policy based on the user’ssubscription profile. In addition, the Home SubscriberServer (HSS) contains subscription information for eachuser.Today’s cellular network architectures have several

major limitations. The contradiction between centralizeddata-plane functions and decentralize control plane is themain reason. For instance, centralized data-plane func-tions force all traffic through the P-GW, while the QoSfunctionality at the P-GW introduces scalability chal-lenges. In respect of the mobile operator, the existingcellular network architecture has the following problems.

• High Capital Expenditure (CAPEX)/OperatingExpense (OPEX)Dedicate hardware are designed to support differentRadio Access Technologies (RATs) or services. Inorder to support advanced features or new functionsof network devices, operators need update andreplace the network hardware. Therefore, itintroduces extra CAPEX and OPEX in networkdeployment and update.

• Low resource utilization efficiencyThe resource (spectrum, computing, power andstorage resource) for each RAT is unable to beintegrated and virtualized. Thus, the problem ofinefficient resource allocation and sharing is severefor the next generation wireless network. Also theperformance of distributed resource management isfar away from optimal.

• Complex interworking and interoperationThe network equipment has vendor-specificconfiguration interfaces, and understanding offunctionalities and implementation of interfacesbetween network devices is inconsistent. Therefore,interoperation among multiple RATs, evencoordination among multiple operators iscomplicated and inefficient based on the currentstructure.

Correspondingly,the SDN-like RAN architecture pro-vides following interesting features from the operator’sside.

• Reduce operation CAPEX/OPEXOn one hand, common network hardware could beconfigured and updated programmatically to act asany kinds of network devices. On the other hand, the

dummy and simple hardware of network device maylower the entry threshold of network device industry.

• Simpler interworking and interoperation by networkvirtualization, slice and isolationPhysical wireless network resources can be abstractedinto virtual wireless network resources holdingcertain corresponding functionalities, and shared bymultiple parties. Firstly, the centralized controllerfacilitates the coordination and cooperation ofvarious RATs. Secondly, the network resource andinfrastructure sharing will raise new security risk ininterworking and interoperation, while virtualnetwork isolation would alleviate the possiblesecurity risk.

• Optimized resource utilization efficiency by globalview and centralized controlA SDN-enabled centralized controller will have aglobal view of the resource allocation, interferencedistribution, and resource usage. In addition, a SDNcontroller running on a commodity server wouldhave much more computing resources than currentnetwork devices. As a result, a SDN controller canmake a more efficient management of radio resources.

Meanwhile, the successful story and commercial matu-rity of SDN technology in wired network made all thethings possible in technical aspect. The timing is ripefor the introduction of SDN to the future radio accessnetwork.

2.2 What is the architecture?Figure 2 shows the proposed SDN-based cellular net-work architecture. Without loss of generality, we useeNodeB to represent the NodeB+RNC in 3G, etc. Com-pared with traditional cellular network architecture inFig. 1, the main difference consisted of the following threepoints.

• Separate data and control pathAs shown in Fig. 2, the control path (green dash line)and data path (red full line) are separated in bothRAN side and core network (CN) side. In RAN side,the eNodeB could be logically split into two parts:eNodeB(U) and eNode(C), where the formerimplements the radio transmission withconfigurations interpreted by eNode(C) according todelivered policy. Besides, eNode(C) reports local view(i.e., network status, such as network load andinterference) to the central controller. Note thateNode(C) may also implement some controlfunctions that are not suitable for virtualization orcentralization. In the CN side, the P-GW and S-GWcould also be virtualized into P-GW(U) andS-GW(U) which are managed by the CN controller.

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 4 of 12

Fig. 2 SDRAN architecture and the modules, interfaces

The separation of data and control path is useful forthe deployment and update of new service, facilitatethe realization of centralized controller. However, thescalability (responsible for hundreds of data planeelements), reliability (what if a controller in failure),security (executing the network functions in thisvirtualized multi-tenant environment) andconsistency (ensure consistency between controllers)issues should be taken into account.

• Centralized regional controller in RANIn the proposed architecture, the controllers could belogically divided into RAN controller and CNcontroller depending on the location of controlentities. Only SDRAN controller is discussed in thearticle. As shown in Fig. 2, the SDRAN controllerobtains the local view from eNode(C), and constructsthe global resource and network topology view fromthe collected local views and subscriber profiles.

There are four interfaces defined for SDRANcontroller.Northbound interface to external API controlled byoperators allowing them to dynamically change theshare of resources and policy at any time or location.Northbound internal interface to CN controller.Information between SDRAN controller and CNcontroller is exchanged via the interface, enabling thecooperation and coordination of RAN and CNfunctions.Southbound interface to RAN entity (eNodeB). Thisinterface is used by the SDRAN controller to enforcedifferent policies according to requests from thevirtual operators and realize effective virtualization ofthe access network and resource, enable quickconfiguration of network parameters.East-West bound interface to other SDRANcontrollers. Considering the scalability and capability

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 5 of 12

of SDRAN controller, the east-west bound interfacesprovide policies exchanged among inter-domainadjacent SDRAN controllers.

• Flow-like policy controlThe flow paradigm of SDN is particularly well suitedto provide end-to-end communications acrossmultiple distinct technologies, such as 4G, and Wi-Fi.Flows can have granular policies for effective trafficisolation and QoS management. At the data plane, aflow is a programmable set of packets that share thesame properties. The per-flow policy control allowsthe policy decision/enforcement for flow accordingto the user profile, network topology and status. Thearchitecture could provide per-fow QoS and controlactions.

2.3 Which function could be virtualized/centralized?The programmable configuration of the RAN functionsallows the best dynamic use of available resource. How-ever, not all control functions are suitable for virtualizedor centralized management. Firstly, the criteria whichdecide the functions could be virtualized/centralized areinvestigated as following.

• Need of coordination (in brief, NC)All decisions that influence the decision-making atneighbouring cells should be made at the controller,since such decisions need to be coordinated acrosseNodeBs.

• Not delay sensitive (in short, DS)The decisions which depend on rapidly varyingnetwork parameters or status can only be optimizedat the eNodeB due to inherent delay between theSDRAN controller and eNodeB.

• Not bandwidth constrained (briefly, BC)Similar with delay constraint, the limited linkbandwidth between controller and eNodeB is anotherbottleneck, especially considering the high networkdensity and scalability.

• With Abstracted View (in brief, AV)The SDRAN controller obtains the global view fromlocal view (information including but not limited to,network status and load situation) generated byeNodeBs. However, due to the feedback latency andbandwidth constraint, the abstraction of local detailview is vital for controller. Abstracted view is theabstraction of collected statistics and events, such asnetwork status, surrounding environment, traffic loadand network overload. However, some informationrequired by some control functions is not suitable toabstract as local view, such as the transmitted data.

Among all functions in RAN side (not only limitedto the functions implemented by RAN device(eNodeB)but also including functions with the cooperation ofeNodeB), Table 1 shows the conclusion after analysis,where three situations are categorized: “Y,” “N” and “A.”Literally, “Y” and “N” presents the case that functioncould or could not be virtualized respectively. “A” standsfor the function which could be assisted by centralizedcontroller.For instance, the resource allocation function does not

have an impact on the decision-making at the neighbour-ing eNodeBs (Not NC) and will be influenced by thechannel quality measurement reported by UEs in shortperiod (DS); therefore, resource allocation task should bedone by the eNodeB rather than at the SDRAN controller.In Coordinated Multiple Point Transmission (CoMP),joint transmission(JT) relies on the data sharing amongdifferent transmission points. As mentioned before, thetransmission data is unsuitable to abstracted into localview.

2.4 How to realize the SDRAN architecture?Considering the challenges to realize the SDRAN archi-tecture based on current networks, the following issues inthe implementation are discussed from the view of mobileoperator.

Table 1 Summary of the feasibility of virtualized/centralized control functions

Color function Enforcement entity Suitable for virtualized? Positive reasons Negative reasons

Power allocation and control eNodeB N NC DS

Resource allocation eNodeB N Not NC,DS

CoMP eNodB N NC DS, BC, not AV

Mobility management (handoff) MME eNodeB A NC not BC DS

Access control eNodeB Y NC not DS, not BC, AV

Interface management eNodeB Y NC, not DS, not BC, AV

Inter-RAT coordination eNodeB via X2 Y NC not DS, not BC, AV

Loading balancing eNodeB via X2 Y NC not DS, not BC, AV

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 6 of 12

2.4.1 Evolutional or clean slate?To solve the challenge of consistency, two different mod-els can be adopted to implement the SDRAN architecture:“evolutionary” and “clean slate.” The evolutionary modelallows for gradually deployment based in existing net-works: legacy control plane entities from operators canconnect to the core network without modifying the exist-ing interfaces. SDRAN controller implements backward-compatible interfaces to support the interworking withlegacy network entities. There are two modes to couplethe legacy architecture and future evolutional architec-ture: loose couple (fully decouple the legacy control planeentities and evolved control plane entities) and tight cou-ple (reuse the legacy control plane entities with evolvedstructure as much as possible).In the clean slate model, the centralized control func-

tions are constructed directly independent of legacy net-work. Although the deployment cost of this approach isrelatively high, it enables the deployment of novel functionand service easier and faster, as it can be directly imple-mented on the new controller and does not have an impacton legacy interfaces and devices. The illustrations of thesemodels are given in Fig. 3a.

2.4.2 Centralized, distributed or hybrid?Considering the scalability and limited capability of con-troller, with regarding the relationship between SDRAN

controllers, there are three possible ways: Central-ized, distributed and hybrid structure, as illustrated inFig. 3b.The centralized approach simplifies managing flows

that are related to specific applications. But as networksscale and become more distributed, centralized controlrequires hardware with stronger computation and storagecapability; therefore, it is less flexible and costly for mobileoperator.In the distributed structure, a centralized controller

manages distributed data planes. From the implementa-tion view, there is a lot more complexity in deployinga distributed network. It may not be possible to get toall of the domains to create the same policy since thereare latency and synchronization issues. The good aspectsabout distributed structure is that they are evolutionaryand easy to scale, rather than replace treatment in thecentralized structure.The hybrid structure utilizes the benefits of the sim-

ple control of managing specific data flows as in thecentralized structure with the scalability and resiliencyof the distributed structure. The hybrid approach allowsfor more flexible policy definition and enforcement invarious scenarios (no matter in local scenario or large-scale deployment). However, the multiple-level structurewill introduce inevitable latency and higher operationcost.

(a)

(b)

Fig. 3 a Possible models to implement the SDRAN based on legacy network. b Illustration

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 7 of 12

3 Relationships with alternative solutions andtechnologies

C-RAN and NFV are different ways to realize the effectiveusage of resource as SDRAN. In addition, there are sometechnologies that could be employed to assist the SDRANarchitecture.

3.1 C-RANCloud-RAN(or centralized RAN) is a centralized, cloudcomputing-based cellular network architecture that sup-ports efficiently centralize computational resource. Byseparating the BBUs from the radio access devices andmigrating baseband units(BBUs) to the cloud forming abaseband resource pool for centralized processing, radiosignals are exchanged over fronthaul between remoteradio heads (RRHs) and the data center [9].However, full-scale coordination leads to high computational overheadand control signaling flooding in the BBU pool, espe-cially for a large-scale network. Therefore, C-RAN canbe regarded as a specific component of SDN. With theassistance of SDN, the above control signaling problem inC-RAN could be possibly alleviated.Literally, both C-RAN and SDRAN are featured by cen-

tralized resource management and sharing. However, thedifferent contents are centralized. In C-RAN, basebanddata are collected by fronthaul, while SDRAN controllerimplements centralized control functions by gathered sig-naling and information rather than the data.EmbeddingC-RAN into the SDRAN architecture is a promisingway to reduce the burden of controllers and integratesthe wired and wireless sides of SDN seamlessly. Tak-ing C-RAN into account along with the SDN approacheswill help in terms of reducing the overall signaling andsimplifying the network topology from the controllerperspective [14].

3.2 NFVSDN and NFV are always linked together. NFV refersto the implementation of network functions by softwarerunning on general purpose computing platforms. Com-pared with the conventional network which implementsnetwork functions on dedicated and specific hardware,the main idea behind NFV is to reduce life and innova-tion cycles within telecommunication networks throughsoftware update rather than hardware update or replace.European Telecommunications Standards Institute (ETSI)has initiated working group and published whitepaperon NFV to show the benefits and challenges for NFVtechnologies to be deployed by network operators [12].Naturally, SDN and NFV are complementary: The for-

mer decouples control from the data path network func-tions, and the later decouples placement of the networkfunctions from the underlying hardware. SDN owns theprogrammatic control and configuration of the devices

and network functions while NFV deals with the lifecycleof those functions. From an operator’s perspective, on theone hand, less update cost could be expected by the fea-ture of NFV due to the share of general computation andstorage resource. On the other hand, virtual operators andnovel services could be flexibly supported by the networkabstraction and virtualization as shown in Fig. 4.Physical wireless infrastructure and radio resources that

belong to one or more mobile network operators(MNOs)could be abstracted to a virtual resource pool via NFV.Afterward, the virtual resources are isolated into mul-tiple virtual resource slices, which then can be offeredto different virtual network operators(VNOs) or serviceproviders(SPs) by the orchestration of SDRAN controller.SDRAN controller receives the application-level com-mand from SPs, inner orchestrator delivers delicate pol-icy and allocates or release dedicated resource slice forthe SP.With the joint cooperation of SDRAN and NFV, VNOs

can lease virtual networks from MNOs, and MNOs cangain greater number of subscribers and revenues attractedfrom flexible services provided by VNOs or SPs. ForMNOs themselves, since the network can be isolated intoseveral slices, any upgrading and maintenance in one slicewill not affect other services in other slices. For SPs,renting virtual networks helps them flexibly control ofunderlay infrastructure without vast investment, so thatcustomized and more flexible services can be providedmore easily and the QoS guarantee can be enhanced aswell.

3.3 Self-organizing network (SON)SON [10] is an automation technology designed to makethe planning, configuration, management, optimizationand healing of RAN simpler and faster. SON has recentlyattracted interests from operators since SON providescapabilities such as significant reductions in CAPEX andOPEX during both network deployment and operationperiod, improvements in network quality and quality ofexperience for users and provides optimized performancefrom adapting the network to environments and otheroperational conditions.SON is not a replacement, but rather a supplement to

provide additional adaptability for SDRAN. The SON-empowered SDN controller is therefore designed to be anexecution environment for automatic functions, i.e. applythe configuration of output parameters to network phys-ical elements. On the other side, as a proprietary add-on,SON is typically only used effectively in the environmentwith devices provided by the same manufacturer. SDRANarchitecture provides a clean abstraction and networkcontrol functions via a set of APIs; therefore, it realizesSON operation via the southbound interface to supportdevices from different vendors.

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 8 of 12

Fig. 4 The joint operation of SDRAN and NFV

3.4 Cognitive radioFew efforts are paid on the monitoring and learn-ing ability of SDN. Compared with wired networks, itcould be harder to monitor the status of wireless net-works because the users and traffic patterns are alwayschanging. Cognitive radio is a promising technologyto improve the spectrum utilization efficiency, and itadapt dynamically to the outside RF environment byobserving the radio scene in real-time and making deci-sions based on reasoning (facilitated by the learningcapability).Similarly, the basic functions for cognitive cycle, i.e.

awareness of the environment and intelligence learningcould be stretched from spectrum utilization to generalresource utilization. Awareness of the radio environmentstands for monitoring and sensing the radio system sta-tus. For instance, instead of reporting from user termi-nals, local status could be collected by dedicate cognitivenodes deployed in the area. Learning engine plays animportant role in the system as it introduces intelligenceto reduce complexity. The networks are configured tothe optimal states by the reinforcement learning processautomatically.

3.5 Big data analyticsThe emerging big data paradigm will help to not onlycollect and filter network information but also presentvaluable sight and understanding for the operation ofthe network. Big data analytics can extract much moreinsightful information than traditional data analytics andcan help improve the performance mobile cellular net-works and maximize the revenue of operators.Particularly, with the global view of the wireless net-

work, the logically centralized controller in SDRAN cancollect kinds of large data from different componentsand layers with arbitrary granularity, such as channelstate information at physical layer, packets informationat link/network layers. Although sharing informationamong different layers can improve network performance,the network will become so complex that traditionalapproaches are inadequate to design and optimize suchnetworks. Fortunately, big data analytics, which leveragesanalytical methods to obtain insights from data to assistdecisions, can help the design and operation of controland management in SDRAN.Big data analytics is an effective way to guarantee

SDRAN evolution in the following two aspects.

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 9 of 12

Firstly, unlike the common policy delivered for all usersin the traditional wireless network, data analytic tools(clustering or classification algorithms) enable the per-user policy based on user preference and service featuresanalyzed by history data stored in SDRAN controller.Forinstance, after classification based on user’s preferencedata on access networks (Wi-Fi or LTE), different userscould be classified into several groups, the controllercould make smart access decision for the users belong tothe same group.Secondly, mobile cellular networks’ operational deci-

sions are usually made manually or depending on thealgorithms inside the hardware. With the development ofbig data analytics, the operations of mobile cellular net-works can be higher availability, higher precision, dynamicand importantly in real-time. Real-time reactions bring inbetter performance, not only for the optimization of thenetwork but also for the quality of user experience.Furthermore, the prediction of user mobility and traf-

fic patterns based on previous data is more accurateand efficient. Accurate prediction is necessary for adap-tively configuring the network policies and managingnetwork resources, which are of importance for the pro-posed architecture.For example, the mobility predictionis useful for the mobility management in RAN, andSDRAN controller could allocate resource ahead of timefor possible handover based on the prediction of users’position.

4 Case study of software-defined RANThe following case studies provide an illustration of intro-ducing SDN into radio access networks. As pointed outbeforehand, logically centralized control allows for effi-cient resource coordination among cells, which is partic-ularly useful for solving inter-cell interference problem asdescribed in the use cases below.With the increase of network density and traffic density,

inter-cell interference can lead to a significant degrada-tion in user throughput and service quality. Overlappingadjacent cells need to coordinate their resource alloca-tions (time/frequency/spatial dimension) to reduce harm-ful interference to users in overlapping area. There are anumber of techniques proposed in 3GPP to alleviate theimpact of inter-cell interference.

• Inter-cell interference coordination (ICIC; for release8/9)ICIC reduces the power for some sub-channels in thefrequency domain. Relative narrowband transmitpower indicator (RNTP), high interference indicator(HII) and interference overload indicator (OI) aredefined to negotiate the interference betweenoverlapping cells by X2 interface. RNTP, HII and OIare indication of possible downlink interference,

uplink interference (proactively triggered) and uplinkinterference (actively triggered), respectively.

• Enhanced inter-cell interference coordination(eICIC; for release 10)eICIC is designed for the scenario where macrocellare complemented with low-power picocells in thehotspot area by coordination in time domain. Almostblank subframe (ABS) is designed that during ABSaggressor macrocell inflicting high interference ontovictim picocell users are muted. In this way, thepicocell users that are suffering from interferencefrom the aggressor macrocell have a chance to beserved in ABS.

There are some drawbacks for current inter-cellinterference coordination techniques. Firstly, currentinter-cell interference management is performed in a dis-tributed manner [11]. Generally, the processing complex-ity and overhead of distributed interference managementimpose high computing and resource burden on the RANside. Secondly, the lack of global view and informationexchanged lead to poor performance.SDN-enabled access network offers the possibility

to overcome the limitations described for inter-cellinterference management. As shown in Fig. 2, the log-ically centralized controller enables radio resource allo-cation decisions to be made with global visibility acrossmany base stations, which is far more optimal than thedistributed interference management. By centralizing net-work intelligence and computation resources, resourceallocation decisions can be adjusted based on the dynamicpower and subcarrier allocation profile of each base sta-tion. In addition, scalability is improved because as newusers are added, the required computing capacity at eachbase station remains low because these processing is cen-tralized in the SDRAN controller.As given in Fig. 5, up-left and up-right figures show

the difference of traditional distributed interference man-agement and logically centralized interference manage-ment. The SDRAN controller collects the necessaryinformation such as CQI/RSRP/RSRQ measurement andbandwidth requirement from macrocells and picocells.The centralized interference management algorithm inSDRAN controller makes decision on frequency sub-channels/ABS configuration/power level (interferencepolicy) for macrocells and picocells. Examples of suchcentralized algorithms are developed based on graph col-oring and optimization theory.For instance, there are two picocells support ICIC and

EICIC separately, denoted as PCf and PCt . The two pic-ocells coexisted with one macrocell (denoted by MC).The macrocell and picocells report their locations, trans-mission power(RNTP), support protocol version, loadstatue and interference statue to SDRAN controller in

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 10 of 12

(a)

(b)Fig. 5 Illustration and system evaluation of inter-cell interference management scheme. a Interference management mechanisms and b Simulationscenario and results

the first time. SDRAN controller records the informationinto database. In eICIC, ABS patterns are configured insemi-statically and signaled between the macrocell andpicocells over the X2 interface. The SDRAN controlleris supposed to decide which subframes are to be set asABS depending on different information. The picocell ormacrocell initiate ABS request by sending load informa-tion (LI) to SDRAN controller. In addition, the SDRANcontroller can ask the macrocell to communicate theutilization of the allocated ABS resources by starting aresource status (RS) reporting initialization mechanism.The macrocell responds and provides the required infor-mation with a RS update with ABS status.Based on the ABS status from the macrocell, the

SDRAN controller has sufficient information to deter-mine whether to usemore or less subframes as ABS before

deciding on a new ABS muting pattern. The SDRANcontroller answers by sending back LI message withABS information (ABS ratio) to macrocell, and macro-cell will send another LI message which includes the ABSmuting pattern. If the macrocell makes the decision ofchanging the ABS muting pattern, it informs the pico-cells within the cluster by means of an ABS informationmessage.Periodically, SDRAN implement following algo-rithm to determine the ABS ratio(denoted by β) and/orsub-channel ratio(denoted by α) among macrocell andpicocells.Step 1: To identify the edge UEs, signal-to-interference-

plus noise (SINR) instead of SNR (signal-to-noise ratio)is used as the metric in the algorithm. Without interfer-ence, the SNR in a PRB(physical resource block) f andsub-frame t can be calculated as

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 11 of 12

γk = ps(f , t) ∗ |Hs,kf (f , t)|2N0 ∗ B

(1)

where |Hs,k(f , t)| is the channel gain from the servingcell to user k, Ps(f , t) denotes the corresponding transmitpower, N0 and B represents the noise power density andbandwidth, respectively.In the presence of inter-cell interference, SNR is

replaced by SINR, which can be obtained as

γk(f , t) = ps(f , t) ∗ |Hs,kf (f , t)|2N0 + ∑

i�=s pi(f , t) ∗ |Hi,k(f , t)|2 (2)

where ‖Hi,k(f , t)‖ is the channel gain from the interferingcell i to user k,Pi(f , t) denotes the transmit power fromthe interfering cell i.Step 2: Label the edge UEs in victim cell based on

following criteria:

γk(f , t) ≤ γth (3)

where γth is the minimal SINR threshold for UE. Oth-erwise, the UEs are labeled as center UEs of victim cell.Calculate the traffic belong to the edge UEs in victimcell, denoted as uedge, while the total traffic of macrocell(interference cell) is umacro.Step 3: SDRAN controller determines the ABS

ratio(β) and sub-channel ratio(α) according to followingalgorithm.

1: Initiate α = β = 02: Input uedge, umacro Protocol version3: Switch(Protocol version)4: Case Release 8 or 9:5: α = uedge/umacro6: Case Release 10:7: β = uedge/umacro8: End switch9: Output α,β

Step 4:Macrocell eNB determine the ABS pattern or fre-quency sub-channels allocation based on parameters β/α

from SDRAN controller.Some system-level simulations are implemented to

prove the performance of proposed scheme. The simu-lation scenario is configured according to case 1 config-uration 4a (corresponds to the macrocell and picocellsdeployment with 6 dB range expansion) defined in the3GPP [Ref TBA]. The scenario is composed of 21 macro-cells (i.e. seven three-sector co-sites cells) and 2 picocellsper macrocell. Thirtty UEs are dropped per macrocellarea, where four UEs are located in each picocell. Theremaining UEs are dropped randomly in the macrocellarea without overlapping with picocell. Figure 5 showsthe simulation scenario and the cumulative distribution

function (CDF) of signal-to-interference-plus-noise ratio(SINR) obtained in this scenario.The distributed interference management scheme

denotes the traditional approaches with the decision ineach macrocell. We observe that the improvement onthe SINR distribution for logically centralized interfer-ence management facilitated by the SDRAN architectureis about 2 dB compared with the distributed scheme.

5 ConclusionsIn this article, we have identified the key benefits andchallenges that software-defined networking can bring toradio access networks. We have presented a high-levelgeneral architecture of SDRAN along with some of itsmost relevant design and implementation details. Next,some alternative technologies such as C-RAN and NFVwere discussed with the emphasis on the differences andcomplementation. Finally, a simple use case (interferencemanagement in heterogeneous network) is discussed toshow the practical implementation and benefits of logi-cally centralized control.This paper is only the first step towards the completion

of a complete solution. In fact, several issues remain open,such as the compatibility with the current standards andthe feasibility of virtualized control functions. The com-plete design of SDN-based RAN architecture is an excitingavenue for the future work.

AcknowledgementsThis work is supported by the Key Program of the National Natural ScienceFoundation of China (Grant No. 61431008), BUPT Youth Research andInnovation Project (2015RC01).

Competing interestsThe authors declare that they have no competing interests.

Received: 1 August 2016 Accepted: 26 October 2016

References1. MY Arslan, K Sundaresan, S Rangarajan, Software-defined networking in

cellular radio access networks: potential and challenges. IEEE Commun.Mag. 53(1), 150–156 (2015)

2. X Zhou, Z Zhao, R Li, Y Zhou, T Chen, Z Niu, H Zhang, Toward 5G: Whenexplosive bursts meet soft cloud. IEEE Netw. 28(6), 12–17 (2014)

3. C Liang, Yu FR, Wireless network virtualization: a survey, some researchissues and challenges. IEEE Commun. Surv. Tutorials. 17(1), 358–380 (2015)

4. K-K Yap, M Kobayashi, R Sherwood, N Handigol, T-Y Huang, M Chan, NMcKeown, OpenRoads: Empowering Research in Mobile Networks. (ACMSIGCOMM, Spain, Barcelona, 2009)

5. A Gudipati, D Perry, LE Li, S Katti, SoftRAN: Software defined radio accessNetwork, HotSDN’13, (Hong Kong, 2013)

6. CJ Bernardos, A De La Oliva, P Serrano, A Banchs, LM Contreras, H Jin, JCZu?niga, An Architecture for software defined wireless networking. IEEEWirel. Commun. 21(3), 56–61 (2014)

7. K Pentikousis, Y Wang, W Hu, MobileFlow: Toward Software-DefinedMobile Networks. IEEE Commun. Mag. 51(7), 44–53

8. T Chen, Zhang H, X Chen, O Tirkkonen, SoftMobile: Control Evolution forFuture Heterogeneous Mobile Networks. IEEE Wirel. Commun. 21(6), 70–78

9. R Wang, H Hu, X Yang, Potentials and Challenges of C-RAN SupportingMulti-RATs Toward 5G Mobile Networks. IEEE Access. 2, 1187–1195 (2014)

Xu et al. EURASIP Journal onWireless Communications and Networking (2016) 2016:264 Page 12 of 12

10. A Imran, A Zoha, Challenges in 5G: How to Empower SON with Big Datafor Enabling 5G. IEEE Netw. 28(6), 27–33

11. OpenFlow-enabled mobile and wireless networks (2013). https://www.opennetworking.org/images/storiessdn-resources/solution-briefs/sb-wireless-mobile.pdf

12. Network functions virtualisation in ETSI (2015). http://www.etsi.org/technologies-clusters/technologies/nfv

13. 3GPP TS36.101 Specification (2016). http://www.3gpp.org/ftp/Specs/archive/36_series/36.101/36101-am0.zip

14. C Yang, Z Chen, B Xia, J Wang, When ICN meets C-RAN for HetNets: anSDN approach. IEEE Commun Mag. 53(11), 118–125 (2015)

Submit your manuscript to a journal and benefi t from:

7 Convenient online submission

7 Rigorous peer review

7 Immediate publication on acceptance

7 Open access: articles freely available online

7 High visibility within the fi eld

7 Retaining the copyright to your article

Submit your next manuscript at 7 springeropen.com