Sie sind auf Seite 1von 14

The Evolution of SON to Extended SON

John M. Graybeal and Kamakshi Sridhar


The promise of Long Term Evolution (LTE) is its ability to deliver high data rates with the high quality of service (QoS) needed for a variety of applications without extensive operational overhead. To this end, selfoptimizing networks (SONs) provide a 3rd Generation Partnership Project (3GPP) standardized framework for self-conguration, self-optimization, and self-healing in LTE networks. Alcatel-Lucent is positioning extended SON as an extension of the SON concept to include the end-to-end network environment, including the application domain and user equipment (UE) clients, as well as associated network and user policies to enable complex optimizations to be applied for specic users and/or applications. This paper presents the drivers for LTE SON, Alcatel-Lucents vision for SON, and the evolution towards extended SON with representative use cases and a discussion of the proposed architecture. 2010 Alcatel-Lucent.

Drivers for SON and Extended SON


Service providers today face a mixed blessing. With the advent of smartphone devices, wireless data services have nally gone mainstream, thereby providing a growing new revenue source. In addition to personal computers (PCs) with wireless data cards, a diverse range of popular wireless data devices has emergedsuch as the iPhone*, Blackberry* and Droid*, as well as others. As a result, a very wide range of applications are now being served wirelessly, which includes always-on synchronization software (e.g., Microsoft Outlook*, Blackberry e-mail), Web browsing, video (real time and buffered), peer-to-peer, gaming, and social-networking. However, the amount of wireless data being consumed by users, as well as the associated growth rate, has been explosivewith signicant implications for the service operators networks, their architecture, and their operations. First, the growth rate of the total data throughput (i.e., measured in total bits delivered) far exceeds the growth rate for wireless revenue; consequently networks today and in the future must deliver everincreasing data at an ever-decreasing cost per bit. This need for a lower cost per bit [4] and the demand for broadband data services have played important roles in driving the rapid introduction of Long Term Evolution (LTE) in North America, and the rest of world may soon follow. As a result, service providers are increasingly having to manage overlapping second generation (2G), third generation (3G), and fourth generation (4G) technologies such as orthogonal frequency division multiplexing (OFDM)-based LTE or WiMAX wireless networks. Second, the roughly exponential growth rate of wireless data outstrips both the projected outlook for continued spectral efciency improvements [2] and the availability of new wireless spectrum. Since present spectral efficiency performance for best-effort data is close to the theoretical Shannon limit,

Bell Labs Technical Journal 15(3), 518 (2010) 2010 Alcatel-Lucent. Published by Wiley Periodicals, Inc. Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/bltj.20454

Panel 1. Abbreviations, Acronyms, and Terms 2GSecond generation 3GThird generation 3GPP3rd Generation Partnership Project 4GFourth generation ANRAutomatic neighbor relations CDMACode Division Multiple Access CoMPCollaborative multi-point DLDownlink E2EEnd-to-end E911Enhanced 911 eNBEnhanced NodeB HARQHybrid automatic repeat request HLNHigh Leverage NetworkTM HOHandover ICICInter-cell interference coordination IDIdentier IPInternet Protocol IRATInter-radio access technology LTELong Term Evolution LTE-ALTE Advanced MIMOMultiple input multiple output MMEMobility management entity MPLSMultiprotocol label switching NetMIMONetwork MIMO NGMNNext Generation Mobile Networks OAMOperations, administration, and maintenance OFDMOrthogonal frequency division multiplexing OPEXOperational expenditures PCPersonal computer PCCPolicy and charging control PCIPhysical cell ID PCMDPer call measurement data PCRFPolicy charging rules function PGWPacket data network gateway PHYPhysical layer QCIQoS class identier QoEQuality of experience QoSQuality of service RACHRandom access channel RANRadio access network RFRadio frequency SGWServing gateway SONSelf-optimizing network TCPTransmission Control Protocol UEUser equipment ULUplink VoIPVoice over Internet Protocol WBWideband W-CDMAWideband Code Division Multiple Access WNGWireless Network Guardian

we anticipate that this growth will drive greater spatial reuse through large numbers of wireless femtocells and picocells, as well as Wi-Fi access points. Such a combination of macrocells and small cells is often referred to as a layered or heterogeneous network. Together with the presence of multiple technologies, they imply growing radio access network (RAN) and core network complexity. Third, the growth of wireless data is increasing the level of network congestion in many networks. Consequently, quality of service (QoS) will finally become widely deployed. The great breadth of wireless data applications will pose complex challenges for service providers in the areas of QoS and policy. Fourth, optimizing the performance of a multitechnology, heterogeneous network implies simultaneous optimization of a large number of parameters. Often changes in these network parameters will need to be coordinated across technologies and layers,
6 Bell Labs Technical Journal DOI: 10.1002/bltj

and varied over a wide range of loads, applications, and time scales. In our view, the combined implications of the wireless data explosion will break the present network management paradigm. Traditional network managementinvolving both centralized and supervised operationswill prove to be inadequate and too expensive to serve the need. This was also the view of the recent 3G Americas white paper on LTE self-optimizing network (SON) [1]. We view the selfconguration and self-optimizing features of the SON and extended SON as being an essential path towards reducing operational expenses as well as increasing network performance, one that can ensure the continuing viability of wireless networks.

Why Automate Network Operations?


For an operator, the network is their key investment. It is thus naturally in their interest to maximize

the performance of the network under highly variable and unpredictable conditions, while minimizing the level of effort and degree of risk. Network automation is a reality in many parts of the network today. However, this automation is typically localizedeither within a specic network element and/or within a given layer. Examples of automation already present within the network include air interface scheduling and rate control within the base station, or fast re-routing within the packet core. They clearly operate stably with signicant performance benets. With proper design, SON and extended SON will extend automation to many other operations and functionse.g., cross-layer, inter-element, or per-userwith signicant additional operational and performance benets. Where feedback of actual network data is utilized to tune design parameters, it is all too often manual in nature and modest in scope. It is important to note that manually set parameterswhether slowly varying or static in natureare typically assigned compromise values. By this we mean that the parameter would ideally take on a range of values for different conditions, but being slowly varying, it is assigned a single choice which is the best compromise or tradeoff across the various options. As a consequence, the resulting network and user performance are likely sub-optimal most of the time. Thus at a high level, we see two overlapping rationales for automation. Simply stated, some processes are repetitive, while others are too fast or difcult to be performed manually. The rst category is automating repetitive processes, which clearly saves time and reduces effort. Auto-conguration and self-conguration clearly fall into this category. There also is a second large category of performance enhancements that cannot be performed manually, as the required response times are too fast and/or the granularity (per-user, per-application, per-ow, as a function of time or loading) is too ne. The results of network optimization can only be as good as the quality of the input data being supplied to it. Recent advances in network monitoring provide signicant potential benets in this endeavor. Automatically collected measurements from multiple

sources (e.g., from user devices, individual network elements, and on an end-to-end basis from advanced monitoring tools) will provide accurate real time and near real time data upon which these algorithms can operate. Consequently, substantial opportunities exist for cross-layer, end-to-end, and per-user/per-application/ per-ow optimizations for extracting additional performance benets and management exibility.

Operations Concerns and Issues


Operators have pointed to many operational areas in need of improvement in their present networks. These areas include: Reducing time-consuming and repetitive tasks, Reducing the time spent manually analyzing network data coming from multiple databases or operations, administration, and maintenance (OAM) systems, and Reducing error-prone operations due to the time lag in updating values to reect the rapidly changing network conditions. Consequently, operators have set clear goals for SON capabilities: Significant reductions in operational expenses (OPEX), during deployment as well as for continuing operations, Higher end user quality of experience (QoE) and reduced churn, Improvements in network quality and reliability, and Higher performance from adapting the network to variations in loading and other dynamic operational conditions. At the same time, operators have expressed legitimate concerns in several areas. They have noted the importance of stability for SON algorithms, that they need to understand the reasoning behind the SON algorithm decisions or outputs, and the potential impacts that SON may pose upon their operations personnel. Alcatel-Lucent believes the best way to address these opportunities and challenges is by introducing SON and extended SON capabilities in a phased manner in concert with extensive eld verication. In this way, service providers can develop condence in SON algorithms through sequential rollout of SON features
DOI: 10.1002/bltj Bell Labs Technical Journal 7

of increasing sophistication, which are aligned with the evolving needs of their network. As SON standards presently are focused upon LTE, let us illustrate this approach with an LTE example. Early LTE networks are expected to be deployed initially as data-only services. For example, Release 8 does not include important voice functionality such as enhanced 911 (E911) and legal intercept. Also, Voice over Internet Protocol (VoIP)-capable devices are not expected until Release 9. Such a data-only LTE network may consist of clusters in higher-density regions, with surrounding lower-density regions being served by a 3G technology such as Code Division Multiple Access 2000 (CDMA2000) or Wideband Code Division Multiple Access (W-CDMA). During this phase, the emphasis will be on quickly establishing services and LTE coverage (as opposed to capacity). Hence the most important SON features will be in the self-conguration and self-organizing categories (e.g., enhanced NodeB [eNB] self-conguration, automatic neighbor relations [ANR], and physical cell ID [PCI] configuration). During this phase, inter-technology handover will also be important. As LTE customer acceptance grows and capacity increases, the network operations emphasis will soon evolve to managing more capacity-related issues: cell loading (and thus interference) will grow, cell-splitting will begin, deployment of femtocells and picocells may begin in earnest, and core network expansion will commence. Voice services will also begin with the advent of Release 9, further highlighting the importance of seamless mobility. The device ecosystem will also increase in diversity, and simultaneous voice and data usage will increase significantly. Hence the most important SON feature will be handover optimization. Cell outage detection and compensation will also gain in importance for reducing dropped calls during infrequent cell outages. Managing interference between macrocells and small cells will also be important when operated in the same spectrum. With increasing load, the various capacity optimization algorithms will gain importance, such as inter-cell load balancing, inter-cell interference coordination (ICIC), and inter-technology load balancing.

As LTE deployments gain additional maturity, we expect LTE-advanced features to become available, including collaborative multi-point (CoMP) transmission, relays, carrier aggregation, and full-blown heterogeneous networks with extensive numbers of femtocells and picocells. Widespread market acceptance will also lead to signicant demand for broadband data. Consequently, management of the user experience, end-to-end QoS and policy, rogue ows, and heavy users will all become important.

SON Use Cases


The 3rd Generation Partnership Project (3GPP) has generated a number of compelling SON use cases which will provide both operational and performance benefits. These are described in the 3GPP Release overview descriptions and other standards publications [3], as well as in the 3G Americas white paper on LTE SON [1]. The use cases include: Self-establishment of eNodeBs. Newly deployed nodes (eNodeBs) are configured by automatic installation procedures. Automatic neighbor relation. eNodeBs capable of automatically generating and managing their own neighbor relation tables, based on measurement reports from user equipment (UE). Automatic conguration of physical cell ID. Automatic selection of the eNodeB physical cell ID, ensuring no redundancy among neighboring cells to which handover may occur. Coverage and capacity optimization. Consists of interference reduction, mobility load balancing, mobility handover (HO) optimization, mobility robustness optimization, and random access channel (RACH) optimization. Interference reduction (ICIC). Inter-cell interference coordination via fractional frequency reuse techniques to improve sector throughput, especially near the cell edge. Mobility robustness optimization. Optimization of handover parameters, detection of problematic cell relationships, and reduction of HO ping pong. Mobility load balancing optimization. Optimization of both intra-/inter- carrier load balancing as well as load balancing between LTE and 2G/3G.

Bell Labs Technical Journal

DOI: 10.1002/bltj

LEGEND Capacity optimization Coverage optimization Coverage impact Capacity impact RACH optimization Cell outage detection/comp Energy savings OPEX impact

ICIC

Load balancing optimization

Self-configuration

Hand over optimization

Automatic neighbor relation

Physical cell ID configuration

Terrain & clutter 2G/3G layout 2GSecond generation 3GThird generation CompCompensation ICICInter-cell interference coordination OPEXOperational expenditures RACHRandom access channel SONSelf-optimizing network

Figure 1. Various SON use cases shown with their dominant optimization purpose (coverage, capacity, or OPEX reduction).

RACH optimization. Optimization of the random access performance, which impacts the call setup delay/success rate and HO delay/success rate. Cell outage detection and compensation. Automatic detection of an eNodeB outage, auto-recovery methods for the failed cell, and coverage compensation via neighboring eNodeBs. The corresponding SON features are being addressed in LTE Releases 8, 9, and 10. The Next Generation Mobile Networks (NGMN) industry group has also generated an additional set of SON use cases and requirements, and they can be found in NGMN white papers [5, 6]. The SON use cases are also listed in the 3G Americas white paper [1].

SON Use Case Interactions


While the various SON use cases clearly have individual merit, their associated algorithms can operate simultaneously on the same and/or adjacent cells

and sectors. As illustrated in Figure 1, when thinking more closely about various goals and optimization functions, it becomes apparent that there are important differences. Some of the use cases are dominantly coverageoriented, some are more capacity-oriented, while others are more operationally oriented. Even among use cases with the same optimization purpose, there can be important differences in the detailed optimization function. For example, ICIC seeks to maximize the sector throughput and edge rate, while load balancing within a carrier seeks to maximize the throughput of a cluster of cells. As a result, there is considerable potential for adverse interactions among the various algorithms. Clearly, these interactions must be understood and properly managed to ensure stability and proper operation. We cannot have a situation where one cell experiences an outage, and the various coverage and

DOI: 10.1002/bltj

Bell Labs Technical Journal

capacity algorithms on neighboring eNodeBs come into conict. Observations From Prior Dynamic Optimization Research The issue of potential conicts between different optimization algorithms is of course not new to wireless. Consider the case of dynamic rate control and soft handoff (or handover). These algorithms have different purposes and could come into conict if not properly handled. Of course, interactions between them are minimized because of the different intrinsic time scales on which they need to operate (but imagine the consequences of trying to design a system which handed-off on the time scale of fast fading!). It is useful to observe that separated time scales can minimize interactions. Figure 2 shows a Bell Laboratories slide from 2003 on relevant time scales for dynamic optimization. Note the logarithmic time axis, whose labeled range extends from 10 milliseconds to 10 8 seconds (~3 years).

Also displayed are the relevant time scales for longterm load variations, per-mobile variations, base station response, and manual optimization. We can see there is considerable range in the time scalesperhaps six decades or morebetween present manual optimization times and the characteristic timing of relevant mobile phenomena (e.g., beginning with typical data session durations, and extending to the time scale for fast power or rate control). Consequently, for many phenomena there can be considerable low-lying fruit for dynamic optimization, even on intermediate time scales of minutes to hours. It is also important that SON algorithms not interact with feedback and control mechanisms already existing within the network, e.g., for rate control or handover. When a selected algorithm operates on a given time scale (e.g., one hour), the dynamics of much faster processes and fluctuations (e.g., 10 minutes) are largely averaged away, and thus may have signicantly reduced impact.

Relevant time scales Selected temporal scales relevant to dynamics (12 decades) Per-mobile variations SHO Voice call Fast-fading Shadow fading Daily Weekly Busy hour Long-time load variations Subscriber growth Seasonal Time (sec)

10

10

10

10

10

3G packet 3G pwr control 2G pwr control Dyn rate control

Packet call Data session Anchor swap Hourly Drop timer Standard service measurements

Network changes, cell splitting/additions In-service optimization Manual optimization

Existing cell response

Dynamic optimization window 6 decades

2GSecond generation 3GThird generation DynDynamic

PwrPower SecSeconds SHOSoft handover

Figure 2. A Bell Laboratories slide from 2003 showing relevant time scales for dynamic optimization.

10

Bell Labs Technical Journal

DOI: 10.1002/bltj

Evolution to a High Leverage NetworkTM


The SON principles of self-configuration, selfoptimization, and self-healing that have been developed for the RAN allow a network operator to set up, manage, and operate the radio access network in a cost-effective manner while maximizing its performance. While the various SON mechanisms are very useful, they are limited to the RAN and the OAM systems only. From an operator perspective, true value is achieved if additional revenue can be generated, thus monetizing the underlying infrastructure by providing differentiated services to a given user. This translates to a need to extend the SON mechanisms from the RAN into the core, because all the requirements that drive the need for SON in the RAN also apply equally to the core network. The converged network of the future will have an Internet Protocol/Multiprotocol Label Switching (IP/MPLS) core, aggregation, and access that will improve scalability reliability and QoS and lower OPEX. In a converged network, wireless and wireline networks will be seamlessly connected through IP. The provisioning of end-to-end QoS mechanisms will be transparent to the underlying physical layer (PHY). To truly optimize the network performance and thus provide value to the service provider, SON principles must be applied across the LTE network including the RAN and into the core. By coupling the RAN and the core network elements, better decisions are made in the base stations not just based on radio channel conditions but also upon end-to-end conditions and performance, as well as the user prole. This requires close coordination among the different RAN and core elements to meet the high QoS needs. Our vision of SON includes the entire converged system with the radio access, edge aggregation, and core networks, and across optical and IP layers to allow for efcient provisioning of optimized end-to-end operations. Recognizing that most service provider networks are multi-vendor, extended SONs design leverages the 3GPP policy charging and control (PCC) architecture. This enables delivery of end-to-end SON capability and coordination across multiple domains (radio, edge, and core), layers (radio and xed transport, IP) and traditional system boundaries for optimized delivery and enhanced user experience.

We now briey describe a few end-to-end SON examples. Consider the case where load balancing is based not just upon RAN resources, but also upon available core resources. This would preclude situations where load is transferred to a cell with limited backhaul capacity which only shifts the bottleneck from the RAN to the core, rather than offering any relief. Likewise, knowledge of delay in the network overall can help the base station optimize its capacity while providing the necessary QoS needed for latency-sensitive users (e.g., voice, real time video, or gaming). Currently, scheduling decisions for voice users are made in the base stations based on QoS class and radio channel conditions. While there is an endto-end latency budget for voice and real time video, today the base station only monitors its own assigned contribution to the latency; it is not aware of the actual end-to-end (E2E) performance. Awareness of an end-to-end delay would provide the base station scheduler more latitude in scheduling air interface resources when there is little end-to-end network delay. Precious air interface resources can be allocated to users who need them because their margin for air interface delay is much lower due to longer backhaul and/or transport delay. For latency-sensitive users experiencing little or no network delay, hybrid automatic repeat request (HARQ) retransmission targets can be adjusted upwards so that additional air interface resources are available for other users in the cell. This scheme allows for more efficient and flexible support for latency-sensitive users with a range of network delays. For end-to-end SON to be scalable, select information must be made available at the base station in a timely manner to allow for better decision making at the eNB. End-to-end SON mechanisms that require close coordination between the RAN and core network form the beginnings of extended SON.

Alcatel-Lucent SON/Extended SON Vision: Optimizing LTE and 3G Performance


Extended SON is the extension of SON concepts across the network to include the end-to-end network environmentincluding the application domain,

DOI: 10.1002/bltj

Bell Labs Technical Journal

11

Key SON features near-term Self-configuration Automatic neighbor relation list management Automatic physical cell ID configuration Cell outage detection Interference reduction/ICIC SON evolving to extended SON

Making real-time optimization decisions based upon a policy-enabled infrastructure

Data measurement Extended SON Enforcement point

Data analysis reduction

Policy-enabled decision

IP channel eNode B

Evolved packet core (all-IP)


IP transport (backhaul and backbone)

3GThird generation eNode BEnhanced node B ICICInter-cell interference coordination

IPInternet Protocol LTELong Term Evolution SONSelf-optimizing network

Figure 3. Optimizing LTE and 3G performance with extended SON.

UE clients, and associated network elementsso as to allow complex optimizations to be applied for specific users and/or applications based on policy. Extended SON, shown in Figure 3, allows the network to make real time optimization decisions based on a policy-enabled infrastructure and is composed of four key aspects that work in concert with each other to allow for network optimization. These four aspects are: Network data measurement, Data analysis and reduction, Policy-enabled decision, and Policy enforcement. Supporting multimedia and advanced applications requires a system approach to instrumentation, conguration, and management. Extended SON automation creates opportunities for further OPEX and performance improvements through a distributed analysis, decision, and enforcement architecture that allows for intelligence to be distributed and applied in different parts of the network in a scalable manner.

Above all, learning must take place so that the network can adapt to dynamic changes without direct human supervision. LTE will evolve to support significantly higher carrier bandwidths, peak rates, and spectral efciencies through the introduction of technologies such as LTE-Advanced (LTE-A) and heterogeneous cells. LTE-A could, for example, dynamically apply network multiple input multiple output (NetMIMO), a form of CoMP, primarily for gold usersthereby avoiding the significant increase in backhaul throughput that would result from applying it to all users. Heterogeneous cells would allow an operator to make optimized decisions regarding which cells to use and when. Active antenna arrays would allow the network to reinforce radiation in a certain direction. Multi-standard radio would allow for exibility in supporting different radio rates. Flat packet architectures would allow for optimized IP routing by reducing the number of network elements in the data path. Rich voice and communications would

12

Bell Labs Technical Journal

DOI: 10.1002/bltj

allow for optimized air interface support for specic applications based on the subscriber profile. It is important to recognize that these enablers and capabilities cannot be provided to all users at all times since network resources are scarce. In an overconstrained environment, extended SON provides the ability to use policy information to make differentiated decisions and therefore allows an operator to identify opportunities to derive revenue by providing a differentiated portfolio of advanced applications. This allows the monetization of the underlying infrastructure without a large crew of engineers to constantly monitor and adjust key architectural network parameters. Extended SON Functional Architecture An open loop system, however well conditioned, does not have a target operating point it needs to reach. An open loop system could potentially reach an operating point or network state different from that expected by the service provider. A network tuned to maximize the throughput of its premium users reaches a different operating point than one tuned to maximize the total number of users in the system. The premise of extended SON is that the implementation of a closed loop system with monitoring, feedback, and control will allow an operator to steer the network towards a target operating point that could be decided based on time of day, user applications and QoS environment, radio channel conditions, network loading, or network topology. The 3GPP PCC architecture allows the introduction of policies (charging policies, user policies, QoS policies) in the network to help an operator manage network resources to best serve a particular user. Sensing the network state and utilizing that information allow the operator to dynamically tweak specic policies in near-real time so that the network can optimize a specic objective as decided by the operator. Figure 4 shows the extended SON functional architecture as applied to an LTE network, although the principles of extended SON apply to 2G/3G networks as well. Real time data collected from various monitoring tools from single or multiple nodes are combined and compressed with persistent network

data such as network topology information, subscriber policies, and dynamic network data including network load, network latency, and subscriber policy information distributed across the network. This combined data is then ltered in the extended SON decision element to derive a parsimonious subset of key relevant variables, which are then used to make decisions that are then enforced at various points in the network by the extended SON control element. For each SON/extended SON algorithm, the network will use optimal data sources and optimal location for the analysis and decision functions. Key to the concept of extended SON is the availability of end-to-end measurement tools including the Alcatel-Lucent 9900 Wireless Network Guardian, and per-call measurement data (PCMD) that help provide a view into data aggregated from across multiple network elements for near-real time proactive monitoring and data signature analysis. Each of these tools provides different kinds of information on different time scales at different layers of the network. Feedback mechanisms are not new to wireless networks, but their reduction for better automation is. Currently, UE measurements allow the eNB to deduce the radio channel conditions and schedule the downlink/uplink (DL/UL) radio resources to meet certain QoE. Through advanced monitoring tools, extended SON extends the notion of feedback to include the entire end-to-end network to provide a mechanism for automated optimal response to dynamic variations in load, applications, policies, and network conditions. The collection of data coupled with the ability to apply real time network policies to tune specific parameters will result in the ability to make better decisions and thus apply optimization across the network. As a result, the whole network does better. This allows operators to provide their gold subscribers with a higher over the air bandwidth through selective NetMIMO. The extended SON architecture is conformant to 3GPP principles and leverages existing 3GPP mechanisms in place to support a broad range of use cases in a multi-vendor environment. Extensions for select optimizations will be introduced in 3GPP with compliance to the 3GPP principles.

DOI: 10.1002/bltj

Bell Labs Technical Journal

13

Internet
DPI

HSS

PCRF

PGW

Data collection from UE, NEs, various clients

Dynamic data Network Load, latency Subscriber Location, app

Persistent data Network Topology Subscriber Policies

SGW LTE/EPC network elements

Monitoring Data analysis & reduction function Control


Enforcement within the eNB, MME, SGW, PGW

MME

eNB

eNB

eNB

Decision function (algorithms)

Legend:
Enforcement
Advanced monitoring

Network element management Advanced E2E monitoring

Policy resource

AppApplication DPIDeep packet inspection E2EEnd-to-end eNBEnhanced NodeB EPCEvolved packet core

HSSHome subscriber server LTELong Term Evolution MMEMobility management entity NENetwork element PCRFPolicy charging rules function

PGWPacket data network gateway SGWServing gateway SONSelf-optimizing network UEUser equipment

Figure 4. Extended SON functional architecture.

Scope of Extended SON Use Case Categories


Extended SON covers a broad category of use cases that appeal to different service provider user segments, depending on their role and function. Figure 5 illustrates a broad grouping of various extended SON uses, organized by QoE and volume of usage. On one hand, there are the platinum users who tend to be fewer in number but require high QoE and potentially more network resources for short durations. At the other end, there are the low QoE but high volume users who provide an opportunity for signicant OPEX savings if performance can be optimized. Furthermore, there are the heavy and/or abusive users who impose a signicant load on the entire network and consume network capacity that could otherwise be accessible to other users. Finally

there are several new applications such as machineto-machine devices and communications which may have low bit rate but very high volumes and have the potential for significant new revenue generation. Extended SONs broad scope provides multiple opportunities to provide signicant benets to the service provider. Monitoring traffic in near-real time and building an internal state engine allow networks to build intelligence about the state of the network which could then be used for decision making via the 3GPP PCC architecture.

Example Extended SON Use Cases


Extended SON permits the network to become a dynamic entity that is able to sense end-to-end network conditions and optimize network and/or user

14

Bell Labs Technical Journal

DOI: 10.1002/bltj

High volume

Volume of users

High volume existing users new m2m applications

Disruptive users OPEX reduction Enhanced QoE platinum users

Low volume

Low QoE m2mMachine to machine OPEXOperational expenditures

High QoE QoEQuality of experience SONSelf-optimizing network

User QoE

Figure 5. Broad scope of extended SON use cases.

performance, based upon user and network policies. Examples of extended SON optimization use cases include: Trafc management. Prioritize and manage ows in the network by: Managing heavy ows (in both trafc and/or signaling planes) which adversely impact other users, and automatically identifying and managing virus-driven and other rogue ows. Optimal cell (macro/pico/femto) or technology (LTE/W-CDMA/Wi-Fi) selection based upon NodeB/eNB capacity, backhaul capacity, application needs, and/or network topology (e.g., to avoid mobility management entity [MME] or serving gateway [SGW] relocation). Dynamically selecting packet data network gateway (PGW)/anchor points in order to optimize transport cost, congestion, and/or QoS. Real time quality/capacity trade-off. Base station scheduler awareness of real time end-to-end latency (VoIP, video, gaming). Wideband (WB) codec support as signal conditions and backhaul capacity permits. User QoE. Various optimizations are possible based on subscriber prole and priority, application, and end-to-end network state.

Providing high value optimizations to select users, e.g., NetMIMO (CoMP) rate enhancements and carrier aggregation enhancements. Application of Extended SON Functional Architecture to Example Use Cases Figure 6 shows the extended SON architecture with monitoring, decision, and control forming the closed loop feedback mechanism that is implemented in an automated manner. As noted earlier, the extended SON framework can be applied to any operator network with multi-vendor elements since the extended SON decision function feeds into the policy charging rules function (PCRF), which is the sole 3GPP arbiter of policy decisions. Without requiring proprietary enhancements to the RAN eNB/NodeB elements or the core SGW, PGW, or MME or SGSN and GGSN elements, extended SON exibly enables a broad range of use cases. These use cases would in general be implemented via extended SON optimizing the end-to-end network on a longer time scale than the existing fast inner-loop optimizations (e.g., rate control within the eNB). This natural time scale separation allows the outer loop to set the network operating point on a longer time scale, which is then tracked by the fast inner loop at the eNB using UE measurements as inputs.
DOI: 10.1002/bltj Bell Labs Technical Journal 15

HSS WNG-monitored elements Control to MME Control to PCRF/PGW interfaces OAM S6a WNG Extended SON MME

Operator IP services

Rx

PCRF SGi

S1-MME S11 LTE macro/ pico/femto SGW

Gx

Femtos

Macros

Picos

S1-U

S5/S8

PGW

3G macro

BSC/ RNC

PDSN

*Schematic 3G network shown


3GThird generation BSCBase station controller HSSHome subscriber server IPInternet Protocol LTELong Term Evolution MMEMobility management entity OAMOperations, administration, and maintenance PCRFPolicy charging rules function PDSNPacket data serving node PGWPacket data network gateway RNCRadio network controller SGWServing gateway SONSelf-optimizing network WNGWireless Network Guardian

Figure 6. Extended SON architecture: closing the control loop with the Alcatel-Lucent Wireless Network Guardian.

Consider some specic examples. First, in many wireless data networks, a small subset of users use a disproportionate amount of the network resources. Extended SON provides a range of options for the service provider, from generating additional revenue to intelligent deprioritization of users when network congestion is present. In the latter case, extended SON can manage large data ows within the 3G/LTE core and RAN by monitoring the source and destination of user ows and their cell sectors, and deprioritizing or ofoading trafc of the heaviest users. This surgical deprioritizing of a few massive ows would be triggered only when network congestion (user or control plane) exists which impacts other users QoE. Constraining the trafc for the heaviest ~2 percent of users can result in a substantial decrease in loading for the macrocell RAN and core. This can benet the operator two ways: 1) through deferrals of RAN and core capital expenditures (CAPEX), or 2) through reduced churn brought on by improved QoE for the

remaining 98 percent of users. Both options allow service providers to focus on serving protable data. This approach does not require extended SON aware user applications, and there is no impact to third party application developers. Furthermore, this would work in a multi-vendor implementation since the decision to throttle is made at the PCRF and enforced at the PGW, consistent with the principles of 3GPP PCC architecture. Similarly, with the detection capabilities of the Wireless Network Guardian, extended SON can identify various types of rogue ows in the network and quickly take action against them (e.g., throttle or block them). Such ows include virus-laden or virusgenerated traffic and/or denial of service (DoS) attacks. Removing these ows benets service providers through improved network performance and benets users through greater security and QoE. Extended SON allows for the optimization of LTE and 3G network performance through dynamic load

16

Bell Labs Technical Journal

DOI: 10.1002/bltj

balancing among 3G, 4G, and potentially Wi-Fi. Through the dynamic adjustment of network policies aligned with E2E operating conditions (e.g., based upon detailed network load, UE capabilities, user application, radio frequency [RF] conditions, and bandwidth requirements), an operator could for example ofoad select users from a locally overloaded 3G NodeB cluster onto another 3G carrier or the LTE RAN (inter-radio access technology load balancing). Signicant capacity gains can ensue as a result of better network utilization. This form of intelligent IRAT load balancing would also minimize ping-pong effects which can lead to radio link failures or reduced QoE. Extended SON also allows the optimization of network resources given the availability of macrocells, picocells, and femtocells by offloading traffic from macros to picos and femtos for low mobility users, thereby freeing up macrocell capacity for high mobility users. Extended SON allows the network to support a broad range of QoS class identiers (QCIs) on each of its cells to allow for better operation of internal scheduling algorithms on the LTE RAN. Finally, we turn to an example of extended SON analysis and decisions extending out from the core into the RAN: specically, the introduction of user policies within the eNB that permit the base station to make optimized trade-offs between throughput and delay for Transmission Control Protocol (TCP) and/or latency-sensitive applications, thereby enabling improved utilization of air interface resources. In summary, extended SON architecture enables the network view comprising end-to-end network topology and end-to-end performance to be aligned with the subscriber view to deliver an enhanced user experience through the optimization of the underlying network.

networks utilizing the standard 3GPP policy infrastructure to enable cross-layer, cross-element, and/or end-to-end optimizations for specific users and/ or applications. Its end-to-end network focus provides extensive additional opportunities for performance enhancement and OPEX reduction. Extended SON covers a wide scope of use case categories to provide benefits to multiple segments of service providers. Extended SON requires appropriate real time and near-real time input data sources that allow for feedback from the network to be used to derive controls to help optimize the network performance. Through the implementation of SON and extended SON, service providers will be able to monetize the underlying infrastructure while lowering OPEX and improving capacity and performance. Acknowledgements The authors gratefully acknowledge useful discussions with Tom Anderson, Konstantin Livanos, Vasu Subramanian, Jim Seymour, and Kevin Sparks. *Trademarks
Blackberry is a registered trademark of Research in Motion Limited. Droid is a trademark of Lucaslm, Ltd. iPhone is registered trademark of Apple Computer, Inc. Microsoft and Outlook are registered trademarks of Microsoft Corporation.

Conclusion
Recent wireless trends have made SON and extended SON functionality essential for cost-effective network operations. SON, whose present focus is primarily RAN- and OAM-based, is standardized within 3GPP for LTE and will evolve in future releases. Alcatel-Lucents vision for extended SON consists of extending SON concepts to the complete 3G and 4G

References [1] 3G Americas, The Benets of SON in LTE: SelfOptimizing and Self-Organizing Networks, Dec. 2009, White Paper, http://www.3gamericas.org/ documents/2009_%203GA_LTE_SON_white_ paper_12_15_09_Final.pdf . [2] 3G Americas and Rysavy Research, EDGE, HSPA and LTE Broadband Innovation, White Paper, Sept. 2008, p. 46, http://www.3gamericas.org/ documents/EDGE_HSPA_and_LTE_Broadband_ Innovation_Rysavy_Sept_2008.pdf . [3] 3rd Generation Partnership Project, Evolved Universal Terrestrial Radio Access Network (E-UTRAN), Self-Conguring and Self-Optimizing Network (SON) Use Cases and Solutions, 3GPP TR 36.902, http://www.3gpp.org/ftp/ Specs/html-info/36902.htm . [4] H. Kafka, Business Drivers for Selecting LTE Technology, Proc. 3G Americas HSPA & LTE Executive Brieng (Dallas, TX, Jan. 27, 2009), www.3gamericas.org/documents/03_Hank% 20Kafka.pdf .

DOI: 10.1002/bltj

Bell Labs Technical Journal

17

[5] NGMN Alliance, Next Generation Mobile Networks Beyond HSPA and EVDO, White Paper, Dec. 5, 2006, http://www.ngmn.org . [6] NGMN Alliance, NGMN Recommendation on SON and O&M Requirements, Dec. 5, 2008, http://www.ngmn.org . (Manuscript approved June 2010)
JOHN M. GRAYBEAL was a director in the AlcatelLucent Wireless Chief Technology Ofce, Murray Hill, New Jersey, when this paper was written. Originally a condensed matter experimental physicist, he received a Ph.D. in physics from Stanford University and is a past recipient of a Sloan Research Fellowship and an Ofce of Naval Research Young Investigator award. He previously served on the physics faculties at the University of Florida and the Massachusetts Institute of Technology. At Bell Labs, Dr. Graybeals research interests included predictive and dynamic RF optimization algorithms for wireless networks, and dynamic optimization research including extensive real time measurements and analysis of operating wireless networks (as an original member of the Celnet Xplorer team). Within Wireless CTO, Dr. Graybeals areas of expertise included SON, extended SON, spectrum issues, and advanced air interface technologies. KAMAKSHI SRIDHAR is a director in the Alcatel-Lucent Wireless Chief Technology Ofce and is located in Plano, Texas. She has held various research and management positions in Alcatel-Lucent Bell Labs and Nortel Networks with signicant contributions to Ethernet, Internet Protocol television (IPTV), mobile backhaul, and Code Division Multiple Access (CDMA) systems engineering. Her current interests are in the areas of LTE, WCDMA and CDMA network architectures and enablers, SON, extended SON, and QoS for 3G and 4G networks. Dr.Sridhar received her S.M. and Ph.D. degrees in electrical engineering and computer science from Massachusetts Institute of Technology and a B.Tech. degree in electronics engineering from Institute of Technology, Varanasi, India. She is a Distinguished Member of Technical Staff and has been a member of the Alcatel Lucent Technical Academy since 2004.

18

Bell Labs Technical Journal

DOI: 10.1002/bltj