IEEE Communications Magazine â€¢ February 2011 Vol. 49, No. 2

A Publication of the IEEE Communications Society®

Special Supplement

Passive Optical Networks

•Next-Generation Mobile Networks•Synchronization over Next Generation

Packet Networks

FreeComSocTutorial

BroadbandVideo

SeePage

9

IEEE

M A G A Z I N E

February 2011, Vol. 49, No. 2

www.comsoc.org

February 2011 Cover 1 1/20/11 3:12 PM

Director of MagazinesAndrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)

Editor-in-ChiefSteve Gorshe, PMC-Sierra, Inc. (USA)

Associate Editor-in-ChiefSean Moore, Centripetal Networks (USA)

Senior Technical EditorsTom Chen, Swansea University (UK)

Nim Cheung, ASTRI (China)Nelson Fonseca, State Univ. of Campinas (Brazil)

Torleiv Maseng, Norwegian Def. Res. Est. (Norway)Peter T. S. Yum, The Chinese U. Hong Kong (China)

Technical EditorsSonia Aissa, Univ. of Quebec (Canada)

Mohammed Atiquzzaman, U. of Oklahoma (USA)Paolo Bellavista, DEIS (Italy)

Tee-Hiang Cheng, Nanyang Tech. U. (Rep. Singapore)Jacek Chrostowski, Scheelite Techn. LLC (USA)Sudhir S. Dixit, Nokia Siemens Networks (USA)

Stefano Galli, Panasonic R&D Co. of America (USA)Joan Garcia-Haro, Poly. U. of Cartagena (Spain)Vimal K. Khanna, mCalibre Technologies (India)

Janusz Konrad, Boston University (USA)Abbas Jamalipour, U. of Sydney (Australia)

Deep Medhi, Univ. of Missouri-Kansas City (USA)Nader F. Mir, San Jose State Univ. (USA)

Amitabh Mishra, Johns Hopkins University (USA)Sedat Ölçer, IBM (Switzerland)

Glenn Parsons, Ericsson Canada (Canada)Harry Rudin, IBM Zurich Res.Lab. (Switzerland)Hady Salloum, Stevens Institute of Tech. (USA)Antonio Sánchez Esguevillas, Telefonica (Spain)

Heinrich J. Stüttgen, NEC Europe Ltd. (Germany)Dan Keun Sung, Korea Adv. Inst. Sci. & Tech. (Korea)Danny Tsang, Hong Kong U. of Sci. & Tech. (Japan)

Series EditorsAd Hoc and Sensor Networks

Edoardo Biagioni, U. of Hawaii, Manoa (USA)Silvia Giordano, Univ. of App. Sci. (Switzerland)

Automotive Networking and ApplicationsWai Chen, Telcordia Technologies, Inc (USA)

Luca Delgrossi, Mercedes-Benz R&D N.A. (USA)Timo Kosch, BMW Group (Germany)

Tadao Saito, University of Tokyo (Japan)Consumer Communicatons and Networking

Madjid Merabti, Liverpool John Moores U. (UK)Mario Kolberg, University of Sterling (UK)

Stan Moyer, Telcordia (USA)Design & Implementation

Sean Moore, Avaya (USA)Salvatore Loreto, Ericsson Research (Finland)

Integrated Circuits for CommunicationsCharles Chien (USA)

Zhiwei Xu, SST Communication Inc. (USA)Stephen Molloy, Qualcomm (USA)

Network and Service Management SeriesGeorge Pavlou, U. of Surrey (UK)

Aiko Pras, U. of Twente (The Netherlands)Networking Testing Series

Yingdar Lin, National Chiao Tung University (Taiwan)Erica Johnson, University of New Hampshire (USA)Tom McBeath, Spirent Communications Inc. (USA)

Eduardo Joo, Empirix Inc. (USA)Topics in Optical Communications

Hideo Kuwahara, Fujitsu Laboratories, Ltd. (Japan)Osman Gebizlioglu, Telcordia Technologies (USA)

John Spencer, Optelian (USA)Vijay Jain, Verizon (USA)

Topics in Radio CommunicationsJoseph B. Evans, U. of Kansas (USA)

Zoran Zvonar, MediaTek (USA)Standards

Yoichi Maeda, NTT Adv. Tech. Corp. (Japan)Mostafa Hashem Sherif, AT&T (USA)

ColumnsBook Reviews

Andrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)History of Communications

Mischa Schwartz, Columbia U. (USA)Regulatory and Policy Issues

J. Scott Marcus, WIK (Germany)Jon M. Peha, Carnegie Mellon U. (USA)

Technology Leaders' ForumSteve Weinstein (USA)

Very Large ProjectsKen Young, Telcordia Technologies (USA)

Publications StaffJoseph Milizzo, Assistant Publisher

Eric Levine, Associate PublisherSusan Lange, Online Production ManagerJennifer Porcello, Publications Specialist

Catherine Kemelmacher, Associate Editor

2 IEEE Communications Magazine • February 2011

IEEE

M A G A Z I N EFebruary 2011, Vol. 49, No. 2

www.comsoc.org/~ci

SPECIAL SUPPLEMENT

ADVANCES IN PASSIVE OPTICAL NETWORKSGUEST EDITORS: MAHMOUD DANESHMAND, CHONGGANG WANG, AND WEI WEI

GUEST EDITORIAL

OPPORTUNITIES FOR NEXT-GENERATION OPTICAL ACCESSNext-generation optical access technologies and architectures are evaluated based on operators’ requirements. The study presented in this article compares different FTTH access network architectures.DIRK BREUER, FRANK GEILHARDT, RALF HÜLSERMANN, MARIO KIND, CHRISTOPH LANGE, THOMAS MONATH, AND ERIK WEIS

COST AND ENERGY CONSUMPTION ANALYSIS OF ADVANCED WDM-PONSThe authors compare several WDM-PON concepts, including hybrid WDM-PON with integrated per-wavelength multiple access, with regard to these parameters. They also show the impact and importance of generic next-generation bandwidth and reach requirements.KLAUS GROBE, MARKUS ROPPELT, ACHIM AUTENRIETH, JÖRG-PETER ELBERS, AND MICHAEL EISELT

TOWARD ENERGY-EFFICIENT 1G-EPON AND 10G-EPON WITH SLEEP-AWARE MACCONTROL AND SCHEDULINGThe authors briefly discuss the key features of 10G-EPON. Then, from the perspectiveof MAC-layer control and scheduling, they discuss challenges and possible solutions to put optical network units into low-power mode for energy saving. JINGJING ZHANG AND NIRWAN ANSARI

MULTIRATE AND MULTI-QUALITY-OF-SERVICE PASSIVE OPTICAL NETWORK BASEDON HYBRID WDM/OCDM SYSTEMThe authors present a new scheme to support multirate and multi-quality-of-servicetransmission in passive optical networks based on a hybrid wavelength-division multiplexing/optical code-division multiplexing scheme. The idea is to use multi-length variable-weight optical orthogonal codes as signature sequences of a hybrid WDM/OCDM system. HAMZEH BEYRANVAND AND JAWAD A. SALEHI

PASSIVE OPTICAL NETWORK MONITORING: CHALLENGES AND REQUIREMENTSThe authors address the required features of PON monitoring techniques and reviewthe major candidate technologies. They highlight some of the limitations of standardand adapted OTDR techniques as well as non-OTDR schemes. MOHAMMAD M. RAD, KERIM FOULI, HABIB A. FATHALLAH, LESLIE A. RUSCH, AND MARTIN MAIER

IMT-ADVANCED AND NEXT-GENERATION MOBILE NETWORKSGUEST EDITORS: WERNER MOHR, JOSE F. MONSERRAT, AFIF OSSEIRAN, AND MARC WERNER

GUEST EDITORIAL

EVOLUTION OF LTE TOWARD IMT-ADVANCEDThe authors provide a high-level overview of LTE Release 10, sometimes referred to as LTE-Advanced. First, a brief overview of the first release of LTE and some of its technology components is given, followed by a discussion on the IMT-Advanced requirements. The technology enhancements introduced to LTE in Release 10, carrier aggregation, improved multi-antenna support, relaying, and improved support for heterogeneous deployments, are described. STEFAN PARKVALL, ANDERS FURUSKÄR, AND ERIK DAHLMAN

ASSESSING 3GPP LTE-ADVANCED AS IMT-ADVANCED TECHNOLOGY: THEWINNER+ EVALUATION GROUP APPROACHThe authors describe the WINNER+ approach to performance evaluation of the 3GPP LTE-Advanced proposal as an IMT-Advanced technology candidate. The officialregistered WINNER+ Independent Evaluation Group evaluated this proposal againstITU-R requirements. The authors provide an overview of the ITU-R evaluation process, criteria, and scenarios, and focus on the working method of the evaluation group.KRYSTIAN SAFJAN, VALERIA D’AMICO, DANIEL BÜLTMANN, DAVID MARTIN-SACRISTAN, AHMED SAADANI, AND HENDRIK SCHÖNEICH®

S12S16

S25

S33

S39

S45

8284

92

LYT-TOC-FEB 1/20/11 12:03 PM

2011 Communications SocietyElected Officers

Byeong Gi Lee, PresidentVijay Bhargava, President-Elect

Mark Karol, VP–Technical ActivitiesKhaled B. Letaief, VP–Conferences

Sergio Benedetto, VP–Member RelationsLeonard Cimini, VP–Publications

Members-at-LargeClass of 2011

Robert Fish, Joseph EvansNelson Fonseca, Michele Zorzi

Class of 2012Stefano Bregni, V. Chan

Iwao Sasase, Sarah K. WilsonClass of 2013

Gerhard Fettweis, Stefano GalliRobert Shapiro, Moe Win

2011 IEEE OfficersMoshe Kam, President

Gordon W. Day, President-ElectRoger D. Pollard, Secretary

Harold L. Flescher, TreasurerPedro A. Ray, Past-President

E. James Prendergast, Executive DirectorNim Cheung, Director, Division III

IEEE COMMUNICATIONS MAGAZINE (ISSN 0163-6804) is published monthly by The Institute ofElectrical and Electronics Engineers, Inc.Headquarters address: IEEE, 3 Park Avenue, 17thFloor, New York, NY 10016-5997, USA; tel: +1-212-705-8900; http://www.comsoc.org/ci. Responsibility forthe contents rests upon authors of signed articles andnot the IEEE or its members. Unless otherwise speci-fied, the IEEE neither endorses nor sanctions any posi-tions or actions espoused in IEEE CommunicationsMagazine.

ANNUAL SUBSCRIPTION: $27 per year print subscrip-tion. $16 per year digital subscription. Non-member printsubscription: $400. Single copy price is $25.

EDITORIAL CORRESPONDENCE: Address to: Editor-in-Chief, Steve Gorshe, PMC-Sierra, Inc., 10565 S.W.Nimbus Avenue, Portland, OR 97223; tel: +(503) 431-7440, e-mail: [email protected].

COPYRIGHT AND REPRINT PERMISSIONS:Abstracting is permitted with credit to the source. Librariesare permitted to photocopy beyond the limits of U.S.Copyright law for private use of patrons: those post-1977articles that carry a code on the bottom of the first page pro-vided the per copy fee indicated in the code is paid throughthe Copyright Clearance Center, 222 Rosewood Drive,Danvers, MA 01923. For other copying, reprint, or republi-cation permission, write to Director, Publishing Services,at IEEE Headquarters. All rights reserved. Copyright © 2011by The Institute of Electrical and Electronics Engineers, Inc.

POSTMASTER: Send address changes to IEEECommunications Magazine, IEEE, 445 Hoes Lane,Piscataway, NJ 08855-1331. GST Registration No.125634188. Printed in USA. Periodicals postage paid at NewYork, NY and at additional mailing offices. Canadian PostInternational Publications Mail (Canadian Distribution)Sales Agreement No. 40030962. Return undeliverableCanadian addresses to: Frontier, PO Box 1051, 1031 HelenaStreet, Fort Eire, ON L2A 6C7

SUBSCRIPTIONS, orders, address changes — IEEEService Center, 445 Hoes Lane, Piscataway, NJ08855-1331, USA; tel: +1-732-981-0060; e-mail:[email protected].

ADVERTISING: Advertising is accepted at the dis-cretion of the publisher. Address correspondence to:Advertising Manager, IEEE Communications Magazine,3 Park Avenue, 17th Floor, New York, NY 10016.

SUBMISSIONS: The magazine welcomes tutorial orsurvey articles that span the breadth of communica-tions. Submissions will normally be approximately 4500words, with few mathematical formulas, accompaniedby up to six figures and/or tables, with up to 10 careful-ly selected references. Electronic submissions are pre-ferred, and should be sumitted through ManuscriptCentral http://mc.manuscriptcentral.com/commag-ieee.Instructions can be found at the following: http://dl.com-soc.org/livepubs/ci1/info/sub_guidelines.html. For furtherinformation contact Sean Moore, Associate Editor-in-Chief ([email protected]). All submissions will bepeer reviewed.


COORDINATED MULTIPOINT: CONCEPTS, PERFORMANCE, AND FIELD TRIAL RESULTSCoordinated multipoint or cooperative MIMO is one of the promising concepts to improve cell edge user data rate and spectral efficiency beyond what is possible with MIMO-OFDM in the first versions of LTE or WiMAX. Interference can be exploited or mitigated by cooperation between sectors or different sites. Significantgains can be shown for both the uplink and downlink.RALF IRMER, HEINZ DROSTE, PATRICK MARSCH, MICHAEL GRIEGER, GERHARD FETTWEIS, STEFAN BRUECK, HANS-PETER MAYER, LARS THIELE, AND VOLKER JUNGNICKEL

EVOLUTION OF UPLINK MIMO FOR LTE-ADVANCEDThe evolution of LTE uplink transmission toward MIMO has recently been agreed in 3GPP, including the support of up to four-layer transmission using precoded spatial multiplexing as well as transmit diversity techniques. The authors provide an overview of these uplink MIMO schemes, along with their impact on reference signals and DL control signaling. CHESTER SUNGCHUNG PARK, Y.-P. ERIC WANG, GEORGE JÖNGREN, AND DAVID HAMMARWALL

A 25 GB/S(/KM2) URBAN WIRELESS NETWORK BEYOND IMT-ADVANCEDThe authors present a survey on the technical challenges of future radio access networks beyond LTE-Advanced, which could offer very high average area throughputto support a huge demand for data traffic and high user density with energy-efficient operation. They highlight various potential enabling technologies and architectures to support the aggressive goal of average area throughput 25 Gb/s/km2

in beyond IMT-Advanced systems. SHENG LIU, JIANJUN WU, CHUNG HA KOH, AND VINCENT K. N. LAU

SYNCHRONIZATION OVER ETHERNET AND IP INNEXT-GENERATION NETWORKS

GUEST EDITORS: STEFANO BREGNI AND RAVI SUBRAHMANYAN

GUEST EDITORIAL

EVOLUTION OF THE STANDARDS FOR PACKET NETWORK SYNCHRONIZATIONThe authors summarize the work done by ITU-T Q13/15 over the last six years to standardize the transport of timing over packet networks. They provide a summary of the published documents in this area from ITU-T while providing some of the background that went into each document including the specification of synchronous Ethernet and IEEE 1588 telecom profiles. JEAN-LOUP FERRANT AND STEFANO RUFFINI

SYNCHRONIZATION OF AUDIO/VIDEO BRIDGING NETWORKS USING IEEE 802.1ASThe Audio/Video Bridging project in the IEEE 802.1 working group is focused on thetransport of time-sensitive traffic over IEEE 802 bridged networks. Current bridged networks do not have mechanisms that enable meeting these requirements under general traffic conditions. IEEE 802.1AS is the AVB standard that will specify requirements to allow for transport of precise timing and synchronization in AVB networks. GEOFFREY M. GARNER AND HYUNSURK (ERIC) RYU

NGN PACKET NETWORK SYNCHRONIZATION MEASUREMENT AND ANALYSISAs the transport of data across the network relies increasingly on Ethernet/IP methods and less on the TDM infrastructure, the need for packet methods of synchronization transport arises. Evaluation of these new packet methods of frequency and time transport requires new approaches to timing measurement and analysis. LEE COSART

PERFORMANCE ASPECTS OF TIMING IN NEXT-GENERATION NETWORKSCircuit-switched networks based on time-division multiplexing require synchronizationto deliver information, whereas packet-switched networks can deliver information in an asynchronous environment. However, all real-time services require that synchronization and timing information be delivered over the network. Performanceof timing distribution is quantified using particular metrics and adherence to requirements determined by using masks. KISHAN SHENOI

USING IEEE 1588 AND BOUNDARY CLOCKS FOR CLOCK SYNCHRONIZATION INTELECOM NETWORKSThe authors describe the use of IEEE 1588 and boundary clocks for clock distributionin telecom networks. The technology is primarily used to serve the radio interface synchronization requirements of mobile systems such as WiMAX and LTE, and to reduce the deployment and dependence of GPS systems in base stations. MICHEL OUELLETTE, KUIWEN JI, SONG LIU, AND HAN LI

102

112

122

164

130132

140

148

156

President’s Page 6Letters to the Editor 12Certification Corner 14New Products 16

Conference Calendar 17Product Spotlights 20Global Communications Newsletter 21Advertisers’ Index 176

LYT-TOC-FEB 1/20/11 12:03 PM Page 4

IEEE Communications Magazine • February 20116

THE PRESIDENT’S PAGE

COMSOC MARKETING: VALUED OFFERINGS FOR VALUED CUSTOMERS

arketing, in general, is a term thatwe often use, but are not sure of the

exact definition. For the purposes of this arti-cle, we cite the definition provided by theAmerican Marketing Association.1

Marketing is the activity, set of institutions,and processes for creating, communicating,delivering, and exchanging offerings that havevalue for customers, clients, partners, andsociety at large.

Key to this definition is the concept that“offerings” must have value for the customer.Therefore, determining what products andservices customers find of value is a crucialaspect of marketing. Our members are “cus-tomers,” with the Society’s marketing effortshaving primary focus on satisfyingtheir needs while meeting thebroader goals of their Society. Sinceany organization must operate with-in the boundaries of its missionand/or goals, we revisit the goals ofthe IEEE Communications Society(ComSoc) first, which are two-fold:

•Scholarly — scientific anddirected toward the advancement ofthe theory, practice, and applicationof communications engineering andrelated arts and sciences.

•Professional — promoting highprofessional standards, develop-ment of competency and theadvancement of the standings ofmembers of the professions.

These goals are implemented through a diverse set ofactivities by members and the business actions managed byvolunteer leaders and paid staff. Our main business includespublishing journals and magazines, holding meetings and con-ferences, offering education and training, and selling adver-tisements. Our business is to serve ComSoc’s customers,including members, and to achieve the goals of the Society.

Next, we need to understand the background of ComSoc’scustomers, which include members, publication subscribers,conference attendees, and recipients of other ComSoc ser-vices. Most current members have advanced educations: post-graduate degrees in EE, physics, mathematics, computersciences, business, or related fields.

Serving ComSoc’s “customers” is the mission of ComSoc’sMarketing and Creative Services Department, which promotesall ComSoc products and provides creative services to Com-Soc officers, volunteers, and all other departments of Com-Soc. The department completes over 300 marketing projectsper year in order to refine and renew ComSoc’s offerings.ComSoc offerings can be grouped into four major areas:membership, publications, conferences, and education &training.

In this article we will describe the marketing of ComSoc interms of ComSoc offerings, marketing process, and the issues

and challenges that must be addressed inorder to best serve ComSoc customers. Ishare this issue with Stan Moyer, ComSocDirector of Marketing and Industry Rela-tions, and John Pape, (staff) Director of Mar-keting and Creative Services.

Stan Moyer is an executive director andstrategic research program manager in theApplied Research area of Telcordia Tech-nologies, where he has worked since 1990.Currently, he is leading a business develop-ment effort for end-user information privacyprotection for mobile services. In the past heled research and business development activi-ties related to digital content services andhome networking. He has also worked on

ATM switch hardware, broadbandnetwork architectures and proto-cols, middleware, Internet networkand application security, InternetQoS, and voice over IP. He is cur-rently President of the OSGiAlliance. He served as a member ofthe IEEE Technical ActivitiesBoard Finance Committee. WithinComSoc, he is currently serving asDirector-Marketing and IndustryRelations, a member of the Com-Soc Standards Board, Vice-Chair ofthe IEEE CCNC steering commit-tee, and co-Chair of the Ad hocIndustry Promotion Committee.Stan has a ME degree in Electrical

Engineering from Stevens Institute of Technology and anMBA degree in Technology Management from the Universityof Phoenix.

John Pape has served as the (staff) Director of Marketingand Creative Services since 1997. His responsibilities includeplanning and implementing the society’s marketing activitiesfor membership, publications, continuing education, and con-ferences. During his tenure, products have migrated fromprint to electronic media, and marketing tactics have evolvedfrom direct mail and manual processing to complicated e-mailcampaigns and social media outreach. Recently, he has ledComSoc’s efforts to provide members with a digital option forIEEE Communications Magazine and to create and executethe plan to offer a virtual course in wireless communicationsengineering. From 1989 to 1997, he managed the PublicationsMarketing Department of the American Society of Civil Engi-neers. He has managed marketing activities for more than 30years with international publishers including S. Karger Pub-lishers, Methuen, and Springer-Verlag.

COMPETITORS AND ADDRESSABLE MARKETComSoc, just like most businesses, has competitors for its

products and services. ComSoc competes for time, prestige,money, authors, attendees, readers, volunteers, subscribers,and resources with other organizations, events, publishers, andinformation sources in the communications field. Some mightconsider communications websites or self-organized socialmedia as directly competitive as they can provide alternative

M

BYEONG GI LEE

STAN MOYER JOHN PAPE

1 American Marketing Association website, http://www.marketingpow-er.com/aboutama/pages/definitionofmarketing.aspx

LYT-PRES PAGE-FEB 1/20/11 3:46 PM Page 6

IEEE Communications Magazine • February 2011 7


resources for members and potential members. Some knowncompetitors, in that context, include trade or corporate-cen-tric organizations such as the ITU, GSMA, IET, ACM, TIA,ATIS, CTIA, AFCEA, OSA, ISOC, and other national orga-nizations. Publishers such as John Wiley and Sons, Springer-Verlag, CMP, Cambridge University Press, and Elsevierprovide alternative global publications (books, journals, maga-zines) as information sources and legitimate venues for schol-arly authors and practical technical publishing. Certain forumsand/or special interest groups (SIGs) deal with rapid techno-logical developments, such as the WiMax Forums, Telecom-munications Management Forum, NGN, IMS, and the FemtoForum. Trade shows and technical conferences (sponsored bycorporate entities such as the Yankee Group or non-profitssuch as PCIA) on communications topics can be foundthroughout the year at locations around the globe. Within theIEEE, other societies such as Antennas and Propagation, Sig-nal Processing, Vehicular Technology, Information Theory,Computer, Photonics, Consumer Electronics, and MicrowaveTheory and Techniques all have some technical overlap withComSoc. However, ComSoc does not have a broad-baseddirect global competitor serving individuals within the globalcommunity in its technical scope.

When marketing products and services, it helps to under-stand the addressable market – that is, the entire space forwhich our products and services would be of interest. Definingthe estimated universe of potential members or communica-tions related subject matter experts can be a challenge. Basedon published data, there have been about 800,000 EE BSdegrees granted in the US in the past 40 years. Less than halfof them attained a Master’s degree or Ph.D. level degree. His-torically about 14% enter communications-centric employ-ment, resulting in about 110,000 individuals with anundergraduate degree and about 50,000 individuals holding anEE Master’s degree or higher in the US. With 20,000 mem-bers in the US, ComSoc member demographics imply thatComSoc has captured about 25% of the potential US marketholding Master’s degrees or higher. Of those earning a Bache-lor’s Degree, data would suggest that ComSoc has capturedabout 10% of the potential US universe. The potential mem-ber universe could include other disciplines such as physics,mathematics, computer science, and business management,but all EE graduates do not pursues careers in EE fields.Data sources for the international higher education area andemployment markets are unreliable and inconsistent; it is notpossible to estimate realistically the global member universe,although someone may estimate the size to be double the US.

The US Department of Labor maintains employment datafor the telecom industry for all employees (including non-degreed employees). Employment in the traditional telecom-munications industry has declined by 35% since reachinghighs of more than 1.4 million in 2000; there have been otherareas of employment growth. While telecom employers havebeen shedding traditional full-time employees, technical con-sulting and scientific research employment have increased by50% in the last decade. As a result, the US membership mar-ket is not as apparent or easily accessible as it once was. Com-munications specialists can be found in a much broader arrayof companies and working scenarios. Figure 1, drawn basedon the US Bureau of Labor statistics charts, represents thisshift graphically.

MEMBERSHIPOne of the primary “products” that ComSoc provides to

“customers” is membership in the society. As a society of theIEEE, ComSoc membership is offered for a fee in addition toIEEE membership dues. ComSoc sets the price and definesthe benefits of society membership. The prime justificationsfor membership are to maintain technical competence, receivedesired services at discounted rates, and network with col-leagues. Each member receives monthly issues of IEEE Com-munications Magazine, the leading technical periodicaldevoted to communications technologies on an advancedlevel. ComSoc membership dues are $25 in 2011, with digitaldelivery of IEEE Communications Magazine; in 1999 ComSocmembership dues were $23 with print delivery of IEEE Com-munications Magazine. From the end of February to the endof September each year, dues for new members are half theregular price.

ComSoc membership was about 8,800 when the Societywas founded in 1963 with the establishment of IEEE. Mem-bership has increased rapidly over the past 15 years, whichwas influenced by the half-year free membership campaignstarted in 1998 and the technology bubble of the late 1990’s.However, it began to decline from the early 2000’s due to therapid decline of traditional telecommunications employmentand full implementation of IEEE Xplore, which resulted inonline IEEE (and ComSoc) content availability. The member-ship decline stopped at the 40,000+ level in the late 2000’sand maintained that level until it began to increase in 2010 toabout 48,000.

The majority of ComSoc members reside outside the US,whereas the majority of IEEE members reside in the US. Thisis the result of several phenomena. The communicationsindustry diffused globally and became successful in manyAsian and European countries. IEEE Xplore sales penetra-tion in the US market has resulted in many potential USmembers satisfying their need for technical publications with-out joining ComSoc. In the late 1990’s, our surveys indicatedthat more than 62% of members were industry-employed; in2008 that number decreased to 45%. This indicates that anymembership growth in the future is most likely to come withthe support of industry, consulting, or government areas.

Each year ComSoc recruits about 11,000 new members. Themajority of these new members result from offering to newIEEE members a free ComSoc membership for the first year.Some new members join during the annual IEEE renewal pro-cess. To recruit new members, marketing executes variousprint, online, and e-mail direct response campaigns; trade showexhibits; free book premiums; conference registration offers;and monthly new IEEE member e-campaigns. In addition,membership recruiting is supported by extensive web pageupdating; distributing promotion material to Sister Societies;

1/00

600

400

800

1000

US

empl

oym

ent

tren

ds (

1000

s) 1200

1400

1/99 1/01 1/02 1/03 1/04 1/05 1/06 1/07 1/08 1/09

TelecommunicationsManagement and technical consultingScientific R&D services

FIGURE 1. U.S. employment trends.


local Chapter support with promotion material, sample copies,posters and special offers; cover wraps; free offers of a briefcommunications history book highlighting Society notables; anup-to-date Society PowerPoint presentation for group presenta-tions; back-office coordination/support for data and othermembership development activities such as Chapter Chair Con-gresses, Distinguished Lecturer Tours, and volunteer visits; Bestof the Best and other book/DVD/product special offers; andspecial opportunistic efforts. In recent years, the Industry Nowprogram has been established to offer bulk or multiple mem-berships to companies in emerging economies that are notacquainted with the advantages of ComSoc association.

Retaining members is a constant process of providingreminders and opportunities to members that illustrate thevalue of membership. The annual ComSoc Community Direc-tory and letter from the President are sent to each new Com-Soc member on a biweekly basis. Members are surveyed forsatisfaction and needs. Articles and columns dealing withmember issues appear every month in IEEE CommunicationsMagazine, as do advertisements specifically aimed at members.Monthly issues of e-News spotlight the President’s monthlymessage and present member-only special offers, includingconference registrations, the Book of the Month, free tutori-als, technically sponsored Webinars, new product offers, andother useful information such as the Top Ten list of ComSocpapers appearing in IEEE Xplore, and content announce-ments for optional publications. Support for volunteer com-mittees, Chapters, Distinguished Lecturer Tours, premiums,and other programs also contribute to retention efforts.

PUBLICATIONSComSoc publishes magazines, journals/transactions, pro-

ceedings, DVDs, books (with IEEE’s publisher John Wileyand independently) and newsletters. There are several meth-ods to measure the popularity and effectiveness of periodicalpublications, e.g., subscription data, electronic PDF down-loads, submissions, the ISI Journal Citation report, and read-er/member surveys. Most periodicals are available in print orelectronic format.

There are three general categories for the subscriptionmarket: subscription agents; libraries; and individuals. Com-Soc relies heavily on IEEE Sales and Marketing for sales tolibraries and to subscription agents; all electronic packagesales, including consortia licensing sales, are handled by theIEEE. The IEEE offers several packages of periodicals. TheAll Societies Periodical Package (ASPP), Enterprise, theIEEE Electronic Library (IEL), and the new IEEE Communi-cations Library are among the offerings that include ComSocperiodicals and conference proceedings. IEEE participates inlibrary conferences such as the ALA and SLA annual tradeshows. All ComSoc periodicals are included in the annualSociety brochure, on Society Membership applications, onlineweb and PDF formats, and in the ComSoc Community direc-tory. IEEE Communications Magazine (CommMag) is themost important Society publication, which all membersreceive monthly. The editorial data reflects hot topics of inter-est to members and is written in a style to be accessible to allmembers, with academic and corporate interests alike. Comm-Mag is a hybrid publication, containing editorial material thatcan be described as scholarly with sufficient industry attrac-tion to generate $1 million plus in advertising sales each year.No other IEEE society magazine can claim the distinction ofmost non-member subscriptions, most 2009 magazine Xploreviews, high ISI Journal Citation Report impact factor (rated#5 in telecommunications in 2009), and $1+ million generatedin advertising revenue. CommMag is the most significant schol-arly/industry publication in the specialty of communications. Toemphasize the uniqueness of ComSoc members, a survey ofreaders found that no competitive trade publication was readby more than 25% of CommMag readers (see Table 1).

IEEE Wireless Communications Magazine and IEEE Net-work Magazine complement IEEE Communications Magazineand focus on two strong areas of communications technology.These are optional bi-monthly publications with technical co-sponsorship with the Computer Society. Both are similar toCommMag in layout and offer a broader readership concept.

Core ComSoc archival journals/transactions includingIEEE Transactions on Communications, IEEE Journal onSelected Areas in Communications, IEEE Communications Let-ters, IEEE Communications Surveys and Tutorials (e-only), andIEEE Transactions on Network and Service Management (e-only) reflect the scope of the scholarly activity. And financiallyco-sponsored journals such as IEEE Transactions on WirelessCommunications, IEEE/ACM Transactions on Networking,IEEE/OSA Journal of Lightwave Technology, IEEE/OSA Jour-nal of Optical Communications and Networking, and IEEETransactions on Mobile Computing demonstrate relationshipswith other related specialties.

CONFERENCESAll conferences sponsored and co-sponsored by ComSoc

are marketed and promoted by mixed media through differentchannels. The degree of marketing effort increases as thelevel of financial ownership and budget increases. For a spe-cific event, a separate marketing plan or strategy is created.There are some common areas. Most of these events includeconference proceedings with papers also appearing online inIEEE Xplore. This results in authors/presenters wanting tosubmit papers and contribute to the conferences and ComSocrevenue. Conferences generate more gross revenue than anyother product in the portfolio, and conferences commandmore marketing resources than any other products in theComSoc portfolio.

Conferences have different levels of financial co-sponsor-



(Continued on page 10)

Publication % of CommMag Readersthat Read

EE Times 23.50%

Telecommunications 20.60%

EDN 15.90%

Electronic Design 15.10%

Wireless Design & Development 13.60%

Microwave & RF 13.20%

Network World 13.10%

Microwave Journal 12.60%

Lightwave 8.90%

Business Communications Review 7.00%

Test & Measurement World 6.50%

Internet Telephony 5.60%

Photonics Spectra 4.90%

Telephony/ Connected Planet 4.10%

Urgent Communications 2.70%

TABLE 1. Percentage of CommMag readers that read other maga-zines.


10



ship that range from fully financially sponsored by ComSoc(e.g., ICC, GLOBECOM, CCNC), financially co-sponsored byComSoc (e.g., OFC/NFOEC, MILCOM), and conferenceswith no financial sponsorship (IEEE Sarnoff 2009, WTS 2009,IEEE Policy 2009). Those conferences are typically technicallyco-sponsored by ComSoc.

Depending on budgets, most fully-owned ComSoc confer-ence marketing includes web site development, Call-for-Papers (CFP) assistance, advance and final program designand production, online and media advertising, flyers, and aseries of e-mail efforts. Recently, there have been additionalactivities such as recording live sessions and other programevents. Under the ComSoc Webcasts brand, access to theseevents can be purchased for live participation or recorded lis-tening. Often, a keynote speech can be enjoyed free of charge.

At some recent flagship events, there have been increasedvisibility activities with press releases and local, national, andglobal media coverage. Some ComSoc volunteer leaders havemade trips to local high schools and colleges to explain thebenefits of careers in the communications sciences. At arecent MILCOM, high school science students visited theexhibit floor to sample the experience and view the scope ofthe industry.

Social media – e.g., blogs, Facebook, Linkedin, Twitter –are now playing a larger role in conference marketing. Com-Soc has more than 35 social media sites dedicated to confer-ences! Social media also provides new ways for registrants toparticipate.

Marketing efforts related to conferences not only help topromote the conference, but also utilize the conference tomarket other ComSoc products and services. Conferencesserve as a great forum for the ComSoc marketing staff andvolunteers to meet ComSoc members and conference atten-dees to get feedback and input on what they do and do notlike about ComSoc, the conference, and other products andservices.

EDUCATION AND TRAININGContinuing education and/or educational products/product

development are important but least developed areas withinComSoc. In reality, everything ComSoc produces falls underthe subject of education, and most member surveys indicatesupport for additional educational opportunities. In response,we plan to invest intensive efforts to fully develop the educa-tion and training areas in accordance with the progress of themobile converged communications era.

Conference Tutorials are developed with ComSoc events,but they are developed under the banner of the event itself, somarketing for these tutorials falls under the domain of theindividual event.

Tutorials Now represent an online portfolio of individualhalf-day or full-day tutorials that had been given at ComSocevents. The total number of titles accumulated so far is 84,but it grows every year. Selected presenters record voice overslide presentations after the event and forward the completedfiles to ComSoc for quality control testing and uploading. Pre-senters receive an honorarium and/or royalties for acceptedtutorials. Recently these tutorials have been indexed foraccess through the ComSoc Digital Library. Some TutorialNow modules are offered as potential sponsorship to compa-nies. These can be offered to ComSoc members free of chargefor a limited time when a sponsor has been secured.

Courses/sessions can be developed for non-ComSoc con-ferences and trade shows. In 2009 and 2010, a course on

Wireless Communications (specifically the areas covered bythe WCET certification program) was held at 4G World inChicago. A five-day virtual intensive course on Wireless Com-munications was held in September 2010. This first offeringwas very successful, with 75 registrants from 15 countries.Each day’s sessions were held over the Internet; participantsnever had to leave their computers. For 2011, the five-day vir-tual intensive course on wireless communications is scheduledfor multiple offering times. Eventually this event could be set-tled as a quarterly or bimonthly course, demand permitting.

ComSoc’s WCET (Wireless Communication EngineeringTechnologies) Certification Program was officially launched inearly 2008. It was developed by ComSoc and an internationalcollection of industry experts to address the worldwide wire-less industry’s growing need for professionals with real-worldproblem-solving skills. Industry consensus, obtained throughan industry survey with more than 1,300 individual responsesrepresenting more than 65 countries from around the world,was that a wireless certification program is necessary. Thepurpose of the WCET is to certify individuals in wireless com-munications. Two testing windows are offered each year, dur-ing which individuals can sit for the exam. There are morethan 500 testing sites located in 75 countries around theworld. The exam is administered at a Computer Based Test-ing facility and consists of 150 multiple-choice questionsencompassing seven major wireless areas: RF engineering;propagation and antennas; access technologies; network andservice architecture; network management and security; facili-ties infrastructure; agreements, standards, policies, and regula-tions; and fundamental knowledge.

The WCET program has evolved for the past two yearswhile passing through a learning curve. Marketing strategy forthe WCET exam originally started with the objective to gener-ate direct individual applications, but the focus was changedfrom the individual to the company and from the exam to train-ing. The five-day virtual intensive course thus developed hasproved popular with industry and provides a natural sequencefor those wishing to successfully navigate the WCET exam.

IN CLOSINGWith a membership of more than 48,000 global individuals,

ComSoc is the second largest IEEE society and has strong vol-unteer commitments, a dedicated staff, expert operational sup-port, and a global reputation of excellence. ComSoc excels atproducing technical publications, organizing technical confer-ences, as well as fostering educational programs. ComSoc hasa potential to grow much more in the future while undergoingtransformation toward the converged communications era.

ComSoc’s marketing has adapted to many changes in thepast decade, thus enabling ComSoc to reach its current sta-tus. Anticipating future opportunities, the ComSoc volun-teer/staff partnership for marketing will support increasedindustry patronage, new partnerships with organizations andcompanies that can help enhance ComSoc’s position as the“go-to” resource for the communications industry. Further, itwill help promote new publications and conferences inemerging communications areas, develop new services gearedto industry, attract non-US members, expand digital deliveryof information and virtual meetings, explore social mediaopportunities, and prepare for the unexpected. ComSoc’smarketing will keep playing a pivotal role while ComSoc nav-igates through the newly emerging converged communica-tions era, creating new offerings that have value for theComSoc’s customers and thereby making ComSoc a valuablehome for all the communications communities and profes-sionals of the world.

(Continued from page 8)



LETTERS TO THE EDITOREDITED BY MISCHA SCHWARTZ

Comments on “An Early History of theInternet”by Leonard Kleinrock

Greg Adamson, MelbourneTo the Editor,I read your “History of Communica-

tions” pages with interest, and particu-larly the August 2010 article by LeonardKleinrock. This article creates a chal-lenge for the reader: how to weigh anaccount of historical events by a majorparticipant in those events. I have sepa-rately seen the problem described as“military history written by generals.”

You would be aware that DonaldDavies in part covered the same groundin an article published in 2001 (TheComputer Journal, vol. 44, no. 3, pp.152–62). I found Davies’ account verymoving: a renowned researcher in hisdying months establishing his view of acontested period of discovery (and notto assert his own claim). After review-ing the work of Kleinrock and PaulBaran, he summed up his finding in thefollowing way:

“My contention is that the work ofKleinrock before and up to 1964 giveshim no claim to have originated packetswitching, the honour for which mustgo to Paul Baran. The passage in hisbook on time-sharing queue discipline,if pursued to a conclusion, might haveled him to packet switching, but it didnot.”

I appreciate that Leonard Kleinrockwould not agree with this perspective,yet I feel his article is too oblique. Iwould be very interested in seeing a

more specific response to the pointsthat Davies made.

Perhaps you should have an occa-sional column titled “Debates in theHistory of Communications.”

Response to Greg Adamson:by Leonard Kleinrock

In my August 2010 article, I statethat the detail I afford to my perspec-tive is based on personal experienceand is not a claim to importance. I alsocall attention to many more histories ofthe Internet in need of study; Adamsoncalls this whole enterprise “military his-tory written by generals” and has askedme to respond to claims he cites fromthe late Donald Davies in which mywork on time-sharing is addressed.

A major goal of packetization is toprevent long messages from hoggingthe channel and thereby causing shortermessages to wait inordinately. This wasraised by Davies as a major concernsince it provided the network operatorthe ability to control network delayrather than being at the mercy of theend user. In my 1962 dissertation Iclearly considered the role of messagepriority classes and priority queueingdisciplines in accomplishing this. Relat-ing this to network delay, I devotedChapter 5 to studying the “...manner inwhich message delay is affected whenone introduces a priority structure (orqueue discipline) into the set of mes-sages....” I isolated the effect of queuediscipline by looking at a single node,as “An understanding of the effects of a

priority discipline at the single-nodelevel is necessary before one can makeany intelligent statements about themultinode case.” Among the classes ofdiscipline I studied were the preemptivedisciplines where the transmission of amessage can be interrupted and thencontinued later. I devoted an entire sec-tion to time-shared servicing of datatraffic in which I broke messages intosmaller, fixed size pieces. I also provid-ed a mathematical analysis and showedthat the deleterious effects of channelhogging were indeed avoided.

It would take far more space todelineate the properties of packetswitching, since it involves much morethan just chopping messages into small-er fixed length segments (the issueaddressed by Adamson to which I haveherein responded). Briefly, it alsoinvolves network efficiency which Iaddressed with the broader issue ofdemand access; it involves robustnessand reliability which comes about fromdistributed adaptive routing in a meshnetwork, which I also presented in myearly work.

In the August 2010 article, I explainhow my work informed the technologyof the ARPANET as well as its timingrelative to that of Baran’s. Prior to thewriting of Davies’ posthumous article,he contacted me about these topics,and I did respond accordingly. Dr.Adamson has raised these same issues,which I have already explained in myarticle.

LYT-LETTERS EDITOR-February 1/21/11 10:07 AM Page 12

14

In many of the months this column hasappeared, it has often included somewords about the volunteers who havemade WCET certification possible. It’sworth taking more time to describe themany roles that volunteers have played –and continue to play – because of thevalue they bring to the program.

It started when the original PracticeAnalysis Task Force (PATF) convened inDecember 2006 to develop the Delineation(the description of wireless practice andunderlying knowledge). The PATF includ-ed more than 15 industry experts fromaround the world who gave their time toget the program off to a strong start. With-out their commitment, there would not bea WCET certification program. Several ofthese volunteers remain actively involvedtoday, especially in championing the valueof certification within the industry.

The next volunteers were the partici-pants in Focus Groups and the Indepen-dent Reviewers. Dozens of industry expertsstudied the Delineation and offered con-structive feedback that helped make it

even more representative of wireless com-munications practice in a broad range ofcompanies and countries. Again, some ofthose volunteers remained active in theWCET program for years, taking on specif-ic roles and responsibilities, and their con-tinued involvement has been valuable inguiding and growing the program.

Special thanks go to the volunteers whohave served on the Industry AdvisoryBoard. They have given focused feedbackon all aspects of the WCET program, rang-ing from promoting the program to indus-try to critiquing the WCET website. Thefact that Board members represent compa-nies in all segments of the wireless commu-

nications industry, based in countriesaround the world, has helped maintain thevendor neutrality and trans-national natureof WCET certification.

Other volunteers have served as ques-tion writers, question reviewers, and on theexam committee that has used the bestquestions to create exams covering thebreadth of wireless communications. Theyhave maintained the balance among theseven technical areas that was developed bythe PATF and reinforced by industry feed-back as to the relative importance of vari-ous tasks and knowledge in the workplace.

Some of these volunteers recently joinedwith others to form a “mini-PATF” toreview all the detailed feedback on theDelineation that we have received. Theyinvested their time and effort to refresh andupdate the Delineation to reflect changes inthe industry over the past few years.

A Core Team of volunteers has providedsteady leadership to the program throughoutits history. They have led committees thatlooked at issues of policy, marketing, theHandbook, strategy, training, the WEBOK,recertification, and question writing andexam creation. A couple dozen volunteershave made a significant commitment tothese leadership positions over the years.

The Steering Committee is responsiblefor the long-term direction of the WCETprogram. More than a dozen volunteershave served on this committee, helping toidentify strengths and weaknesses and alsoareas where ComSoc can build on the cer-tification program. An example of the lat-ter is the development of ComSoc trainingofferings, an area where there was a cleardemand in the industry and a path withinComSoc to address the need.

The title of this month’s column sumsit up: the hundreds of volunteers who haveplayed many different roles in the devel-opment, growth, guidance, and success ofWCET certification have been our mostvaluable asset. We have expressed ourthanks to each as they have relinquished arole or responsibility, but we owe onelarge THANK YOU to all of them.

Attention WCP certificate holders inparticular: volunteer opportunitiesabound! Please let us know how you wouldlike to contribute to continuing to growthe WCET program to its full potential.


CERTIFICATION CORNER

THE VALUE OF VOLUNTEERSBY ROLF FRANTZ

IENYCM2830.indd 1 1/17/11 3:31:21 PM

LYT-CERTIFICATION-FEB 1/19/11 4:25 PM Page 14


CHIP FAMILY SUPPORTS THE ITU-TG.HN GLOBAL STANDARD

Lantiq

Lantiq has introduced a chip familysupporting the ITU-T G.hn global stan-dard for next generation wired homenetworks. Lantiq XWAY HNX devicesprovide manufacturers of consumer,computing and smart home electronicswith the foundation for in-home net-works that can be connected using anycombination of phone, power and cablewiring.

Endorsed by the 191 member coun-tries of the ITU in June 2010, the G.hnstandard defines technology to providenetwork connectivity across all commonin-home wiring with data rates as highas 1 Gigabit per second (Gbps). AsG.hn becomes an integral feature inresidential gateways, consumer elec-tronics devices, personal computers andInternet-connected smart home devices,service providers will be able to realizesignificantly reduced installation andoperations costs as a result of plug-and-play network operation and greaterdevice connectivity.

Lantiq XWAY HNX chips can beused in standalone G.hn node applica-tions or as part of multi-service plat-forms. The device is provided tocustomers with a software package thatincludes pre-integrated drivers for thebroad range of Lantiq system-level sili-con devices, including Gigabit speedgateway processors, 802.11n WLANsupporting carrier-grade video,DECT/CAT-iq, VoIP and analog voice.

http://www.lantiq.com/hnx

LOW NOISE, HIGH-OUTPUT XPONVIDEO RECEIVER

RF Micro Devices

RFMD’s new RFRX8888 videoreceiver performs transimpedanceamplification of the differential inputfrom a high performance 1550nm opti-cal wavelength photo detector (PD),all with best-in-class noise perfor-mance. This IC's output is linear lowdistortion RF from 48 MHz to 1002MHz. RFRX8888 is ideal for 1550nmoptical wavelength RF analog or digi-tal overlay video receive circuitryemployed in xPON FTTP ONT triplex-er and quadplexer modules. Its firststage features integrated bias circuitrysimplifying external-to-IC end productdesign and lowering overall end prod-uct assembly cost. Optimized for oper-ation from a +12 VDC power supplywith a highly efficient power consump-tion of just 1.4 W, it eliminates theneed for a supplemental ONT power

supply in most xPON applications.RFRX8888’s ultra-low noise perfor-mance, combined with high outputpower, extends the performance andlifetime of wired networks by improv-ing the link margin and/or allowingmore passive optical splits.

For FTTP applications requiring +5VDC power supply operation, theRFRX8890 video receiver is also avail-able. Features include:

•+12V Single Supply Operation•On-Die Bias Circuitry Reduces

Cost and Board Area•Best-in-Class Low Noise (<3.0

pA/rtHz Equivalent Input Noise Cur-rent)

•Low Power: 1.4 W at +12V•Best-in-Class +23 dBmV per Chan-

nel RF Output Capability•Linearity Better Than -63 dBc CSO

and -66 dBc CTB at +23 dBmV RFOut Per Channel (79-NTSC EquivalentChannels)

•48 MHz to 1002 MHz OperationalBandwidth

•30 dB AGC Rangehttp://www.rfmd.com

SG384 RF SIGNAL GENERATOR

Stanford Research Systems, Inc.

Introducing the SG384, a 4 GHzRF Signal Generator from SRS. Itoffers a DC to 4 GHz frequency rangewith 1 μHz resolution, AM, FM, andPM, with -116 dBc/Hz phase noise at20 kHz offset from 1 GHz, full octavefrequency sweeps, an OCXO timebaseand standard RS232, GPIB and Ether-net interfaces. Options include clockoutputs, analog I/Q inputs and a rubid-ium timebase.

http://www.thinkSRS.com

DIGITAL I/Q DATA RECORDER

Rohde & Schwarz

Rohde & Schwarz introduuced theR&S IQR digital I/Q data recorder atelectronica in Munich. The recordercan record, store and replay digital RFsignals loss-free and in realtime overthe I/Q interface developed by Rohde& Schwarz. When used in combinationwith RF scanners, generators and net-work analyzers from Rohde & Schwarz,the recorder completes a high-perfor-mance, continuous analysis system fordigital RF signals. This system willprove to be of particular benefit forusers in broadcasting, mobile radio,aerospace & defense, and the automo-bile industry. The compact recorder –in half 119-inch format – currentlyoffers transmission rates of up to 66

Msample per second. It comes with arobust, interchangeable solid-statedrive with one Tbyte of storage capaci-ty and a recording rate of 270 Mbyteper second.

The R&S IQR I/Q data recorderrecords digital RF signals in realtime.Thanks to its unique combination ofspeed, compactness and robustness, itis ideal for use in drive tests in broad-casting and mobile radio networks.For instrument tests or electronic com-ponent testing, the recorder can beused to supply previously generatedtest signals. In addition, broadbandspectra or sporadic signals can berecorded in realtime for later offlineanalysis.

To obtain a continuous analysis sys-tem for digital RF signals, the user canconnect the data recorder with either aspectrum or radio network analyzerand with a signal generator from Rohde& Schwarz via the digital I/Q interface.This I/Q interface simplifies bothparameter exchange between theinstruments and the setup of the datarecorder. During configuration, theuser can access several trigger modesfor the start/stop function that rangefrom manual quick start to the trigger-ing of recording via a previouslyentered I/Q signal level. The integratedEthernet interface permits remote con-trol of the instrument as well as thetransfer of the measurement data viaLAN. Two extra USB interfaces on thefront panel and a touch screen roundout the user-friendly concept of thedata recorder.

http://www.rohde-schwarz.com

MODEL-DRIVEN CONFIGURATIONMANAGEMENT

Tail-f Systems

Tail-f Systemshas announced thefirst model-driven configuration man-agement application for provisioningCarrier Ethernet services. NCS for Car-rier Ethernet will benefit both serviceproviders and networking equipmentproviders by enabling the activation ofcomplex services in less time and withfewer resources. NCS also radically sim-plifies the development of new andenhanced management systems, allow-ing developers to bring new products tomarket much faster.

NCS is a general applications frame-work for building configuration man-agement systems. NCS for CarrierEthernet extends the value of NCS byincorporating service and device modelsplus a Web UI optimized for imple-menting Carrier Ethernet managementsystems. http://www.tail-f.com

NEW PRODUCTS

LYT-PRODUCTS-FEB 1/20/11 12:06 PM Page 16


2011

F E B R U A R Y

• NTMS 2011 - 4th Int’l. Conferenceon New Technologies, Mobility andSecurity, 7-10 Feb.Paris, France.http://www.ntms-conf.org/innovative-pro-jects.htm

• ONDM 2011 - 15th Int’l. Conferenceon Optical Networking Design andModeling, 8-10 Feb.Bologna, Italy.http://www.ondm2011.unibo.it/

• ICACT 2011 - 13th Int’l. Conferenceon Advanced Communication Tech-nology, 13-16 Feb.

Phoenix Park, Korea. http://www.icact.org/

♦ IEEE CogSIMA 2011 - IEEE Confer-ence on Cognitive Methods in Situ-ation Awareness and DecisionSupport, 22-24 Feb.Miami, FL.http://www.ieee-cogsima.org

• ISWPC 2011 - Int’l. Symposium onWireless Pervasive Computing, 23-25Feb.Hong Kong, China.http://www.iswpc.org/2011/

• WSA 2011 - Int’l. ITG Workshop onSmart Antennas, 24-25 Feb.Aachen, Germany.http://www.wsa2011.rwth-aachen.de/

M A R C H

♦ OFC/NFOEC 2011- Optical FiberCommunication Conference, 6-10MarchLos Angeles, CA.http://www.ofcnfoec.org/

♦ IEEE WCNC 2011 - IEEE WirelessCommunications and NetworkingConference, 28-31 MarchCancun, Mexico.http://www.ieee-wcnc.org/

• ICCIT 2011 - Int’l. Conference onCommunications and InformationTechnology, 28-31 MarchAqaba, Jordan.http://iccit-conf.org/

A P R I L

♦ IEEE ISPLC 2011 - 15th IEEE Int’l.Symposium on Power Line Commu-nications and Its Applications, 3-6AprilUdine, Italy.http://www.ieee-isplc.org/

CONFERENCE CALENDAR

♦ Communications Society portfolio events are indicated with a diamond before the listing;• Communications Society technically co-sponsored conferences are indicated with a bulletbefore the listing. Individuals with information about upcoming conferences, calls for papers, meetingannouncements, and meeting reports should send this information to: IEEE CommunicationsSociety, 3 Park Avenue, 17th Floor, New York, NY 10016; e-mail: [email protected]; fax:+1-212-705-8996. Items submitted for publication will be included on a space-available basis.

IENYCM2846.indd 1 1/17/11 5:48:38 PM

LYT-CALENDAR-FEB 1/20/11 12:25 PM Page 17


♦ IEEE INFOCOM 2011 - IEEE Con-ference on Computer Communica-tions, 10-15 AprilShanghai, China.http://www.ieee-infocom.org

• IEEE RFID 2011 - IEEE Int’l. Confer-ence on RFID 2011, 12-14 AprilOrlando, FL.http://ewh.ieee.org/mu/rfid2011/

• WTS 2011 - Wireless Telecommuni-cations Symposium 2011, 13-15 AprilNew York, NY.http://www.csupomona.edu/~wtsi/wts/index.htm

• IIT 2011 - Int’l. Conference on Inno-vations in Information Technology,25-27 AprilDubai, United Arab Emirates.http://www.it-innovations.ae/iit10/index.html

• ISPS 2011 - 10th Int’l. Symposiumon Programming and Systems, 25-27AprilAlgiers, Algeria.http://www.isps2011.dz/

M A Y

• IEEE SARNOFF - 34th Sarnoff Sym-posium 2011, 2-4 MayPrinceton, NJ.http://sarnoff-symposium.ning.com/

• MC-SS 2011 - 8th Int’l. Workshopon Multi-Carrier Systems and Solu-tions, 3-4 MayHerrsching, Germany.http://www.mcss.dlr.de/

♦ IEEE DySPAN 2011 - IEEE Int’l.Symposium on Dynamic SpectrumAccess Networks, 3-6 MayAachen, Germany.http://www.ieee-dyspan.org/

• ICT 2011 - 18th Int’l. Conference onTelecommunications, 8-11 MayAyia Napa, Cyprus.http://www.ict2011.org/

♦ IEEE CQR 2011 - 2011 Annual IEEECQR Int’l. Workshop, 10-12 MayNaples, FL.http://committees.comsoc.org/cqr/

♦ IEEE ICSOS 2011 - IEEE Int’l. Con-ference on Space Optical Systemsand Applications, 11-13 MaySanta Monica, CA.

http://icsos2011.nict.go.jp/• CONATEL 2011 - 2nd Nat’l. Confer-ence on Telecommunications (Peru),

17-20 MayArequipa, Peru.http://conatel.ucsp.edu.pe/

♦ IEEE/IFIP IM 2011 - 12th IFIP/IEEEInt’l. Symposium on Integrated Net-work Management, 23-27 MayDublin, Ireland.http://www.ieee-im.org/

J U N E

♦ IEEE IWQoS 2011 - 18th IEEE Int’l.Workshop on Quality of Service, 5-7JuneSan Jose, California.http://www.ieee-iwqos.org/

♦ IEEE ICC 2011 - IEEE Int’l. Confer-ence on Communications, 5-9 JuneKyoto, Japan.http://www.ieee-icc.org/2011/

• IEEE POLICY 2011 - IEEE Int’l. Sym-posium on Policies for Distributed Sys-tems and Networks, 6-8 JunePisa, Italy.http://www.ieee-policy.org/

♦ IEEE CAMAD 2011 - IEEE Int’l.Workshop on Computer-AidedModeling Analysis and Design ofCommunication Links and Networks2011, 10-11 JuneKyoto, Japan.http://www.nprg.ncsu.edu/camad/

♦ IEEE HEALTHCOM 2011 - 13thIEEE Int’l. Conferece on e-HealthNetworking, Application & Services,13-15 JuneColumbia, MO.http://www.ieee-healthcom.org/

• ConTEL 2011 - 11th Int’l. Confer-ence on Telecommunications, 15-17JuneGraz, Austria.http://www.contel.hr/

• ICUFN 2011 - 3rd Int’l. Conferenceon Ubiquitous and Future NetworksDalian, China.http://www.icufn.org/main/

♦ IEEE CTW 2011 - IEEE Communi-cation Theory Workshop, 20-22JuneSitges, Spain.http://www.ieee-ctw.org

♦ IEEE SECON 2011 - 8th AnnualIEEE Communications Society Con-ference on Sensor, Mesh and AdHoc Communications and Net-works, 27-30 June

Salt Lake City, Utah.http://www.ieee-secon.org/2011/

• IEEE ITMC 2011 - IEEE Int’l. Technol-ogy Management Conference, 27-30JuneSan Jose, CA.http://www.ieee-itmc.org/

♦ IEEE ISCC 2011 - 16th IEEE Sym-posium on Computers and Commu-nications, 28 June-1 JulyKerkyra, Greece.http://www.ieee-iscc.org/2011/

• ICL-GNSS 2011 - Int’l. Conferenceon Localization and GNSS 2011, 29-30JuneTampere, Finland.http://www.icl-gnss.org/2011/index.php

J U L Y

♦ IEEE HPSR 2011 - 12th IEEE Int’l.Conference on High PerformanceSwitching and Routing, 4-6 JulyCartagena, Spain.http://www.ieee-hpsr.org/

• OECC 2011 - 16th Opto-Electronicsand Communications Conference, 4-8JulyKaoshlung, Taiwan.http://www.oecc2011.org/

♦ IEEE ICME 2011 - 2011 IEEE Int’l.Conference on Multimedia andExpo, 11-15 JulyBarcelona, Spain.http://www.icme2011.org/

A U G U S T

• ICCCN 2011 - Int’l. Conference onComputer Communications and Net-works 2011, 1-4 Aug.Maui, Hawaii.http://www.icccn.org/ICCCN11/

♦ ATC 2011 - 2011 Int’l. Conferenceon Advanced Technologies for Com-munications, 3-5 Aug.Da Nang City, Vietnam.http://rev-conf.org/

• ICADIWT 2011 - 4th Int’l. Confer-ence on the Applications of DigitalInformation and Web Technologies, 4-6 Aug.Stevens Point, WI.http://www.dirf.org/DIWT/

♦ IEEE P2P 2011 - IEEE Int’l. Confer-ence on Peer-to-Peer Computing,31 Aug.-2 Sept.Tokyo, Japan.http://p2p11.org/

CONFERENCE CALENDAR


♦ IEEE EDOC 2011 - 15th IEEE Int’l.Enterprise Distributed Object Com-puting Conference, 31 Aug.-2 Sept.Helsinki, Finland.http://edoc2011.cs.helsinki.fi/edoc2011/

S E P T E M B E R

• ITC 23 2011 - 2011 Int’l. TeletrafficCongress, 6-8 Sept.San Francisco, CA.http://www.itc-conference.org/2011

♦ IEEE PIMRC 2011 - 22nd IEEE Int’l.Symposium on Personal, Indoorand Mobile Radio Communications,11-14 Sept.Toronto, Canada.http://www.ieee-pimrc.org/2011/

• ICUWB 2011 - 2011 IEEE Int’l. Con-ference on Ultra-Wideband, 14-16Sept.Bologna, Italy.http://www.icuwb2011.org/

♦ IEEE GreenCom 2011 - OnlineConference, 26-29 Sept.Virtual.http://www.ieee-greencom.org/

O C T O B E R

• DRCN 2011 - 8th Int’l. Workshop onDesign of Reliable CommunicationNetworks, 10-12 Oct.Krakow, Poland.http://www.drcn2011.net/index.html

N O V E M B E R

• ISWCS 2011 - 8th Int’l. Symposiumon Wireless Communication SystemsAachen, Germany.http://www.ti.rwth-aachen.de/iswcs2011/

• COMCAS 2011 - 2011 IEEE Int’l.Conference on Microwaves, Commu-nications, Antennas and ElectronicSystems, 7-9 Nov.Tel Aviv, Israel.http://www.comcas.org/

♦ MILCOM 2011 - Military Commu-nications Conference, 7-10 Nov.Baltimore, MD.http://www.milcom.org/index.asp

D E C E M B E R

♦ IEEE GLOBECOM 2011 - 2011 IEEEGlobal Communications Confer-ence, 5-9 Dec.

Houston, TX.http://www.ieee-globecom.org/2011/

2012J A N U A R Y

♦ IEEE CCNC 2012 - IEEE ConsumverCommunications and NetworkingConference, 8-11 Jan.

Las Vegas, NV.http://www.ieee-ccnc.org/

M A R C H

♦ IEEE INFOCOM 2012 - IEEE Int’l.Conference on Computer Commu-nications, 25-30 MarchOrlando, FL.http://www.ieee-infocom.org/2012/


CONFERENCE CALENDAR

(Continued from page 17)

(UK): +44 141 552 8855(US): +1 805 413 4127

[email protected]

Realize a new level ofconfidence in LTE PHY design

Verification Golden Reference Performance Testing

Comprehensive set of functions/blocks modeling the LTE physical layer transmit and receive processing

Supports FDD and TDD duplexing

Standard compliant propagation channel models (EPA, EVA, ETU, Moving, HST)

Channel estimation, synchronization and MIMO receivers (ZF,MMSE,SFBC)

Product roadmap includes releases through LTE-Advanced

GUI based tools for LTE conformance test simulation and ARB waveformgeneration

Extensive help, background documentation and demo set including HARQand MIMO performance, PDCCH/DCI decoding and MIB/SIB recovery

LTE Toolbox/Blockset for MATLAB®/Simulink®

Evaluate the complete solution atwww.steepestascent.com/lte/mathworks

IENYCM2844.indd 1 1/17/11 5:47:34 PM



PRODUCT SPOTLIGHTS

Omicron LabsOmicron Labs’ Vector Network Ana-lyzer Bode 100 now provides a 360°view on switched mode power sup-plies and voltage regulators. In com-bination with the Picotest SignalInjector series all key parameterssuch as Stability, PSRR, cross-talk,reverse transfer and impedance canbe measured easily and accurately.

http://www.omicron-lab.comhttp://www.picotest.com

GL CommunicationsGL’s PacketProbe™ is an advancedCPE-based VoIP monitoring, report-ing and diagnostic appliance. Packet-Probe™ passively monitors VoIPtraffic carried over the WAN/LAN by

producing per call and per-stream voice quality metrics. Call Detail Records(CDR’s) along with voice quality statistics and other vital diagnostic informationprovide network managers immediate visibility into service quality, call volumesand call details. Service providers are able to rapidly drill down and diagnosevoice related issues. Standards based Real-time Monitoring, Reporting andDiagnostic tools fit seamlessly into any existing standards based Management orReporting environment, such as SNMP and RADIUS. Vital voice call qualitystatistics, Call Detail Records and Quality of Service metrics are available at theend of each call. Optionally GL offers its own Monitoring and Reporting Sys-tem, PacketScanWEB™.

http://www.gl.com/packetprobe.html

VectronThe solution for all applications withno room for compromise in g-sensi-tivity, noise or stability performancebut small size is required. Ideal forequipment used in harsh environ-ments such as tactical weapons,avionics and portable equipment.The 508 product family of oscillatorsoffers the industry's leading aging,stability, phase noise and g-sensitivi-ty in a miniature 9 x 14 package.

http://www.vectron.com

RSoft DesignRSoft releases version 5.2 of its awardwinning Optical CommunicationDesign Suite OptSim and its multi-mode companion ModeSYS. The lat-est release features enhancedmodeling capabilities for 100G Coher-ent PM-QPSK, Polarization inducedall-optical switching, Bi-directionalEthernet in the First Mile, Interfero-

metric systems, mode-coupling based Multimode systems, and Plastic OpticalFiber systems.

http://www.rsoftdesign.com

Silicon LabsLearn how to simplify your timingdesign using glitch-free frequencyshifting to address low-power designchallenges and the complexity of gen-erating a wide range of frequencies inconsumer electronics applicationssuch as audio, video, computing orany application that requires multiplefrequencies. Download this in-depthwhite paper from Silicon Labs.

http://www.silabs.com/frequency-shifting

GL CommunicationsGL’s MAPS™(Message Automation& Protocol Simulation)-LTE supportsscripted LTE simulation with the abil-ity to simulate entities such as eNodeB(Evolved Nodeb), MME (MobilityManagement Entity), SGW (ServingGateway), and PGW (Packet DataNetwork Gateway). Supported inter-faces include S1, S11 and S5/S8 (LTE-S1 and LTE-eGTP). Support for other

interfaces such as S4, S11, and S12 is coming soon. The application gives theusers the unlimited ability to edit S1-AP/NAS and eGTP-C (Evolved GPRS Tun-neling Protocol for Control Plane (eGTP-c) messages and control scenarios(message sequences).

http://www.gl.com/maps-lte.html

SPOT PAGE-11-02 1/20/11 12:13 PM Page 20

Global Communications Newsletter • February 2011

G l o b a l

N e w s l e t t e rFebruary 2011

1

Telecommunications and networks are well recognizedenabling technologies for telemedicine applications in remoteand rural locations, but are also more and more becomingfacilitating technologies for continuous health monitoring outof hospital, for p-health at home, and during various humanactivities, professional, leisure, or sports.

A new field of investigation, never entered before, is nowmade possible: to continuously collect health information withcontext awareness. The field of e-health is also a remarkablemelting point which gives the opportunity to bring togetherinterested parties from around the world working in the vari-ous fields of healthcare and engineering to exchange ideas,discuss innovative and emerging solutions, and develop collab-orations around operational projects.

The 12th International Conference on e-Health Network-ing, Application & Services, Healthcom2010, was held inLyon, France, on 1–3 July 2010. It was an important forum fordiscussions on e-health projects sponsored by world bodiessuch as the European Community (FP6 and FP7 Europeanprojects on AAL and e-inclusion, etc.).

Each year, a broad variety of topics are presented atHealthcom, addressing the different levels of e-Health, fromtechnologies to applications:•Network and Communications Infrastructures and Architec-

tures for Healthcare Delivery•New Models for Healthcare Delivery•e-Health for Public Health•e-Health for Aging •m-Health•Field Applications•Education•Ethics

The conference gathered more than 130 attendees, comingfrom 35 countries, including clinicians, IT professionals,researchers, healthcare solutions vendors, and consultants.About 90 papers were presented, addressing a broad variety oftopics within e-Health. The submission rate was high, with 122papers out of which 62 were selected for presentation, includ-ed in the Proceedings accessible through IEEEXplore, and sixselected papers are to appear in a special issue of the Interna-tional Journal of e-Health and Medical Communications (Pro-fessor Joel Rodrigues, Editor-in-Chief).

The first keynote speaker, Professor Louis Lareng, Direc-tor of the European Telemedicine Society, reviewed the histo-ry of developing telemedicine in Europe since the 1950s. DrLoukianos Gatzoulis, from the European Commission inBrussels, presented the goals of the EU in supporting e-health, societal and economical. Jean Schwoerer, from OrangeLabs, depicted the current normalization efforts in the field of

body sensor networks. Professor André Dittmar from CNRSin Lyon demonstrated the need for new ubiquitous body sen-sors to support e-health efforts and deployments.

The HC10 was articulated around three main sessions,“Enabling Technologies,” “Enabling Information Systems,”and “Enabling Applications.” The variety and density of infor-mation were very high. Some main points are the followings.

Body sensor networks allow collecting and aggregatingdata from several sensors in a mobile context. This mobilityoffers continuous monitoring of patient status, imrpovingpatients’ quality of life. Existing femtocellular networkresources, already available on site, may be used for rapidprovisioning of mobile broadband data connectivity indoorsfor emergency telemedicine applications. This approachresults in a reduction in service outage rates.

Remote telemonitoring of elderly people in their ownhomes is a major challenge to face the fast growing populationof elderly people. Health Smart Homes allow monitoring thebehavior of a person with non-intrusive sensors. The majortrends in the activity reflect the global homeostasis of the sub-ject. High-level decision tools are used to classify scenarios ofdaily living and eventually to build an index of activities of dailyliving. As these smart homes will benefit people in preventingloss of autonomy, disabled people or elderly people with cogni-tive deficiencies, it is essential to facilitate their interactionswith Smart Homes through dedicated interfaces, such as sys-tems reactive to vocal orders. Audio recognition is also apromising way to ensure detection of distress situations.

E-healthcare and telemedicine applications, when deployedto provide healthcare to remote locations in developing coun-tries, must carefully take into account the existing healthcareand communications facilities, but also socio-economic condi-tions of populations.

Telemedicine can also allow deployment of real-time dis-ease surveillance and notification systems in developing coun-tries. The communication technologies, adopting globalstandards for structured messages (SMS, email, web), willreduce the delays in communicating field data to central epi-demiology units, which can therefore detect disease outbreaksin a timely manner, and allow health system to effectivelyrespond and mitigate the consequences for populations.

The domain of e-health is currently demonstrating highvitality. It is a living laboratory for cooperation between thefields of health and engineering. It is also a chance to betterunderstand the health of humans in their living contexts.

I want to thank Assistant Professor Pradeep Ray, directorof the Asia-Pacific Ubiquitous Healthcare Research Centre(APuHC) at the University of New South Wales, who kindlyinvited me to organize IEEE Healthcom 2010.

IEEE Healthcom 2010: “Ambient Assisted Living” for Better HealthBy Norbert Noury, Healthcom 2010 General Chair, University of Lyon, France

LYT-NEWSLETTER-FEB 1/19/11 3:43 PM Page 21


The First international FOKUS FUSECO Forum (FFF) on“Business and Technical Challenges of Seamless Service Provi-sion in Converging Next Generation Fixed and Mobile Net-works” was held in Berlin, Germany, on 14-15 October 2010 andwas attended by around 150 experts from industry and academia.The FFF represents a technologically focused follow-up to thefamous FOKUS IMS Workshop series, as the future role of IPMultimedia Subsystem (IMS) in mobile and fixed next-genera-tion networks (NGNs) deserves some critical considerations inregard to slow industry adoption and rapidly emerging new con-trol protocols and platforms from the Internet domain. There isno doubt that IMS has consolidated the various views on Ses-sion Initiation Protocol (SIP) and Diameter-based NGNs, andmany standards have been established for voice over IP (VoIP),rich communications services (RCS), and IPTV. But there is lotof pressure from emerging over-the-top (OTT) applicationsoriginating from the Internet, challenging IMS business case cal-culations and extensive deployments beyond NGN VoIP. Inaddition, IMS is limited to SIP-based control and thus HTTP(Hypertext Transport Protocol) and other protocol-based appli-cations, forming the majority of current fixed and mobile broad-band traffic are out of control. Here the 3GPP Evolved PacketCore) (EPC) has the potential to emerge as a common controlplatform for any IP application. From the technical maturity andglobal recognition points of view, EPC stands today where IMSstood five years ago; thus, it is time to launch a new workshopseries to create global awareness in academia and industry aboutthis promising technology.

Thus, the FFF discussed for two days the relationshipbetween the EPC, IMS, and OTT approaches. The first daystarted with a technical tutorial about Long Term Evolution(LTE) and EPC standards, and introduced the FraunhoferHHI LTE-Advanced Testbed and Fraunhofer FOKUSOpenEPC toolkit, which have been integrated into the FutureSeamless Communication (FUSECO) Playground, the globallyfirst open testbed uniting wireless LANs, 3G networks, LTE,and EPC technologies for early prototyping of new seamlessapplications. Practical demonstrations including seamless hand-overs of Skype and video services with quality of service assur-ance from the FUSECO Playground and an OpenEPC Release2 preview were also provided to the audience. The second dayfeatured a conference with presentations from various interna-tional network operators and service providers. A vendor panelsession and an associated exhibition presented the state of theart and upcoming products in this field. In the following somemore details are provided for both days.

The tutorial “Understanding the Next Generation of MobileBroadband Communications: LTE and EPC Concepts, Archi-tectures, Protocols and Applications” on the first day was pre-sented by Dr. Thomas Haustein, Fraunhofer HHI, and Prof.Dr. Thomas Magedanz, Fraunhofer FOKUS and TU Berlin. Itstarted by pointing out the continuously increasing mobile datatraffic demand and motivated the need for LTE, EPC, and IMStechnologies in order to allow smooth evolution from existingcircuit- and packet-switched mobile networks to a next-genera-tion mobile network. A session about LTE presented detailsabout the radio part, covering standards and architectures, andgave an outlook on LTE-Advanced. The correlated networkpart was covered in the following session by introducing EPCterminology, key concepts, and architecture, as well as the relat-ed Third Generation Partnership Project (3GPP) standards.Subsequently, potential applications and related platforms werediscussed, including operator IMS platforms for voice over LTE(VoLTE), as well as over the top Internet service platforms. Anoutlook onto global Future Internet research and related appli-cation areas has concluded this tutorial section. The tutorial

ended with a presentation of the experiences from the BerlinLTE-Advanced testbed, the OpenEPC testbed toolkit, and theFUSECO Playground. Demonstrations during the breaksshowed current proof of concept realizations, including seam-less handovers between LTE-A and WLAN, service composi-tion, and transparent mobility. The demonstrations coveredessential challenges in the scope of LTE and EPC, and illustrat-ed practical solutions based on these emerging technologies.

On day 2 the conference started with a session on “Compet-ing Mobile Broadband Access Network Technologies” chairedby Thomas Haustein, Fraunhofer HHI, and addressed chal-lenges emerging with the introduction of LTE and EPC. Thesecond session, “Access Network Integration and ServiceEnabling,” chaired by Hans Schotten, University of Kaiser-slautern, addressed technical problems during the deploymentof EPC into existing network infrastructures, in particular IMSroaming and different QoS signaling, as well as the complexIMS and EPC interoperability, which might take until2014/2015. A vendor panel, “Standards, Products, and BusinessCases for Future Seamless Communication,” chaired byThomas Magedanz, TU Berlin, discussed the LTE businesscase, voice over LTE, the importance of open application pro-gramming interfaces (APIs) for VoLTE and RCS, and so on.The fourth session, “FUSECO Telco Applications: Voice, RCSand More,” chaired by Hans Joachim Einsiedler, DeutscheTelekom Laboratories, addressed mobile broadband servicesand M2M opportunities, and application challenges regardingfast deployment of web services and slow agreement on inter-operability. The overall tenor forecasts no LTE/mobile broad-band killer application. The last session, “FUSECO OTTApplications: Beyond Smart Bit Pipes,” chaired by ThomasMichael Bohnert, SAP Research, presented opportunities forwholesale and enterprise operators, OTT services, and theusability of LTE for vehicles. VoIP in mobile networks is grow-ing in acceptance, increasingly challenging the operators aroundthe globe. As Facebook has now announced interworking withSkype, the question of using Facebook as the main interface forlaunching new applications in the future has been raised.

Alongside the workshops and conference, vendor exhibitionsshowed 4G Subscriber Data Management/Communications as aService, Enhancements of Mobility Management for the 3GPPEPS — smart mobile devices in a dense wireless network envi-ronment, IBM Software Strategy for CSPs — start planningand implementing smarter communications systems, and smartnetworks for user-centric broadband. In addition, the newest

1st FOKUS FUSECO Forum 2010, Berlin, GermanyBy Prof. Dr. Thomas Magedanz, General Chair, TU Berlin/Fraunhofer FOKUS, Germany

(Continued on Newsletter page 4)

The 1st International FOKUS FUSECO Forum (FFF) wasattended by approximately 150 experts from industry andacademia.


Global Communications Newsletter • February 2011 3

Universal telecom service is a concept defined at the Euro-pean level with the objective to guarantee all citizens the rightof access to a defined set of basic electronic communicationsservices, independent of their geographical location, with aminimum quality and at a reasonable price. Until now, inSpain, only a functional Internet access was considered as aservice offered by the public phone wired network, with down-load speeds of 256 kb/s.

During October 2009, the Spanish Ministry of Industryperformed a public survey [1] to determine the minimum fea-tures that would require the Internet universal service to beupdated to the growing needs of the Information Society. Thesurvey outcomes concluded that choosing broadband accesswith a bit rate of 1 Mb/s would conciliate the current demandsof the Information Society and the requirements to impel themodernization of the digital infrastructures. As a conse-quence, the Minister of Industry announced that the universalservice would include the 1 Mb/s downlink broadband connec-tion in 2011 as a minimum requirement to increase competi-tiveness in the broadband business. Recently, during thecelebration of CEBIT 2010 in Hannover, Spanish Govern-ment President Mr Rodriguez Zapatero confirmed this aim.

The operator in charge of providing the universal servicewill be selected during 2010. The inclusion of the broadbandaccess with a minimum downlink speed of 1 Mbps makesSpain to go in head of the digital European policies. Thisactuation can be considered an epilogue to the AVANZA-PEBA plan [2], designed to extend the penetration of thebroadband access in Spain and deployed by the Governmentfrom 2005 to 2008. Now, that the plan is over, it is time tomeasure the achievements before designing the new telecompolicy.

We can compare the Spanish case to other Europeancountries such as Finland. Traditionally, the latter has beenfound positioned in first place of the pro-right career intelecommunications, reaching the point of including in its con-stitution the right of a nationwide 1 Mb/s broadband connec-tion provided by any kind of technology, hoping to reach thespeed of 100 Mb/s by 2015. However, it was not Finland butSwitzerland that, in 2008, defined broadband access as a con-nection of 100 kb/s/600 kb/s for up-/downlink speeds. Thus,Spain will become the third European country adopting aquantitative definition of broadband access. Despite this, thereality and the political intentions follow paths dangerouslyseparated.

The European Competitive Telecommunications Associa-tion (ECTA) Broadband Scorecard is a recognized bench-mark being used regularly by industry, the EuropeanCommission, national regulators, and institutions. Biennially,ECTA collates and publishes data tracking the progress onbroadband penetration and local loop unbunding in the 25European Member States [3]. Some of the latest publishedstatistics are plotted in Figure 1. According to this source,goals seem not to be reached, placing Spain under the averageof European countries in broadband penetration, with a pene-tration rate of 21 percent, which is two points under the EUaverage placed at 23.5 percent. Along the same lines, a recentreport published by the Regional Government of Galicia(Northwestern Spain) demonstrated that this situation is evenmore critical [4]. The determining factor of this circumstancecan be found in the situation experienced by rural areas,which present 70 percent nonexistent or low-quality networkaccess, under 512 kb/s, even after an investment of€225,000,000 provided by public funds. The same happens inother Spanish regions, showing a deep imbalance betweenurban and rural state areas.

Closer insight into tje data provided by ECTA reveals thatSweden remains Europe’s fiber leader, with 7.5 percent of thepopulation benefiting from high-speed modern access linescompared to an average of just 0.4 percent across the EU.Despite “regulatory holidays” for incumbents in countriessuch as Germany and Spain, the survey showed little evidenceof increased fiber deployment in those countries. It also high-lights a strong link between effective economic regulation andinvestment levels in the telecom sector, where a regulatoryframework helps alternative operators compete against thenational incumbent, but also does not encourage investmentson new infrastructure since new deployments require manyresources and are only profitable in the long term.

The EU presidency by part of Spain in 2010 would seem tobe a suitable moment to support the consideration of broad-band Internet as an indispensable service in the InformationSociety. However, 1 Mb/s may not be enough and can hardlybe considered broadband Internet in 2010. Moreover, no onehas explained how this service will be provided since there arelarge areas in Spain without any kind of Internet access. Andthe most important thing is the need to define how the provi-sion of this service will be financed. Probably, without publicfunds, it will not be possible to accomplish these goals,although achieving them would doubtlessly place Spain on topof the most developed information societies.

References[1] http://www.mityc.es[2] http://www.planavanza.es[3] http://www.ectaportal.com[4] http://imit.xunta.es/

Internet for Everybody in Spain: The 1 Mb/s Universal ServiceBy Ana Vázquez Alejos, Rafel Asorey Cacheda and Felipe José Gil Castiñeira, University of Vigo, Spain

2009 April total connections per technology for Spain includingradio [2].


4

STEFANO BREGNIEditor

Politecnico di Milano - Dept. of Electronics and InformationPiazza Leonardo da Vinci 32, 20133 MILANO MI, Italy

Ph.: +39-02-2399.3503 - Fax: +39-02-2399.3413Email: [email protected], [email protected]

IEEE COMMUNICATIONS SOCIETY

KHALED B. LETAIEF, VICE-PRESIDENT CONFERENCESSERGIO BENEDETTO, VICE-PRESIDENT MEMBER RELATIONSJOSÉ-DAVID CELY, DIRECTOR OF LA REGIONGABE JAKOBSON, DIRECTOR OF NA REGIONTARIQ DURRANI, DIRECTOR OF EAME REGIONNAOAKI YAMANAKA, DIRECTOR OF AP REGIONROBERTO SARACCO, DIRECTOR OF SISTER AND RELATED SOCIETIES

REGIONAL CORRESPONDENTS WHO CONTRIBUTED TO THIS ISSUE

THOMAS M. BOHNERT, SWITZERLAND([email protected])JOSEMARIA MALGOSA SANAHUJA, SPAIN ([email protected])JOEL RODRIGUES, PORTUGAL ([email protected])EWELL TAN, SINGAPORE ([email protected])

®�

A publication of the IEEE Communications Society

G l o b a l

N e w s l e t t e rwww.comsoc.org/pubs/gcn


FOKUS FUSECO FORUM/continued from page 2

toolkits from Fraunhofer FOKUS, the universal client frame-work myMONSTER TCS (www.opensoaplayground.org/tcs)and the newest release of the OpenEPC toolkit(www.openepc.net), were presented.

This new forum will be continued next year in November2011 to establish a regular meeting point for internationalresearchers from academia and industry (www.fuseco-forum.org/2011).

can Park,Hi-Tech City, Hyderabad, from 5 p.m. to 7 p.m. Therewere 154 audience members for this lecture, including students,research scholars, participants from industry, and faculty of uni-versities and colleges. In this lecture the speaker answered issuesraised by participants related to protocols and infrastructure,IETF.org, streaming media, VoIP, IPTV, packet delay, packetlosses, acceptable delays (< 150 ms), the difference betweenInternet phone and IP phone, telemedicine prospects in India,IPv6 adoption, QoS, QoE, WiMAX, LTE, virtual device (embed-ded) services, and more.

The single biggest phenomenon that is transforming theglobal telecom industry is convergence. Internet users are nowexposed to different modes of communication than basic voicetelephony. Communication now includes pictures and videos,and is not limited to person-to-person communication; com-munities and user groups are being created, and informationexchange is not limited to and from people known to eachother. Content is driving service subscription, and identifyingthe right content in the right format at the right cost and deliv-ering it in a secure manner to any kind of business model forthe communication industry are the main issues. Convergencehas been visible on the horizon for the last couple of years, buthas yet to arrive in developing countries in a major way.

The lecture focused on convergence of communication ser-vices like data, voice, and video in IP-based networks. Thespeaker stressed the importance of quality of experience(QoE) apart from quality of service (QoS) in IPTV user judg-ment and acceptance. Experience needs to be preserved asvideo traffic is transported across IP infrastructure. Serviceproviders need IP-based next-generation network (IP-NGN)infrastructure solutions that are intelligent and video-aware.An outstanding video experience requires excellent solutionsin the customer home to decode, decrypt, share, and displaythe content the way it was intended.

In feedback, the participants expressed satisfaction withthe event organization, suggesting increasing the time forthese kinds of lectures, more explanations of security, andsome demonstrations using MATLAB interfacing and Lab-views.

The Distinguished Lecturer Program is one of the best ini-tiatives of the IEEE Communications Society. It brings distin-guished experts to give lectures at Chapters on all continents.A DL tour of Dr. Bhumip Khasnabish, ZTE, United States,was held in India in July 2010. Lectures entitled “Services overIP: Implementation Options and Challenges” and “ConvergedServices and a New Generation of Networking” were given inIndia from 9 July to 17 July 2010 with the following schedule:1) Mumbai, 9 July 2010 (two lectures); 2) Pune, 10 July, 12July 2010 (two lectures); 3) Hyderabad, 13 July 2010 (two lec-tures); 4) Kharagpur, 15 July 2010 (one lecture); 5) Kolkata,17 July 2010 (one lecture).

Dr. Khasnabish’s lectures in Hyderabad were organized bythe Communications and Signal Processing Societies Joint Chap-ter of the IEEE Hyderabad Section, and his accommodation andtravel within Hyderabad were arranged by TCS Hyderabad.“Services over IP: Implementation Options and Challenges” is atutorial and was held on 13 July 2010 at the Research and Train-ing Unit for Navigational Electronics (NERTU) auditorium,University College of Engineering, Osmania University, Hyder-abad from 9 a.m. to 1 p.m. There were 55 audience members forthe tutorial, including students, research scholars, participantsfrom industry, and faculty of colleges. During this lecture, issuesrelated to mouth-to-ear delay calculation or estimation, GPSintegration, location services, current status and future infras-tructure deployment, speed of IPTV and the status of currentcompression technology, cooperating multimode devices, netenabled health services, security issues like DRM, Skype, andemerging trends were raised by the participants. These were wellanswered by the speaker; furthermore, he emphasized the needfor interoperability, standardization, and protocol integration.

“Converged Services and a New Generation of Networking”was held on 13 July 2010 at the Gadavari auditorium, TCS, Dec-

Distinguished Lecturer Tour of Bhumip Khasnabish in IndiaBy Deergha Rao Korrai, Chair, Communications and Signal Processing Societies Joint Chapter, Hyderabad, India

Mr.MGPL Narayana(Hyderabad section chair, third from left),Dr.Bhumip Khasnabish (fourth from left), Dr.Deergha Rao Kor-rai (chapter chair, fifth from left) and other IEEE volunteers ofthe Hyderabad section after the lecture at the Godavari auditori-um of TCS.


A Publication of the IEEE Communications Society®

Special Supplement

Passive Optical Networks

Official Co-Sponsor

IEEE

M A G A Z I N E

February 2011, Vol. 49, No. 2

www.comsoc.org

February 2011 Supplement Cover 1 1/20/11 3:19 PM Page 1

Director of MagazinesAndrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)

Editor-in-ChiefSteve Gorshe, PMC-Sierra, Inc. (USA)

Associate Editor-in-ChiefSean Moore, Centripetal Networks (USA)

Senior Technical EditorsTom Chen, Swansea University (UK)

Nim Cheung, ASTRI (China)Nelson Fonseca, State Univ. of Campinas (Brazil)

Torleiv Maseng, Norwegian Def. Res. Est. (Norway)Peter T. S. Yum, The Chinese U. Hong Kong (China)

Technical EditorsSonia Aissa, Univ. of Quebec (Canada)

Mohammed Atiquzzaman, U. of Oklahoma (USA)Paolo Bellavista, DEIS (Italy)

Tee-Hiang Cheng, Nanyang Tech. U. (Rep. Singapore)Jacek Chrostowski, Scheelite Techn. LLC (USA)Sudhir S. Dixit, Nokia Siemens Networks (USA)

Stefano Galli, Panasonic R&D Co. of America (USA)Joan Garcia-Haro, Poly. U. of Cartagena (Spain)Vimal K. Khanna, mCalibre Technologies (India)

Janusz Konrad, Boston University (USA)Abbas Jamalipour, U. of Sydney (Australia)

Deep Medhi, Univ. of Missouri-Kansas City (USA)Nader F. Mir, San Jose State Univ. (USA)

Amitabh Mishra, Johns Hopkins University (USA)Sedat Ölçer, IBM (Switzerland)

Glenn Parsons, Ericsson Canada (Canada)Harry Rudin, IBM Zurich Res.Lab. (Switzerland)Hady Salloum, Stevens Institute of Tech. (USA)Antonio Sánchez Esguevillas, Telefonica (Spain)

Heinrich J. Stüttgen, NEC Europe Ltd. (Germany)Dan Keun Sung, Korea Adv. Inst. Sci. & Tech. (Korea)Danny Tsang, Hong Kong U. of Sci. & Tech. (Japan)

Series EditorsAd Hoc and Sensor Networks

Edoardo Biagioni, U. of Hawaii, Manoa (USA)Silvia Giordano, Univ. of App. Sci. (Switzerland)

Automotive Networking and ApplicationsWai Chen, Telcordia Technologies, Inc (USA)

Luca Delgrossi, Mercedes-Benz R&D N.A. (USA)Timo Kosch, BMW Group (Germany)

Tadao Saito, University of Tokyo (Japan)Consumer Communicatons and Networking

Madjid Merabti, Liverpool John Moores U. (UK)Mario Kolberg, University of Sterling (UK)

Stan Moyer, Telcordia (USA)Design & Implementation

Sean Moore, Avaya (USA)Salvatore Loreto, Ericsson Research (Finland)

Integrated Circuits for CommunicationsCharles Chien (USA)

Zhiwei Xu, SST Communication Inc. (USA)Stephen Molloy, Qualcomm (USA)

Network and Service Management SeriesGeorge Pavlou, U. of Surrey (UK)

Aiko Pras, U. of Twente (The Netherlands)Networking Testing Series

Yingdar Lin, National Chiao Tung University (Taiwan)Erica Johnson, University of New Hampshire (USA)Tom McBeath, Spirent Communications Inc. (USA)

Eduardo Joo, Empirix Inc. (USA)Topics in Optical Communications

Hideo Kuwahara, Fujitsu Laboratories, Ltd. (Japan)Osman Gebizlioglu, Telcordia Technologies (USA)

John Spencer, Optelian (USA)Vijay Jain, Verizon (USA)

Topics in Radio CommunicationsJoseph B. Evans, U. of Kansas (USA)

Zoran Zvonar, MediaTek (USA)Standards

Yoichi Maeda, NTT Adv. Tech. Corp. (Japan)Mostafa Hashem Sherif, AT&T (USA)

ColumnsBook Reviews

Andrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)History of Communications

Mischa Schwartz, Columbia U. (USA)Regulatory and Policy Issues

J. Scott Marcus, WIK (Germany)Jon M. Peha, Carnegie Mellon U. (USA)

Technology Leaders' ForumSteve Weinstein (USA)

Very Large ProjectsKen Young, Telcordia Technologies (USA)

Publications StaffJoseph Milizzo, Assistant Publisher

Eric Levine, Associate PublisherSusan Lange, Online Production ManagerJennifer Porcello, Publications Specialist

Catherine Kemelmacher, Associate Editor

S2 IEEE Communications Magazine • February 2011

IEEE

M A G A Z I N EFebruary 2011, Vol. 49, No. 2

www.comsoc.org/~ci

SPECIAL SUPPLEMENT

ADVANCES IN PASSIVE OPTICAL NETWORKSGUEST EDITORS: MAHMOUD DANESHMAND, CHONGGANG WANG, AND WEI WEI

CONFERENCE PREVIEWOFC/NFOEC 2011: LEADING THE WAY IN OPTICAL COMMUNICATIONSLYNDSAY BASISTA

CONFERENCE REPORT36TH EUROPEAN CONFERENCE ON OPTICAL COMMUNICATIONFABIO NERI

GUEST EDITORIAL

OPPORTUNITIES FOR NEXT-GENERATION OPTICAL ACCESSNext-generation optical access technologies and architectures are evaluated based on operators’ requirements. The study presented in this article compares different FTTH access network architectures.DIRK BREUER, FRANK GEILHARDT, RALF HÜLSERMANN, MARIO KIND, CHRISTOPH LANGE, THOMAS MONATH, AND ERIK WEIS

COST AND ENERGY CONSUMPTION ANALYSIS OF ADVANCED WDM-PONSThe authors compare several WDM-PON concepts, including hybrid WDM-PON with integrated per-wavelength multiple access, with regard to these parameters. They also show the impact and importance of generic next-generation bandwidth and reach requirements.KLAUS GROBE, MARKUS ROPPELT, ACHIM AUTENRIETH, JÖRG-PETER ELBERS, AND MICHAEL EISELT

TOWARD ENERGY-EFFICIENT 1G-EPON AND 10G-EPON WITH SLEEP-AWARE MACCONTROL AND SCHEDULINGThe authors briefly discuss the key features of 10G-EPON. Then, from the perspectiveof MAC-layer control and scheduling, they discuss challenges and possible solutions to put optical network units into low-power mode for energy saving. JINGJING ZHANG AND NIRWAN ANSARI

MULTIRATE AND MULTI-QUALITY-OF-SERVICE PASSIVE OPTICAL NETWORK BASEDON HYBRID WDM/OCDM SYSTEMThe authors present a new scheme to support multirate and multi-quality-of-servicetransmission in passive optical networks based on a hybrid wavelength-division multiplexing/optical code-division multiplexing scheme. The idea is to use multi-length variable-weight optical orthogonal codes as signature sequences of a hybrid WDM/OCDM system. HAMZEH BEYRANVAND AND JAWAD A. SALEHI

PASSIVE OPTICAL NETWORK MONITORING: CHALLENGES AND REQUIREMENTSThe authors address the required features of PON monitoring techniques and reviewthe major candidate technologies. They highlight some of the limitations of standardand adapted OTDR techniques as well as non-OTDR schemes. MOHAMMAD M. RAD, KERIM FOULI, HABIB A. FATHALLAH, LESLIE A. RUSCH, AND MARTIN MAIER

S12

S4

S8

S16

S25

S33

S39

S45

FOR INFORMATION ABOUT ADVERTISING CONTACT

ERIC LEVINE

ADVERTISING MANAGER

[email protected]

®

LYT-SUPPLEMENT-TOC-FEB 1/20/11 12:04 PM Page 28

S4

Once again, optical communica-tions and networking professionalsfrom around the world will cometogether for an intense week of net-working, high-quality science, andthe most comprehensive exhibit inthe industry at the Optical Fiber Communication Conferenceand Exposition (OFC) and the National Fiber Optic Engi-neers Conference (NFOEC). This year’s conference will fea-ture hundreds of peer-reviewed technical and invitedpresentations, a rump session focusing on “green” networking,a special symposium on computer components and architec-tures, an exhibit hall with more than 500 leading optical com-munications companies and so much more. Areas that will behighlighted include: Datacom/Data Centers, FTTx/In-home,Next Generation Data Transfer Technology, Photonic Inte-gration and Wireless Backhaul.

OFC/NFOEC will be held at the Los Angeles ConventionCenter in Los Angeles, Calif. from March 6 – 10 with theexposition taking place March 8 – 10.

PLENARY SESSIONThis year’s plenary speakers are leaders at the forefront of

the optical communications industry. As always, OFC/NFOECdraws speakers who are industry veterans and their widerange of expertise is certain to be a main attraction at thisyear’s OFC/NFOEC. The event will take place on Tuesday,March 8 from 8 – 11 a.m.

Olivier Baujard, Chief Technology Officer at DeutscheTelekom has had a career spanning all facets of the telecom-munications industry. From his time at France Telecom wherehe held engineering and managerial roles to his 20-yeartenure at Alcatel-Lucent, France where he rose to the level ofchief executive officer and chairman, Baujard has been a lead-er in the ever-changing telecommunications industry. In hiscurrent position at Deutsche Telekom, Baujard is responsiblefor the group-wide engineering, deployment and operation offixed, mobile and carrier infrastructures. During his plenarytalk, Baujard will discuss the future data and media-centricworld with a particular emphasis on the challenges it presentsfor network operators. Additionally, he will cover networktransformation and strategy and practical examples of how toenable efficient broadband. Baujard will also provide insightsabout Deutsche Telekom and their vision for the future ofwireless services.

Providing an insider’s perspective on the advances anddevelopments that optical technologies have on supercomput-ing is Alan Gara, IBM Fellow and Blue Gene Chief Archi-tect. Gara’s talk will discuss the challenges of reachingcomputing at the exaflop level with a special emphasis onareas where traditional communication solutions will not suf-fice. Gara has been leading high performance computingarchitecture and design efforts at IBM Research since 1999.Before then, Gara worked on the Superconductor supercol-lider (SSC) in Texas and at the Large Hadron Collider(LHC) at CERN. He has been the recipient of many presti-gious awards for his supercomputing efforts including twoGordon Bell awards.

With a career spanning more than 30 years in the telecom-munications industry, Kristin Rinne, Senior Vice President –Architecture & Planning at AT&T, will share with attendeesher outlook on the explosive growth of mobile data and howthat growth impacts optical networks including cell site back-

haul. Rinne is responsible for theIT and Network architecture andplanning for AT&T, including set-ting the direction for the Network,BSS/OSS Systems, Services, etc. Sheis also responsible for the IPTV and

the wireless network infrastructure and device technology ven-dor selection and first office implementation. Prior to joiningAT&T, Rinne served as Cingular’s chief technology officerand before that was vice president–Technology Strategy atSBC Wireless and managing director–Operations at South-western Bell Mobile Systems.

The plenary session will be held on Tuesday, March 8 from8 – 11 a.m.

RUMP SESSIONBack for its third year is the always popular Rump Session,

an audience-driven discussion designed to be as interactiveand dynamic as possible. This year’s Rump Session will bedevoted to the question: “Is green networking revolutionary?”Organizers are looking for attendees to discuss whether amore energy-efficient network is crucial to continued growthof the Internet or if “green” is merely today’s buzzword forsomething quite familiar to us in the past, whether energy bot-tlenecks will constrain network growth if more attention is notpaid to energy efficiency, and whether consumers and carrierswill pay more to go “green,” and whether “green” networkingwill reduce costs in the long run. Any audience member ispermitted to participate in the discussion and can use slides tofurther demonstrate points (no more than two can be accept-ed). The Rump Session will take place Tuesday, March 8from 7:30 – 9:30 p.m.

EXHIBIT HALLThe OFC/NFOEC exhibit hall attracts some of the

biggest players in the optical communications industry fromfiber equipment and components manufacturers to systemsand cable vendors, participating companies include industryheavyweights like Juniper Networks, JDSU, Huawei, Finis-ar, Ciena and Nokia Seimens. The exhibit hall will be theplace to see the latest in product demonstrations and inno-vation while also providing an opportunity for networkingwith industry leaders.

The exhibit hall is also home to a number of programmingevents and activities. New this year, EXPO Theater II willhave an optical networking focus and will feature events suchas the Ethernet Alliance Program (panel discussion on thefuture of high-speed Ethernet and new storage facilities forefficiency and flexibility), the Green Touch Panel Presenta-tion (panel discussion on when the energy crunch will affectoptical networks), and the Optical Internetworking Forum(panel discussions on market requirements for the next gener-ation optical network, considering 400G vs.1 Terabit speeds;and building blocks for high-speed, on-demand services).

THE OPTICAL BUSINESS FORUM – TUESDAY, MARCH 8Optical technologies have become integral to the solutions

needed to ensure the speed – and capacity – of transactionscan keep pace with the needs of customers in the businessworld. The Optical Business Forum will discuss the latestadvances in optical business services with sessions including:


CONFERENCE PREVIEW

OFC/NFOEC 2011: LEADING THE WAY IN OPTICAL COMMUNICATIONSBY LYNDSAY BASISTA

(Continued on page S6)

LYT-CONF PREVIEW-FEB 1/20/11 12:17 PM Page 30

•Who’s Buying Optical Bandwidth Services?•The Economics & Business Case for Connecting Data Cen-

ters•Carrier Ethernet Exchanges

Additional information can be found online:http://www.ofcnfoec.org/OBF

SERVICE PROVIDER SUMMIT – WEDNESDAY, MARCH 9Also in the exhibit hall is the perennial favorite Service

Provider Summit. The Service Provider Summit is gearedtoward CTOs, network architects, network designers and tech-nologists within the service provider and carrier sector. Theprogram will include a keynote presentation, exhibit time andnetworking time. Keynote Presentation: The Financial Industry’s Race to ZeroLatency and Terabit Networking – Andrew Bach, Senior VicePresident and Global Head of Network Services, NYSEEuronextPanels:•Evolution to Higher Speed•What’s Going on in Wireless?

Additional information can be found online:http://www.ofcnfoec.org/Service_Provider_Summit

MARKET WATCH – TUESDAY, MARCH 8 – THURSDAY,MARCH 10

This three-day series of panel sessions engages the applica-tions and business communities in the field of optical commu-nications. Presentations and panel discussions featureesteemed guest speakers from industry, research and theinvestment communities.Tuesday Panels:•State of the Optical Industry•Implications of Converged Wireline Wireless for Network

EvolutionWednesday Panel:•100G Ecosystem: Enabling Technology and EconomicsThursday Panels:•Data Center: Traffic and Technology Drivers•What’s Next for Optical Networking

More information can be found online: http://www.ofcn-foec.org/Market_Watch

SHORT COURSESShort Courses cover a broad range of topic areas at a variety

of educational levels. The courses are taught by highly regardedindustry experts on subjects such as 40 Gb/s transmission sys-tems, optical transmission systems, photonic integrated circuits,and ROADM technologies. New topics for 2011 include com-putercom interconnects and data center networking. More infor-mation on the Web: http://www.ofcnfoec. org/Short_Courses/

INVITED SPEAKERSOFC/NFOEC invited speakers are chosen through a highly selec-

tive nominations process to keep attendees at the forefront of opticalcommunications. This year’s exciting lineup of speakers will covervarious topics in 14 categories, from Optical Network Applicationsand Services to Transmission Subsystems and Network Elements.•Regulation Environments around the World: Impacts on

Deployments, Fabrice Bourgart, France Telecom, France.•Google Optical Network Deployment, Vijay Vusirikala,

Google, USA.•Energy-Efficient Optical Access Network Technologies,

Junichi Kani, NTT Access Service Systems Labs, Japan.•Cloud Computing over Telecom Network, Dominique

Verchere, Alcatel-Lucent Bell Labs France, France.•Scaling Networks in Large Data Centers, Donn Lee, Face-

book, USA.•Optical Networking Trends and Evolution, Christoph Glin-

gener, ADVA Optical Networking, Germany.A full list of invited speakers is available online:

http://www.ofcnfoec.org/Invited_Speakers

WORKSHOPSWorkshops and panel discussions will take place through-

out the conference and will cover all topical areas ofOFC/NFOEC. Some topics include:•Beyond 100G, Options and Implications for Today’s Net-

works TJ Xia, Verizon USA; Milorad Cvijetic, NEC andFrank Chang, Vitesse.

•FTTH Around the World: Today and Tomorrow ShoichiHanatani, Hitachi.

•FTTX and Technical Challenges of Emerging Applications:How Do We Keep Up with the Pace of Digital EvolutionDavid Li, Ligents Photonics.A full list of workshops and panel discussions is available

online at: http://www.ofcnfoec.org/Workshops_ and_Panels

CONFERENCE PREVIEW


(Continued from page S4)

LYT-CONF PREVIEW-FEB 1/20/11 12:17 PM Page 32

The annual European Conference on Optical Communi-cation (ECOC) is the largest conference on optical commu-nication in Europe, and one of the largest and mostprestigious events in this field worldwide. ECOC travelsfrom one European country to another each year and thisyear has visited Turin, Italy, for the first time, after the twoprevious successful ECOCs, 2008 in Brussels and 2009 inVienna. As a not-for-profit conference, ECOC focuses onthe dissemination of new research results, the education ofengineering and business leaders, and the exposition of cut-ting-edge optical fiber communication and networkingproducts in the associated exhibition. ECOC brings togeth-er professionals in several fields related to optical commu-nications, covering the areas of fiber design, opto-electronicdevices, optical transmission systems, optical transport net-works, and access technologies, all the way to future opticalrouting architectures and quantum information applica-tions. Most of the major international telecommunicationservice providers and optical network system vendors par-ticipate and present their most recent developments atECOC each year.

ECOC 2010 was held from 19 to 23 September in Turin,Italy, at the Lingotto Conference and Exhibition Center,former location of Fiat’ s first major car factory, builtbetween 1917 and 1920. Lingotto houses today a shoppingcenter, a hotel, an art gallery, a concert hall and a confer-ence center: a large modern structure designed especiallyfor conventions. The city of Turin is a major business andcultural center in northern Italy. Founded by the Romansover 2000 years ago, in 1861, it became the first capital ofItaly, playing a leading role in the nation’s history as a cityof many vocations: political, social, and cultural. Although itowes its recent development mainly to industry, today itoffers a variety of attractions: Roman ruins, Baroque monu-ments, urban landscapes, world-class museums — like theEgyptian Museum, featuring the second most important col-lection after that of Cairo — a vibrant cultural life, and sev-eral opportunities for sport and recreation. The region isalso renowned for its varied and refined cuisine, and thefamous wines from the surrounding area. For what concernsthe scientific history of the town, important contributions tothe early developments of fiber optic communications weremade by the CSELT laboratories of Telecom Italia, locatedin Turin. September 2010 was indeed the 33rd anniversary

of an important event in the history of optical communica-tions: in September 1977 the first optical communicationcable, called COS 2, was laid between two urban exchangesof SIP (now Telecom Italia).

ECOC 2010 was the 36th conference edition, and con-firmed ECOC’s reputation and attractiveness as one of theworld’s major forums for discussion of the most recentadvances in research, development, and industrial applicationsof optical communication technologies and networks. Thegeneral co-Chairmen of ECOC 2010 were Pierluigi Franco(PGT Photonics), Fabio Neri (Politecnico di Torino), andGiancarlo Prati (Scuola Superiore Sant’Anna and CentroNazionale Interuniversitario per le Telecomunicazioni —CNIT). The conference was organized by the Stilema compa-ny, based in Turin. ECOC 2010 was attended by 1111 dele-gates from all over the world; the countries with the largestrepresentation were Japan, the United States, Germany, andItaly. As in the ECOC tradition, a major exhibition was collo-cated with the conference, offering a showcase for the mostrecent advances in optical communications products. Threehundred companies worldwide participated as exhibitors, andthe countries with the largest representation among exhibitorswere China and the United States; 3775 exhibition visitorswere registered in addition to conference delegates. The con-ference was organized by Nexus Business Media, UnitedKingdom.

IEEE Communications Magazine • February 2011S8

ECOC 2010 CONFERENCE REPORT

36TH EUROPEAN CONFERENCE ON OPTICAL COMMUNICATIONBY FABIO NERI, POLITECNICO DI TORINO

Delegates at ECOC 2010.

Plenary session at the Lingotto Auditorium, designed by RenzoPiano.

The ECOC 2010 Technical Program co-Chairmen: P. Poggiolini,M. Schiano, and A. Galtarossa (left to right).

(Continued on page S10)

LYT-ECOC-Conference Report 1/20/11 12:09 PM Page 34


ECOC 2010 was sponsored by Telecom Italia, Cisco, Erics-son, Istituto Superiore Mario Boella, and ZTE, and by thelocal authorities and agencies Regione Piemonte, Camera diCommercio Artigianato e Agricoltura di Torino, UnioneIndustriale Torino, and Fondazione CRT.

A rich technical program was organized by technical pro-gram co-Chairmen Andrea Galtarossa (Università di Padova),Pierluigi Poggiolini (Politecnico di Torino), and Marco Schi-ano (Telecom Italia). The conference program started onSunday with 11 well attended (over 650 delegate badges werecollected on Sunday) half-day workshops. The opening sessionwas held in the Lingotto Auditorium, and featured threeworld-renowned plenary speakers from industry and academia:Menahem Kaplan (former CTO of Alcatel-Lucent Optics),with a talk on “Technology Opportunities Beyond and Besides100G,” Masataka Nagazawa (Tohoku University, Japan), witha talk on “Giant Leaps in Optical Communication Technolo-gies Towards 2030 and Beyond,” and John Bowers (Universityof California atSanta Barbara), with a talk on “Challenges inSilicon as a Photonic Platform.”

The bulk of the conference program was based on con-tributed technical papers, carefully selected for oral or posterpresentation by an outstanding technical program committee,comprising around 100 well-known experts of the field, andorganized in six subcommittees: “Fibers, Fiber Devices andAmplifiers,” “Waveguides and Optoelectronic Devices,” “Sub-systems and Network Elements for Optical Networks,”“Transmission Systems,” “Backbone and Core Networks,” and

“Access Networks and LANs.” The 245 oral and 131 posterpaper presentations were complemented by 38 invited talksfrom renowned experts, six tutorial presentations, and sevensymposia focused on hot topics, organized in 71 sessions run-ning in seven parallel tracks for the four conference days. Theposter session on Wednesday afternoon was very popular,with intense informal discussions among authors and otherconference delegates. The most recent research achievementsand breakthroughs in the field of optical communicationswere presented in post-deadline paper sessions at the end ofthe conference. The European Physical Society and CLEOEurope-EQEC organized within ECOC 2010 a special CLEOFocus Meeting on New Frontiers in Photonics, aimed atbridging the gap between basic science and optical telecom-munications applications. The CLEO Focus Meeting coveredfive sessions in the ECOC program.

Seven hundred ninety-three regular papers and 66 post-deadline papers were submitted to ECOC 2010. The accep-tance ratio was 51 percent for regular papers (33 percent fororal presentations and 18 percent for posters), and 27 percentfor post-deadline papers. Accepted ECOC 2010 papers arenow available on IEEE Xplore, thanks to the technical spon-sorship of the IEEE Photonics Society.

The ECOC 2010 conference organization has enriched thetechnical program with a wide range of social events and addi-tional services. A get-together cocktail was offered on Sundayevening, and a welcome reception was held at the LingottoConference Center on Monday evening, with around 1000participants. The gala dinner was organized at the La VenariaReale royal castle, part of the UNESCO World Heritage, andwas a memorable event for over 500 participants.

In the closing session, the token was passed to the organiz-ers of ECOC 2012 in Geneva, who offered chocolate to par-ticipants, representing the worldwide reputation ofSwitzerland in chocolate making.

ECOC 2010 confirmed the ECOC tradition of high-qualitytechnical contributions, and very active and interactive partici-pation of the lively research community on optical communi-cations and technologies.

The CLEO Focus Meeting, the ECOC conference, and theECOC exhibition together provided an exciting inter- andmultidisciplinary forum for people from basic research, R&D,industry, and telecom operators interested in optical commu-nications. The rich social program and conference servicescontributed to making ECOC 2010 a memorable edition.

More details on ECOC 2010 are available athttp://www.ecoc2010.org.

ECOC 2010 CONFERENCE REPORT

Welcome reception, “Taste of Piemonte.”

Gala dinner at La Venaria Reale royal castle.


LYT-ECOC-Conference Report 1/20/11 12:09 PM Page 36


ADVANCES IN PASSIVE OPTICAL NETWORKS

s an ultimate broadband access solution for futureInternet, the passive optical network (PON) brings

many advantages such as cost effectiveness, energy sav-ings, service transparency, and signal security over otherlast-/first-mile technologies. Over the past several years,we have witnessed significant development and deploy-ment of time-division multiple access (TDMA) PONssuch as IEEE 802.3ah Ethernet PONs (EPONs) and ITU-T G.984 Gigabit PONs (GPONs) to provide high-qualitytriple-play services for residential users. However, futureInternet applications, apart from triple-play service (e.g.,peer-to-peer [P2P] social networking, online video shar-ing, grid computing, and mobile Internet), along withtheir unique traffic characteristics and huge bandwidthrequirements, pose big challenges for current PON designand migration, which in turn are driving legacy TDMAPONs toward ultra-high-speed flexible next-generationPONs such as wavelength-division multiplexed (WDM)PONs and optical orthogonal frequency-division multi-plexed (OFDM) PONs, and/or a hybrid WDM/OFDM/TDM PON.

This special issue features recent and emergingadvances in PONs. Of the large number of submittedpapers, five were selected for this issue . The selected arti-cles cover topics including next-generation PON architec-ture, energy-efficient PONs, layer 2 medium access control(L2 MAC), quality of service (QoS) provisioning in futurePONs, and PON monitoring techniques. The first article,“Opportunities for Next Generation Optical Access,” co-authored by Dirk Breuer, Frank Geilhardt, Ralf Hülser-mann, Mario Kind, Christoph Lange, Thomas Monath,and Erik Weis, discusses the impact of the new businessmodels on network architecture based on the comparisonof different optical access network variants. It also pro-

vides perspective on access node consolidation for networkoperators.

One of the PON’s advantages is the potential to pro-vide high energy efficiency toward future green communi-cations. The second article, “Cost and EnergyConsumption Analysis of Advanced WDM-PONs” con-tributed by Klaus Grobe, Markus Roppelt, Achim Auten-rieth, Jörg-Peter Elbers, and Michael Eiselt, focuses onthe analysis of cost and energy-consumptions of futureadvanced WDM-PON options. The authors conclude thatit is essential to carefully clarify the requirements fornext-generation access with regard to per-PON clientcount and maximum reach. In particular, if client countdoes not exceed ~320, and a passive filter-based opticaldistribution network (ODN) is accepted, the most effi-cient solution, with regard to both cost and power con-sumption, is a simple WDM-PON. The article “TowardEnergy-Efficient 1G-EPON and 10G-EPON with Sleep-Aware MAC Control and Scheduling,” co-authored byJingjing Zhang and Nirwan Ansari, presents L2 tech-niques, proposing sleep-aware MAC control and schedul-ing approaches for EPON. Two sleep-mode control andsleep-aware scheduling schemes are analyzed: sleep forover one DBA cycle and sleep within one DBA cycle.

In addition to reducing energy consumption, multirateand multi-QoS provision is a critical feature next-genera-tion PONs shall possess naturally to cater for existing andemerging Internet applications. The article titled “Multi-rate and Multi-Quality-of-Service Passive Optical NetworkBased on Hybrid WDM/OCDM System” by HamzehBeyranvand and Jawad A. Salehi proposes a new schemeto guarantee multi-QoS in WDM/OCDM system. Thebasic idea is to use multilength variable-weight opticalorthogonal codes (MLVWOOC) as the signature sequenceof an OCDM system. The code weight and code length ofMLVWOOC are designed based on the characteristics ofthe requested classes of services.

The last article, “Passive Optical Network Monitoring:

A

GUEST EDITORIAL

MahmoudDaneshmand

Chonggang Wang Wei Wei

1 Please note that several more papers were accepted as the second part ofthe special issue and will be published in the September 2011 issue ofIEEE Communications Magazine. (Continued on page S14)

LYT-GUEST EDIT-Daneshmand 1/20/11 12:10 PM Page 38


Challenges and Requirements,” co-authored by Moham-mad M. Rad, Kerim Fouli, Habib A. Fathallah, Leslie A.Rusch, and Martin Maier, touches on a different problem.In addition to the discussion of challenges and require-ments for PON monitoring, it presents a comprehensivereview of techniques for in-service monitoring PONs todetect and localize faults. The authors recommend thehybrid techniques as promising solutions for delivering themaintenance and protection functionalities required bycurrent and next-generation PONs.

We would like to take this opportunity to thank ourreviewers for their effort in reviewing the manuscripts. Wealso thank the Editor-in-Chief, Dr. Steve Gorshe, for hissupportive guidance during the entire process.

BIOGRAPHIESMAHMOUD DANESHMAND ([email protected]) is a Distinguished Memberof Technical Staff, AT&T Labs Research; executive director of the UniversityCollaborations Program and assistant chief scientist of AT&T Labs; adjunctprofessor of computer science at the Stevens Institute of Technology; andadjunct professor of electrical engineering at Sharif University of Technolo-gy. He has more than 35 years of teaching, research and publications, andmanagement experience in academia and industry, including Bell Laborato-ries, AT&T Labs, the University of California at Berkeley, the University ofTexas at Austin, Tehran University, Sharif University of Technology, NationalUniversity of Iran, New York University, and Stevens Institute of Technolo-

gy. He has published more than 70 journal/conference papers and bookchapters, co-authored two books, given several keynote talks, and servedas general chair and TPC chair of many IEEE conferences. His current areasof teaching and research include artificial intelligence, knowledge discoveryand data mining, complex network analysis, sensor network and RFID sys-tem reliability and performance, and mining of sensor and RFID data. Hehas Ph.D. and M.A. degrees in statistics from the University of California,Berkeley, and M.S. and B.S. degrees in mathematics from the University ofTehran.

CHONGGANG WANG ([email protected]) is a senior staff engineer in InterDigi-tal Communications. Before joining InterDigital Communications, he con-ducted research with NEC Laboratories America, AT&T Labs Research, theUniversity of Arkansas, and Hong Kong University of Science and Technolo-gy. His research interests include future Internet, machine-to-machine(M2M) communications, and wireless networks. He has published morethan 80 journal/conference articles and book chapters. He is on the editori-al boards of IEEE Communications Magazine, IEEE Network, ACM/SpringerWireless Networks, and Wiley Wireless Communications and Mobile Com-puting. He has served as a TPC member for numerous IEEE conferencesincluding ICNP, INFOCOM, GLOBECOM, ICC, and WCNC. He received hisPh.D. in computer science from Beijing University of Posts and Telecommu-nications in 2002.

WEI WEI [SM] ([email protected]) is a senior engineer at Ciena Corpora-tion. Before joining Ciena, he conducted research with NEC LaboratoriesAmerica and the State University of New York at Buffalo. His research inter-ests include cognitive optical networks, network virtualization, and futureInternet. He has published more than 60 journal/conference articles andbook chapters. He also has rich engineering experiences in developing anddesigning broadband optical networks and IP networks. He is the holder ofthree patents with five others pending. He has served as a TPC member forseveral IEEE conferences including GLOBECOM and ICC. He received hisPh.D. degree in electrical engineering from Shanghai Jiao Tong University.

S14

GUEST EDITORIAL


LYT-GUEST EDIT-Daneshmand 1/20/11 12:10 PM Page 40

IEEE Communications Magazine • February 2011S16 0163-6804/11/$25.00 © 2011 IEEE

MOTIVATION

It is expected that in the near future an end userwill require much more guaranteed bandwidththan is available today [1]. There is a commonunderstanding that fiber to the home (FTTH)will overcome the bandwidth limitations oftoday’s copper-based and hybrid fiber accesssolutions (e.g., fiber to the cabinet [FTTCab]).FTTH is seen as the ultimate and most future-proof access solution. In consequence, thismeans building a completely new access net-work, thus requiring enormous investment andpotentially allowing new business models. In thelong run this will enable next-generation opticalaccess (NGOA) networks where the access net-work is not limited to the first mile but willpotentially further extend beyond today’s centraloffices (COs).

The target function when optimizing thestructure of such a new network is rather simple:satisfy all the needs of the customers along withminimized total costs of ownership (TCO) forbuilding and operation of the whole network.

When looking at a cost-optimal structure ofthe access/aggregation network, its structure canbe subdivided into different building blocks, andthe cost of the majority of the building blocksdepends significantly on the number of accesssites. This is shown in Fig. 1. In this model theaccess site is the demarcation point between theaccess network and the aggregation network pro-viding active network elements, which terminatethe access lines, aggregate the traffic, and forwardthe aggregated traffic via the aggregation networktoward the core network. Nevertheless, accessand aggregation networks can merge seamlesslywhen making use of appropriate optical-fiber-based network architectures and related systems.

Considering the different building blocks, thefollowing interdependencies can be seen.

The costs of the first mile and in-housecabling, used for connecting the customerpremises, are mostly driven by the number ofcustomer connections and the customer densityin a given area, and are independent of the num-ber of access sites.

The cost of the feeder links, which connectthe first mile with the access sites, are mostlydriven by the length of the feeder links, first dueto the building and material costs of the cables,and second due to rising costs of the used accessnetwork systems when increasing the reachrequirements. Since the length of the feeder linkswill decrease with an increasing number of accesssites, the costs of this link are inversely propor-tionally related to the number of access sites.

The cost of building and operating the accesssites scales with the number of sites. Whendecreasing the number of access sites, the meannumber of customers per site will increase, andin consequence the size of the traffic switchesinside the access sites can be increased.

The cost for maintaining the network equip-ment installed in the access sites is also relatedto the number of access sites: maintaining cen-tralized equipment installed in a small numberof access sites will cause smaller traveling timesand a smaller number of required maintenancepersonal compared to distributed equipmentinstalled in a higher number of access sites.

The costs of the aggregation links will increasewith an increasing number of access nodes. Inour model the aggregation link includes theegress interfaces of the switches inside the accessnodes, the transmission systems (typically wave-length-division multiplexing [WDM] systems),and the ingress interfaces of the core networkequipment. The mean number of customers con-nected to an access site increases with a decreas-ing number of access sites, and the number ofcustomers per switch will also increase. A highernumber of customers per switch will result in ahigher amount of aggregated traffic at the egressof the switch; therefore, the required capacity ofthe egress interfaces also increases.

The cost of these four building blocks con-tributes significantly to the total cost of theaccess/aggregation network. When analyzing thesum of the three cost types, the cost function hasa minimum defining the cost-optimal number ofaccess nodes. The characteristics of the cost

ABSTRACT

Next-generation optical access technologiesand architectures are evaluated based on opera-tors’ requirements. The study presented in thisarticle compares different FTTH access networkarchitectures. Additionally, the impact of newbusiness models on network architectures is dis-cussed.


Dirk Breuer, Frank Geilhardt, Ralf Hülsermann, Mario Kind, Christoph Lange, Thomas Monath, and Erik

Weis, Deutsche Telekom Laboratories

Opportunities for Next-GenerationOptical Access

BREUER LAYOUT 1/19/11 3:25 PM Page 42

IEEE Communications Magazine • February 2011 S17

function of each network segment are stronglyrelated to the deployed technology in the respec-tive network segment (e.g., copper lines vs. FTTxin the access network). Therefore, it is expectedthat the cost-optimal number of access nodes forNGOA will differ from the cost optimum in cop-per-based access networks.

Particular attention should be paid to the costsof the feeder link section since only the feederlink cost decreases with increasing number ofaccess sites. In copper-based access networks thecost of the feeder link section is strongly relatedto the pure material cost of the cables, whichdepends on the length and the cross-sectionalarea of the copper wires. The maximum transmis-sion distance is limited by the copper line loopresistance and can only be increased by increasingthe cross-sectional area of the copper wires orintroducing intermediate repeaters, which bothwould drive feeder link costs. In optical accessnetworks the situation is different: Due to thesuperior transmission characteristics of opticalfibers — high transmission bandwidth and lowloss — the interdependency between feeder cablelength and feeder section costs is much morerelaxed. Furthermore, point-to-multipoint opticalaccess network architectures provide multiplexingof several subscriber lines, which decreases thenumber of required feeder fibers significantlycompared to a copper-like point-to-point archi-tecture. Therefore, we expect that the cost-opti-mal number of access sites in optical accessnetworks will be below the number of access sitesin today’s copper-based access networks.

These cost considerations affect a number ofplayers who maintain an active interest in futureaccess networks. Even under competition, thedeployment and operation of broadband networksis an attractive business area for a still increasingnumber of players. This includes mobile as well asfixed operators, but also utility companies likeenergy providers; even construction companies areentering this business environment.

This article gives an overview of today’sFTTH approaches and the related economics,and an outlook toward network consolidation

evolution and the enabling optical access net-work technologies. In addition, the impact ofnew business models on network architectureand the requirements resulting from node con-solidation concepts are qualitatively discussed.

TODAY’S FTTH APPROACH

FIBER RICH VS. SHARED APPROACHFTTH networks can be deployed using differentarchitectures as shown in Fig. 2. In a point-to-point (PtP) architecture all subscribers are con-nected to an access node (e.g., CO) via dedicatedfibers. Today’s PtP deployments are mainly basedon Ethernet technology using Ethernet switchesin the access node with a high port density. Thenetwork termination at the subscriber site is real-ized with media converters (e.g., 100Base-TX ⇔100Base-BX) or mini-switches. The PtP architec-ture with its dedicated fiber connections requiresa very high number of fibers in the whole accessnetwork, which causes high costs for fiber rolloutand handling. In addition, each connectionrequires two interfaces, which cause high foot-print and power consumption.

In order to reduce the high number of fibersin the access network, point-to-multipoint(PtMP) architectures can be used. A PtMP archi-tecture offers one or more additional aggrega-tion layers between the subscriber location andCO. In general, two PtMP architectures can bedistinguished: the active optical network (AON)and passive optical network (PON).

The AON is determined by an active aggre-gation element (e.g., Ethernet switch) in the firstmile. Figure 2 shows two AON variations. In oneAON concept an Ethernet switch is located atthe street cabinet (Fig. 2a), whereas in the sec-ond variant an Ethernet switch is used at thebuilding location (Fig. 2b). On one hand theAON allows a reduction of the fiber count in theaccess network compared to a PtP solution, buton the other hand it is not able to decrease thenumber of required interfaces, so it is virtuallyimpossible to reduce the footprint and powerconsumption.

Figure 1. Principal building blocks of access/aggregation networks and corresponding cost structure as afunction of the number of access sites.

Total costs

Feeder linkcosts

Site costs

First mile costs

Aggregation link costs

Number of access sites

Cos

ts

Acc

ess

netw

ork

Agg

rega

tion

netw

ork

First mile

Feeder links

Aggregationlinks

Optical distribution network/customer premises

Accesssite

PoP

Accesssite

These cost

considerations affect

a number of players

who maintain an

active interest in

future access

networks. Even

under competition

the deployment and

operation of

broadband networks

is an attractive

business area for a

still increasing

number of players.



In contrast to an AON, PON aggregation isbased on passive components such as opticalpower splitters or WDM (de)multiplexers.Today, typically 32–64 subscribers are connectedto one PON port at the optical line termination(OLT). This means that a PON architectureenables fiber reduction as well as optimization ofthe footprint and power consumption comparedto a PtP architecture.

AN OPERATOR’S VIEW ON ECONOMICSThis analysis is based on the FTTH architecturesdepicted in Fig. 2. The scenarios have beenmodeled based on commercially available tech-nology from leading system vendors.

The PtP and AON scenarios are predicted onswitched Ethernet technology. The AON switchesin the first mile are connected to an Ethernetswitch at the CO via optical gigabit Ethernet links(1GbE or 10GbE). The Ethernet optical networktermination (E-ONT) at the subscriber site islinked to an AON switch via single-fiber Ethernetlines (100Base-BX, 1000Base-BX). Two transmis-sion data rates have been considered for the PtPand AON scenarios, 100 Mb/s and 1 Gb/s.

The PON scenario has been modeled with aG-PON system and a splitting ratio of 1:32,enabling a data rate of 2.5 Gb/s (downstream)and 1.25 Gb/s (upstream).

The results refer to a dense urban servicearea with one CO, 100 street cabinets, 2000buildings, and 16,000 subscribers. This servicearea is a brownfield area with a high number ofempty ducts that can be used for fiber rollout.

This analysis assumes a zero touch deploy-ment. This means that the fiber infrastructureand the system equipment of the first mile are

deployed at the beginning for final demand butthe ONTs and the system equipment of COlocation are considered demand-orientated. Allscenarios allow a guaranteed downstream datarate of 100 Mb/s. This means that maximum 25subscribers can be connected to one G-PON. Apeak data rate of 1000 Mb/s at the User Net-work Interface (UNI) of the ONT is supportedby the G-PON and PtP/AON solutions with atransmission data rate of 1 Gb/s. The PtP/AONscenarios with a transmission data rate of 100Mb/s allow a peak data rate of 100 Mb/s.

The techno-economical analysis considers thecost of the active system technology and the fiberinfrastructure including installation. The fiberinfrastructure takes into account the civil works(digging), fiber cables, optical passive splitters,optical distribution frame (ODF) at the CO, out-door cabinets, and power for the active technolo-gy in the field. The system technology has beenmodeled on the basis of the price informationprovided by the vendors. Table 1 shows, as anexample, the relative price information for theG-PON and PtP/AON solutions with a transmis-sion data rate of 1 Gb/s. The difference betweenprices for the same network element from differ-ent vendors (e.g., pluggable 10GbE transceiver)can be explained by various business strategies.

Figure 3a shows the total cost per line overthe number of subscribers for the FTTH archi-tectures described in Fig. 2. The results are nor-malized to the total cost per line of the G-PONsolution at the end of the rollout (16,000 sub-scribers).

The chart shows that the G-PON results inthe lowest total cost per line independent of thenumber of connected subscribers. Especially in

Figure 2. FTTH architectures and reference points.

1. PtP

2a. AON

2b. AON

3. PON

4. Optical access and node consolidation

Cab CO Metro access nodePoP

Home/building

Inhouse

1 or 2

1 or 2

1

1

1

First mile network Feeder network Aggr. network

OLT

OLT/switch / ...

Powersplitter

e.g. GPON

Splitter/switch/...

ONTONTONT

ONTONTONT

ONTONTONT

ONTONTONT

ONTONTONT

Not considered in theeconomic analysis

This analysis assumes

a zero touch

deployment. This

means that the fiber

infrastructure and

the system

equipment of the

first mile are

deployed at the

beginning for final

demand but the

ONTs and the system

equipment of CO

location are

considered demand-

orientated.



Table 1. Relative price information.

Network element Description Relative price

GPON

ONT Data only 1.00

CO OLT GPON OLT with 16 PON card slots —

Basic costs incl. chassis, fan, power supply, switch fabric 78.67

Optics uplink 10GBASE-LR X2 module 11.00

Line card uplink 2 × 10GbE 6.91

PON card 4 × G-PON ports incl. class B optics 80.00

PtP with GbE interface

ONT Data only 0.87

CO switch Ethernet switch with 8 Line card slots —

Basic costs Incl. chassis, fan, power supply, switch fabric 179.66

Optics uplink 10GBASE-LR X2 Module 20.26


Optics downlink 1000BASE-BX 6.58

Line card downlink 48 × 1000BASE-X 83.53

AON with Cab switch and GbE interface

PtP ONT Data only 0.87

CO switch Ethernet switch with 8 line card slots —


Optics (up and down) 10GBASE-LR X2 Module 20.26

Line card 4 × 10GbE 101.28

Cab switch Ethernet switch with 5 Line card slots —




Optics uplink 10GBASE-LR X2 module 20.26

AON with basement switch and GbE interface

PtP ONT Data only 0.87

CO switch Ethernet switch with 11 line card slots —


Optics uplink 10GBASE-LR X2 Module 20.26




Basement switch Ethernet switch —

Basic costs 12-port 1000BASEs-X Ethernet switch 40.48

Optics (up & down) 1000BASE-BX 6.58

The fiber infra-

structure takes into

account the civil

works, fiber cables,

optical passive split-

ters, optical distribu-

tion frame (ODF) at

the CO, the outdoor

cabinets and the

power for the active

technology in the

field. The system

technology has been

modeled on the

basis of the price

information provided

by the vendors.



the economically sensitive initial deploymentperiod with a low number of subscribers, thetotal cost per line decreases very fast due to thehigh level of sharing of the PON architecture.Though the system costs of the PtP solutions arelower for small subscriber numbers (less than250–750 users per service area), the total cost arehigher due to the infrastructure effort (fibers).

For 100 percent coverage the total cost per lineof the AON scenario with an Ethernet switch atthe street cabinet (100 Mb/s transmission rate) isabout 1.5 times more expensive than the G-PONsolution. The total cost per line of the AON sce-nario with an Ethernet switch at the street cabinetand a transmission rate of 1 Gb/s is almost 2.3times more expensive than the G-PON variant.

The worst result has been calculated for theAON scenario with Ethernet switch at the build-ing location (Fig. 2b). For 100 percent coveragethe total costs per line are about 2.7 times morecostly than the G-PON solution. This is mainlycaused by the high cost of the building switchthat has to fulfill network operator require-ments. The cost of this switch cannot be com-pared with the cost of a simple LAN switch.

At the beginning of the FTTH deploymentwith a low number of subscribers the PtP solutionhas a lower total cost per line than the AON vari-ant because the AON equipment in the field caus-es very high initial investment due to zero touchrollout. The optical interfaces of the PtP solutionhave been considered demand-orientated, but theequipment of the AON has been deployed at thebeginning for 100 percent coverage.

Another important aspect within the compar-ison of different FTTH architectures is the ener-gy consumption, since over the lifetime of adeployed technology it is a major contribution tothe operational expenditures and also has adirect environmental impact. The energy con-

sumption has been analyzed for the consideredFTTH architectures (Fig. 2). It has been mod-eled on the basis of the power consumption perport for different interface types and networkelements, as shown in Table 2. These valueshave been extracted from vendor data sheets. Itincludes the energy consumption of thetransceiver and the appropriate portion of theline card and basic components (chassis, switchfabric, uplink, etc.). The difference between thepower consumption of interfaces with the sametype can be explained by the fact that a nodewith a high port density is more energy efficientthan a node with a low port density.

Figure 3b shows the maximum energy con-sumption of the different FTTH architecturesfor one dense urban service area with 100 per-cent coverage, neglecting the ONT at the cus-tomer site. It is assumed that all subscribers areconnected.

The PtP deployment with a Gigabit-Ethernetcustomer interface was used as a reference case.The GPON solution needs about 84 percent lessenergy compared to the reference. It is also evi-dent from Fig. 3b that a PtP deployment with areduced bandwidth of 100 Mb/s would not leadto significant power savings. In the AON sce-nario the power consumption would evenincrease due to field deployed active switch tech-nology. Also, in the AON case a speed reductionhas almost no influence on power saving.

EVOLUTION TOWARDNETWORK CONSOLIDATION

TOPOLOGY SCENARIOSIn the case of reducing the total numbers ofaccess sites when deploying optical access net-work technologies, new (NGOA) service areas

Figure 3. a) Total costs per line for different FTTH architectures; b) energy consumption of different FTTH variants.

1 2

20

P/kW

03

1. PtP with GbE interface (UNI)2. PtP with 100BT interface3. AON with cab switch and GbE interface4. AON with cab switch and 100BT interface5. AON with basement switch and GbE interface6. G-PON with GbE interface

4

Power consumption of FTTH variants in a denseurban service area (without ONT)

5 6

Building (basement)Cabinet (air conditioning)CabinetCO

Subscribers

Total costs per line (equipment + infrastructure)

2000

1

0

2

3

4

5

6

40

60

80

100

120

140

160

0 4000 6000 8000 10,000 12,000 14,000 16,000

AON with cab switch and GbE interfaceAON with cab switch and 100BT interfaceAON with basement switch and GbE interfacePtP with 100BT interfacePtP with GbE interfaceG-PON



have to be identified according to a number ofboundary conditions:• Economic considerations: A cost-optimal

cut of service areas has to be identified.• Maximum reach of access network tech-

nologies.• Limitation of the number of fibers that can

be terminated in the remaining CO build-ings. This refers mainly to the availablecapacity of the house lead-ins and feedingtrunks as well as the available floor spacefor the optical distribution frame and activeequipment, mainly the OLT devices.

• Resilience requirements: The maximumnumber of subscribers connected to a cer-tain CO may be limited.To identify potential new service areas by

consolidation of some existing traditional serviceareas, we used a clustering algorithm for devel-oping four exemplary network scenarios differingin the number of NGOA service areas. We useda reference model for Germany, and in all sce-narios we started from 8000 COs and reducedthe remaining number to 500, 1000, 2000, and4000 COnew, respectively. It is obvious that otherrelated parameters like the number of house-holds connected to a certain access node or thecovered area per NGOA service area vary in thedifferent scenarios too.

The whole NGOA service area is served bythe COnew, which, however, now comprises sev-eral traditional service areas, one directly linkedto the COnew and the others remotely linked tothe COnew by elongated feeder links. The feed-er links are interconnecting the former COs inremote service areas to the COnew. The dis-tance for this feeder fiber depends on the num-ber of COnew and with this on the degree ofCO consolidation, and is up to about 40 km inmost cases for working and backup paths (Fig.4a). Figure 4a shows the feeder length anddemand as a function of the degree of nodeconsolidation, whereby solid lines represent theshortest working path and dashed lines showthe shortest protection path, respectively. Dis-tances above 40 km are observed only in thecase of a very high degree of node consolida-tion (< 500 COnew/> 95 percent node consoli-dation) and when considering the most distantcustomer premises (95 percent quantile). Sincein Fig. 4a only the additional feeder link length,caused by the consolidation of the CO sites, isshown, the length of the subscriber lines in thefirst mile (inside the traditional service area)must be added when estimating the totalrequired reach budget of the access line system.In Germany the subscriber line length is typi-cally below 5 km.

Besides the length of the feeder links also thefiber demand in the feeder link section is animport evaluation criterion since this parameterimpacts the costs of the cable infrastructure. InFig. 4b the amount of feeder fibers needed forconnecting 100 percent of all households isshown for all scenarios assuming bidirectionaltransmission on a single fiber and different split-ting ratios. We use the splitting ratio 1:1 (point-to-point systems), 1:32 (widely used GPONsplitting ratio) as well as 1:512 (feasible splittingratios for NGOA systems). The ordinate is in

logarithmic scale and it is obvious that the fiberdemand for p-t-p systems compared to systemsproviding 1:512 splitting ratio is increased bymore than two orders of magnitude.

POTENTIAL TECHNOLOGIESTechnologies for NGOA networks must be ableto connect subscribers by optical fibers over typi-cal distances ranging from several hundredmeters up to about 60 km or even higher, takinginto account protection scenarios at high bitrates and high splitting ratios of up to 1:1024.There are different options to meet theserequirements: dedicated point-to-point fiberlinks from the CO to every subscriber, AONswith intermediate active equipment in the field,and PONs relying on a fully passive optical out-side plant with power splitters. Additionally,these options can be realized in different waysregarding topology, architecture, and used tech-nology [2]. Incumbent network operators arevery much in favor of PON-based FTTH accessnetworks due to their cost advantages in large-scale network deployments with inherited pas-sive infrastructure (proved also by the detailedcost considerations for current PON architec-tures in the previous section).

In the case of PONs, the optical fiber as thetransmission medium is shared between multipleusers. There are several principal options forensuring the necessary multi-user access: time-division multiple access (TDMA), frequency-division multiple access (FDMA), andcode-division multiple access (CDMA), wherebyfrom the principal point of view FDMA encom-passes both wavelength-division multiple access(WDMA) and orthogonal FDMA (OFDMA) asoptical technologies discussed in the PON envi-ronment. In addition, besides the multi-useraccess property, some of these options can beseen as efficient modulation formats, as current-ly discussed in case of OFDM(A).

Today’s gigabit-class PON technology is basedon time-division multiplexing (TDM) and a pas-sive power splitter in the field in order to collecta number of subscribers into a single OLT port.Higher-rate TDM-PON systems have been con-sidered for the next generation of access net-works, and they rely on the same opticaldistribution network as the gigabit-class PONsystems. PON systems with a downstream data

Table 2. Assumed energy consumption per interface.

Interface type Network element Portdensity

Energy consumptionper port (W)

1000BASE-BX CO switch High 4.4

1000BASE-BX Cabinet switch Medium 4.8

1000BASE-BX Building switch Low 6.7

100BASE-BX CO switch High 4.3

100BASE-BX Cabinet switch Medium 4.8

G-PON-OLT CO switch Low 22.3



rate of 10 Gb/s are standardized by IEEE andfull-service access network (FSAN)/InternationalTelecommunication Union — Telecommunica-tion Standardization Sector (ITU-T) [3, 4]. The10 Gigabit Ethernet PON (10G-EPON) specifi-cation has been finalized by IEEE task group802.3av, and the 10 Gigabit PON (XG-PON) [5]has been specified by the ITU-T within theG.987 Recommendation series. Long-reachextensions of TDM-type PONs utilize intermedi-ate reach extenders amplifying or regeneratingthe signal; for example, amplified long-reachGPON or 10G-PONs systems have been demon-strated.

Another option for obtaining longer reachare WDM-type PONs partly relying on arrayedwaveguide gratings (AWGs) as distribution ele-ments with less attenuation — compared topower splitters — in the field. However, whetherWDM PONs may reach resource allocation flex-ibility and necessary cost targets for mass marketdeployments — compared to TDM-PONs — isquestionable from today’s point of view.

Hybrid PONs combining WDM and TDMapproaches seem to be promising solutions inorder to obtain long-reach high-rate and high-split access systems. Further interesting andpromising PON approaches currently underinvestigation are OFDM PONs and OCDMPONs (optical CDMA). OFDM and OCDMPONs are intensely discussed in the researchcommunity. They are in an early stage of devel-opment and laboratory tests using demonstratorsetups have been reported.

In conclusion, there are different and verypromising technology solutions for optical accesssystems: The main question is how these differ-ent access technology options may suit the oper-ator’s needs to provide cost-efficient and reliablebroadband access over long periods of time.

IMPACT OF NEW BUSINESS MODELS

GENERAL BUSINESS ENVIRONMENT

NGOA, as it is defined in this article, assumes aFTTH network. In practice, this might be not theavailable network for the upcoming years. Accord-ing to market data from the Organization forEconomic Cooperation and Development(OECD), current digital subscriber line (DSL)coverage is about 88 percent related to the popu-lation [6]. The same report outlines that the aver-age coverage for broadband cable is about 60percent, but this differs across countries inEurope, and not all cable networks are upgradedyet. In addition, mobile, wireless, and satellitenetworks are present today with coverage of 80percent. Fiber networks are, besides some coun-tries, increasing but not available to a large partof the population. So starting from today, thereare a number of networks available, and fiber willhave to increase market share in order to becomethe relevant base infrastructure for NGOA.

Comparing in detail the situation of today withthe announcements, again a mixed view can befound. Most incumbents in large European coun-tries have connected only few households withfiber. BT wants to connect 2.5 million homes by2012, Deutsche Telekom announced 4 millionhomes for the same time frame, and TelecomItalia 13 percent of the households FTTx in 2013and 15 percent FTTH in the long term, just toname some of the plans. In addition, a number ofsmaller players will start to deploy their own net-works, but those numbers are so far relativelysmall compared with total population. Whileextrapolating these announcements to the 2015 or2020 time frame (under the assumption that thetargets will by accomplished), the authors assumethat 20–30 percent of all households will be cov-ered by fiber in 2015 and 30–50 percent in 2020.

Figure 4. a) Feeder link length; b) feeder fiber demand as functions of the access network node consolidation degree.

Number of metro access nodes

Feeder link length (related to number of households)

500

98.75%

2.5

Feed

er li

nk le

ngth

[km

]

1

5

7.510

25

5075

100

250

100

90% 80% 70% 60% 50%

1000 1500 2000

Consolidation degree γ

2500 3000 3500 4000 4500

Number of metro access nodes

500

98.75%

Feed

er f

iber

dem

and

[km

]

109

108

107

106

105

100

90% 80% 70% 60% 50%

1000 1500 2000

Consolidation degree γ

2500 3000 3500 4000 4500

Feeder fiber demand

1:1 splitting ratio1:32 splitting ratio1:512 splitting ratioShortest working pathShortest disjoint backup path

Average values80% quantile95% quantileShortest working pathShortest disjoint backup path



Political interest and regulation are demand-ing certain coverage, and this might increase thedeployment speed [7].

Other important aspects are changes in thebusiness relationships. In copper networks theincumbent has built the network and was forcedeither to provide appropriate wholesale productsand/or to open the infrastructure. Nonetheless, theonly available systems and infrastructure were pro-vided by one player. According to the announce-ments, the infrastructure business will be moredivided in the future, not only in terms of technol-ogy but also in terms of ownership. So the businessenvironment might need new rules and mecha-nisms to work together and coexist. This willinclude organization of everyday processes likeprovisioning and forums with the aim to developstandards for the technology to be deployed.

For example, in Sweden alternative networkoperators have already established such an orga-nization, the Swedish Urban Network Associa-tion; in the United Kingdom BT Openreach isoperating a platform called the EquivalenceManagement Platform (EMP: Openreach’s trad-ing platform for Local Loop Unbundling). Inaddition, the trust relationship between the newalternative operators and the incumbents willchange as the incumbent will be no longer theprimary infrastructure provider. Overall, it canbe seen that the situation is split in a number ofaspects and directions, and a business frameworkfor cooperation in NGOA networks is needed.

BUSINESS CASE ASPECTS FOR ANGOA PROVIDER

The previous section outlined that there will becompetition by different infrastructures and tech-nologies. On the other hand, the active infra-structure of xDSL is no longer provided by a singleplayer on a per country basis, and it can be assumedthat this competition will still be present in theNGOA case, at least when a player reaches a signif-icant market power and regulation comes into play.An important aspect is here cooperation whetherforced, in the case of regulation, or on an openbasis, like sharing active infrastructure with otherproviders. In any case, this demands a business rela-tionship, which provides trust between the differentparties in order to work together on a well-knownbasis. In addition, competition might be indirectlyincreased by infrastructure providers, who have thehighest capital expenditures while deploying opticalnetworks, and are looking to share these costs andretain revenues as fast as possible.

Extending the viewpoint from an NGOAprovider to a combined service and/or infra-structure provider/operator (like all incumbents aretypically), the analysis will become even more com-plicated and will depend more on country specifics.

DISCUSSION OFNGOA REQUIREMENTS AND

ECONOMIC OUTLOOK

TECHNICAL REQUIREMENTS

NGOA networks are intended to allow opera-tors to bring down the network and productioncosts while at the same time ensuring high net-

work quality, availability, security, and signifi-cantly increased bandwidth per user. A numberof requirements are aligned with these aspects.It is expected that the data rate demand willcontinuously grow over the next decades [8].This will drive the peak rates per user to at least1 Gb/s or even more and the committed datarate to above 0.3 Gb/s, and will force much moresymmetry between downstream and upstreamtraffic [1].

As shown, structural network changes will bethe key to optimize the network and to bringdown costs. Merging access and aggregation net-works into a simplified network an NGOA net-work will lead to cost savings resulting formbetter utilization of network resources (e.g.,interfaces, fibers, aggregation nodes), reducingsignificantly the number of aggregation networkelements, thus avoiding expensive signal adapta-tions. This restructuring of the network, howev-er, will require much higher access reach up to100 km, and a high customer concentration perfiber (e.g., 1024). Subsequently per cable routeeffective redundancy concepts and protectionmechanisms for service and network availabilitywill be required.

Over the next years significant FTTH deploy-ments are expected. NGOA systems have towork on existing first mile fiber infrastructureswithout requiring changes on the deployed infra-structure and deployed components. Migrationto the NGOA network should not affect runningservices, already deployed systems and usedspectra.

Another key to enable further cost savings iscommon access for residential customers, busi-ness customers (small and medium enterprises),and mobile radio backhaul on a single NGOAnetwork avoiding multiplied network operationsand network resources is required. Efficient useof network resources requires furthermore suit-able quality of service (QoS) mechanisms,resource control, and management functions toaddress the requirements of different user typesand the mobile backhaul as well as functionsenabling efficient content distribution (e.g., mul-ticast).

A huge effort is to bring down network oper-ation cost. For the systems and architecture part,much higher energy efficiency is expected inNGOA networks. For the network operationitself, a zero touch network is expected, withhigh automation and support functions for basicoperation processes like provisioning, mainte-nance, fault, and management needs to be devel-oped that minimize manual switching efforts andincrease process efficiency. A customer networktermination (NT), for example, should allow do-it-yourself (DIY) installation (plug and play) andbe customer-unspecific (colorless) to simplifylogistics. Changes in the service portfolio or theaccess product should be done through a self-service approach enabled by auto-configurationfunctions in the network. Functions for easy,fast, and efficient maintainability and restorabili-ty are needed (e.g., seamless software upgradewithout service interruption, end-to-end service/traffic performance monitoring per customerand service). Resiliency including automaticreconnections through redundant network con-

NGOA networks are

intended to allow

the operators to

bring down the net-

work and production

costs while ensuring

at the same time

high network quality,

availability, security,

and significantly

increased bandwidth

per user. A number

of requirements are

aligned with these

aspects.



cepts should minimize the effect of failures, butto prevent and clear failures, suitable fault man-agement supported by optical diagnosis andmeasurement solutions for fault detection andlocalization up to the home are necessary.

There are a variety of requirements forNGOA networks. Therefore, the challenge is todefine and select the requirements in a way tofind the right balance, enabling cost optimizationand establishing a network with minimum TCO.

BUSINESS CONSEQUENCES FORTECHNICAL ASPECTS

With respect to technology, there are a lot ofoptions to establish next-generation access net-works. In this section some points are highlight-ed that are related to the interrelationshipbetween the business and technology aspects.

Provided that there is competition and openaccess in next-generation access networks, it willbe important to consider the market share andthe related number of customers in order to fillpossible long-reach PONs with high splittingratios. Technically, it would be possible to extendthe access up to 100 km, but for business needsand the open access model, new interfaces —possibly active — are necessary and may be con-tradictory to the node consolidation approach.These are important for establishing and operat-ing access networks in changing business envi-ronments cost efficiently.

From the operators’ point of view, one mainrequirement is upgrading either bandwidth orsystems on existing fiber infrastructure as long aspossible since infrastructure investment is thebig bulk. Another point is changing systemsunder almost running conditions. This needs tobe considered from the beginning.

In consequence, cost efficiency in networkinstallation and operation is key for networkoperators. It requires finding a balance betweenpure technical and business problems.

Bringing all these problems together is stillopen from today’s point of view, and requiresfurther research and in-depth investigationswhich will be conducted to a certain extent incurrent FP 7 projects like OASE.

ACKNOWLEDGMENTSThe work leading to these results has receivedfunding from the European Union’s SeventhFramework Program (FP7 2007/2013) undergrant agreement no. 249025 (project: OpticalAccess Seamless Evolution — OASE).

REFERENCES[1] Analysis Mason, “Fibre Capacity Limitations in Access

Networks,” report for OFCOM, Jan. 2010.[2] Optical Access Seamless Evolution (OASE), “Survey of

NGOA system Concepts,” FP7/2007–2013, deliv. D4.1;http://www.ict-oase.eu/.

[3] J. Kani et al., “Next-Generation PON Part I — Technolo-gy Roadmap and General Requirements,” IEEE Com-mun. Mag., vol. 47, no. 11, 2009, pp. 43–49.

[4] F. Effenberger et al., “Next-Generation PON Part II —Candidate Systems for Next Generation PON,” IEEECommun. Mag., vol. 47, no. 11, 2009, pp. 50–57.

[5] F. Effenberger et al., “Next-Generation PON Part II —System Specifications for XG-PON,” IEEE Commun.Mag., vol. 47, no. 11, 2009, pp. 58–64.

[6] A. Díaz-Pinés, “Indicators of Broadband Coverage,”OECD, DSTI/ICCP/CISP(2009)3/FINAL.

[7] Booz & Company, “NGNBN Case Studies — Next Gener-ation National Broadband Network Country Profiles,”July 2009; http://www.booz.com/media/file/NGNBN-Country-Profiles.pdf

[8] Cisco Systems, “Cisco Visual Networking Index: Forecastand Methodology, 2009–2014,” San Jose, CA, 2009;http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-481360.pdf.

BIOGRAPHIESDIRK BREUER ([email protected]) received Dipl.-Ing. andDr.-Ing. degrees in electrical engineering from the TechnicalUniversity of Berlin in 1993 and 1999, respectively. Sincejoining Deutsche Telekom Laboratories, he has mainly beenconcerned with developing optimization strategies for theoptical transport network of Deutsche Telekom. In recentyears he is mainly involved in developing upgrade strategiestoward next-generation broadband access networks.

FRANK GEILHARDT ([email protected]) received aDipl.-Ing. (FH) degree in telecommunication engineeringfrom the University of Applied Sciences of Berlin in 2001.Afterward he started to work in the Fixed Access NetworkGroup of T-Systems Nova GmbH that was merged into T-Systems Enterprise Services GmbH. In 2009 he joinedDeutsche Telekom Laboratories in Berlin. Since joining T-Systems he was mainly concerned with access networkevolution including techno-economic assessments as wellas the operation of access technology demonstrators.

RALF HÜLSERMANN ([email protected]) received aDipl.-Ing. (FH) degree in telecommunication engineeringfrom the University of Applied Sciences of Leipzig in 2001.Since joining Deutsche Telekom AG in 2001 he has beenconcerned with architectures for optical transport net-works. Currently he is with Deutsche Telekom Laboratorieswhere he is mainly involved in planning and optimizingoptical access networks with focus on techno-economicassessment.

MARIO KIND ([email protected]) received a Dipl.-Inf(FH) degree in communications and information technolo-gy at Deutsche Telekom Hochschule für TelekommunikationLeipzig (FH), University of Applied Sciences (HfTL). He isemployed as an expert in the area of broadband accessnetworks. His main working area is the economic evalua-tion of business, technology, and service trends in theInternet, telecommunication, and broadcast industry. He isauthor or co-author of several papers published in interna-tional telecommunications conferences and journals.

CHRISTOPH LANGE ([email protected]) is withDeutsche Telekom Laboratories, the research departmentof Deutsche Telekom AG, Berlin, Germany. He received aDipl.-Ing. degree (diploma) in electrical engineering andDr.-Ing. degree (Ph.D.) in communications engineeringfrom the University of Rostock, Germany, in 1998 and2003, respectively. His current research interests includebroadband networks, emphasizing access network topicsas well as the energy consumption and sustainability oftelecommunication networks.

THOMAS MONATH ([email protected]) received hisDiplom-Ingenieur (M.S.) degree in communication engi-neering from the University of Rostock in 1997. After fin-ishing studies he joined Deutsche Telekom. He is workingas a senior expert in techno-economics of telecommunica-tion networks and as a project Leader (PMP) of strategicaccess network evolution projects. He has been involved inseveral European projects of EURESCOM, ACTS, and ISTfocused on broadband access network evolution. He isauthor or co-author of several papers published in interna-tional telecommunications conferences and journals.

ERIK WEIS ([email protected]) received a Dipl.-Ing.degree in electrical engineering from the Technical Univer-sity of Dresden. He joined the Research Institute ofDeutsche Telekom in 1997, where he was involved innational and international R&D projects on optical andhybrid optical broadband access networks. His currentresearch interests include developing upgrade strategiesand concepts toward next-generation broadband opticalaccess networks.

There are a variety of

requirements on

NGOA networks.

Therefore the

challenge is to define

and select the

requirements in a

way to find the right

balance enabling a

cost optimization

and establishing a

network with

minimum total cost

of ownership.


IEEE Communications Magazine • February 2011 S250163-6804/11/$25.00 © 2011 IEEE

INTRODUCTION

Residential access bandwidth is ever increasing,and today, no change in this trend can be seen.Future applications that can easily fill band-widths in excess of (several) 100 Mb/s consist ofa combination of ultra-high-definition three-dimensional TV with time-shifted or peer-to-peer unicast. It can already be predicted thateven fiber access based on Ethernet/gigabit-capable passive optical network (EPON/GPON)and their successors, 10G-EPON and XG-PON1,will reach its limit and is not suitable for thatkind of application. So a new generation ofbroadband access systems will be required. Theseaccess systems are also referred to as next-gener-ation PON, or NG-PON2 in full-service accessnetwork (FSAN) terminology [1] and are likelyto be more than five years away from now. Dur-ing this time, active site reduction and networkconsolidation will lead to larger reach require-ments for the access and backhaul systems, withremaining sites having to serve a significantlylarger number of customers [2]. The new tech-nology will likely be based on wavelength-divi-sion multiplexing (WDM), possibly inconjunction with suitable wavelength sharingschemes.

Following cost expectations and energy-con-sumption predictions [3], much focus with regardto next-generation access systems is put onadvanced PON concepts.

There is, however, much uncertainty aboutthe respective ranges of these requirements. The

different requirements regarding reach andclient count as well as the requirements withrespect to the optical distribution network(ODN) have severe impact on the potential cap-ital expenditures (CapEx) and also on the opera-tional expenditures (OpEx) of the resultingsystems solutions. This is discussed in detail inthe next sections. Regarding OpEx, we concen-trate on energy consumption, as this constitutesa large part of these costs. Other differences inoperation expenses resulting from operationsand maintenance are hard to quantify, and mightnot even be fully appreciated by the networkoperators.

BROADBAND APPLICATIONSToday, there is much debate about a usefulbroadband definition for residential access.Sometimes, bandwidths starting at 1 Mb/s arereferred to as broadband, sometimes this isincreased to the range of 10 Mb/s, but some-times, a definition of 256 kb/s has been used [4].It is obvious that the broadband definition is fre-quently adopted toward increased bandwidthvalues. This discussion often leaves the down-stream (central office or point of presence tocustomer) vs. upstream (the counter-direction)asymmetry unconsidered.

Here, we refer to broadband access as aninfrastructure which is able to potentially scale tobeyond 1 Gb/s of sustained bandwidth per resi-dential customer. Oversubscription, or statisticalmultiplex bandwidth gain, can be applied in adedicated aggregation layer. This aggregationmay be implemented at the edge to the backhauland core parts of the network and can lead tosignificantly decreased sustained bandwidths.However, the transport and multiple-access partsof the access infrastructure should still enablesustained bandwidths in the range of 1 Gb/s.

On a broad scale, bandwidths beyond, say,sustained 100 Mb/s should be provided via afiber-based infrastructure (fiber to the home[FTTH] or fiber to the building [FTTB]), mostlydue to cost and energy consumption reasons [3].FTTH and FTTB require a massive, very costlyinfrastructure to be overbuilt. Hence, the next-generation infrastructure in turn must not belimited to anything below 1 Gb/s for reasons ofinvestment security.

ABSTRACT

Next-generation access systems will have toprovide bandwidths in excess of 100 Mb/s perresidential customer, in conjunction with highcustomer count and high maximum reach. Poten-tial systems solutions include several variants ofWDM-PONs. These systems, however, differ sig-nificantly in their cost (capital expenditures) andenergy consumption potential. We compare sev-eral WDM-PON concepts, including hybridWDM-PON with integrated per-wavelength mul-tiple access, with regard to these parameters. Wealso show the impact and importance of genericnext-generation bandwidth and reach require-ments.


Klaus Grobe, Markus Roppelt, Achim Autenrieth, Jörg-Peter Elbers, and Michael Eiselt,

ADVA AG Optical Networking

Cost and Energy Consumption Analysisof Advanced WDM-PONs

GROBE LAYOUT 1/19/11 3:40 PM Page 51


It is easy to envisage future services that canfill these bandwidths. This even holds for sym-metric downstream-upstream scenarios. Fromtoday’s viewpoint, one application in particularcan be identified which could make use of band-widths approaching, and possibly exceeding, 1Gb/s — unicast highest-resolution video, in con-junction with symmetric-bandwidth peer-to-peerapplications.

Today’s high-definition television (HDTV)requires bit rates between 8 and 15 Mb/s,depending on the compression efficiency (of theMoving Pictures Experts Group’ MPEG4 orInternational Organization for Standards/Inter-national Electrotechnical Commission[ISO/IEC] 14496 standard). The next stepbeyond HDTV is already clear — ultra-HDTV(UHDTV), with a resolution of today 3840 ×2160, in future 7680 × 4320 pixels, with 16:9 for-mat, 60 Hz frame rate, and 22.2 audio (24-chan-nel audio, including two subwoofer channels).These formats are also referred to as 4k and 8kUHDTV, respectively. They have been devel-oped by the Japanese TV broadcast companyNHK, together with the consumer systemsindustry. According to [5], it is likely that the 8kformat will require up to 200 Mb/s per channel(72 Gb/s uncompressed). Furthermore, 3D ver-sions of UHDTV may require up to 180 percentincreased bit rate compared to 2D. Combinedwith time-shifted unicast or even peer-to-peerapplications, these formats will then fill band-widths up to the 400 Mb/s range. They will haveto be provided — in the busy hour — via sus-tained bandwidth, without further oversubscrip-tion in the access network. The respective(downstream) UHDTV bandwidth will need tobe at least twice that per typical household,assuming two parallel independent downstreamor download sessions. Concerning the upstreamdirection, a similar bandwidth may be requiredin peer-to-peer applications. Here, a chicken-egg problem comes into play: we do not seebroadband peer-to-peer applications today,which is likely due to the fact that the requiredupstream bandwidth is not available in the

majority of today’s access installations. Once thebandwidth is provided, it may thus fuel theapplication.

The downstream and upstream bandwidthrequirements of corresponding applications arecompared in Fig. 1, together with the bandwidthcapabilities of some of today’s access solutionswith the highest bandwidth. Here, we assume anUHDTV bandwidth of only 75 Mb/s, whichcaters to the uncertainty about future codec effi-ciency and 3D overhead. It is also not clear whatbit rate requirements may come after 8k 3D-UHDTV. The authors hence believe that anynew access infrastructure must be able to poten-tially deliver bit rates that are considerably high-er than the requirements shown in Fig. 1.

BROADBAND ACCESS SOLUTIONSAs discussed earlier, we focus on fiber-basednext-generation broadband access. This refers toboth FTTH and FTTB scenarios, where FTTHis more likely to be the broadband accessendgame, and FTTB may be seen as one of themigration scenarios and also as an alternative incases where optical in-house cabling is difficultor impossible. The main reason for this fiberaccess focus can also be derived from Fig. 1: foraccess bandwidth requirements in the 100 Mb/srange and beyond, there are no efficient alterna-tives. Clearly, DOCSIS3.0 and VDSL2 canexceed 100 Mb/s, but the former is a sharedmedium incapable of providing high dedicatedbandwidth to all customers in a cluster at thesame time, and the latter is severely limited inreach for bandwidths exceeding 50 Mb/s. Inaddition, both will lead to higher energy con-sumption than the PON candidates due to thefact that they are based on copper cabling. Fig-ure 1 also gives the comparison of GPON andWDM-PON downstream and upstream capabili-ties. From this comparison it can be derived thatGPON, in particular when it comes to unicastand symmetric-bandwidth applications, cannotmeet the requirements by 2020. The same is truefor EPON and the successors, XG-PON1 and

Figure 1. Downstream and upstream bandwidth requirements and capabilities of different (near-future) ser-vices and solutions.

70

Video phones VoIP

SDTV VoD Multi-gaming e-Commerce, e-Learning

HDTV VoD

ADSL2+ DOCSIS2.0

Downstream rate (Mb/s)Upstream rate (Mb/s)

GPON 1:32 WDM-PON

VSDL2

WDM-PON

VSDL2 Symmetry option

BluRay / Large-file peer-to-peerUHDTV VoD UHDTV P2P

60 50 40 30 20 10 5 5 10 20 30 40 50 60 70 80

A chicken-egg

problem comes into

play: we do not see

broadband peer-to-

peer applications

today, which is likely

due to the fact that

the required

upstream bandwidth

is not available in the

majority of today’s

access installations.

Once the bandwidth

is provided, it may

thus fuel the

application.



10G-EPON. The proposed solution to the scala-bility requirement is an access technology thatscales through the use of WDM. If this is com-bined with a passive ODN for cost and energyconsumption effectiveness, we refer to it as aWDM-PON.

WDM-PONs have been considered for quitea while as promising contenders for next-genera-tion access [6, 7]. To the knowledge of theauthors, only Korea Telecom has deployed anearly implementation of a WDM-PON so far.The lack of further installations is mostlyattributed to missing international standardiza-tion and cost efficiency. This is expected tochange once the International Telecommunica-tion Union (ITU) or IEEE provides relevantstandards. Increased deployment numbersenabled by standardization will in turn lead todecreased WDM-PON cost.

REQUIREMENTSAs mentioned in the introduction, the require-ments for next-generation PONs vary a lot. Thekey performance indices are dedicated data rateper customer, total customer count, and maxi-mum reach. With regard to customer count, any-thing between 64 and several thousand is stated[8]. Concerning maximum reach, the range from50 to >100 km is covered. Only the bandwidthrequirements seem to be clear: whatever thenext generation of access is, it will have to sup-port per-client bandwidth, which may go beyond1 Gb/s in the long term.

In order to limit the candidates, the next-gen-eration access configurations considered herewere required to support >100 customers with asustained bandwidth of >500 Mb/s each. Mini-mum reach requirement for the ODN withoutadditional reach extension was set to at least 50km. To the best of our knowledge, these num-bers represent what can be considered sufficientfor 2020. It also has to be noted that, in particu-lar, the sustained per-client bandwidth can beincreased to 1 Gb/s (and potentially more)through reconfigurations of the respective solu-tions.

OPTICAL DISTRIBUTION NETWORKTaking reach extension into account, the keyrequirements will also support reduction ofactive sites by an order of magnitude, and hencelonger distances in the ODN. It is not clear yet ifa fully passive ODN is required, or if the ODNmay accommodate a certain amount of activeequipment. This is considered very importantsince it has an obvious impact on the accumulat-ed insertion or path loss of the end-to-end ODN.Combining an optical power splitter/combiner-only ODN with the requirement to support veryhigh customer count, say in clear excess of 100,leads to high insertion/path loss. On the otherhand, in order to keep any next-generationaccess technology competitive, it must berestricted in its available power budget. This isdemonstrated in Fig. 2.

Figure 2 shows the accumulated insertion lossfor PONs with power splitters/combiners only,filters (arrayed waveguide gratings [AWGs]),band-splitters and interleavers only (i.e., a pureWDM-PON), and hybrid PON with both power

splitters/combiners and filters. The light-greyarea indicates the region for a pure WDM-PONthat can be covered with cheap (positive-intrin-sic-negative [PIN] photo-diode-based)transceivers and leaves enough power budget forthe ODN fiber to cover distances of 40–60 km.For the hybrid infrastructure, it is assumed thatthe interfaces have a total bandwidth of 10 Gb/sin order to provide sufficient per-client band-width after the power splitter. It becomes clearthat this infrastructure does not allow use of thecheapest transceivers since these have to support10 Gb/s accumulated bandwidth, a multipleaccess mechanism, and also higher total powerbudgets for per-wavelength split ratios exceeding1:4. It is also obvious that higher numbers ofwavelengths decrease the insertion loss forhybrid infrastructure.

The black squares and diamonds in Fig. 2indicate the range for the split ratio where up to1 Gb/s per client can be guaranteed when using10 Gb/s total per-wavelength capacity. For splitratios >1:8 (white squares and diamonds), this isnot possible anymore at 10 Gb/s per wavelength.

For the splitter-only infrastructure, very highsplit ratios result in very high insertion lossexceeding 30 dB. This infrastructure can only besupported by means of coherent ultra-denseWDM (UDWDM). Then ultra-densely spaced(i.e., < 12.5 GHz) wavelengths with dedicatedper-client bandwidth of 1 Gb/s can be used.Coherent technology can provide the necessarypower budget to cope with the insertion loss.However, it is doubtful that it can achieve thecost points of the much simpler filter-based (orwavelength-routed) WDM-PON. It must benoted that in Fig. 2, we calculated the insertionloss for the ODN power splitters/combiners only.In an ultra-dense WDM-PON (UDWDM-PON),the transceivers in the optical line terminal(OLT, i.e., the central office equipment) have tobe combined by means of splitters and/or filtersas well. These additional combiners must beconsidered when designing the respective OLTtransceiver array.

Figure 2. Accumulated insertion loss of power-splitter PON, filter-based(WDM-) PON, and hybrid filter-plus-splitter PON. TFF is a thin-film filter,S/C is an additional S/C-band splitter, IL is an interleaver, and AWG is anarrayed waveguide grating.

Number of clients10

10

Inse

rtio

n lo

ss (

dB)

0

20

30

40

1 100 1000 10000

Splitter only1:8 TFF + Splitter1:80 AWG + S/C + IL1:80 AWG + Splitter



COMPARED SOLUTIONS

Several WDM-PONs and derivatives have beendiscussed so far [6, 7]. These include plainWDM-PON, which only makes use of wave-length-division multiple access (WDMA), as wellas more sophisticated so-called hybrid PON andUDWDM-PON. Hybrid PON refers to combina-tions of WDM transport and multiple access(WDMA) and additional per-wavelength multi-ple access mechanisms like time-division multi-ple access (TDMA), subcarrier multiple access(SCMA), and code-division multiple access(CDMA).

WDM-, hybrid, and UDWDM-PON are notequally capable of supporting all next-generationrequirements (high client count, high dedicatedper-client bandwidth, long reach), regardless ofthe exact numbers. In addition, they differ signif-icantly in estimated per-client cost (CapEx) andper-client end-to-end energy consumption. Thesetwo parameters are very important on a broadscale with millions of subscribers. Hence, adetailed analysis is required, and is providedhereinafter. We restrict our analysis to fiber-based solutions that use WDM filters in theODN due to the considerations in the last sec-tion. The schematic diagrams of the consideredarchitecture alternatives are compared in Fig. 3.

The WDM-PON shown in Fig. 3a makes useof the C-, L-, and S-bands, which are accessiblevia a cyclic AWG. Today, such AWGs (whichalso cover further bands like the extended L-band) are available for 32 ports and channelspacing in the 100 GHz range. Non-cyclic low-loss AWGs are available that support 96 chan-nels at 50 GHz. We expect that cyclic 50 GHzAWGs with port counts between 64 and 96 willbecome available very soon. The WDM-PON isfurther based on highly integrated photonic inte-grated circuits (PICs), which contain the OLTtransceiver arrays, and low-cost tunable lasers inthe optical networking units (ONUs, i.e., theclient equipment). PICs are considered neces-sary in order to reduce both power consumptionand form factor in the OLT.

The tunable ONU transmitters can be basedon monolithic multisection distributed Braggreflector (DBR) lasers (e.g., SG DBR or DSDBR lasers). Uncooled operation without ther-mo-electric cooling (TEC) of these devices hasalready been demonstrated [9]. TEC-less opera-tion is required in order to achieve the lowestpower consumption. It is also frequently requiredby large network operators. For lowest cost, theONU lasers will also not get their own dedicatedwave lockers. They will then have to be wave-length-tuned via closed-loop control, whichincorporates feedback signals from either theOLT or the PON remote node.

The active/passive hybrid PON shown in Fig.3b connects access switches (active part) via pas-sive WDM (passive backhaul part) running 10Gb/s per wavelength. Tunable extended-form-factor pluggables (XFPs) can be used for theWDM backhaul, and lowest-cost grey smallform-factor pluggables (SFPs) can be used forthe active point-to-point access. The passiveWDM part can also be used for generalizedbackhaul. This combination has potential for

very high per-client and total capacity and longreach in excess of 60 km. It requires, however,the accessibility of the respective active sites forthe access switches.

Figure 3c shows a hybrid WDM/TDMA-PON.For our analysis, we considered DWDM-coloredTDMA at symmetric 10 Gb/s (per pair of wave-lengths). This can be regarded as an extension ofthe stacked XG-PON approach. It can enableboth high split ratio and high capacity. However,it leads to the necessity for 10 Gb/s TDMA(burst-mode) transceivers, which have up to 35dB power budget. In addition, the transceiversmust be tunable or be based on seeded reflectivetechnology [6] (i.e., reflective electro-absorptionmodulator-semiconductor optical amplifier[REAM-SOA] combinations). The latter, howev-er, lack long-distance capability because ofRayleigh crosstalk or require dedicated seedfibers [10]. The OLT again makes use of photon-ic integration.

Finally, Fig. 3d shows a diagram of a coher-ent UDWDM-PON. Heterodyne detection isused in order to reuse the local oscillator laseralso for the respective upstream or downstreamsignal transmission. In the OLT several channelscan be integrated in broadband multichanneltransceivers, thus reducing cost. In the ONUs,narrow-line-width lasers are required. If quater-nary phase shift keying (QPSK) at 1 Gb/s isused, the line width must not exceed ~200 kHz[11]. The lasers must be very precisely tunable tomaintain the ultra-dense spacing. This isachieved by heterodyne closed-loop control andmay require additional environmental control. Itis therefore doubtful that uncooled low-costlasers can be used in this scheme. The coherentdetection scheme also requires polarization con-trol, diversity, or scrambling (Pol. Scr. in Fig. 3)mechanisms [12], which further increase com-plexity and cost. On the other hand, power bud-get of >45 dB is possible which allows thecombination of very high customer count (up toand exceeding 1000), dedicated 1 Gb/s per cus-tomer, and long reach in excess of 50 km. Notethat in Fig. 3d, a hybrid infrastructure with 100GHz AWGs and subsequent power splitters/combiners is shown. The AWG first routesgroups of tightly spaced wavelengths to therespective ports. These wavelength groups arethen split in the splitters/combiners. Thisapproach requires gaps between the wavelengthgroups to allow for filtering, but leads to signifi-cantly decreased insertion loss compared to thesplitter/combiner-only configuration.

COST AND ENERGY CONSUMPTIONFor a comparison of cost and energy consump-tion, the respective contributions of all majorcomponents must be considered. One major dif-ficulty results from the fact that not all compo-nents are commercially available today. Parts ofthese components are not even technicallymature yet. This leads to the necessity of predict-ing both cost and power consumption that thesecomponents will have. This prediction has beendone based on the complexity (functionality),optical power budget, and bandwidth processingrequirements of the respective components.

The lasers must be

very precisely tunable

to maintain the

ultra-dense spacing.

This is achieved by

heterodyne closed-

loop control and

may require

additional environ-

mental control. It is

therefore doubtful if

uncooled, low-cost

lasers can be used in

this scheme.



Where possible, we considered the nearest ormost sophisticated components available today,and estimated their cost and power consumptiondecrease over the next years, assuming mass pro-duction (examples: tunable XFPs, grey SFPs).Also note that, for example, the tunable XFPswill be produced in significantly smaller numbersthan the lowest-cost tunable ONU transceivers

(TRXs). The respective cost impact has beenconsidered. The cost and power consumptionassessment has been performed, to the best ofour knowledge, with equally challenging assump-tions for all components. The resulting cost andpower consumption figures were discussed withcomponent and system vendors, and also net-work operators at FSAN.

Figure 3. Schematic diagrams of broadband next-generation access solutions: a) a basic C+L+S-band WDM-PON for 128-192 bidirec-tional channels — L/C, C/S, and R/B are the respective band filters, with R, B the Red and Blue sub-bands in the S-band; RXA, TXA:transmit and receive arrays, RN: remote node, SFF: small form factor transceiver, including a tunable laser diode (T-LD); note that onlythe WDM-PON in 3a is based on a filter-only ODN; b) an active/passive WDM hybrid PON, where the backhaul part is based on pas-sive WDM and the last-mile access is based on active (Ethernet) point-to-point; TXFP, SFP: tunable extended and small form-factorpluggables, ARN and PRN: active and passive RN; c) a WDM/TDMA hybrid PON running (symmetrical) 10 Gb/s TDMA on eachwavelength pair. MDXM:mux/demux, APD: avalanche photo diode, SOA: semiconductor optical amplifier, CLK Rec.: clock recovery;d) a coherent UDWDM-PON with cascaded filters and power splitters/combiners in the ODN. ADC and DAC: analog-digital and digi-tal-analog conversion, US: upstream, DS: downstream, RF: radio frequency, Pol. Scr.: polarization scrambler.

OLT

TDMAMDXM

TXFP

TXFPSFPONU

10G-TDMAONUn

L2

ARN / LX

λDλU

SFP

SFPTXFP

PoP

SFF

ONU

T-LD

LTX

A

AW

G

AW

G

AW

GA

WG

C

OLT(a)

(b)

(c)

(d)

L/SR-Band 50 GHz

C/SB-Band ~50 GHz

TDMAMDXM

Tx/RxArray

TXFP

TXA

AW

GC

SC

RN 1

RN 2

AW

GPRN

RN 1 RN2

APD

SOA

Data FEC

Data FEC T-LD

CLK Rec.

1:k

Cyclic AWG

S

RBPIC 2 S,

50 GHz

Scalableuniversalswitch

PIC 1 C/L,50 GHz

1xN

AW

G

Rx

OLT

UDWDMMDXM

1G-UDWDMONUnλDλU

AW

G

UDWDMMDXM

Pol. Scr.

Tx/RxArray

RN 1 RN2

1:k

3dB

Pol.Scr.

DS

US

3dB

3dB

3dB

DA

C

RF M

odul

atio

n

AD

C 3dB

1xN

AW

G

RF

CLK Rec.

T-LD

-+

DigitalSCMA

Burst M.



Table 1 gives an overview of cost and powerconsumption figures of the most relevant com-ponents for the systems shown in Fig. 3. Costfigures have been normalized to the commonbaseline costs of the solutions, which are thecosts for common parts such as shelves and man-agement controllers.

From Table 1, the most relevant componentscan be identified. In the first place the transceivers,which are required on a per-client basis, deter-mine the resulting cost and also part of the powerconsumption. With regard to cost, the commoncontribution from shelves, power supply units,management controllers, interfaces, and so on arein second place. Regarding power consumption,these common components are already the maincontributor. Solution-specific application-specificintegrated circuits (ASICs), switches, and the per-port contribution from filters (AWGs) and split-ters/combiners are of least importance.

The results for cost and power consumptionof the four solutions compared in Fig. 3 are list-ed in Table 2. They are split into contributionsfrom the ONU, OLT, and remote node. Theresults are also visualized in Fig. 4.

From the analysis, it can be seen that a pureWDM-PON clearly has the lowest per-client(end-to-end) power consumption, and almostlowest per-client cost. Again, plain WDM-PONrefers to a system without any further added

multiple access mechanism or coherent detec-tion, using filters rather than splitters/combiners.

The per-client end-to-end connection includesthe complete ONU, the respective portion of theremote node (RN; with regard to power consump-tion, there is only a contribution for the active/pas-sive hybrid PON), and the respective portion ofthe OLT node. The latter includes transceivers,any electronics necessary for modulation, multipleaccess and signal processing, and also the respec-tive portion of an aggregation switch. We attributethe WDM-PON performance to the possibility touse cost-effective simple transceivers (27 dB classwith PIN photo diodes, 1 Gb/s bandwidth, mono-lithically integrated lasers, no complicated mediumaccess control [MAC] layer or multiple access). Inturn, such transceivers are allowed because filtersare used in the ODN rather than powersplitters/combiners.

According to the numbers in Table 2, theactive/passive hybrid system leads to lowest cost(although the difference against the WDM-PONis small), and also has second best power con-sumption. Per-client power consumption isincreased by 0.8 W. Note that this difference isproduced mainly in the added active sites. Therelatively good cost and power consumption per-formance of this solution can mainly be attribut-ed to the use of lowest-cost lowest-complexityvery-low-power-consuming grey transceivers

Table 1. Cost and energy-consumption parameters.

Figure Part Component Energy consumption Cost

3a OLT 40 × REAM/Rx array, plus MFL and circulator 20 W 2400

3a ONU 1 Gb/s tunable TRX (PIN-PD, no TEC, no WL) 1.0 W 75

3b OLT 40 λ 1G Laser/Rx array 20 W 2000

3b RN 10 Gb/s tunable XFP (TEC, WL, 25 dB) 3.5 W 1200

3b ONU 1 Gb/s grey SFP, 10 dB power budget 0.5 W 15

3c OLT 30 GHz TRX (no TEC, no WL, 32 dB) 2.5 W 175

3c ONU 10 Gb/s Burst-mode tunable TRX (ADP, SOA, FEC, no TEC, no WL, 35 dB) 2.5 W 175

3d OLT 30 GHz TRX (coherent, TEC, WL, 16 channels, 1G/3 GHz) 8.0 W 1600

3d ONU 1 Gb/s coherent (heterodyne) TRX (polarization diversity or scrambling) 2.0 W 175

3a–3c ONU ASIC 1 GHz (ONU) 1.0 W 10

3d OLT ASIC 50 GHz UDWDMA (OLT, 16 channels) 16 W 320

3a OLT EDFA booster/pre-amplifier combination (OLT) 25 W 2000

All AWG port/power splitter/combiner port — 20/10

All OLT/PoP Layer-2 switch, per 1 Gb/s 1.0 W 5

All Baseline cost per client (CPE housing, OLT shelf, etc.) 5.0 W 100

MFL: multi-frequency laser, PD: photo diode, WL: wavelength locker, ADP: avalanche PD, FEC: forward error correction, ASIC: applica-tion-specific integrated circuit, EDFA: erbium-doped fiber amplifier, CPE: Customer-Premises Equipment.



(SFPs) for the active access links. Today, suchSFPs are available with power consumptiongoing down to 0.4 W and cost going down intothe range of US$20. These numbers cannot beachieved with any other transceiver class (10G,added multiple access, high power budget, etc.).We note, however, that this solution will not beaccepted by every network operator due to thenecessity of active sites.

The hybrid WDM/TDMA-PON is higher incost and power consumption than the two firstsolutions. Compared to the WDM-PON, cost is~27 percent higher. This can be regarded as areasonable cost markup given the fact that theWDM/TDMA-PON can support a larger num-ber of clients over a purely passive ODN. Powerconsumption, however, is increased by 1.3W/client. This difference is partly spent in theONU and is likely to be paid by the customers,not the network operator or service provider. Ona global scale, at high take rate of the respectivetechnology, the question remains if such a differ-ence is acceptable in the context of green IT. Anadvantage of this solution is the support ofGPON access infrastructure, together with thefilter-based backhaul. It also has to be notedthat other WDM hybrid schemes (based onadded SCMA or CDMA) from today’s perspec-tive do not seem to offer advantages over theWDM/TDMA hybrid PON.

According to our analysis, the UDWDM-PON leads to highest cost and also highest powerconsumption. Cost compared to the simplerDWDM-PON is 165 percent, and per-clientpower consumption is higher by 2 W. Weattribute these differences to the necessaryadded complexity of coherent detection, whichincludes tight wavelength control, added (digitalor analog) signal processing, ultra-narrow-linewidth lasers, polarization control, diversity,or scrambling, and also the need for balancedreceivers with multiple photo diodes. Clearly, aUDWDM-PON has very high optical perfor-mance and is able to potentially support morethan 1000 clients via a splitter infrastructure or,using WDM filters, over very long access dis-tances in excess of 100 km. Here, the mostimportant question is whether or not this tech-nology is overengineered for the majority ofapplications or clients (which will require signifi-cantly less reach performance than 100 km).

So far, we have not considered means otherthan PICs and TEC-less lasers for reduction ofpower consumption in WDM-based PONs.Besides general power consumption reduction in(opto-) electronic components, reduction inWDM-PON can be achieved through so-calleddoze or (cyclic) sleep modes, where either trans-mitters are deactivated when not used, or com-plete transceivers are periodically deactivated.Such power-saving modes are expected to besimpler in implementation and more power-sav-ing-efficient in WDM-PONs than the onesknown from GPON since no common MAClayer has to be kept alive.

CONCLUSIONWe have shown that a simple WDM-PON is mostefficient for next-generation access with up to 1Gb/s (sustained) per client and client numbersnot exceeding 128–192. Access distances of 40–60km can also be supported. This can clearly beattributed to the inherent simplicity of thetransceivers and the multiple access mechanismthat can still be used for these numbers. Othersolutions — hybrid WDM/TDMA, UDWDM, oractive-plus-passive hybrid — exist, but either leadto higher cost and power consumption, or requireactive sites which may contradict site consolida-tion programs of certain network operators.

We conclude that it is essential to carefullyclarify the requirements for next-generationaccess with regard to per-PON client count andmaximum reach. The question of whether WDMfilters are allowed in the ODN (instead of powersplitters/combiners) also has to be answered, andthe same is true for active equipment in theaccess network. In particular, if client count doesnot exceed 128–192, and a passive filter-basedODN is accepted, the most efficient solutionwith regard to both cost and power consumptionis a simple WDM-PON. This client-count rangemay even be increased to 256–384 by using inter-leavers in order to effectively use a 25 GHzDWDM grid.

ACKNOWLEDGMENTThe research leading to these results hasreceived funding from the European Communi-ty’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 249025

Table 2. Results of cost and power consumption analysis for next-generation access.

Energy consumption Cost

Common baseline 5 W 100

Solution individualcontribution

ONUTRX

OLT TRX portincl. amplification

DSP,switching

AWG + splitterports

OLT TRX portincl. amplification

ONUTRX

DSP,switching

WDM-PON 1.0 W 0.5 W 1.0 W 20 50 75 5

Active/passive hybrid 1.0 W 0.3 W 2.0 W 2 100 30 15

WDM/TDMA-PON 2.5 W 0.3 W 1.0 W 12 25 175 5

UDWDM-PON 2.0 W 1.5 W 1.0 W 12 100 175 25


(ICT-OASE) and from the German Ministry forEducation and Research (BMBF) under Grant13N10864.

REFERENCES[1] J. Kani and R. Davey, “Requirements for Next Genera-

tion PON,” ITU-T/IEEE Wksp. NGOA Sys., Geneva,Switzerland, June 2008.

[2] M. Fricke, “Examining the Evolution of the Access Net-work Topology,” IEEE GLOBECOM ‘09, Honolulu, HI,Dec. 2009.

[3] R. Tucker, “Optical Packet-Switched WDM Networks: aCost and Energy Perspective,” OFC/NFOEC ‘08, SanDiego, CA, Feb. 2008.

[4] OECD, “Broadband Statistics”;http://www.oecd.org/sti/ict/broadband.

[5] T. Kuroda, “Current Status on Super HDTV Developmentin Japan,” IEEE Int’l. Symp. Broadband Multimedia Sys.Broadcasting, 2010.

[6] A. Banerjee et al., “Wavelength-Division MultiplexedPassive Optical Network (WDM-PON) Technologies forBroadband Access: A Review,” J. Optical Net., vol. 4,no. 11, Nov. 2005, pp. 737–58.

[7] K. Grobe and J.-P. Elbers, “PON in Adolescence: FromTDMA to WDM-PON,” IEEE Commun. Mag., vol. 46, no.1, Jan. 2008, pp. 26–34.

[8] R. P. Davey et al., “Long-Reach Access and FutureBroadband Network Economics,” ECOC ‘07, Berlin, Ger-many, Sept. 2007.

[9] Y. Liu et al., “Directly-Modulated DS dBR Tunable Laserfor Uncooled C-band WDM System,” OFC ‘06, Ana-heim, CA, Mar. 2006.

[10] G. Talli and P. D. Townsend, “Hybrid DWDM-TDMLong-Reach PON for Next-Generation Optical Access,”IEEE J. Lightwave Tech., vol. 24, no. 7, July 2006, pp.2827–34.

[11] M. Seimetz, “Laser Linewidth Limitations for OpticalSystems with High-Order Modulation Employing FeedForward Digital Carrier Phase Estimation,” OFC/NFOEC‘08, San Diego, CA, Mar. 2008.

[12] J. M. Fabrega and J. Prat, “New intradyne Receiverwith Electronic-Driven Phase and Polarization Diversity,”OFC ‘06, Anaheim, CA, Mar. 2006.

BIOGRAPHIESKLAUS GROBE [M‘94] received Dipl.-Ing. and Dr.-Ing. degreesin electrical engineering from Leibniz University, Hannover,Germany, in 1990 and 1998, respectively. Since 2000 hehas been with ADVA AG Optical Networking. He hasauthored or co-authored three book chapters on WDM andPON technologies, and more than 70 scientific papers. Heis a member of the IEEE Photonics Society, OFC Subcom-mittee F, the German VDE ITG, and ITG Fachgruppe 5.3.3(Photonic Networks).

JÖRG-PETER ELBERS received his diploma and Dr.-Ing. degreesin electrical engineering from Dortmund University, Ger-many, in 1996 and 2000, respectively. In 1999–2001 hewas with Siemens AG — Optical Networks, as director ofnetwork architecture. In 2001 he joined Marconi Communi-cations as director of technology. Since September 2007 hehas been with ADVA AG Optical Networking, where he isvice president of advanced technology. He authored andco-authored more than 70 scientific publications and holds14 patents.

MARKUS ROPPELT received a diploma degree in electricalengineering from the Karlsruhe Institute of Technology(KIT), Germany, in 2009. He is currently working toward a

Ph.D. degree in electrical engineering at ADVA AG OpticalNetworking. His current primary research interest is in next-generation optical networks.

MICHAEL EISELT [SM] received his Dr.-Ing. degree in pho-tonics from the Technical University of Berlin in 1994.During his 20-year career in optical communications, hehas worked at various companies and research organiza-tions in Germany and the United States. As director ofadvanced technology at ADVA Optical Networking, he iscurrently leading physical layer research for high-speedand access applications. He is a Fellow of the OpticalSociety of America.

ACHIM AUTENRIETH [M] received his Dipl.-Ing. and Dr.-Ing.degrees in electrical engineering and information technolo-gy from Munich University of Technology (TUM), Germany,in 1996 and 2003, respectively. From 2003 to 2010 he waswith Siemens AG, Siemens Networks GmbH & Co KB, andNokia Siemens Networks, last as head of BCS R&D Innova-tions. Since 2010 he is with ADVA AG Optical Networking,Advanced Technology. His research interests include multi-layer transport networks and control planes. He is a mem-ber of VDE/ITG.

Figure 4. Per-client end-to-end power consumption (top) and relative cost (bot-tom) of the four solutions, broken down into the main contributions of each.

WDM

Base

Active/passive

Base

WDM/TDMA

Base

UDWDM

Base

2

Pow

er c

onsu

mpt

ion

(W)

0

DSP, switchOLT TRX portONU TRX

DSP, switchOLT TRX portONU TRXAWG + splitter ports

4

6

8

10

WDM

Base

Active/passive

Base

WDM/TDMA

Base

UDWDM

Base

Cos

t (r

elat

ive)

0

400

300

200

100



IEEE Communications Magazine • February 2011 S330163-6804/11/$25.00 © 2011 IEEE

INTRODUCTIONAs one of the major fiber to the home/curb/cabi-net/and so on (FTTx) technologies, Ethernetpassive optical network (EPON) is developedbased on Ethernet technology, and enablesseamless integration with IP and Ethernet tech-nologies [1]. Due to the advantages of fine scala-bility, simplicity, and multicast convenience, aswell as the capability of providing full-serviceaccess, EPON has been rapidly adopted in Japanand is also gaining momentum with carriers inChina, Korea, and Taiwan since the IEEE rati-fied EPON as the IEEE 802.3ah standard inJune 2004.

On the other hand, video-centric applicationsand services such as HDTV are growing andemerging in the network [2]. As compared totraditional voice and data traffic, these multime-dia applications are more bandwidth-hungry. Forexample, one HDTV channel requires as muchas 10 Mb/s bandwidth. Motivated by satisfyingthese emerging high bandwidth demands, theIEEE 802.3av 10G-EPON task force was chargedto increase the downstream bandwidth to 10Gb/s, and to support two upstream data rates: 10Gb/s and 1 Gb/s.

While the line rate is significantly increased

to satisfy subscribers’ demands, the power con-sumption of 10G-EPON may be increased aswell [3]. Power consumption of 10G-EPON hasbecome a big concern of network service pro-viders as it contributes to part of their opera-tional expenditure (OPEX). Moreover, energyconsumption is becoming an environmental andtherefore social and economic issue because onebig reason for climate change is the burning offossil fuels and the direct impact of greenhousegases on the Earth’s environment [4]. Previously,Baliga et al. [5] estimated that the Internet cur-rently consumes around one percent of the totalelectricity consumption in broadband enabledcountries. It is also shown that currently and inthe medium term future, access networks con-sume the majority of the energy in the Internet.The analysis in [5, 6] showed that, among vari-ous access technologies, such as WiMAX, fiberto the node (FTTN), and point-to-point opticalaccess networks, PON is the most power-effi-cient solution in terms of energy consumptionper transmission bit attributed to the nearestapproach of optical fibers to users.

Although PON consumes the smallest poweramong all access network technologies, it isdesirable to further reduce its power consump-tion, especially when the line rate is increased to10 Gb/s. With the increase in line rate, the opti-cal dispersion increases as well. Compensatingfor higher dispersion exerts higher requirementson optical lasers, which may incur an increase ofpower consumption of lasers. In addition, theelectronic circuit should be sufficiently poweredsuch that it can process 10 times faster than thatof 1G-EPON. Thus, 10G-EPON will consumemore energy than 1G-EPON [4, 5].

Reducing power consumption of 10G-EPONrequires efforts across both the physical andmedium access control (MAC) layers. Effortsare being made to develop optical transceiversand electronic circuits with low power consump-tion. Besides, multi-power-mode devices with theability of disabling certain functions can alsohelp reduce the energy consumption of the net-work. However, low-power-mode devices withsome functions disabled may result in degrada-tion of network performances. To avoid servicedegradation, it is important to properly design

ABSTRACT

To rapidly meet increasing traffic demandsfrom subscribers, the IEEE 802.3av task forcewas charged to study the 10G-EPON system toincrease the data rate to 10 times the line rate of1G-EPON. With the increase of the line rate,the energy consumption of 10G-EPON increasesaccordingly. Achieving low energy consumptionwith 10G-EPON has attracted broad researchattention from both academia and industry. Inthis article we first briefly discuss the key fea-tures of 10G-EPON. Then, from the perspectiveof MAC-layer control and scheduling, we discusschallenges and possible solutions to put opticalnetwork units into low-power mode for energysaving. More specifically, we detail sleep-modecontrol and sleep-aware traffic schedulingschemes in two scenarios: sleep for over oneDBA cycle, and sleep within one DBA cycle.


Jingjing Zhang and Nirwan Ansari, New Jersey Institute of Technology

Toward Energy-Efficient 1G-EPON and10G-EPON with Sleep-AwareMAC Control and Scheduling

ANSARI LAYOUT 1/19/11 3:29 PM Page 59


MAC-layer control and scheduling schemes thatare aware of the disabled functions. This is thefocus of this article.

We first discuss the evolution from 1G-EPONto 10G-EPON. Then, we discuss the challengesin empowering optical network units (ONUs) inlow-power mode. Finally, we detail our proposedcontrol and scheduling schemes to achieve ener-gy saving without degrading user services.

EVOLUTION FROM1G-EPON TO 10G-EPON

10G-EPON supports both symmetric 10 Gb/sdownstream and upstream, and asymmetric 10Gb/s downstream and 1 Gb/s upstream datarates, while 1G-EPON provides only the 1 Gb/ssymmetric data rate. With a focus on the physicallayer, the IEEE 802.3av Task Force specifies thereconciliation sublayer (RS), symmetric andasymmetric physical coding sublayers (PCSs),physical media attachments (PMAs), and physicalmedia-dependent (PMD) sublayers. Table 1 listsseveral key physical layer features of 10G-EPON[7]. Instead of using the 8B/10B line codingadopted in 1G-EPON, 10G-EPON employs64B/66B line coding, with which the bit-to-baudoverhead is reduced to as small as 3 percent. Torelax the requirements for optical transceivers,Reed-Solomon code (255, 223) is chosen as themandatory forward error correction (FEC) codein 10G-EPON to enhance the FEC gain, whileReed-Solomon code (255, 239) is specified asoptional for 1G-EPON. 10G-EPON defines thePRX power budget for asymmetric-rate PHY of10 Gb/s downstream and 1 Gb/s upstream, andthe PR power budget for symmetric-rate PHY of10 Gb/s both upstream and downstream. Eachpower budget further contains three power bud-get classes: low power budget (PR(X)10), medi-um power budget (PR(X)20), and high powerbudget (PR(X)30). PR(X)10 and PR(X)20 powerbudget classes are defined in 1G-EPON as well,while PR(X)30, which can support 32-split with adistance of at least 20 km, is an additional onedefined in 10G-EPON. Due to limited space, weonly list the transmitter (Tx) type along with itslaunch power of 10G-EPON in Table 1. As com-pared to 1G-EPON, advanced transmitters andhigher launch power are employed in 10G-EPON

to guarantee a sufficient signal-to-noise ratio(SNR) at the receiver side for accurate recoveryof data with a rate of 10 Gb/s. Because of theincreased launch power, the power consumptionof the optical transmitter should be increasedaccordingly. Also, due to the mandatory FECmechanism and increased line rate, the electroniccircuit has to enable more functions and processfaster than that in 1G-EPON, thus consequentlyincurring high power consumption and possiblylarger heat dissipation. Therefore, to accommo-date 10 Gb/s in the physical layer, the power con-sumption of the optical line terminal (OLT) andONU may increase significantly.

For the MAC layer and layers above, in orderto achieve backward compatibility such that net-work operators are encouraged to upgrade theirservices, 10G-EPON keeps the EPON frame for-mat, MAC layer, MAC control layer, and all thelayers above almost unchanged from 1G-EPON.This further implies that similar network manage-ment system (NMS), PON-layer operations,administrations, and maintenance (OAM) system,and dynamic bandwidth allocation (DBA) used in1G-EPON can be applied to 10G-EPON as well.

Next, with a focus on MAC layer control andDBA, we propose a scheme to reduce energyconsumption of 1G-EPON and 10G-EPON. Inthis article, we focus on reducing the energy con-sumption of ONUs.

CHALLENGES IN SAVING ENERGY OF1G-EPON AND 10G-EPON

Formerly, the sleep mode was proposed to beintroduced into ONUs to save energy whenONUs are idle [8–10]. ITU-T RecommendationG. sup 45 [11] specified two energy saving modesfor ONUs in GPON. One is doze mode, in whichonly the transmitter can be turned off when pos-sible. Another one is cyclic sleep mode, in whichboth transmitter and receiver can be turned off.Since the access network traffic is rather bursty[12], ONUs may be idle for quite long periods,implying that putting idle ONUs into sleep modeis an effective way to reduce energy consump-tion. However, it is challenging to wake up sleepONUs in time to avoid service disruption whendownstream or upstream traffic arrives in 1G-EPON and 10G-EPON.

Table 1. Comparison between 1G-EPON and 10G-EPON.

Data rate (Gb/s)FEC Line

coding Tx type and launch power (dbm)Upstream Downstream

1G-EPON 1.25 1.25 OptionalRS(255, 239) 8b/10b

PX10:OLT: [–3,2]ONU: [–1,4]

PX20:OLT: [2,7]ONU: [–1,4]

10G-EPON

10.3125 10.3125 EnabledRS(255, 223) 64b/66b

PR10:OLT: EML [1,4]ONU: DML [–1,4]

PR20:OLT: EML+AMP [5,9]ONU: DML [–1,4]

PR30:OLT: EML [2,5]ONU: HP DML [4,9]

1.25 10.3125 EnabledRS(255,223) 64b/66b

PRX10:OLT: EML [1,4]ONU: DML [–1,4]

PRX20:OLT: EML+AMP [5,9]ONU: DML [–1,4]

PRX30:OLT: EML [2,5]ONU: DML [.6,5.6]



The major challenge lies in downstream trans-mission. In EPON, the downstream data trafficof all ONUs is time-division multiplexed (TDM)into a single wavelength, and is then broadcastedto all ONUs. An ONU receives all downstreampackets and checks whether the packets are des-tined to itself. An ONU does not know when thedownstream traffic arrives at the OLT and theexact time the OLT schedules its downstreamtraffic. Therefore, without proper sleep-awareMAC control, receivers at ONUs need to beawake all the time to avoid missing their down-stream packets.

To address this problem, Mandin [9] pro-posed to implement a three-way handshake pro-cess between OLT and ONUs before puttingONUs to sleep. Since an OLT is aware of thesleep status of ONUs, it can queue the down-stream arrival traffic until the sleep ONU wakesup. However, to implement the three-way hand-shake process, extended multipoint control pro-tocol (MPCP) is required to introduce newMPCP packet data units (PDUs). In addition,the negotiation process takes at least severalround-trip times that further incurs large delay.Lee et al. [13] proposed to implement fixedbandwidth allocation (FBA) when the network islightly loaded. By using FBA, the time slots allo-cated to each ONU in each cycle are fixed andknown to the ONU, and thus ONUs can go tosleep in the time slots allocated to other ONUs.However, since traffic of an ONU changesdynamically and from cycle to cycle, FBA mayresult in bandwidth under- or overallocation,and consequently degrade services of ONUs tosome degree.

Besides the downstream scenario, an effi-cient sleep control mechanism should also con-sider upstream traffic and MPCP controlmessage transmission. For upstream transmis-sion, the wake-up of a sleep ONU can be trig-gered by the arrival of upstream traffic.However, this arrival traffic cannot be transmit-ted until the ONU is notified of the allocatedtime from the OLT. Before the OLT allocatesbandwidth to an ONU, the newly awake ONUfirst needs to request upstream bandwidth. Torealize this, some periodic time slots may needto be allocated to ONUs to enable them toaccess the upstream channel in time even whenthey are asleep.

Regarding the MPCP control message trans-mission, to keep a watchdog timer in the OLTfrom expiring and deregistering the ONU, bothIEEE 802.3ah and IEEE 802.3av specify thatONUs should send MPCP REPORT messagesto the OLT periodically to signal bandwidthneeds as well as to arm the OLT watchdog timereven when no request for bandwidth is beingmade. The longest interval between two reportsas specified by report_timeout is set as 50ms in both 1G-EPON and 10G-EPON. Besides,the OLT also periodically sends GATE messagesto an ONU even when the ONU does not havedata traffic. The longest interval between twoGATE messages, specified by gate_timeout,is set as 50 ms. Therefore, to comply with MPCP,sleeping ONUs need to wake up every 50 ms tosend the MPCP REPORT messages and receivethe GATE messages.

SLEEP-AWARE MAC CONTROL ANDSCHEDULING

In this section, we discuss our proposals on sav-ing energy in ONUs. Our basic idea is stillputting ONUs into sleep mode whenever possi-ble. Different from existing proposals, which putthe whole ONU to sleep, we investigate the con-stitution of an ONU and put different compo-nents of an ONU to sleep under differentconditions.

SLEEP STATUS OF ONUSFigure 1 illustrates the typical constitution of anONU. The optical module consists of an opticaltransmitter (Tx) and an optical receiver (Rx).The electrical module mainly contains serializ-er/deserializer (SERDES), ONU MAC, net-work/packet processing engine (NPE/PPE),Ethernet switch, and user-network interfaces(UNIs). When neither upstream nor downstreamtraffic exists, every component in the ONU canbe put to sleep. When only downstream trafficexists, the functions related to upstream trans-mission can be disabled. Similarly, the functionsrelated to receiving downstream traffic can bedisabled when only upstream traffic exists. Evenwhen upstream traffic exists, the laser driver andlaser diode (LD) do not need to function all thetime, but only during the time slots allocated tothis ONU. Thus, each component in the ONUcan likely sleep, and potentially higher powersavings can be achieved.

By putting each component of an ONU tosleep, an ONU ends up with multiple power lev-els. Figure 2 shows the power levels of an ONUand the transition between different power lev-els. The wakeup of UNI, NPE/PPE, and switchcan be triggered by the arrival of upstream traf-fic and the forwarding of downstream trafficfrom the ONU MAC [14]. They are relativelyeasily controlled as compared to the other com-ponents. Thus, we only focus on the ONU MAC,SERDES, Tx, and Rx. As shown in Fig. 2a, twopower levels, all:awake and all:sleep, result fromputting the whole ONU to sleep. In our proposaltwo more sleep statuses, Rx:sleep and Tx:sleep,are introduced; thus, four power levels are gen-erated. When the ONU is in the all:awake status,if Tx does not need to work, it enters intoTx:sleep status, and further enters into all:sleepstatus if Rx does not need to function either. Inall:sleep status, besides Rx and Tx, the ONUMAC and SERDES sleep as well. Similarly, tran-sitions happen between the all:awake, Tx:sleep,and Rx:sleep statuses.

Transitioning among these statuses should beproperly designed so as to maximize energy sav-

Figure 1. The constitution of an ONU.

NPE/PPE: network/packet processing engine

Ethernet switch

ONU MAC

SER DES

NPE and PPE

UNI Laser driver LD

CD R

APD/ PIN

L A

TI A

UNI

UNI



ing without degrading services. We present solu-tions in determining the transitions under tworespective scenarios: sleep for more than oneDBA cycle, and sleep within one DBA cycle.

SCENARIO 1: SLEEP FORMORE THAN ONE DBA CYCLE

In this scenario the transition is decided by theincoming traffic status. Tx/Rx is put to sleep ifno upstream/downstream traffic exists for sometime.

Whether or not downstream/upstream trafficexists can be inferred based on the informationof the time allocated to ONUs and queue lengthsreported from ONUs, which is known to bothOLT and ONUs. If no upstream traffic arrives atan ONU, the ONU requests zero bandwidth inthe MPCP REPORT message. Then, the OLTcan assume that this ONU does not haveupstream traffic. If no downstream traffic for anONU arrives at the OLT, the OLT will not allo-cate downstream bandwidth to the ONU. Assumethat, out of fairness concerns, an OLT allocatessome time slots in a DBA cycle to every ONUwith downstream traffic. Then, considering theuncertainty of the exact time allocated to anONU in a DBA cycle, the ONU can infer that nodownstream traffic exists if it does not receiveany downstream traffic within two DBA cycles.

The next question is to decide the transitionbetween different statuses. In this scenario, thestatus Rx:sleep actually does not exist since therewill still be some downstream MPCP controlpackets for an ONU to facilitate the upstreamtransmission even when no downstream datatraffic exists. Hence, we next discuss the transi-tion between all:awake and Tx:sleep, and thetransition between Tx:sleep and all:sleep.

Formerly, Kudo et al. [10] proposed periodicwakeup with sleep time adaptive to the arrivaltraffic status. We also decide the sleeping timebased on traffic status. More specifically, we setthe sleep time as the time duration in which traf-fic stops arriving. When putting Tx to sleep, forexample, Algorithm 1 describes the transitionbetween all:awake and Tx:sleep. We assume thatAlgorithm 1 is known to the OLT as well. Then,the OLT can accurately infer the time that Tx isasleep or awake.

Let idle_threshold be the maximum timeduration a transmitter stays idle before being putto sleep, short_active be the time taken foran ONU to check its queue status and send outthe report, and sleep_time be the time dura-tion each time an ONU sleeps. If the transmitteris idle for idle_threshold, Tx will be put intosleep status, and the sleep_time for the firstsleep equals idle_threshold. Then, Tx wakesup to check its queue status and sends a report tothe OLT, which takes short_active time dura-tion. If there is no upstream traffic being queued,Tx will enter sleep status again. Until now, theelapsed time since the last time Tx transmitteddata packets equals idle_threshold + timeduration of the first sleep + short_active. Sofor the second sleep, the sleep time durationsleep_time is set as idle_threshold +time duration of the first sleep +short_active. According to MPCP, ONUssend MPCP REPORT messages to the OLTevery 50 ms when there is no traffic. Thus, we setthe upper bound of sleep_time as 50 ms to becompatible with MPCP and also to avoid intro-ducing too much delay of the traffic which arrivesduring sleep mode. This process repeats untilupstream traffic arrives. For the sth sleep, thesleep_time equals idle_threshold + thetotal time durations of the former s – 1 sleep + (s– 1)* short_active, which also equals 2 s – 1 *short_active + (2s–1 – 1) * idle_thresh-old.

For the transition between Tx:sleep andall:sleep, the transitioning algorithm is similar toAlgorithm 1 with the exception that line a shouldbe changed to: “If the Rx has not received down-stream traffic destined to the ONU for the timeduration of idle_threshold.” The remainingcodes are similar.

Figure 3 shows an example of the sleep timecontrol process with short_active = 2.5 msand idle_threshold = 10 ms. Then,sleep_times of the first, second, third, andfourth sleeps are as follows:• First sleep: 10 ms• Second sleep: idle_threshold + 10 ms

+ short_active = 22.5 ms• Third sleep: idle_threshold + 32.5 ms

+ 2 * short_active = 47.5 ms• Fourth sleep: min{50 ms, idle_thresh-

Figure 2. Multi-power-level ONUs.

4

31

2

5

6

7

8

Power

Level 2All:awake

All:awake

All:sleep

Tx:sleep

Rx:sleep

All:sleep Level 1

Power

Level 2

Level 3

Level 4

Level 1

Algorithm 1. Decide the transition between all:awake and Tx:sleep.

a: If the Tx has not transmitted traffic for the time duration of idle_thresholds = 1;b: Tx enters into sleep status;sleep_time = 2s – 1*short_active + (2s–1 – 1) * idle_threshold;If sleep_time > 50 ms

sleep_time = 50 msEndifTx wakes up after sleep time durationThe ONU checks the queue length and reports the queue statusIf there is queued traffic

Keep Tx awakes = 0;go to line a;

Elses = s + 1;go to line b;Endif

Endif



old + 80 ms + 3 * short_active} = 50ms

In deciding the sleep time, idle_thresholdand short_active are two key parameterswhich are set as follows.

idle_threshold — Setting idle_thresholdneeds to consider the time taken to transitbetween sleep and awake. Considering the transi-tion time, the net sleep time will be reduced bythe sum of the transit time from awake to sleepand the transit time from sleep to awake. Hence,idle_threshold should be set longer thanthe sum of two transit times in order to saveenergy in the first sleep. Currently, the timetaken to power up the whole ONU is around2–5 ms [8]. Hence, idle_threshold shouldbe greater than 4 ms in this case. In addition, weassert that the upstream/downstream trafficqueue is empty if no bandwidth is allocated toupstream/downstream traffic foridle_threshold. To ensure this assertion iscorrect, idle_threshold should be at leastone DBA cycle duration, which typically extendsless than 3 ms to guarantee delay performancefor some delay-sensitive service. So Tx/Rx mustsleep for over one DBA cycle with this scheme.

short_active — During the short awake time ofTx, an ONU checks its upstream queue statusand reports to the OLT. Thus, short_activeshould be long enough for an ONU to completethese tasks. In addition, using some upstreambandwidth for an ONU to send a report affectsthe upstream traffic transmission of other ONUs.In order to avoid interruption of the traffictransmission of other ONUs, we setshort_active to be at least one DBA cycleduration such that an OLT can have freedom indeciding the allocated time for an ONU to sendits report. For Rx, during the short awake time,the OLT begins sending the queued downstreamtraffic if there is any. Similar to the Tx case,short_active is set to be at least one DBAcycle to avoid interrupting services of ONUs inthe Rx case.

SCENARIO 2: SLEEP WITHIN ONE DBA CYCLEIn the former scenario, the sleep and awakedurations of Tx and Rx are greater than oneDBA cycle. In this section, we discuss the schemeof putting Tx and Rx to sleep within one DBAcycle.

Consider a PON with 16 ONUs. During aDBA cycle, on average, only 1/16 of time dura-tion is allocated to an ONU. This means thateven if upstream/downstream traffic exists,Tx/Rx need only be awake for 1/16 of the timeand can go to sleep for the other 15/16 of thetime. Therefore, significant energy savings canbe achieved.

To enable an ONU to sleep and wake upwithin a DBA cycle, the transit time betweenawake and sleep should be less than half theDBA cycle duration such that the net sleep timecan be greater than zero, and thus energy can besaved. Formerly, Wong et al. [15] reduced thetransition time to as small as 1–10 ns by keepingsome of the back-end circuits awake. Thus, withadvances in speeding up transition time, it is

physically possible to put an ONU to sleep with-in one cycle to save energy.

For the upstream case, waking up Tx can betriggered by ONU MAC when the allocatedtime comes. Tx can go to sleep after data trans-mission. For the downstream case, however, it isdifficult to achieve since Rx does not know thetime when the downstream traffic is sent and hasto check every downstream packet. To addressthis problem, we propose the following sleep-aware downstream scheduling scheme.

For downstream transmission, an OLT sched-ules the downstream traffic of ONUs one byone, and the interval between two transmissionsof an ONU is determined by the sum of thedownstream traffic of all other ONUs. Again,due to the bursty nature of ONU traffic, theONU traffic in the next cycle does not varymuch from that in the current cycle. According-ly, we can make an estimation of traffic of otherONUs and put this particular ONU to sleep forsome time.

More specifically, for a given ONU, denote Δas the difference between the ending time of itslast scheduling and the beginning time of its cur-rent scheduling. Then we set the rule that theOLT will not schedule this ONU’s traffic untilf(Δ) time after the ending time of the currentscheduling. As long as the ONU is aware of thisrule, it can go to sleep for f(Δ) time durations.

Figure 4 illustrates one example of putting anONU to sleep within one DBA cycle. In thisexample, one OLT is connected to four ONUs,and f(Δ) is set as 0.8 * Δ. The interval betweenthe first two schedulings of ONU 4 is 9. Hence,the OLT will not schedule the traffic of ONU 4until 7.2 time duration later; ONU 4 can sleepfor 7.2 time duration and then wake up. Howev-er, this wakeup is an early wakeup since theactual transmission of the other ONUs takes 9.5time durations, which is 2.3 longer than the esti-mation. Similarly, the duration of the secondsleep is set as 7.6. However, this wakeup is a latewakeup since the actual time taken to transmitthe other ONUs’ traffic is 6.5. The late wakeupincurs 1.1 idle time duration on the downstreamchannel.

As can be seen from the example, early wake-up and late wakeup are two common phenome-na of this scheme. Early wakeup implies thatenergy can be further saved, while late wakeupresults in idle time durations, and thus possiblyservice degradation. From the network serviceproviders’ perspective, avoiding late wakeup andthe subsequent service degradation is more

Figure 3. An example of sleep time control of the transmitter.

Transmitting Idle

Sleep 1: 10

1 x

1 x

1 x

Idle threshold: 10

1 x

1 x

Sleep 2: 22.5 Sleep 3: 47.5

Sleep 4: 50 Sleep 5: 50

Time

Time


desirable than avoiding early wakeup. Hence, 0< f(Δ) < Δ is suggested to be set. If f(Δ) is set assmall as 0.5Δ, on average an ONU can still sleep15/32 of the time when a PON supports 16ONUs and traffic of ONUs is uniformly dis-tributed. Therefore, significant power savingscan be achieved with this scheme.

CONCLUSIONWith the increase of line rate, energy consump-tion of 10G-EPON may be significantly increasedas compared to that of 1G-EPON. It is impor-tant to reduce power consumption of 10G-EPONfor low OPEX and environmental friendliness.In this article, with the focus on saving energy ofONUs, we have proposed to disable some func-tions of ONUs whenever possible, and intro-duced four power levels for ONUs. Then, fromthe perspective of MAC control and scheduling,we have investigated schemes to properly transitbetween power levels of ONUs. Two schemeshave been proposed for two respective scenarios:sleep for over one DBA cycle, and sleep withinone DBA cycle. For the former scenario, wehave described the scheme of implicitly convey-ing sleep information between ONU and OLT,and proposed an algorithm to adapt the sleepingtime to user traffic. For the latter scenario, toaddress the challenging issue in the downstreamcase, we have designed a sleep-aware down-stream scheduling scheme to achieve energy sav-ing without degrading user service.

REFERENCES[1] Y. Luo and N. Ansari, “Bandwidth Allocation for Multi-

service Access on EPONs,” IEEE Commun. Mag., vol. 43,no. 2, 2005, pp. S16–S21.

[2] K. Cho et al., “The Impact and Implications of theGrowth in Residential User-to-User Traffic,” Proc. Conf.Apps., Tech., Architectures, Protocols for Comp. Com-mun., 2006.

[3] R. Kubo et al., “Sleep and Adaptive Link Rate Controlfor Power Saving in 10G-EPON Systems,” Proc. 2009IEEE GLOBECOM, 2009, pp. 1573–78.

[4] J. Baliga et al., “Energy Consumption in Optical IP Net-works,” IEEE/OSA J. Lightwave Tech., vol. 27, no. 13,2009, pp. 2391–2403.

[5] J. Baliga et al., “Energy Consumption in Access Net-works,” Proc. Optical Fiber Commun. Conf., 2008.

[6] C. Lange and A. Gladisch, “On the Energy Consumptionof FTTH Access Networks,” Proc. Optical Fiber Commun.Conf., 2009

[7] K. Tanaka, A. Agata, and Y. Horiuchi, “IEEE 802.3av

10G-EPON Standardization and Its Research and Devel-opment Status,” IEEE/OSA J. Lightwave Tech., vol. 28,no. 4, Feb.15, 2010, pp. 651–61.

[8] S. Wong et al., “Sleep Mode for Energy Saving PONs:Advantages and Drawbacks,” IEEE GreenCom, 2009.

[9] J. Mandin, “EPON Power Saving via Sleep Mode,” IEEEP802. 3av 10GEPON Task Force Meeting, 2008.

[10] R. Kubo et al., “Adaptive Power Saving Mechanism for10 Gigabit Class PON Systems,” IEICE Trans. Commun.,vol. E93-B, no.2, 2010, pp. 280–88.

[11] ITU-T Rec. G. sup. 45.[12] G. Kramer, B. Mukherjee, and G. Pesavento, “Ethernet

PON (ePON): Design and Analysis of an Optical AccessNetwork,” Photonic Net. Commun., vol. 3, no. 3, 2001,pp. 307–19.

[13] S. Lee and A. Chen, “Design and Analysis of a NovelEnergy Efficient Ethernet Passive Optical Network,”Proc. 9th Int’l. Conf. Net., 2010.

[14] E. Trojer and P. Eriksson, “Power Saving Modes forGPON and VDSL,” Proc. 13th Euro. Conf. Netw. & Opti-cal Commun., Austria, June 30–July 3, 2008.

[15] S. Wong et al., “Demonstration of Energy ConservingTDM-PON with Sleep Mode ONU Using Fast Clock Recov-ery Circuit,” Proc. Optical Fiber Commun. Conf., 2010.

BIOGRAPHIESJINGJING ZHANG [S‘09] received a B.E. degree from Xi’anInstitute of Posts and Telecommunications, China, in 2003,and an M.E. degree from Shanghai Jiao Tong University,China, in 2006, both in electrical engineering. She is work-ing toward her Ph.D. degree in electrical engineering at theNew Jersey Institute of Technology (NJIT), Newark. Herresearch interests include planning, capacity analysis, andresource allocation of broadband access networks, QoEprovisioning in next-generation networks, and energy-effi-cient networking. She received the New Jersey InventorsHall of Fame Graduate Student Award in 2010.

NIRWAN ANSARI [S‘78, M‘83, SM‘94, F‘09] ([email protected]) received his B.S.E.E. (summa cum laude,with a perfect GPA) from the New Jersey Institute of Tech-nology (NJIT) in 1982, his M.S.E.E. degree from the Univer-sity of Michigan, Ann Arbor, in 1983, and his Ph.D. degreefrom Purdue University, West Lafayette, Indiana, in 1988.He joined NJIT’s Department of Electrical and ComputerEngineering as an assistant professor in 1988, tenuredassociate professor in 1993, and full professor since 1997.He has also assumed various administrative positions atNJIT. He authored Computational Intelligence for Optimiza-tion (Springer, 1997, translated into Chinese in 2000) withE.S.H. Hou, and edited Neural Networks in Telecommunica-tions (Springer, 1994) with B. Yuhas. His current researchfocuses on various aspects of broadband networks andmultimedia communications. He has also contributed over350 technical papers, over one third of which were pub-lished in widely cited refereed journals/magazines. Forexample, one of his published works was the sixth mostcited article published in IEEE Transactions on Parallel andDistributed Systems, according to the journal EIC report inFebruary 2010. He has also guest edited a number of spe-cial issues, covering various emerging topics in communica-tions and networking. He was visiting (chair) professor atseveral universities. He has served on the Editorial andAdvisory Boards of eight journals, including as a SeniorTechnical Editor of IEEE Communications Magazine(2006–2009). He has served the IEEE in various capacitiessuch as Chair of the IEEE North Jersey CommunicationsSociety Chapter, Chair of the IEEE North Jersey Section,member of the IEEE Region 1 Board of Governors, Chair ofthe IEEE Communications Society Networking TechnicalCommittee Cluster, Chair of the IEEE Communications Soci-ety Technical Committee on Ad Hoc and Sensor Networks,and Chair/Technical Program Committee Chair of severalconferences/symposia. Some of his recent recognitionsinclude a 2007 IEEE Leadership Award from the CentralJersey/Princeton Section, the NJIT Excellence in TeachingAward in Outstanding Professional Development in 2008, a2008 IEEE MGA Leadership Award, the 2009 NCE Excel-lence in Teaching Award, several best paper awards (IC-NIDC 2009 and IEEE GLOBECOM 2010), a 2010 ThomasAlva Edison Patent Award, and designation as an IEEECommunications Society Distinguished Lecturer(2006–2009, two terms).

Figure 4. One example of putting ONUs into sleep within one DBA cycle.

9

Awake

Late wake upEarly wake up

Sleep Sleep 2Awake

Awake

OLT

ONU 1 ONU 2 ONU 3 ONU 4

4 4 4

1 2 3

1 2 3

4

1 2 3

9.5

7.2

7.6 Idle 6.5




INTRODUCTION

The increasing growth of Internet Protocol (IP)and the popularity of the web are resulting inaltering the traffic pattern in data networks suchthat voice- and text-oriented traffic have changedto data- and image-based traffic. Furthermore,the emergence of IP based multimedia applica-tions such as VoIP, IPTV, video conferencing,and so on, diversifies data traffic having differentdata rate and quality of service (QoS) require-ments. Therefore, designing a high-capacity net-work to handle diverse and bulky data traffic isan essential challenge toward next-generationdata networks.

Optical networks exploiting tremendous fiberoptic bandwidth is a promising solution to trans-mit bulky data traffic. Various techniques havebeen proposed to utilize fiber optic capacity inthe access and backbone of the network. Fiberto the home (FTTH) is an interesting solutionproposed to exploit fiber optic capacity in accessnetworks. The passive optical network (PON) isa promising scheme to implement FTTH costeffectively [1].

Currently time-division multiplexing (TDM)-PON has been implemented. Ethernet PON(EPON) based on IEEE 802.3ha, asynchronoustransfer mode (ATM) PON (APON) based on

ITU-T G.983.1, and Gigabit PON (GPON)based on ITU-T G.984 are typical examples ofthe implemented TDM-PON [1]. However, dueto the uplink time-sharing, TDM-PON systemsare limited in supporting bursty traffic and pro-viding multirate transmission.

Wavelength-division multiplexing (WDM)-PON is another technique introduced to resolveTDM-PON’s shortcomings. Furthermore, matur-ing key optical technologies and the emergenceof advanced optical devices are reducing the costof WDM-PON deployment. Therefore, it isexpected that WDM-PON schemes will be stan-dardized and widely implemented. However, inWDM-PON the number of available wavelengthsis not adequate to support users of access net-works. Moreover, assigning an individual wave-length to a user decreases bandwidth efficiencyand increases the coarseness of data granularity.

Optical code-division multiplexing (OCDM)as a viable multiplexing technique is receivingmuch attention as a promising access techniqueto share common resource among asynchronoususers without any central controller [2]. TheOCDM technique is becoming an attractive can-didate in the next-generation optical networkand has been considered to be used in a PON.This is mainly due to the attractive properties ofOCDM such as flexible and asynchronous band-width sharing, statistical multiplexing, provision-ing differentiated QoS at the physical layer, andthe capability to secure data transmission using apseudo random signature. Hybrid OCMD/WDM-PON is another interesting scheme pro-posed to resolve WDM-PON and utilize OCDMcapabilities in future PONs [3].

In this article we introduce a novel hybridOCDM/WDM PON scheme to support multirateand multi-QoS transmission in PON. The idea isbased on utilizing multilength variable-weightoptical orthogonal codes (MLVW-OOCs) as thesignature sequence of the OCDM scheme. Thelength and weight of OOCs are designed basedon the characteristics of the supported classes ofservice. Furthermore, in order to improve thethroughput of the presented scheme, we proposeto employ a multilevel signaling technique and aninterference remover structure based onadvanced optical logic gates [4, 5].

0163-6804/11/$25.00 © 2011 IEEE

This work was supportedin part by Iran NationalScience Foundation(INSF).

ABSTRACT

In this article we present a new scheme tosupport multirate and multi-quality-of-servicetransmission in passive optical networks basedon a hybrid wavelength-division multiplexing/optical code-division multiplexing scheme. Theidea is to use multilength variable-weight opticalorthogonal codes as signature sequences of ahybrid WDM/OCDM system. To provide therequested classes of service, the code weight andcode length of MLVW-OOCs are designed basedon the characteristics of the requested classes ofservice. In order to mitigate multiple accessinterference, we propose to utilize a multilevelsignaling technique and interference removerstructure based on advanced optical logic gateelements. We show that utilizing such a tech-nique improves the QoS of the proposed scheme.


Hamzeh Beyranvand and Jawad A. Salehi, Sharif University of Technology

Multirate and Multi-Quality-of-ServicePassive Optical Network Based onHybrid WDM/OCDM System

SALEHI LAYOUT 1/19/11 3:29 PM Page 65


The rest of the article is organized as follows.In the next section we review conventionalTDM-PON, WDM-PON, and WDM/OCDM-PON. Our proposed multirate multi-QoSWDM/OCDM-PON is then presented. We intro-duce multilevel signaling and the interferenceremover via optical logic gates. The article isconcluded in the final section.

PASSIVE OPTICAL NETWORKAs mentioned above, the ultimate solution tohandle increasing data traffic at the access net-work is FTTH. Basically, three architectures maybe used to implement FTTH as shown in Fig. 1

[1]. The possible architectures are point-to-point,active star, and passive star. In the point-to-pointarchitecture an individual fiber runs between thecentral office (CO) and each end user. Althoughpoint-to-point architecture provides the ultimatecapacity and can support possible future high-data-rate applications, it needs many fibers,which increases the installation cost. Further-more, for each fiber (home) we need a terminalat the CO, which complicates the CO architec-ture and raises scaling and powering issues. Onthe other hand, in the active star architecture asingle fiber runs between the CO and an activenode closed to end users. End users are connect-ed to the active star by individual branchingfibers. In the active star architecture only a sin-gle fiber is needed as a feeder, with a number ofshort branching fibers to connect end users andthe active star, so installation cost is reduced.However, due to the presence of the active star,the powering issue remains. In the passive stararchitecture the active node of the active stararchitecture is replaced by a passive node. Thepassive node acts as a power splitter and powercombiner to split the received signal from thefeeder fiber among the branching fibers andaggregate branching fibers signals into the feed-er fiber, respectively. In such an architecture, inaddition to the cost reduction due to using feed-er fiber, the passive power splitter/combinerresolves the power issue. Therefore, the passivestar architecture as a cost-effective solution hasreceived much attention and is becoming a pop-ular architecture to implement FTTH. It is inter-esting to note that the passive star architecture isreferred to as a passive optical network (PON).

Based on the multiplexing method used toshare the common resource of the feeder fiberamong end users, we have three scenarios:TDM-PON, WDM-PON, and OCDM-PON.These scenarios are compared in Fig. 2. Asshown in the figure, in the TDM-PON scenariobandwidth of the feeder fiber is slotted in thetime domain, and each user (optical networkunit, ONU) is assigned a dedicate time slot. Onthe other hand, in the WDM-PON scenario,bandwidth of feeder fiber is divided into multi-ple bands, and each user is assigned a dedicatedwavelength. As shown in Fig. 2, in the OCDM-PON scenario bandwidth of feeder fiber is divid-ed among end users in code space. In thisscenario each user is assigned a specific codeconsidered as the user address and employed totransmit bitstreams. In an OCDM systememploying on-off keying modulation, to send bit“1,” users transmit the signals encoded by theassigned codeword; and to send bit “0,” theytransmit no signal. At the receiver front-end, byusing the corresponding decoder the bitstreamcan be extracted.

In comparison to TDM-PON, WDM-PONprovides more bandwidth and can support futurelarge amoaunts of data traffic. However, thenumber of available wavelengths is not ade-quate, and data granularity of the access net-work is coarse. On the other hand, OCDM-PONprovides flexible bandwidth, and users can trans-mit asynchronously. In OCDM-PON the QoS ofthe users is limited by multiple access interfer-ence (MAI), which is a function of the number

Figure 1. Different architectures for FTTH: a) point-to-point; b) active star; c)passive star.

Active star

Feeder fiber

Feeder fiber

Power splitter/combiner

Central office (CO)

Central office (CO)

Central office (CO)

To metro network

To metro network

To metro network

(a)

(b)

(c)

End users



of transmitting users [2]. In order to guaranteethe desired QoS, the number of transmittingusers needs to be restricted. So in OCDM-PONthe number of supported end users is limited bythe number of available codewords and MAI.

Hybrid WDM/OCDM-PON is an ultimatesolution to resolve scarceness of the number ofavailable channels and codes, and the coarsenessof the data granularity [6]. To resolve the MAIlimit of the OCDM system, we propose to use arecently introduced multilevel signaling techniquebased on advanced optical logic gates [4, 5].

MULTIRATE AND MULTI-QOSHYBRID WDM/OCDM PON

In hybrid WDM/OCDM-PON, OCDM is usedin each WDM channel to share the availablebandwidth among end users. In Fig. 3a the band-width classification of the hybrid WDM/OCDM-PON is presented. As shown in the fig-ure, in each wavelength Nc channels are avail-able where Nc is the number of availablecodewords.

Generally, OCDM, based on coding princi-ples, is divided into two types, coherent andincoherent. In a coherent OCDM scheme, thephase of an optical signal is encoded by bipolarcodes such as m-sequence, Gold code, orHadamard. On the other hand, in an incoherentOCDM scheme the intensity of an optical signalis encoded by unipolar codes such as OOC orprime code. In this article we employ an inco-herent OCDM scheme based on OOCs.

An OOC is a family of (0, 1) sequences withgood auto- and cross-correlation properties [2].In the literature an OOC is characterized by (L,w, λ) where L is the code length, w is the codeweight that determines the total number of onesin each codeword, and λ is the maximum valueof shifted auto-correlation and cross-correlation.The number of available OOCs (Nc) is limitedby the well-known Johnson bound as follows [2]:

(1)

Equation 1 indicates that the number of avail-able OOCs, Nc, is a function of code parameters(L, w, λ). As an example, for L = 101, w = 5,and λ = 1 we have Nc ≤ 5 while for λ = 2 wehave Nc ≤ 165. Thus, increasing the maximumcorrelation increases Nc at the expense of inter-ference excess and performance degradation.

It is worth noting that the QoS in such anOCDM system depends on the number of inter-fering users (NI) and the code parameters (L, w,λ). The code weight has a direct effect on QoS.By increasing the code weight, QoS of transmit-ting users is improved, while according to theJohnson bound, Nc is reduced. On the otherhand, code length has a reverse relation withtransmission rate and a direct relation with thenumber of available codewords. So for a specificbandwidth, Nc of high data rate is less than Nc oflow data rate.

In conventional OOCs all codes have thesame parameters, so all users have the same

transmission rate and QoS. Multilength OOCs(ML-OOCs) have been designed to support mul-tirate transmission. In ML-OOCs codewords aredivided into multiple classes. Although all code-words have the same weight, each class has aspecific code length. So using ML-OOCs, we cansupport multirate transmission. In order to sup-port multi-QoS transmission, variable-weightOOCs (VW-OOCs) are designed. In VW-OOCsall codewords have the same code length andcodewords are divided into multiple classes hav-ing specific code weight. So using VW-OOCs,we can support multi-QoS transmission in accessnetwork.

In order to jointly support multirate andmulti-QoS transmission, MLVW-OOCs havebeen designed. In MLVW-OOCs codewords aredivided into multiple classes, and codewords ofeach class have a specific code length and codeweight. So utilizing MLVW-OOCs, we can joint-ly support multirate and multi-QoS transmission.In OCDM based on MLVW-OOCs, high-weightcodewords are assigned to high QoS users andshort-length codewords are assigned to high-rateusers. In Fig. 4 different OOC families are com-pared.

Generally, MLVW-OOCs are characterizedby (L = {L1, L2, …, LQ}, w = {w1, w2, …, wQ},Nc = {Nc1, Nc2, …, NcQ}, Q, Γ), where Li, wi,and Nci denote the code length, code weight,and number of available codes in class i, respec-tively. In addition, Q denotes the number ofspecified classes in the network, and Γ indicatesthe cross correlation matrix, which is defined asΓ = {I(n,m), for n,m = 1, 2, …, Q}. I(n,m) denotethe maximum correlation between class n and

NL L L

w w w wc ≤−( ) −( ) −( )−( ) −( ) −( )1 2

1 2

……

λ

λ.

Figure 2. Bandwidth classification in different PON schemes.

Code

ONU: optical network unit

ONU1 ONU2 ONU3 ONUNw

λ1

Wavelength

Wavelength

λi : ith wavelength

Tim

e

Tim

e

WDM-PON

λ2 λ3 λNw

ONU1 ONU2 ONU3 ONUNw

ONUNc

CNc

C2

C1

ONU2

ONU1

Time

Ti : ith time slot

Ci : ith code

Wav

elen

gth

TDM-PON

OCDM-PON

TNtT3 T2 T1



class m codewords. If n = m, I(n,m) is referred toas intra-cross-correlation, which indicates themaximum cross-correlation between the sameclass codewords; if n ≠ m, I(n,m) is referred to asinter-cross-correlation, which shows the maximumcross-correlation between two codes from differ-ent classes.

It is worthy to note that in the proposedmulti-rate hybrid WDM/OCDM-PON the maxi-mum transmission rate is limited by the shortestcode length. Furthermore, the maximum trans-mission rate in hybrid WDM/OCDM-PON isless than that of OCDM-PON. In order to sup-port ultra high rate services we propose to designa PON scheme utilizing both OCDM and hybridWDM/OCDM scenario. In this scheme, a num-ber of codewords are used to encode optical sig-nal along all wavelength, same as OCDM PON.The remained codewords are used in the wave-length windows, the same as in WDM/OCDM-PON. The bandwidth sharing in this scheme is

shown in Fig. 3b. As it can be seen in the figure,codes C1 up to Ck are used in OCDM scenarioto support ultra high rate service and codes Ck+1up to CNc are used in the hybrid WDM/OCDMscenario to support high-, medium-, and low-rateservices. We refer to this scheme asOCDM+WDM/OCDM-PON scenario.

MAI MITIGATION USING AMULTILEVEL SIGNALING TECHNIQUE

Multiple access interference is the dominantfactor degrading QoS of an OCDM system. In[4], multilevel signaling technique has beenintroduced to mitigate MAI. In conventionalincoherent one-level OCDM system, all userstransmit at the same power level. In such a sys-tem, tapped delay lines (TDLs) and an ANDlogic gate (ALG) structure are used as encoderand decoder, respectively.

Figure 3. a) Bandwidth classification in WDM/OCM-PON, b) bandwidth classification in OCDM+WDM/OCDM-PON.

Code

λ1 λ2 λNw

(a)

Wavelength

WDM/OCDM-PON

Tim

e

ONUNc

ONUNc×Nw

ONU2

ONU1

C1

C2

CNc

C1 C1

C2C2

CNc CNc

Code

λ1 λ2 λNw

(b)

Wavelength

OCDM + WDM/OCDM-PON

Tim

e

ONUNc

ONUk+1

ONUk

ONU1

CNcCNc CNc

Ck+2 Ck+2

Ck+1 Ck+1

Ck

C1



Figure 4. Different OOCs families.

Multi-length OOC (ML-OOC)

Multi-length variable weight (OOC (MLVW-OOC)

Regular OOC (equalweight equal length)

L1

L1

L1

L1

L2

L2

L2

L1

L1

Variable weight OOC (VW-OOC)

L1

Figure 5. a) Multi-stage receiver structure, b) interference remover structure.

Input

Output

H2

H2

P0

P1

(a)

(b)

τ1

τ2

OpticalAND

Multi-stageinterference remover

H

OutputInput

τw

P2 2P1 2P2

Input

Output

P0

P1

H1

P2 2P1 2P2

2P0

P2

2P1

H1

OutputInput3P0

P1

2P2

P2

In a multilevel signal-

ing technique, users

are categorized into

multiple groups and

users of each group

transmit at a specific

power level. In such

a system a multi-

stage interference

remover based on

optical logic gates is

an essential element

to mitigate interfer-

ence of users trans-

mitting at other

power levels.


In a multilevel signaling technique, users arecategorized into multiple groups, and users ofeach group transmit at a specific power level. Insuch a system a multistage interference removerbased on optical logic gates is an essential ele-ment to mitigate interference of users transmit-ting at other power levels. The structure of atypical receiver based on multi-stage interfer-ence remover is shown in Fig. 5a. The structureof interference remover relates to the number ofpower levels and the depth, i.e., the number ofstages, of interference removing. In Fig. 5b thestructure of two-stage interference remover in atwo-level system is shown [4]. In the two-levelsystem users are divided into two groups, group1 and group 2. The users of group 1 and group 2transmit at power levels P1 and P2, respectively(assume P2 > P1). In the figure, H1 and H2 arethe interference remover of group 1 and group 2users, respectively. From Fig. 5b we can observethat H1 removes interferences at power levels P2and 2P2 and H2 removes interferences at powerlevels P1 and 2P1. So, in such a two-stage two-level system for group 1 users pulses at powerlevel P1 and 3P2 have the same effect. On theother hand, for group 2 users pulses at powerlevel P2 and 3P1 have the same effect. As a mat-ter of fact, in multilevel signaling technique bytransmitting at different power levels and utiliz-ing multi-stage structure, pulses at the otherpower levels can be distinguished and removed.

Obviously increasing the number of powerlevel results to the performance improvementdue to the increase of the interference mitiga-tion capability of a multistage structure. In Fig. 6the probability of error of two-class OCDM sys-tem using MLVW-OOC characterized by (L ={400, 400}, w = {8, 12}, Nc = {12, 12}, Q = 2)is shown. As it can be observed in the figure,using multilevel signaling technique the proba-

bility of error is decreased. Furthermore, theincrease of the number of stages of interferenceremover results to the performance improve-ment.

CONCLUSIONIn this article we have presented a new schemeto support multirate multi-QoS transmission inoptical passive networks. Utilizing MLVW-OOCin hybrid WDM/OCDM-PON we can providethe requested classes of services. The codeweight and the code length of MLVW-OOC aredesigned based on the characteristics of therequested classes of services. Furthermore, tosupport ultra high rate service, we have pro-posed to use a combination of OCDM andWDM/OCDM scheme in PON, OCDM+WDM/OCDM-PON.

In order to mitigate MAI and to increase net-work throughput sufficiently, we have utilized amultilevel signaling technique and interferenceremover based on advanced optical logic gateselements. In such a technique interferenceremover, mitigate inference based on the powerlevel on the input signals. We have showed thatusing this multilevel signaling technique, theQoS of the system is improved.

REFERENCES[1] T. Koonen, “Fiber to the Home/Fiber to the Premises:

What, Where, and When?” Proc. IEEE, vol. 94, no. 5,May 2006, pp. 911–34.

[2] J. A. Salehi, “Code Division Multiple-Access Techniquesin Optical Fiber Networks — Part I: Fundamental Princi-ples,” IEEE Trans. Commun., vol. 37, no. 8, Aug. 1989,pp. 824–33.

[3] K. Kitayama, X. Wang, and N. Wada, “OCDMA OverWDM PON-Solution Path to Gigabit-Symmetric FTTH,”IEEE J. Lightwave Tech., vol. 24, no. 4, Apr. 2006, pp.1654–62.

[4] B. M. Ghaffari and J. A. Salehi, “Multiclass, Multistage,and Multilevel Fiber-Optic CDMA Signaling TechniquesBased on Advanced Binary Optical Logic Gate Ele-ments,” IEEE Trans. Commun., vol. 57, no. 5, May2009, pp. 1424–32.

[5] H. Beyranvnad, B. Ghaffari, and J. A. Salehi, “Multirate,Differentiated-QoS, and Multilevel Fiber-Optic CDMASystem via Optical Logic Gate Elements,” IEEE J. Light-wave Tech., vol. 27, no. 19, Oct. 2009, pp. 4348–59.

[6] H. Beyranvand and J. A. Salehi, “All-Optical Multi-Ser-vice Path Switching in Optical Code Switched GMPLSCore Network,” IEEE J. Lightwave Tech., vol. 27, no. 12,June 2009, pp. 2001–12.

BIOGRAPHIESHAMZEH BEYRANVAND ([email protected]) received aB.S. degree (with honors, first rank) in electrical engineer-ing from Shahed University, Tehran, Iran, in 2006 and anM.S. degree from Sharif University of Technology (SUT),Tehran, Iran, in 2008. He is currently working toward aPh.D. degree in the Department of Electrical Engineering atSUT. Since summer 2007, he has been working as a mem-ber of the Optical Networks Research Laboratory (ONRL),SUT.

JAWAD A. SALEHI [M‘84, SM‘07, F‘10] ([email protected])received a B.S. degree from the University of California,Irvine, in 1979, and M.S. and Ph.D. degrees from the Uni-versity of Southern California (USC), Los Angeles, in 1980and 1984, respectively, all in electrical engineering. From1984 to 1993 he was a member of technical staff of theApplied Research Area, Bell Communications Research (Bell-core), Morristown, New Jersey. He is currently a full profes-sor at the ONRL, Department of Electrical Engineering, SUT.

Figure 6. Probability of error of one-level and two-level OCDM systems.

Number of interfering users

Probability of error

50

10-12

10-14

Pe

10-10

10-8

10-6

10-4

10-2

10 15 20 25 30 35 40

Conventional one-level scenarioTwo-level one-stage scenarioTwo-level two-stage scenario




INTRODUCTION

Since the emergence of the passive optical net-work (PON) as a crucial access technology, aconsiderable amount of research has focused onfundamental design issues such as resource allo-cation [1]. PON technologies are constantlyadvancing toward increased capacity, embodiedprimarily by high-speed time-division multiplex-ing (TDM) PONs and wavelength-division multi-plexing (WDM) PONs. In addition, importantadvances have been achieved to extend PONreaches, hence multiplying their subscribercounts. As a consequence, PONs are destined tocarry huge amounts of traffic in the near future.The search for practical and cost-effective sur-vivability and maintenance mechanisms is there-fore becoming key to the continued developmentof viable PON solutions.

The standardization of PON survivabilitymechanisms started within the broadband PON(BPON) standardization effort. InternationalTelecommunication Union — Telecommunica-tion Standardization Sector (ITU-T) G.983.1described a set of four PON protection configu-rations that were subsequently narrowed down

to two protection schemes in ITU-T Recommen-dations G.983.5 (BPON) and G.984.1 (GigabitPON), Type B and Type C protection. Type Bprotection duplicates both the feeder fiber andoptical line terminal (OLT) interface and usesan N:2 splitter at the remote node (RN), whereN is the number of supported optical networkunits (ONUs). The Type B configuration henceoffers protection only against the failure of theOLT interface equipment or a cut in the feederfiber. In contrast, Type C duplicates the wholePON network infrastructure, including ONUand OLT interfaces, as well as the splitter, thusproviding additional protection against ONUequipment failures. EPON has no standardizedprotection scheme but may adopt Type C pro-tection through the adaptation of Ethernet pro-tection switching defined in ITU-T G.8031 [2].In both Type B and C protection configurations,automatic protection switching is typically trig-gered by layer 2 alarms related to the loss of sig-nal intensity or quality. This has two importantconsequences [3]. First, the physical PON infra-structure is not entirely visible to the networkmanagement system (NMS) for fault manage-ment operations. Second, failures within thefiber plant are likely to entail service disruptionbefore being detected, leading to revenue lossesand customer dissatisfaction.

Due to the high capital expenditures incurredby the deployment of such protection mecha-nisms, operators have resorted to troubleshoot-ing and restoration once faults are detected [4].Troubleshooting is an important network main-tenance function that involves locating and iden-tifying any source of fault in the network. Theabove-mentioned ITU-T protection configura-tions make no specific provisions to identify andlocalize faults within the optical infrastructureand defer the task to maintenance standards (Lseries). ITU-T L.53 (2003) is the first standardto specifically address the maintenance of PONsby recommending the use of optical time-domainreflectometry (OTDR)-based techniques fortroubleshooting.

Whether it is for survivability or maintenance

0163-6804/11/$25.00 © 2011 IEEE

ABSTRACT

As PONs carry increasing amounts of data,issues relating to their protection and mainte-nance are becoming crucial. In-service monitor-ing of the PON’s fiber infrastructure is apowerful enabling tool to those ends, and anumber of techniques have been proposed, someof them based on optical time-domain reflec-tometry. In this work we address the requiredfeatures of PON monitoring techniques andreview the major candidate technologies. Wehighlight some of the limitations of standard andadapted OTDR techniques as well as non-OTDR schemes. Among the proposed optical-layer monitoring schemes, we describe our noveloptical-coding-based reflection monitoring pro-posal and report on recent progress. We endwith a discussion of promising solution paths.


Mohammad M. Rad, University of Waterloo

Kerim Fouli, Optical Zeitgeist Laboratory, INRS

Habib A. Fathallah, King Saud University

Leslie A. Rusch, Université Laval

Martin Maier, Optical Zeitgeist Laboratory, INRS

Passive Optical Network Monitoring:Challenges and Requirements

RAD LAYOUT 1/19/11 3:32 PM Page 71


purposes, there is a growing need for the moni-toring of the PON fiber plant. PON monitoringtechnology automatically identifies and localizesfaults of the in-service PON optical infra-structure. In doing so, it provides the NMS withenhanced optical infrastructure visibility in realtime, thus speeding up the detection and local-ization of faults. Monitoring avoids the opera-tional expenditures (OPEX) and large servicerestoration times of offline troubleshooting, thusenabling wider service differentiation andstronger QoS guarantees. In addition, it pavesthe way to potentially enhanced physical layerprotection mechanisms.

Accordingly, PON monitoring has beenreceiving increasing attention, and a variety ofproposals have emerged [3, 5]. To accommodatethe demand for monitoring technology, the ITU-T L.66 (2007) Recommendation standardizes thecriteria for in-service maintenance of PONs. Itreserves the U-band (1625–1675 nm) for mainte-nance and lists several methods to implementPON in-service maintenance functions such asOTDR testing, loss testing, and power monitor-ing (i.e., monitoring a proportion of the signalpower).

Note that PONs need to be tested duringinstallation to ensure that all fiber links andcomponents are properly installed and working.Therefore, link characterization and diagnosisduring network installation is also of greatimportance and can easily be performed usingone of the aforementioned testing methods.However, there is a growing need to monitorfiber link failures and degradations without dis-turbing ongoing services. In this article we focuson the monitoring of in-service live PONs (i.e.,after installation), where a service interruptiondue to monitoring is not permissible.

In this article we review and compare themajor optical-layer PON monitoring proposals,and address advantages and challenges of themonitoring techniques for deployment of high-capacity PONs. In the next section we enumer-ate the desired features and major requirementsof in-service PON monitoring techniques. Wethen briefly review the basic principles of OTDRfor point-to-point monitoring, and outline thechallenges and limitations of standard OTDR inPON (point-to-multipoint) applications. Non-OTDR-based techniques are then addressed.We particularly focus on two recently proposedtechniques: Brillouin frequency shift assignmentand optical-coding (OC)-based reflection moni-toring. We also address in detail the advantagesand disadvantages of each of the mentionedtechniques in PONs. Finally, we discuss promis-ing solution paths before concluding in the finalsection.

REQUIRED FEATURES OFPON MONITORING TECHNOLOGIES

GENERAL REQUIREMENTSBy definition, an effective monitoring technologyshould be able to both detect a fault and providethe NMS with useful information for root causeanalysis. Useful monitoring information enablestechnicians to perform fast network repair,

hence increasing PON reliability and reducingoperational expenses.

The most important issue in PON monitoringtechnology is cost, including capital expenditure(CAPEX, i.e., the initial cost of the monitoringtechnology per customer) and operational expen-diture (OPEX, i.e., the cost of system mainte-nance). The reason is that the PON market ishighly cost-sensitive, especially for the compo-nents not shared between customers, such as dis-tributed monitoring nodes. Therefore, anexpensive technology, even though it may pro-vide in-service full visibility of the optical infra-structure to the network operator, may not beinteresting for PON applications. Consequently,the monitoring technology requires simpledesign, fabrication, and implementation proce-dures to minimize the cost.

Capacity, in terms of the number of PONbranches or distribution fibers that can be simul-taneously monitored, is the second desired fea-ture. Candidate monitoring technologies shouldbe able to support at least the maximum split-ratio of current PON standards (e.g., 1:128 forITU-T G.984 GPON). Accommodating largersplit-ratios increases the number of supportedcustomers, thus amortizing the expenses of theservice provider and generating higher benefits.The monitoring technology should thus be scal-able in order to enable seamless and continuousupgrades of the PON infrastructure (i.e., PONcapacity, reach, and customer base) at low costs.The simplicity of the monitoring architecture andcomponents directly affects the cost, and ishence an important requirement. In addition, asfor any maintenance and protection mechanism,reliability is primordial. Furthermore, to operatein-service, the desired monitoring technologyshould act transparently to the data band signalssuch as the L and C bands. Therefore, strict iso-lation between the data band and monitoringsignals is required.

AUTOMATIC AND CENTRALIZED MONITORINGAn automatic monitoring technique allows thenetwork operator to detect faults without resort-ing to in-field technicians or relying on customerequipment or feedback. This feature is highlydesirable as the deployment of in-field personnelis usually equated with increased PON downtimeand OPEX. Besides, it allows the operator toenhance customer satisfaction by potentiallyreacting to faults before service disruption (e.g.,through automatic protection switching [APS]).A fully automatic monitoring system is usuallycentralized, allowing the NMS, from its locationin the central office (CO), to remotely acquirecomplete live network information withoutrequiring the collaboration of customers or theirONUs, as does traditional OTDR in a point-to-point link.

Both centralized and distributed approacheshave been proposed for monitoring the fiber linkquality of a PON [3–5]. In distributed (decen-tralized) monitoring strategies, active modulesare placed inside the ONUs to measure perfor-mance and report to the NMS. These modulesperiodically evaluate the uplink for a specificfiber branch and may be implemented electroni-cally at the ONU.

An automatic

monitoring tech-

nique allows the

network operator to

detect faults without

resorting to in-field

technicians or relying

on customer

equipment or feed-

back. This feature is

highly desirable as

the deployment of

in-field personnel is

usually equated with

increased PON

downtime and

OPEX.



Although the distributed approach effectivelyidentifies fiber link degradation, it is ineffectivewhen there is an interruption in the fiber link(e.g., a fiber cut) as it requires the real-time col-laboration of ONUs. For instance, a missingmonitoring signal at the NMS can be interpretedas the result of either a fiber fault or an elec-tronic malfunction at the ONU. While the oper-ator may take advantage of information on linkquality provided by the ONU, the case is strongfor a separate, independent, and rapid indicatorof whether the fault occurred in the client’s orthe operator’s domain. Therefore, a centralizedautomatic monitoring technology is highly desir-able for PON applications.

OPTICAL TIME DOMAINREFLECTOMETRY

Optical-time-domain-reflectometry-based moni-toring has been implemented for the first timefor optical carriers in long-distance transmissionsystems. OTDR is an efficient way to character-ize an optical link while accessing only one end,as appropriate for point-to-point links. It oper-ates as follows. The OTDR equipment launchesa short light pulse into the fiber and measuresthe backscattered light. Rayleigh scattering andFresnel reflections are the physical causes of thisscattering behavior [4]. Due to the measuredpower at the OTDR receiver, a trace of thepower vs. the distance may be computed, repre-senting the impulse response of the link undertest, as shown in Fig. 1. This trace can be used toextract information about link faults, includingfiber misalignment, fiber mismatch, angularfaults, dirt on connectors, macro-bends, andbreaks. These faults are usually referred to asevents on the OTDR trace. For instance, thejumps in Fig. 1 correspond to the insertion lossof different network components, whereas thepower reflection peak at 40 km indicates theFresnel reflections at the fiber-air interface, sig-nifying the fiber end. After the fiber end, nobackscattering is detected, and the trace drops toreceiver noise levels.

CHALLENGES OF STANDARD OTDR FOR PONWhile providing automatic monitoring and fullcharacterization of the fiber link, OTDR is inef-fective for PON point-to-multipoint (PMP) net-works [3–6]. This is because a branchbackscattering signal in a PON can be partiallyor totally masked by other branch signals. ForPONs, the total power measured by the OTDRis a linear sum of all powers coming from differ-ent branches. Useful information can be extract-ed from the global backscattering trace whenreturns from individual branches are separatedin time. Otherwise, extracting the desired infor-mation from the OTDR trace may require con-siderable offline signal processing, or simply beimpossible.

OTDR analysis for a branched network com-pares the backscattering trace with referencereturns acquired under controlled conditions. Asimulator interprets any deviation from the ref-erence signals [7, 8]. The accuracy of such soft-ware depends on the quality of the simulator as

well as the uncertainties in both the measuredtraces and the simulated return based on refer-ence measurements. In the event of equidistantbranch terminations, the challenge is severe. Asthe network size increases, analysis complexityincreases, leading to less reliable monitoring.

In addition, the huge loss by passive splitters,typically located at the remote node (RN), leadsto a significant drop in measured power. Forexample, a 1:32 splitter at the RN leads to 15 dBloss in the total backscattered light from eachbranch. The RN then resembles a fiber end, andno useful information can be extracted beyondthe RN. In traditional OTDR, losses higher than3–7 dB are identified as end-of-fiber. However,it is reported that by modifying the OTDR anal-ysis, testing can be performed through splitterswith losses up to 20 dB. This type of OTDR isusually referred to as PON-tuned OTDR [6].

MODIFIED OTDR SOLUTIONSReference Reflector — In order to reducePON OTDR analysis complexity, a variety ofsolutions have been proposed to distinguish indi-vidual fiber branches. The most well-knowntechnique is the use of reference reflectors (RR-OTDR) [5] assigned to each fiber branch to ren-der it distinguishable from others in the totalmeasured OTDR trace.

The principle of the reference reflectors isillustrated in Fig. 2. A reflector can be realizedby different methods [5]. It could be wavelengthselective and inserted in the input of the ONUconnector to act as a stop filter. It also could bea non-wavelength-selective reflector placed on aseparate tap (lower part of Fig. 2). Note that thereflectors at each fiber end are identical, and allreflect the same wavelength, each producing areflection for its corresponding branch. To dis-tinguish between the branches, it is critical toadjust the fiber lengths in each branch to avoidtemporal overlapping. In this way a single OTDRreturn will have each branch return located in anisolated time interval.

By monitoring the stability and level of reflec-tions from reference reflectors placed at each

Figure 1. Typical trace of OTDR of a fiber link.

Distance (km)50

-30

-35

Back

scat

tere

d po

wer

(dB

)

-25

-20

-15

-10

-5

0

15 2010 25

Backscatter

Fusion splice

Connector pair

Front connector

Bend Crack

Fiber end

Noise

30 35 40 45 50



fiber branch end, the integrity of a specificbranch can easily be investigated. The OTDR isexploited for a full characterization of the corre-sponding fiber branch. The shift in the powerlevel of the reference reflection for a desiredbranch provides useful information for theOTDR trace analysis. Checking the stability ofthe strong reflection (located well above thenoise level) is faster and easier than analyzingthe OTDR trace, and these reflectors are oftenused as a first fault indicator in most OTDR-based techniques.

In RR-OTDR, the choice of the fiber lengthsrequires an important trade-off between OTDRsensitivity and resolution. The required fiberlength is proportionally related to the transmit-ted OTDR pulse width as well as the relativedistances between the customers. While for veryshort pulses small fiber lengths are required, theOTDR sensitivity is very poor, limiting allowablesplitter size at the RN. For longer pulses, sensi-tivity improves. However, significantly long delaylines are required, leading to lower OTDR accu-racy and larger dead zones (i.e., the area of anOTDR trace where events are not distinguish-able).

The NMS requires updated information onthe customer distribution in the network; other-wise, customer relocations cause false alarms.The RR-OTDR scheme does not scale well with

large network sizes. In fact, due to the hugesplitter loss at the RN, it is difficult to extractuseful information from the OTDR trace beyondthe RN. In addition, as the network size increas-es, the selection of an optimal delay linebecomes more challenging, and the complexityof the OTDR trace increases.

Multi-Wavelength Approach — One othersimple approach would be employing a multi-wavelength source and an arrayed waveguidegrating (AWG) at the RN. This reduces thePON monitoring problem to point-to-point linkcharacterization, as illustrated in Fig. 3. In thiscase the tunable multiwavelength OTDR sourceshould be very stable for reliable monitoring.Isolation between the monitoring and data sig-nals will be more strict than single-wavelengthOTDR. In addition to its high cost, this tech-nique also has limited capacity due to practicallimitations and very poor spectrum efficiency [9].Its scalability is hence very low. Nevertheless,this approach provides a centralized monitoringsystem that enables the NMS to both detect andlocalize faults.

Electronic Solutions — Note that the function-ality of an OTDR device can be implementedwithin the ONU at the customer side [10]. Thisapproach, known as embedded OTDR, leveragesthe electronics at the ONU for a cost-efficientsolution, such that embedded OTDR within theONUs becomes an integral part of the monitor-ing network. In this scheme the monitoring seg-ment transmits an OTDR trace from the ONUupon request of the NMS at the CO when thecorresponding ONU is idle over the upstreamchannel. Therefore, this solution relies on in-band upstream signaling. As mentioned earlier,this solution is inadequate when a fiber cut hap-pens, as all data and control channels linking theNMS to the ONU are disrupted.

CRITICAL ISSUES FOR THE USE OFOTDR IN PONS

As the basic equipment for the above automatictest systems, OTDR requires suitable technicalcharacteristics. The most important performancecharacteristics of OTDR-based techniques arespatial resolution, dynamic range, dead zone,wavelength stability, and minimum sensitivity[4–7]. Adequate performance requirementsshould be met for an OTDR to be an effectivemonitoring solution for future PONs. Forinstance, as the splitting ratio increases, largerdynamic ranges are required. Increasing thetransmitted pulse width is not an efficient solu-tion, as it decreases the spatial resolution andenlarges the dead zone of the OTDR. Also, thelaunched power is limited due to nonlineareffects. Generally, the capacity of OTDR-basedtechniques are limited to tens of customers, andsystem scalability is a serious concern. Recallthat although cost is an important issue, it is notcritical since the OTDR is shared among net-work clients.

The leakage of the monitoring power fromthe U band to the data band (C and L) maycause performance degradation for data commu-

Figure 2. Use of a reference reflector for OTDR-based automatic monitoring ofPONs.

RM

Data wavelength

Wavelength separator(coupler)

Feeder fiber

CO

OLT

ONU

OTDR

Monitoring wavelengthStop filter for data

Reference reflectorTerminating fiber

ONU

RN

Figure 3. OTDR for PON via one monitoring wavelength per ONU.

ONU1 Λd

Data Monitoring

ΛmCentral office

Array waveguide

Monitoring wavelengths

Tunable OTDR

OLT

Data wavelength WDM coupler Optical filters

ONUN

ONU2

ONUN-1

RN



nications. Hence, strict isolation between thedata and monitoring bands is required. As aresult, optical sources with very high sidebandsuppression ratios and optical filters with highinsertion losses are required [11]. Other criticalissues for OTDR-based monitoring are the useof optical selectors, filters, reflectors, and WDMdevices. These devices should be cost- anddimension-effective (i.e., low cost and high den-sity) in order to be able to monitor a largeamount of fibers in future access networks.While the ITU Recommendations propose theU band for monitoring applications, the behav-ior of passive components is not very well inves-tigated for this wavelength regime. Due to thecontinuous advancement in related fields,OTDR-based techniques are expected to becomemore reliable in the future.

NON-OTDR-BASED TECHNIQUESA variety of non-OTDR techniques have beenproposed recently for the monitoring of linkquality in a PON. In this article we focus on twoof the most interesting, Brillouin frequency shiftassignment (BFSA) and OC-based PON moni-toring, and address their challenges and advan-tages.

BRILLOUIN FREQUENCY SHIFT ASSIGNMENTThis technique uses Brillouin-based OTDRs(BOTDRs) at the CO [11] and deploys specialtyfibers in the distribution segment of the PON, asshown in Fig. 4. Each fiber branch is hence dis-tinguished by a unique Brillouin frequency shiftas a signature, and is called an identificationfiber.

To monitor an individual fiber in a PON, anoptical pulse with center frequency ν is launchedthrough the network from the CO using aBOTDR. After the RN, subpulses are passedthrough different identification fibers, each ofwhich scatters a unique pre-assigned Brillouinfrequency. A specific identification fiber is thenselected by monitoring the spectrum of the

received signal. The frequency shifts aredesigned to have disjoint spectra for differentbranches. By observing peaks at center frequen-cies fk = ν – νk, as shown in Fig. 4, the status ofthe identification fiber is monitored. Further-more, by measuring the filtered backscatteredoptical signal for a specific branch, BOTDRachieves a unique trace that is identical to thetrace provided by traditional OTDR in a point-to-point link. In principle similar to the multi-wavelength OTDR approach, this centralizedtechnique provides a unique OTDR trace foreach fiber branch that lies beyond the RN.Hence, it is capable of both detecting and local-izing a fault at any branch of a PON.

While providing a centralized and completecharacterization of the identification fibers, theBFSA technique imposes significant design chal-lenges for the network infrastructure. This tech-nique requires the identification fibers to bemanufactured with different physical characteris-tics that generate and return different Brillouinfrequencies. Each identification fiber, while scat-tering a unique Brillouin frequency shift, shouldnaturally also operate as a data link to satisfy thedata transmission requirements of PONs. Inaddition to involving high CAPEX, this tech-nique has a dramatic impact on existing fibernetwork infrastructures, as new fibers have to bedesigned and all existing distribution fibersreplaced. As the capacity of the network increas-es, so does the number of required identificationfibers. This leads to more strenuous constraintson the required frequency shifts, implying theuse of more advanced manufacturing technologywith higher cost and complexity. This techniqueis hence not simply scalable and has yet todemonstrate its capability for the monitoring ofcurrently deployed PONs with standard splittingratios (e.g., GPON with 64 and 128 branches).Furthermore, the use of specialty fiber for sub-scriber drop cables adds substantially to the costof network deployment, especially when the sub-scriber take rate (i.e., the anticipated number ofsubscribers) is low. Due to the aforementioned

Figure 4. Performance monitoring based on Brillouin frequency shift assignment.

Brillouin frequency shifted spectrums

Separation

Frequency υ-υ1=f1 υ-υ2=f2

Data wavelength

Specialty fibers, each with a unique Brillouin frequency shift

CO

Standard Fiber

υ

OLT RN

f1 ONU1

BOTDR

Monitoring wavelength

Stop filter for monitoring wavelength Wavelength separator (coupler)

ONU2

ONUN-1

ONUN

f2

fN-1

fN

υ-υN-1=fN-1 υ-υN=fN

While the ITU

recommendations

propose the U band

for monitoring

applications, the

behavior of passive

components is not

very well investigated

for this wavelength

regime. Due to the

continuous

advancement in

related fields,

OTDR-based tech-

niques are expected

to become more reli-

able in the future.



reasons, this technique is very unlikely to beadopted commercially.

OPTICAL-CODING-BASED PON MONITORINGOC exploits signal-coding techniques (inspiredby optical code-division multiplexing) for con-trol- and management-layer signaling operations.In OC-based PON monitoring, passive out-of-band encoders (Encn) are placed at the extremi-ty of each PON distribution fiber to identify andmonitor it, as shown in Fig. 5a [12]. The dataand monitoring signals occupy separate wave-length bands (Λd and Λm, respectively) consis-tent with emerging standards. An optical sourceat the CO transmits the out-of-band pulsesdownstream; an optical or electronic receiver atthe CO processes the aggregate upstream reflect-ed signal.

The encoders both reflect and imprint aunique code (i.e., specific to the PON branch)on the source pulses. Waveband separators splitthe data and monitoring wavebands at the ONUand the OLT. Alternatively, a combination of in-line encoders and monitoring band-stop filtersmay be used at the branch termination pointsprior to the ONUs, as is the case for RR-OTDR.The use of simple fiber Bragg gratings (FBGs)directly inscribed at the termination of dropfibers may be regarded as a particular case ofOC-based PON monitoring. Although the sim-plest approach, the use of FBGs as wavelengthreflectors shares the low scalability and band-width efficiency drawbacks of the multiwave-length technique described earlier. Nevertheless,the use of in-fiber FBGs is particularly attractivesince it reduces monitoring power losses byremoving the requirement for waveband separa-tors at the termination of drop fibers.

While several encoders and receivers havebeen proposed, the most cost-effective and high-performance solution that has emerged is a com-bination of periodic codes [13] and an electronicreceiver [14], illustrated in Figs. 5b and 5c. Peri-odic codes were developed exclusively for thisapplication, and have low loss, low-complexityhardware, and good performance. Previouslyproposed encoders based on optical orthogonalcodes exhibited much lower performance for thisapplication [13].

A particularly attractive feature of this solu-tion is the self-configuring nature of the net-work. Encoders are installed at the drop fiberends without concern for the fiber length fromthe remote node, unlike RR-OTDR. Signal pro-cessing at the receiver differentiates returns evenfor remarkably similar fiber lengths (withinmeters). Customer relocations can be accommo-dated without a re-allocation strategy based onprevious installations.

One of the challenges in evaluating any moni-toring solution is predicting system capacity as itvaries with the specific topology of the PON,whether legacy or greenfield. Simulations of specif-ic topologies can be performed; however, they donot probe the generality of the solution. Statisticalexaminations of topologies can provide outageprobabilities for the monitoring system in general.

Several research topics remain to bring thistechnology to the marketplace. Compact low-cost periodic encoders are essential. While previ-ously proposed fiber delay lines are simple, massproduction is problematic. An integrated solu-tion for the encoder would reduce both cost andbulk. Signal processing challenges also remain toincrease the coverage capability of the decodingalgorithm. A reduced complexity maximum like-lihood receiver has been proposed in [14], but isnonetheless suboptimal and may leave room forperformance improvement.

The use of time- or wavelength-domainreflectors to identify PON branches, as in thereference-reflector and wavelength-based OTDRapproaches, may be treated as particular cases ofOC-based PON monitoring [15]. Compared towavelength-domain reflection monitoring, code-domain reflection monitoring trades its morecomplex reflectors for higher scalability andbandwidth efficiency. Compared to time-domainreflection monitoring, it avoids the use of delaylines to differentiate branch fiber lengths andoffers potentially higher scalability, particularlyin the context of future long-reach PON (LR-PON) applications. Moreover, the extension ofOC-based monitoring to LR-PONs may be facil-itated through the use of in-line reflectors [15].However, this places additional strain on themore stringent power budgeting constraints ofcode-domain monitoring.

Figure 5. OC-based PON monitoring: a) architecture; b) encoder; c) receiver for monitoring.

FPGAor DSP

Low speedelectronics

Offlineprocessing

NetworkmanagerADC

8 bits/sample

GHzsampling

High speed electronics

U banddetector

Fromnetwork

R2=100% for λmR2=0% for λd

R1=38% for λmR1=0% for λd

FBGFBGPeriodic code implementation

DDF:WDM couplerDistribution drop fiberOptical encoder

RN

DDF

Feeder

Dataλd

OLT ONUn

1000100010001

CO

U band short-pulse

Monitoringreceiver

Monitoring

(a)

(b)

(c)

Patchcord

λm

ONU2

ONU1Enc1

Enc2

Encn

EncNONUN



OC-based monitoring does not require forkliftupgrades of the PON distribution infrastructure,as does BOTDR. Consequently, the design ofsimplified and more cost-effective encoder andsystem architectures is a promising researchdirection [13]. Although OC-based monitoringdoes not offer complete fault localization (onlyfault identification on the branch), it is potential-ly more scalable than BOTDR. Therefore, it iswell suited as a component within a hybrid solu-tion, discussed in the next section.

SOLUTION PATHS TO SCALABLECOST-EFFECTIVE PON MONITORINGOur review of proposed optical-layer PON moni-toring technologies reveals two distinct and com-plementary monitoring principles: standardreflectometry and the use of dedicated reflectorsat the termination points of distribution fibers.While reflectors are capable of speedy identifica-tion of faulty distribution fibers, they lack anyaccurate fault localization capability. Conversely,whereas standard reflectometry methods areinefficient for distribution fibers, they are capa-ble of yielding highly accurate fault localizationin point-to-point settings. In addition, adaptingstandard reflectometry techniques for PONapplications is neither economical nor practical,particularly due to their lack of scalability tolarger PON sizes. In contrast, fault detection viathe monitoring of reflected signals is potentiallysimple, cost-effective, and reliable.

Therefore, we expect that comprehensivemonitoring methods will integrate both theaforementioned monitoring principles. To do so,it is necessary to break the monitoring proce-dure into two separate steps, whereby faultdetection and the identification of faulty distri-bution fiber is carried out in real time throughreflection monitoring, and precise fault localiza-tion is implemented subsequently throughOTDR.

In currently deployed PONs, the implementa-tion of reflection-based monitoring will enablefaster troubleshooting, as they will allow theNMS to exclude customer equipment malfunc-tions as the cause for a loss of signal while indi-cating the faulty distribution fiber. Techniciansequipped with high-resolution OTDR can thusbe dispatched immediately for exact fault local-ization and root cause analysis. Hence, fiberplant degradation may be detected long beforetransmission errors occur or services fail. Infuture PON deployments where protection isexpected to play an increasing role, reflection-based monitoring may be integrated with theprotection schemes as triggers for implementedAPS mechanisms, leading to reduced downtimesand higher quality of service.

CONCLUSIONSCost effectiveness and scalability are among themajor requirements for in-service monitoring ofPON fiber infrastructures. OTDR requires costlyarchitectural enhancements to deliver fast auto-matic fault localization in PON tree topologies.In this work we review some of the most promis-

ing OTDR- and non-OTDR-based proposals forPON monitoring, and address the practical chal-lenges facing their potential deployment. Ratherthan being exclusive, OTDR and alternativetechnologies such as reflection-based monitoringare complementary. Therefore, hybrid tech-niques should be investigated as promising solu-tions for delivering the maintenance andprotection functionalities required by currentand next-generation PONs. OC-based methodsare particularly attractive to implement reflec-tion monitoring in the context of increasingPON sizes.

REFERENCES[1] M. P. McGarry, M. Reisslein, and M. Maier, “Ethernet Pas-

sive Optical Network Architectures and Dynamic Band-width Allocation Algorithms,” IEEE Commun. Surveys &Tutorials, vol. 10, no. 3, 3rd qtr. 2008, pp. 46–60.

[2] F. Effenberger et al., “Next-Generation PON-Part III: Sys-tem Specifications for XP-PON,” IEEE Commun. Mag.,vol. 47, no. 11, Nov. 2009, pp. 58–64.

[3] K. Yuksel et al., “Optical Layer Monitoring in PassiveOptical Networks (PONs): A Review,” Proc. ICTON ‘08,2008, pp. 92–98.

[4] D. Anderson, L. Johnson, and F. G. Bell, Troubleshoot-ing Optical Fiber Networks: Understanding and UsingOptical Time-Domain Reflectometers, Academic Press,2004.

[5] F. Caviglia and V. C. Biase, “Optical Maintenance inPONs,” Proc. ECOC, Madrid, Spain, 1998, pp. 621–25.

[6] EXFO, “Application Notes 110 and 201”;http://exfo.com.

[7] I. Sankawa et al., “Fault Location Technique for In-Ser-vice Branched Optical Fiber Networks,” IEEE PhotonicsTech. Letters, vol. 2, no. 10, Oct. 1990, pp. 766–69.

[8] L. Wuilmart et al., “A PC Based Method for the Localizationand Quantization of Faults in Passive Tree-Structured Opti-cal Networks using the OTDR Technique,” Proc. IEEE LEOSAnnual Meeting, vol. 2, Nov. 1996, pp. 121–23.

[9] M. Thollabandi et al., “Tunable OTDR (TOTDR) Based onDirect Modulation of Self-Injection Locked RSOA forLine Monitoring of WDM-PON,” Proc. ECOC, Brussels,Belgium, Sept. 2008.

[10] W. Chen et al., “Embedded OTDR Monitoring of theFiber Plant Behind the PON Power Splitter,” Proc. IEEELEOS Symp., Benelux Chapter, Eindhoven, Netherlands,2006, pp. 13–16.

[11] N. Honda et al., “In-Service Line Monitoring System inPONs Using 1650 nm Brillouin OTDR and Fibers withIndividually Assigned BFSs,” IEEE/OSA J. LightwaveTech., vol. 27, no. 20, Oct. 2009, pp. 4575–82.

[12] H. Fathallah and L. A. Rusch, “Code-Division Multiplexingfor In-Service Out-of-Band Monitoring of Live FTTH-PONs,”OSA J. Optical Net., vol. 6, no. 7, July 2007, pp. 819–29.

[13] H. Fathallah, M. M. Rad, and L. A. Rusch, “PON Moni-toring: Periodic Encoders with Low Capital and Opera-tional Cost,” IEEE Photonics Tech. Letters, vol. 20, no.24, Dec. 2008, pp. 2039–41.

[14] M. M. Rad et al., “Experimental Validation of PeriodicCodes for PON Monitoring,” IEEE GLOBECOM ’09, Opti-cal Net. Sys. Symp., Honolulu, HI, Dec. 2009, paper no.ONS-04.6.

[15] K. Fouli, L. R. Chen, and M. Maier, “Optical ReflectionMonitoring for Next-Generation Long-Reach PassiveOptical Networks,” Proc. IEEE Photonics Society AnnualMeeting, Belek-Antalya, Turkey, Oct. 2009.

BIOGRAPHIESMOHAMMAD M. RAD ([email protected]) receivedboth his B.S.E.E. and M.S.C. from Sharif University of Tech-nology in 2003 and 2005, respectively. In September 2006he joined the Department of Electrical and Computer Engi-neering, Center for Optics, Photonics, and Lasers (COPL),Université Laval as a Ph.D. candidate. His research interestsinclude fiber-optic communications, long haul data trans-mission, multiple access networks, network monitoring,and sensor networks.

KERIM FOULI is a Ph.D. student at Institut National de laRecherche Scientifique (INRS), Montréal, Canada. Hereceived his B.Sc. degree in electrical engineering at Bilkent

Hybrid techniques

should be investigat-

ed as promising

solutions for deliver-

ing the maintenance

and protection

functionalities

required by current

and next-generation

PONs. OC-based

methods are

particularly attractive

to implement

reflection monitoring

in the context of

increasing PON sizes.


University, Ankara, Turkey, in 1998 and his M.Sc. degree inoptical communications at Université Laval, Quebec City,Canada, in 2003. He was a research engineer withAccessPhotonic Networks (Quebec City) from 2001 to2005. His research interests are in the area of opticalaccess and metropolitan network architectures with afocus on enabling technologies. He is the recipient of atwo-year doctoral NSERC Alexander Graham Bell CanadaGraduate Scholarship for his work on the architectures andperformance of optical coding in access and metropolitannetworks.

HABIB FATHALLAH [S‘96, M‘01] received a B.S.E.E degree (withHonors) from the National Engineering School of Tunis in1994, and M.A. and Ph.D. degrees in electrical engineeringfrom Université Laval in 1997 and 2001, respectively. He ini-tiated the use of Bragg grating technology for all-optical/all-fiber coding/decoding in optical CDMA systems. He was thefounder of AccessPhotonic Networks (2001–2006). He iscurrently with the Electrical Engineering Department, Col-lege of Engineering and Prince Sultan Advanced TechnologyResearch Institute of King Saud University (Riyadh, KSA),and adjunct professor with the Electrical and ComputerEngineering Department of Université Laval. His researchinterests include optical communications systems and tech-nologies, metro and access networks, optical CDMA, PONsand long reach PONs, FTTH, network monitoring, andhybrid fiber wireless (FiWi) systems.

LESLIE A. RUSCH [S‘91, M‘94, ‘SM‘00, F‘10] received aB.S.E.E. degree (with honors) from the California Instituteof Technology, Pasadena, in 1980, and M.A. and Ph.D.degrees in electrical engineering from Princeton University,New Jersey, in 1992 and 1994, respectively. She has experi-ence in defense, industrial, and academic communicationsresearch. She was a communications project engineer forthe Department of Defense from 1980–1990. While onleave from Université Laval she spent two years

(2001–2002) at Intel Corporation creating and managing agroup researching new wireless technologies. She is cur-rently a professor in the Department of Electrical and Com-puter Engineering at Université Laval performing researchon wireless and optical communications. Her researchinterests include wavelength-division multiple access usingincoherent sources for metropolitan area networks; analysisof optical systems using coherent detection; semiconductorand erbium-doped optical amplifiers and their dynamics;and in wireless communications, optical pulse shaping forhigh-bit rate ultrawide-band systems (UWB), as well as per-formance analysis of reduced-complexity receivers for UWB.She has served as associate editor for IEEE CommunicationsLetters and on several IEEE technical program committees.She has published over 70 journal articles in internationaljournals (90 percent IEEE/IEE) with wide readership, andcontributed to over 100 conferences. Her journal articleshave been cited over 750 times per the Science CitationIndex (SCI).

MARTIN MAIER ([email protected]) is an associate professor atthe INRS. He was educated at the Technical University ofBerlin, Germany, and received M.Sc. and Ph.D. degrees(both with distinctions) in 1998 and 2003, respectively. Inthe summer of 2003 he was a postdoc fellow at the Mas-sachusetts Institute of Technology (MIT), Cambridge. Hewas a visiting professor at Stanford University, October2006 through March 2007. He is a co-recipient of the 2009IEEE Communications Society Best Tutorial Paper Award.His research activities aim at rethinking the role of opticalnetworks and exploring novel applications of optical net-working concepts and technologies across multidisciplinarydomains, with a particular focus on communications, ener-gy, and transport for emerging smart grid applications andbimodal fiber-wireless (FiWi) networks for broadbandaccess. He is the author of the book Optical Switching Net-works (Cambridge University Press, 2008), which was trans-lated into Japanese in 2009.




IMT-ADVANCED AND NEXT-GENERATION MOBILE NETWORKS

lobally, mobile communications are moving towardbroadband communication systems to meet the

challenges of significantly increasing data traffic such asthat for mobile Internet applications. In order to meet thisincreasing traffic, the International TelecommunicationUnion — Radiocommunication Standardization Sector(ITU-R) initiated in 2000 the process toward the next gen-eration of International Mobile Telecommunications sys-tems, referred to as IMT-Advanced systems. ITU-RRecommendation M.1645 specifies the objectives of thefuture development of the IMT-Advanced family, amongthem to reach 100 Mb/s for mobile access and up to 1 Gb/sfor nomadic wireless access. The research community wasasked to develop concepts and system proposals to meetthese requirements. These activities were the basis for thepreparation of the ITU World Radiocommunications Con-ference 2007 to identify additional frequency spectrum formobile and wireless communications. In 2008 ITU-Rissued a Circular Letter calling for candidate radio accesstechnologies (RATs) for IMT-Advanced by taking intoaccount the identified frequency bands. In parallel, inter-national specification and standardization bodies weredeveloping technology proposals. Finally, two main tech-nologies were submitted to ITU-R for approval within theIMT-Advanced framework: one based on Third Genera-tion Partnership Project (3GPP) Long Term Evolution(LTE)-Advanced, and the other one based on IEEE802.16m. ITU-R decided in October 2010 that both sub-mitted IMT-Advanced system proposals successfully metall of the established criteria for the first release of IMT-Advanced, qualifying them as the first true fourth-genera-tion (4G) systems.

In order to meet these challenging requirements onbroadband mobile communication systems in limited avail-able frequency spectrum, different advanced technologieswere taken into account: advanced antenna concepts, mod-ulation, coding and scheduling algorithms, radio resourcemanagement (including dynamic resource allocation, carri-er aggregation, cross-layer optimization, admission control,

congestion control, mobility management, and interoper-ability), coordinated multipoint schemes, and new deploy-ment elements like relays and femtocells.

This feature topic issue provides an overview of majordevelopments of both IMT-Advanced proposals. The fivearticles included in this issue describe new technologytrends that will have an impact on future standardizationand the evolution of IMT-Advanced technologies.

The first article, “Evolution of LTE toward IMT-Advanced” by Stefan Parkvall et al., provides a high-leveloverview of 3GPP LTE Release 10 (IMT-Advanced) andan analysis of some of its key technologies. IMT-Advancedbased on LTE Release 10 enhances LTE with carrieraggregation, enhanced multi-antenna support, improvedsupport for heterogeneous deployments, and relaying. Sim-ulation results show that LTE Release 10 fulfills and evensurpasses the requirements for IMT-Advanced.

The second article, “Assessing 3GPP Long Term Evolu-tion (LTE)-Advanced as IMT-Advanced Technology: TheWINNER+ Evaluation Group Approach” by Krystian Saf-jan et al., describes the approach followed by the Europeanresearch project WINNER+ to successfully complete theperformance evaluation of the 3GPP LTE-Advanced pro-posal as an IMT-Advanced technology candidate. Exem-plary analytical and simulation results are providedtogether with the procedure followed for simulator calibra-tion, which is essential in order to achieve comparableresults between different evaluation organizations. Theobtained results confirm that the 3GPP LTE Release 10 &Beyond (LTE-Advanced) proposal satisfies all the IMT-Advanced requirements.

The third article, “Coordinated Multipoint: Concepts,Performance, and Field Trial Results” by Ralf Irmer et al.,shows that coordinated multipoint (COMP) or cooperativemultiple-input multiple-output (MIMO) is one of the mostpromising concepts to improve cell edge user data rate andspectral efficiency beyond legacy systems. Interference canbe exploited or mitigated by cooperation between sectorsor different sites. Significant gains can be obtained for

G

GUEST EDITORIAL

Werner Mohr Jose F. Monserrat Afif Osseiran Marc Werner

LYT-GUEST EDIT-Mohr 1/20/11 12:11 PM Page 82


both the uplink and downlink. A range of technical chal-lenges are identified and addressed, such as backhaul traf-fic, synchronization, and feedback design. The feasibility ofCOMP is demonstrated in two field testbeds with multiplesites and different backhaul solutions between the sites.

The fourth article, “Evolution of Uplink MIMO forLTE-Advanced” by Chester Sungchung Park et al., is aboutthe other main enabling technology for next-generationmobile networks, advanced antenna design. Specifically,the article gives an overview of a MIMO approach that hasrecently been adopted in 3GPP, including up to four-layertransmission using precoded spatial multiplexing as well astransmit diversity techniques. Receivers suitable for uplinkMIMO are presented, and their link-level performancesare compared. It is shown that with advanced receivers,single-carrier transmission performs as well as orthogonalfrequency-division multiplexing (OFDM).

The fifth article, “A 25 Gb/s/km2 Urban Wireless Net-work beyond IMT-Advanced” by Sheng Liu et al., presentsa survey on the technical challenges of the future radioaccess network beyond IMT-Advanced, which should offervery high average area throughput in order to supporthuge data traffic demand and high user density with ener-gy-efficient operation. Several potential enabling technolo-gies and architectures from the controlling/processing,radio resource management, and physical layer perspec-tives for dense urban cell deployment are investigated tosupport the aggressive goal of an average area throughputof 25 Gb/s/km2 in beyond-IMT-Advanced systems. Thecombination of various advanced technologies such asinterference mitigation techniques, MIMO, and coopera-tive communications, as well as cross-layer self-organizingnetworks (SONs) could potentially offer high-qualitymobile services in future urban wireless networks and anexperience similar to that of the wired Internet.

Finally, we would like to thank Dr. Steve Gorshe,Joseph Milizzo, Devika Mittra, and Jennifer Porcello fortheir continuous support and valuable comments toimprove this feature topic issue. We hope that the articlesin this issue will encourage the readers of IEEE Communi-cations Magazine to contribute to the further developmentand improvement of IMT-Advanced.

BIOGRAPHIESWERNER MOHR [SM] ([email protected]) graduated from the Universityof Hannover, Germany, with a Master’s degree in electrical engineering in1981 and a Ph.D. degree in 1987. He joined Siemens AG, Mobile NetworkDivision, Munich, Germany, in 1991. He was involved in several EU fundedprojects and ETSI standardization groups on UMTS and systems beyond 3G.In December 1996 he became project manager of the European ACTSFRAMES Project until the project finished in August 1999. This projectdeveloped the basic concepts of the UMTS radio interface. Since April 2007

he has been with Nokia Siemens Networks GmbH & Co. KG, Munich, Ger-many, where he is head of Research Alliances. He was the coordinator ofthe WINNER Project in Framework Program 6 of the European Commission,Chairman of WWI (Wireless World Initiative), and the Eureka Celtic projectWINNER+. The WINNER project laid the foundation for the radio interfacefor IMT-Advanced and provided the starting point for the 3GPP LTE stan-dardization. In addition, he was Vice Chair of the eMobility European Tech-nology Platform in the period 2008–2009 and is now eMobility (now calledNet!works) Chairperson for the period 2010–2011. He was Chair of theWireless World Research Forum from its launch in August 2001 up toDecember 2003. He is a member of VDE (the Association for Electrical,Electronic & Information Technologies, Germany). In 1990 he received theAward of the ITG (Information Technology Society) in VDE. He was a boardmember of ITG in VDE for the term 2006–2008 and was re-elected for theterm 2009-2011. He is coauthor of the books Third Generation MobileCommunication Systems and Radio Technologies and Concepts for IMT-Advanced.

JOSE F. MONSERRAT [M] ([email protected]) received his M.Sc. degreewith High Honors and Ph.D. degree in telecommunications engineeringfrom the Polytechnic University of Valencia (UPV) in 2003 and 2007,respectively. He was the recipient of the First Regional Prize of EngineeringStudies in 2003 for his outstanding student record, also receiving the BestThesis Prize from the UPV in 2008. In 2009 he was awarded the best youngresearcher prize of Valencia. He is currently an associate professor in theCommunications Department of UPV. His research focuses on the applica-tion of complex computation techniques to radio resource management(RRM) strategies and to the optimization of current and future mobile com-munications networks, such as LTE-Advanced and IEEE 802.16m. He hasbeen involved in several European Projects, acting as task or work packageleader in WINNER+, ICARUS, COMIC, and PROSIMOS. He also participatedin 2010 in one external evaluation group within ITU-R on the performanceassessment of the candidates for the future family of standards, IMT-Advanced.

AFIF OSSEIRAN [M] ([email protected]) received a B.Sc. in electricaland electronics engineering from Université de Rennes I, France, in 1995, aDEA (B.Sc.E.E) degree in electrical engineering from Université de Rennes Iand INSA Rennes in 1997, and an M.A.Sc. degree in electrical and commu-nication engineering from École Polytechnique de Montreal, Canada, in1999. In 2006 he defended successfully his Ph.D. thesis at the Royal Insti-tute of Technology (KTH), Stockholm, Sweden. Since 1999 he has beenwith Ericsson, Sweden. In 2004 he joined as one of Ericsson’s representa-tives the European project WINNER. During 2006 and 2007 he led in WIN-NER the spatial temporal processing (i.e. MIMO) task. From April 2008 toJune 2010 he was the technical manager of the Eureka Celtic project WIN-NER+. His research interests include many aspects of wireless communica-tions with a special emphasis on advanced antenna systems, relaying, radioresource management, network coding, and cooperative communications.He is listed in Who’s Who in the World and Who’s Who in Science & Engi-neering. He has published over more 40 technical papers in internationaljournals and conferences. He co-authored a book, Radio Technologies andConcepts for IMT-Advanced (Wiley, 2009). Since 2006 he has been giving afew graduate-level lectures yearly on advanced antennas at KTH.

MARC WERNER ([email protected]) received his InternationalDiploma degree from Imperial College, London, United Kingdom, in 1997,and his Dipl.-Ing. and Dr.-Ing. degrees in electrical engineering from RWTHAachen University, Germany, in 1999 and 2006, respectively. He worked asa research scientist at RWTH Aachen University from 1999 to 2005. Hisresearch activities included capacity optimization and speech qualityimprovement for cellular communication systems. He has also worked inmultiple industry projects in the area of mobile communication, and as aconsultant for the German telecoms regulator. Since January 2006 he hasbeen with QUALCOMM CDMA Technologies GmbH, Nuremberg, Germany.For Qualcomm he worked as work package and task leader in severalmultinational European research projects such as WINNER II and WINNER+.In WINNER+ he coordinated the simulative evaluation of the IMT-Advancedsystem proposal by 3GPP, LTE-Advanced.

GUEST EDITORIAL

LYT-GUEST EDIT-Mohr 1/20/11 12:11 PM Page 83

IEEE Communications Magazine • February 201184 0163-6804/11/$25.00 © 2011 IEEE

INTRODUCTION

Deployment of fourth-generation (4G) mobile-broadband systems based on the highly flexibleLong Term Evolution (LTE) radio access tech-nology [1, 2] defined by the Third GenerationPartnership Project (3GPP) is currently ongoingon a broad scale, with the first systems alreadybeing in full commercial operation. These sys-tems are based on the first release of LTE, 3GPPRelease 8, which was finalized in 2008. Release8 can provide downlink and uplink peak rates upto 300 and 75 Mb/s, respectively, a one-wayradio-network delay of less than 5 ms, and a sig-nificant increase in spectrum efficiency. LTEprovides extensive support for spectrum flexibili-ty, supports both frequency-division duplex(FDD) and time-division duplex (TDD), and tar-gets a smooth evolution from earlier 3GPP tech-nologies such as time-division synchronouscode-division multiple access (TD-SCDMA) andwideband CDMA (WCDMA)/high-speed pakcetaccess (HSPA) as well as 3GPP2 technologiessuch as cdma2000.

The LTE radio access technology is continu-ously evolving to meet future requirements. InRelease 9, finalized at the end of 2009, supportfor broadcast/multicast services, positioning ser-vices, and enhanced emergency-call functionali-ty, as well as enhancements for downlinkdual-layer beam-forming, were added.

Recently, 3GPP has concluded the work onLTE Release 10, finalized at the end of 2010and further extending the performance andcapabilities of LTE beyond Release 8/9. Animportant aim of LTE Release 10 is to ensurethat LTE fulfills all the requirements for Inter-

national Mobile Telecommunications (IMT)-Advanced as defined by the InternationalTelecommunication Union (ITU) [3, 4]. Therelation to IMT-Advanced is also the reason forthe label LTE-Advanced sometimes given toLTE Release 10 and beyond.

This article provides a brief overview of LTERelease 8/9 and a short introduction to the IMT-Advanced work. Following this background, theextensions introduced in Release 10 aredescribed. The article is concluded with resultsfrom system-level evaluations showing that LTERelease 10 can fulfill and even surpass the IMT-Advanced requirements.

OVERVIEW OF LTE RELEASE 8LTE is an orthogonal frequency-division multi-plexing (OFDM)-based radio access technology,with conventional OFDM on the downlink anddiscrete Fourier transform spread OFDM(DFTS-OFDM) [1] on the uplink. DFTS-OFDMallows for more efficient power-amplifier opera-tion, thus providing the opportunity for reducedterminal power consumption. At the same time,equalization of the received signal is straightfor-ward with conventional OFDM. The use ofOFDM on the downlink combined with DFTS-OFDM on the uplink thus minimizes terminalcomplexity on the receiver side (downlink) aswell as on the transmitter side (uplink), leadingto an overall reduction in terminal complexityand power consumption.

The transmitted signal is organized into sub-frames of 1 ms duration with 10 subframes form-ing a radio frame as illustrated in Fig. 1. Eachdownlink subframe consists of a control regionof one to three OFDM symbols, used for controlsignaling from the base station to the terminals,and a data region comprising the remaining partand used for data transmission to the terminals.The data transmissions in each subframe aredynamically scheduled by the base station. Asseen in Fig. 1, cell-specific reference signals arealso transmitted in each downlink subframe.These reference signals are used for data demod-ulation at the terminal (or user equipment, UE),and for measurement purposes (e.g., for channelstatus reports sent from the terminals to thebase station).

Spectrum flexibility is one of the key proper-ties of the LTE radio access technology. A widerange of different bandwidths is defined and

ABSTRACT

This article provides a high-level overview ofLTE Release 10, sometimes referred to as LTE-Advanced. First, a brief overview of the firstrelease of LTE and some of its technology com-ponents is given, followed by a discussion on theIMT-Advanced requirements. The technologyenhancements introduced to LTE in Release 10,carrier aggregation, improved multi-antennasupport, relaying, and improved support for het-erogeneous deployments, are described. Thearticle is concluded with simulation results,showing that LTE Release 10 fulfills and evensurpasses the requirements for IMT-Advanced.

IMT-ADVANCED AND NEXT-GENERATIONMOBILE NETWORKS

Stefan Parkvall, Anders Furuskär, and Erik Dahlman, Ericsson Research

Evolution of LTE toward IMT-Advanced

PARKVALL LAYOUT 1/19/11 3:31 PM Page 84


both FDD and TDD modes of operation aresupported, allowing for operation in both pairedand unpaired spectrum. An important require-ment in the LTE design has been to avoid un-necessary fragmentation and strive forcommonality between the FDD and TDD modesof operation while still maintaining the possibili-ty to fully exploit duplex-specific properties suchas channel reciprocity in TDD. Aligning the twoduplex schemes to the extent possible does notonly increase the momentum in the definitionand standardization of the technology but alsofurther improves the economy of scale of theLTE radio access technology.

Support for multi-antenna transmission is anintegral part of LTE from the first release.Downlink multi-antenna schemes supported byLTE include transmit diversity, spatial multiplex-ing (including both so-called single-user multi-ple-input multiple-output [MIMO] as well asmulti-user MIMO), and beamforming.

ITU AND IMT-ADVANCEDIMT-Advanced is the term used by ITU forradio access technologies beyond IMT-2000. Aninvitation to submit candidate technologies forIMT-Advanced was issued by ITU in 2008 [3].Along with the invitation, ITU has also defineda set of requirements to be fulfilled by any IMT-Advanced candidate technology [4], some ofwhich are shown in Table 1 together with thecorresponding capabilities of LTE Release 10.

Anticipating the invitation from ITU, 3GPPalready in March 2008 initiated a study item onLTE-Advanced, with the task of definingrequirements and investigating potential technol-ogy components for the LTE evolution. Thisstudy item, completed in March 2010 and form-ing the basis for the Release 10 work, aimedbeyond IMT-Advanced [5]. In 2010 3GPP sub-mitted LTE Release 10 to ITU and, based onthis submission, ITU approved LTE Release 10as one of two IMT-Advanced technologies. As

will be seen, Release 10 will not only fulfill theIMT-Advanced requirements but in many caseseven surpass them.

LTE RELEASE 10LTE Release 10, sometimes known as LTE-Advanced, is not a new radio access technologybut the evolution of LTE to further improveperformance. Being an evolution of LTE,Release 10 includes all the features of Release8/9 and adds several new features, the mostimportant of which — carrier aggregation,enhanced multi-antenna support, improved sup-port for heterogeneous deployments, and relay-ing — are discussed in the following sections.Evolving LTE rather than designing a new radioaccess technology is important from an operatorperspective as it allows for smooth introductionof new technologies without jeopardizing exist-ing investments. A Release 10 terminal candirectly connect to a network of an earlierrelease, and a Release 8/9 terminal can connectto a network supporting the new enhancements.Hence, an operator can deploy a Release 8 net-work and later, when the need arises, upgrade toRelease 10 functionality where needed. In fact,most of the Release 10 features can be intro-duced into the network as simple softwareupgrades.

CARRIER AGGREGATIONAlready the first release of LTE, Release 8, pro-vides extensive support for deployment in spec-trum allocations of various characteristics, withbandwidths ranging from around 1.4 up to 20MHz in both paired and unpaired bands. InRelease 10 the transmission bandwidth can befurther extended by means of so-called carrieraggregation (CA) where multiple component car-riers are aggregated and jointly used for trans-mission to/from a single mobile terminal, asillustrated in Fig. 2. Up to five component carri-

Figure 1. LTE time-frequency structure.

Controlsignaling

Cell-specificreferencesymbols

DwPTS UpPTS GP

FDD

TDD

UL DL

#0 #1 #2 #3 #4 #5 #6 #7 #8 #9

fULfDL

One subframe 1 ms

One radio frame 10 ms

(Special subframe) (Special subframe)

One subframe

Control region(1-3 OFDM symbols)

UL DL

fDL/UL

Support for multi-

antenna transmission

is an integral part of

LTE from the first

release. Downlink

multi-antenna

schemes supported

by LTE include

transmit diversity,

spatial multiplexing

(including both

so-called single-user

MIMO as well as

multi-user MIMO),

and beamforming.



ers, possibly each of different bandwidth, can beaggregated, allowing for transmission bandwidthsup to 100 MHz. Backward compatibility iscatered for as each component carrier uses theRelease 8 structure. Hence, to a Release 8/9 ter-minal each component carrier will appear as anLTE Release 8 carrier, while a carrier-aggrega-tion-capable terminal can exploit the total aggre-gated bandwidth enabling higher data rates. Inthe general case, different numbers of compo-nent carriers can be aggregated for the downlinkand uplink.

With respect to the frequency location of thedifferent component carriers, three differentcases can be identified: intra-band aggregationwith contiguous carriers (e.g., aggregation of #2and #3 in Fig. 2), inter-band aggregation (#1and #4), and intra-band aggregation with non-contiguous carriers (#1 and #2). The possibilityto aggregate non-adjacent component carriersenables exploitation of fragmented spectrum;operators with a fragmented spectrum can pro-vide high-data-rate services based on the avail-ability of wide overall bandwidth even thoughthey do not possess a single wideband spectrumallocation. From a baseband perspective, there isno difference between the cases, and they are allsupported by LTE Release 10. However, theradio frequency (RF) implementation complexi-ty is vastly different with the first case being theleast complex. Thus, although spectrum aggrega-tion is supported by the basic specifications, theactual implementation will be strongly con-strained, including specification of only a limitednumber of aggregation scenarios and aggrega-tion over dispersed spectrum only being support-ed by the most advanced terminals. Althoughexploitation of fragmented spectrum and expan-sion of the total bandwidth beyond 20 MHz aretwo important usages of carrier aggregation,there are also scenarios where carrier aggrega-tion within 20 MHz of contiguous spectrum isuseful. One example is heterogeneous deploy-ments, discussed below.

Scheduling and hybrid automatic repeatrequest (ARQ) retransmissions are handledindependently for each component carrier (Fig.2). As a baseline, control signaling is transmittedon the same component carrier as the corre-sponding data. However, as a complement it ispossible to use so-called cross-carrier scheduling

where the scheduling decision is transmitted tothe terminal on another component carrier thanthe corresponding data.

To reduce the terminal power consumption, acarrier-aggregation-capable terminal typicallyreceives on one component carrier only, the pri-mary component carrier. Reception of additionalsecondary component carriers can be rapidlyturned on/off in the terminal by the base stationthrough medium access control (MAC) signal-ing. Similarly, in the uplink all the feedback sig-naling is transmitted on the primary componentcarrier, and secondary component carriers areonly enabled when necessary for data transmis-sion.

ENHANCED MULTI-ANTENNA SUPPORTLTE supports a rich set of multi-antenna trans-mission techniques already in the first release.This includes downlink transmit diversity basedon space-frequency block coding (SFBC) for thecase of two transmit antennas and SFBC in com-bination with frequency shift time diversity(FSTD) for four transmit antennas. In addition,downlink codebook-based precoding, includingthe possibility for multilayer transmission (spa-tial multiplexing) with up to four layers, is sup-ported in LTE Release 8. This includes thepossibility for rank-adaptation down to single-layer transmission, leading to codebook-basedbeamforming, as well as a basic form of multi-user MIMO where different layers in the sametime-frequency resource can be assigned to dif-ferent terminals.

The multi-antenna techniques above rely onthe previously mentioned cell-specific referencesignals for demodulation as well as to acquirechannel-state feedback from the terminal to thebase station. In addition, UE-specific referencesignals are part of Release 8 to support single-layer beam-forming; support that is extended todual-layer transmission in Release 9. UE-specificreference signals are pre-coded together with thedata, implying that the pre-coder weights are notrestricted to a certain codebook and do not needto be known to the receiver. An important appli-cation is beamforming with more than fourantennas and, for TDD, reciprocity-based trans-mission strategies.

In Release 10, downlink spatial multiplexingis expanded to support up to eight transmission

Table 1. Requirements and LTE fulfillment.

IMT-Advanced requirement LTE Release 8 LTE Release 10

Transmission bandwidth At least 40 MHz Up to 20 MHz Up to 100 MHz

Peak spectral efficiency– Downlink– Uplink

15 b/s/Hz6.75 b/s/Hz

16 b/s/Hz4 b/s/Hz

16.0 [30.0]* b/s/Hz8.1 [16.1]** b/s/Hz

Latency– Control plane– User plane

Less than 100 msLess than 10 ms

50 ms4.9 ms

50 ms4.9 ms

*Value is for a 4 × 4 antenna configuration. Value in brackets for 8 × 8.** Values is for a 2 × 2 antenna configuration. Value in brackets for 4 × 4.

LTE supports a rich

set of multi-antenna

transmission

techniques already in

the first release. This

includes downlink

transmit diversity

based on

Space-Frequency

Block Coding (SFBC)

for the case of two

transmit antennas

and SFBC in

combination with

Frequency Shift Time

Diversity (FFSTD)

for four transmit

antennas.



layers together with an enhanced reference sig-nal structure. Relying on cell-specific referencesignals for higher-order spatial multiplexing isless attractive since the reference signal over-head is not proportional to the instantaneoustransmission rank but rather to the maximumsupported transmission rank. Hence, Release 10introduces extensive support of UE-specific ref-erence signals for demodulation of up to eightlayers. Furthermore, feedback of channel-stateinformation (CSI) is based on a separate set ofreference signals broadcasted in the cell, knownas CSI reference signals. CSI reference signals arerelatively sparse in frequency (every 12th subcar-rier, corresponding to 180 kHz) but regularlytransmitted from all antennas at the base station.The periodicity is configurable but is typically onorder of once per 10 ms. UE-specific referencesignals, on the other hand, are denser in fre-quency and only transmitted when data is trans-mitted on the corresponding layer. Separatingthe reference signal structure supporting demod-ulation from that supporting channel state esti-mation helps reduce the reference signaloverhead, especially for high degrees of spatialmultiplexing, and allows for implementation ofvarious beamforming schemes.

Uplink spatial multiplexing of up to four lay-ers is also part of Release 10. The basis is acodebook-based scheme where the scheduler inthe base station determines the precoding matrixto be applied in the terminal. The selected pre-coding matrix is applied to uplink data transmis-sions as well as the uplink demodulationreference signals. To facilitate the selection of asuitable preceding matrix in the terminal, thesounding reference signals are enhanced to sup-port up to four antennas.

IMPROVED SUPPORT FORHETEROGENEOUS DEPLOYMENTS

With the rapidly growing usage of mobile broad-band, the data rates experienced by the users inthe network become increasingly important. Theend-user data rate in a practical deployment ishighly dependent on factors such as the termi-nal-to-base-station distance, whether the user isindoor or outdoor, and so on. As the possibilitiesto improve the link performance or increase thetransmission power are limited, supporting veryhigh end-user data rates requires a denser infra-structure. Not only does a densified networkhave the possibility to increase the data ratesexperienced, it can also increase the overallcapacity as the number of sites increase. Astraightforward densification of an existingmacro network is one possibility, but in scenarioswhere the users are highly clustered, a potential-ly attractive approach is to complement a macrocell providing basic coverage with multiple low-output-power pico cells where needed as shownin Fig. 3. The result of such a strategy is a het-erogeneous deployment with two or more cell lay-ers. The idea of multiple cell layers is in itselfnot new; hierarchical cell structures have beendiscussed since the mid-’90s but then for (low-rate) voice users. It is important to point outthat this is a deployment strategy, not a technol-ogy component, and as such is possible already

in LTE Release 8/9. However, Release 10 pro-vides some additional features, improving thesupport for heterogeneous deployments.

In a heterogeneous deployment, cell associa-tion (i.e., to which cell a terminal should be con-nected) plays an important role. From an uplinkdata rate perspective, it is fundamentally benefi-cial to connect to the cell with the lowest pathloss as this results in a higher data rate at agiven transmit power, instead of the traditionalapproach of connecting to the cell with thestrongest received downlink. The best cell fordownlink association depends on the load; at lowload connecting to the cell with the strongestreceived downlink offers the highest data rates,while at high loads connecting to the low-powernode may be preferable as it provides for down-link resource reuse between the cells served bythe low-power nodes. The backhaul capacity tothe low-power node is also important to consid-er. Cell association strategies in a heterogeneousdeployment are therefore nontrivial where theoverall network performance must be taken intoaccount. Nevertheless, any cell association strat-egy not solely based on maximizing the receiveddownlink signal quality can lead to a new inter-ference situation in the network as, in essence,the uplink coverage area can be larger than thedownlink coverage area, implying that there is aregion around the low-power node (lighter ringin Fig. 3) where downlink transmission from thelow-power node to a terminal is subject to stronginterference from the macrocell. The signal-to-interference ratio experienced by the terminal atthe outermost coverage area of the low-powernode is, due to the difference in output powerbetween the high-power macro and the low-power node, significantly lower than in a tradi-tional homogeneous macro network.

For the data part of a subframe, this is not aserious problem as the intercell interference coor-dination (ICIC) mechanism present in LTE

Figure 2. Carrier aggregation in LTE Release 10.

HARQ HARQ HARQ HARQ

Coding Coding Coding Coding

DFT DFT DFT DFT

OFDM OFDM OFDM OFDM

Frequency band A Frequency band B

RLC

MAC

PHY

Uplink only

Componentcarrier #1 #2 #3

Componentcarrier #4

RLC

MAC



already from Release 8 can be used. With ICIC,different cells can exchange information aboutwhich frequencies they intend to schedule trans-missions on in the near future, thereby reducingor completely avoiding intercell interference.This can be used to more or less dynamicallycoordinate the resource usage between the celllayers and avoid overlapping resource usage.

The control signaling in each subframe ismore problematic as it spans the full cell band-width and is not subject to ICIC. To address this,LTE Release 10 provides enhancements to sepa-rate the control signaling for the different celllayers in either the frequency or time domain.

Frequency domain schemes use carrier aggre-gation to separate control signaling for the differ-ent cell layers. At least one component carrier ineach cell layer is protected from interferencefrom other cell layers by not transmitting controlsignaling on the component carrier in question inthe other cell layers. For example, referring toFig. 3, the macro base station transmits controlsignaling on component carrier f1 but not oncomponent carrier f2, while the situation is theopposite in the low-power nodes located withinthe macrocell. Since Release 10 introduces cross-carrier scheduling, resources on f2 can be usedfor data transmission, scheduled by control sig-naling received on f1, subject to the normal ICICmechanism. In essence, this creates frequencyreuse for the control signaling while still allowingterminals to dynamically utilize the full band-width (and thereby supporting the highest datarates) for the data part. For example, an opera-tor with 20 MHz of spectrum may choose to con-figure two component carriers of 10 MHz eachand use carrier aggregation as described above.Note that carrier-aggregation-capable terminals,in addition to benefits of connecting to the low-

power node, also in the lighter ring in Fig. 3, willhave the same peak data rates as in the case of asingle 20 MHz carrier. Release 8/9 can also bene-fit from seeing a larger picocell but can obviouslyonly access one component carrier.

Time domain schemes use a single compo-nent carrier f in all the cell layers and separatethe control signaling in the different cell layersin the time domain, as seen in Fig. 3. At leastsome subframes in the low-power cell layer areprotected from interference by the macro layermuting the control signaling in those subframes.However, for backward compatibility, cell-specif-ic reference signals still needs to be transmittedfrom the macro cell, resulting in some interfer-ence to the terminals. To provide for accurateCSI feedback, Release 10 provides the possibilityto configure on which subframes the terminalshould base its channel-quality estimates as theinterference experienced by a terminal connect-ed to a low-power node may vary drasticallydepending on the macrocell activity. Note that inthis approach, Release 8/9 terminals will connectto the macro and not to the low-power node inthe lighter area in Fig. 3, but can access the fullbandwidth of the carrier.

The discussion above assumes that the termi-nals are allowed to connect to the low-powernode. This is known as open access, and typicallythe low-power nodes are operator-deployed insuch a scenario. Another scenario, giving rise toa similar interference problem, is user-deployedhome base stations. The term closed subscribergroups (CSGs) is commonly used to refer tocases when access to such a low-power base sta-tion is limited to a small set of terminals (e.g., afamily living in a house where the home basestation is located). CSG results in additionalinterference scenarios. For example, a terminal

Figure 3. Heterogeneous deployment with a macro cell overlaying multiple pico cells.

f1

f2

f

Frequency-domainscheme

Time-domain scheme

Subframe Data Control region

“Almost blank” control region

Macro

Macro Pico

Pathloss border

Signal strength border

Rx power

Pico

(path loss)-1

To provide for accu-

rate CSI feedback,

Release 10 provides

the possibility to

configure which sub-

frames the terminal

should base it chan-

nel-quality estimates

upon as the interfer-

ence experience by a

terminal connected

to a low-power node

may vary drastically

depending on the

macro cell activity.



located close to but not admitted to connect tothe home base station will be subject to stronginterference, and may not be able to access themacrocell. In essence, the presence of a homebase station may cause a coverage hole in theoperator’s macro network; a problem that is par-ticularly worrisome as home base stations typi-cally are user deployed, and their locations arenot controlled by the operator. Similarly, recep-tion at the home base station may be severelyimpacted by uplink transmissions from the ter-minal connected to the macrocell. Therefore, ifclosed subscriber groups are supported, it ispreferable to use a separate carrier for the CSGcells to maintain the overall performance of theradio access network. Interference handlingbetween CSG cells, which typically lack back-haul-based coordination schemes, could rely ondistributed algorithms for power control and/orresource partitioning between the cells.

RELAYINGLTE Release 10 also extends the LTE radioaccess technology with support for relaying func-tionality (Fig. 4). With relaying, the mobile termi-nal communicates with the network via a relaynode that is wirelessly connected to a donor cellusing the LTE radio interface technology. Thedonor cell may, in addition to one or severalrelays, also directly serve terminals of its own.The donor-relay link may operate on the samefrequency as the relay-terminal link (inband relay-ing) or on a different frequency (outband relay-ing). With the 3GPP relaying solution [6], therelay node will, from a terminal point of view,appear as an ordinary cell. This has the importantadvantage of simplifying the terminal implemen-tation and making the relay node backward com-patible (i.e., also accessible to LTE Release 8terminals). In essence, the relay is a low-powerbase station wirelessly connected to the remainingpart of the network. One of the attractive featuresof a relay is the LTE-based wireless backhaul as

this could provide a simple way of improving cov-erage, e.g., in indoor environments by simplyplacing relays at the problematic locations. At alater stage, if motivated by the traffic situation,the wireless donor-relay link could be replaced bye.g., an optical fiber in order to use the preciousradio resources in the donor cell for terminalcommunication instead of serving the relay.

Due to the relay transmitter causing interfer-ence to its own receiver, simultaneous donor-to-relay and relay-to-terminal transmission may notbe feasible unless sufficient isolation of the out-going and incoming signals is provided, forexample, by means of specific well separated andwell isolated antenna structures or through theuse of outband relaying. Similarly, at the relay itmay not be possible to receive transmissionsfrom the terminals simultaneously with the relaytransmitting to the donor cell. In Release 10 agap in the relay-to-terminal transmissions toallow for reception of donor-to-relay transmis-sions is created using MBSFN subframes,1 asshown in Fig. 4. In an MBSFN subframe the firstone or two OFDM symbols in a subframe aretransmitted as usual carrying cell-specific refer-ence signals and downlink control signaling,while the rest of an MBSFN subframe is notused and can therefore be used for the donor-to-relay communication. The benefit of usingMBSFN subframes compared to blanking trans-mission in the whole subframe is backward com-patibility with Release 8/9 terminals. Blankingthe whole subframe would not be compatiblewith Release 8/9 terminals as they assume cell-specific reference signals to be present in (partof) each subframe, while MBSFN subframes aresupported already in Release 8. Similar to thedownlink gaps obtained through the use ofMBSFN subframes, there is a need to creategaps in the terminal-to-relay transmission inorder for the relay to transmit to the donor. Thisis handled by not scheduling terminal-to-relaytransmissions in some subframes.

1 Multicast-broadcast sin-gle-frequency network(MBSFN) subframes, pre-sent already in Release 8,were originally intendedfor broadcast support buthas later been seen as ageneric tool (e.g., to blankparts of a subframe forrelaying support).

Figure 4. Relaying.

Guard for Rx-Tx switch in the relay

Con

trol

BS-to-RN control transmission

BS-to-RN data transmission

Data

DL-related UL-related

Base station transmission

Relay node transmission

Subframe

MBSFN subframe No relay-to-terminal transmission

Relay cell

Donor cell(Donor) base station

In essence, the relay

is a low-power base

station wirelessly

connected to the

remaining part of

the network. One of

the attractive

features of a relay is

the LTE-based

wireless backhaul as

this could provide a

simple way of

improving coverage.



Since the relay needs to transmit cell-specificreference signals in the first part of an MBSFNsubframe, it cannot receive the normal controlsignaling from the donor cell. Therefore, Release10 defines a new control channel, transmittedlater in the subframe as shown in Fig. 4, to pro-vide control signaling from the donor to therelay. This control channel type, of which multi-ple instances can be configured, carries downlinkscheduling assignments and uplink schedulinggrants in the same way as the normal control sig-naling. As the assignments refer to data in thesame subframe and the grants relate to transmis-sions in a later subframe, early decoding of theformer control information is beneficial. For thisreason, downlink assignments are transmitted inthe first part of the donor-to-relay transmission,while the latter part is used for (less time-criti-cal) uplink grants.

PERFORMANCE RESULTSAs discussed in the introduction, ITU hasdefined basic requirements to be fulfilled by anyIMT-Advanced technology [4]. Some of the mostbasic requirements, together with the corre-

sponding capabilities of LTE [7], are summa-rized in Table 1.

From the table it is seen that already the firstrelease of LTE, Release 8, is capable of meetingall of the requirements except the bandwidthand uplink spectral efficiency requirements.These two requirements are addressed inRelease 10 through carrier aggregation anduplink spatial multiplexing, respectively.

For the detailed requirements on averageand cell-edge spectral-efficiency, 3GPP has car-ried out an extensive evaluation campaign toconclude on the performance of the LTE radio-access technology in relation to the IMT-Advanced requirements. Examples of LTEsystem performance for the different test envi-ronments specified by the ITU (indoor hotspot,urban micro, urban macro, and rural) are pro-vided in Fig. 5. In the downlink, a coordinatedbeamforming scheme is used with spatial multi-plexing of two layers to a single terminal ineach beam. Beams are dynamically adapted tolimit interference, allowing reuse of time-fre-quency resources within cells. The beam-form-ing is coordinated between cells belonging tothe same site. This can be seen as a simple

Figure 5. Performance results for FDD (top) and TDD (bottom), and downlink (right) and uplink (left).

InH

1

0Avg

cel

l tp

[bps

/Hz/

cell]

2

3

4

UMi

Downlink 4x2 FDD

Downlink 4x2 TDD

UMa RMa

±2.3%

4.42

±4.9%

2.82

±1.9%

2.36±2.2%

3.36

InH

0.05

0

Cel

l-edg

e us

er t

p [b

ps/H

z]

0.1

0.15

0.2

0.25

UMi UMa RMa

±5.4%

0.205

±14.3%0.081

±13.3%0.072 ±5.1%

0.102

InH

0.05

0

Cel

l-edg

e us

er t

p [b

ps/H

z]

0.1

0.15

0.2

UMi UMa RMa

±6.0%

0.182

±7.3%

0.076

±7.6%0.065 ±5.1%

0.099

InH

1

0Avg

cel

l tp

[bps

/Hz/

cell]

2

3

4

UMi UMa RMa

±2.3%

4.26

±2.4%

2.63

±2.2%

2.26±2.2%

3.25

InH

1

0Avg

cel

l tp

[bps

/Hz/

cell]

2

3

UMi

Uplink 1x4 FDD

Uplink 1x4 TDD

UMa RMa

±0.0%

3.52

±1.2%

2.26

±2.1%

1.71±2.8%

2.13

InH

0.1

0

Cel

l-edg

e us

er t

p [b

ps/H

z]

0.2

0.3

0.4

UMi UMa RMa

±2.8%

0.267

±4.1%0.093

±3.4%0.075 ±6.2%

0.101

InH

0.1

0

Cel

l-edg

e us

er t

p [b

ps/H

z]

0.2

0.3

0.4

UMi UMa RMa

±2.8%

0.254

±3.2%0.085

±3.4%0.070 ±6.5%

0.094

InH

1

0Avg

cel

l tp

[bps

/Hz/

cell]

2

3

UMi UMa RMa

±0.1%

3.34

±1.2%

2.09

±2.1%

1.61±2.9%

1.99



form of coordinated multipoint transmission(CoMP) or multi-user MIMO. In the uplink,single-layer transmission is used. For furtherdetails on the simulation assumptions, pleasesee [8]. These performance results are achievedwithout using any of the features introduced inRelease 10. The IMT-Advanced requirementson average and cell edge spectral efficiency canthus already be fulfilled with LTE Release 8. Itis important to point out that this does notmean that Release 10 features, such as extend-ed downlink multi-antenna transmission andrelaying functionality, are of no use. Rather,these features take the capabilities of the LTEradio access technology even further, beyondIMT-Advanced. Thus, by including moreadvanced features, such as extended multi-antenna transmission, LTE system perfor-mance is further enhanced, beyond what isillustrated above. A wider range of deploymentscenarios is also addressed, including such withrelays and non-contiguous spectrum alloca-tions.

CONCLUSIONThis article has provided a high-level overview ofthe evolution of LTE towards Release 10. Someof the key components — carrier aggregation,enhanced multi-antenna support, and relaying —are described. Numerical results show that LTERelease 10 fulfills and even surpasses the IMT-Advanced requirements. Given the largemomentum behind LTE, this is a very attractiveroute for an operator to meet future demandson mobile broadband. Clearly, LTE is a veryflexible platform and will continue to evolve formany years to come.

REFERENCES[1] E. Dahlman et al., 3G Evolution: HSPA and LTE for

Mobile Broadband, 2nd ed., Academic Press, 2008.[2] D. Astély et al., “LTE: The Evolution of Mobile Broad-

band,” IEEE Commun. Mag., vol. 47, no. 4, Apr. 2009.

[3] ITU-R SG5, “Invitation for Submission of Proposals forCandidate Radio Interface Technologies for the Terres-trial Components of the Radio Interface(s) for IMT-Advanced and Invitation to Participate in theirSubsequent Evaluation,” Circular Letter 5/LCCE/2, Mar.2008.

[4] ITU-R M.2134, “Requirements Related to Technical Per-formance for IMT-Advanced Radio Interface(s)”;http://www.itu.int/dms_pub/itu-r/opb/rep/R-REP-M.2134-2008-PDF-E.pdf.

[5] 3GPP TR 36.913, “Requirements for Further Advance-ments for Evolved Universal Terrestrial Radio Access (E-UTRA) (LTE-Advanced)”.

[6] 3GPP TS 36.216, “Evolved Universal Terrestrial RadioAccess (E-UTRA); Physical Layer for Relaying Operation”.

[7] 3GPP TR 36.912, “Feasibility Study for Further Advance-ments of E-UTRA (LTE-Advanced)”.

[8] A. Furuskär, “Performance Evaluations of LTE-Advanced— The 3GPP ITU Proposal,” 12th Int’l. Symp. WirelessPersonal Multimedia Commun. ‘09, Sendai, Japan, Sept.2009.

BIOGRAPHIESSTEFAN PARKVALL [SM] ([email protected]) joinedEricsson Research in 1999 and is currently a principalresearcher in the area of radio access, working withresearch and standardization of cellular technologies. Hehas been heavily involved in the development of HSPA andLTE and is also co-author of 3G Evolution: HSPA and LTEfor Mobile Broadband. In 2009 he received “StoraTeknikpriset” (one of Sweden’s major technical awards) forhis work on HSPA. He holds a Ph.D. from the Royal Insti-tute of Technology (KTH), Stockholm, Sweden. His previouspositions include assistant professor in communication the-ory at KTH and visiting researcher at the University of Cali-fornia, San Diego.

ANDERS FURUSKÄR is a principal researcher within the field ofwireless access networks at Ericsson Research. His currentfocus is on evolving HSPA and LTE to meet future demandson data rates and traffic volumes. He holds an M.Sc. and aPh.D. from KTH. He joined Ericsson in 1990.

ERIK DAHLMAN joined Ericsson Research in 1993 and is cur-rently senior expert in the area of radio access technolo-gies. He has been deeply involved in the development andstandardization of 3G radio access technologies(WCDMA/HSPA) as well as LTE and its evolution. He is partof the Ericsson Research management team working withlong-term radio access strategies. He is also co-author ofthe book 3G Evolution: HSPA and LTE for Mobile Broad-band and, together with Stefan Parkvall, received “StoraTeknikpriset” in 2009 for his contributions to the standard-ization of HSPA. He holds a Ph.D. from KTH.

Given the large

momentum behind

LTE, this is a very

attractive route for

an operator to meet

future demands on

mobile broadband.

Clearly, LTE is a very

flexible platform and

will continue to

evolve for many

years to come.


INTRODUCTION

The fast growth of mobile traffic volume is oneof the main reasons why the so-called fourth-generation mobile communication systems arebeing investigated and standardized. For thatreason a call for submission of system candidatesfor International Mobile Telecommunications-Advanced (IMT-A) was opened by the Interna-tional Telecommunication Union —Radiocommunication Standardization Sector(ITU-R), while independent groups wereencouraged to register with ITU-R to evaluatecandidate systems. IMT-A systems are meant tosupport low to high user mobility, various datarates, and support for multiple environmentswhile having capabilities for high-quality multi-

media applications and providing a significantimprovement in performance and quality of ser-vice [1].

The predecessors of the WINNER+ project,WINNER I and II, had an important impact onthe Long Term Evolution (LTE) roadmap. TheWINNER I system concept represented animportant contribution toward LTE, while WIN-NER II was involved in the preparation forWorld Radiocommunication Conference 2007(WRC ’07) and had an impact on IMT-A require-ments in terms of spectrum demand, minimumrequirements, and evaluation methodology.

Shortly after WRC ’07, ITU-R issued the Cir-cular Letter [2] with a call for submission ofIMT-Advanced radio interface technology (RIT)proposals to ITU-R. Since WINNER+ predeces-sors were involved in the ITU-R process, WIN-NER+ is covering both competence and toolsfor performing evaluations. In November 2008WINNER+ registered as an Independent Evalu-ation Group (IEG) at ITU-R for IMT-Advancedwith a focus on evaluating the Third GenerationPartnership Project (3GPP) LTE-Advanced pro-posal. Finally, 14 IEGs from the Americas, Asia,and Europe registered at ITU-R.

By highlighting the WINNER+ IEGapproach to simulator calibration and evalua-tion, and providing exemplary evaluation results,this article attempts to address the challenge ofhow to pursue a system-level performance checksupplying relevant and reliable performanceindicators while keeping the performance analy-sis feasible and practical. WINNER+ is a con-sortium of project partners; therefore, manydifferent tools are used for evaluation. Thus, arelevant question appears: is it possible to assesssimilar performance results using different simu-


ABSTRACT

This article describes the WINNER+approach to performance evaluation of the 3GPPLTE-Advanced proposal as an IMT-Advancedtechnology candidate. The official registeredWINNER+ Independent Evaluation Group eval-uated this proposal against ITU-R requirements.The first part of the article gives an overview ofthe ITU-R evaluation process, criteria, and sce-narios. The second part is focused on the work-ing method of the evaluation group, emphasizingthe simulator calibration approach. Finally, thearticle contains exemplary evaluation resultsbased on analytical and simulation approaches.The obtained results allow WINNER+ to con-firm that the 3GPP LTE Release 10 & Beyond(LTE-Advanced) proposal satisfies all the IMT-Advanced requirements, and thus qualifies as anIMT-advanced system.


Krystian Safjan, Nokia Siemens Networks — Research

Valeria D’Amico, Telecom Italia

Daniel Bültmann, RWTH Aachen University

David Martín-Sacristán, Universidad Politécnica de Valencia

Ahmed Saadani, Orange-Labs

Hendrik Schöneich, Qualcomm CDMA Technologies GmbH

Assessing 3GPP LTE-Advanced asIMT-Advanced Technology:The WINNER+ Evaluation Group Approach

SAFJAN LAYOUT 1/19/11 3:32 PM Page 92


lation tools of a complex communication sys-tem? In this article we present the WINNER+evaluation group approach to harmonizing theorchestra of simulators while aligning differentorganizations with a variety of tools to produceconverging system performance evaluationresults.

We also briefly describe a limited set of testscenarios used in the evaluations that directlycorrespond to a typical usage scenario of the sys-tem under consideration. Finally, a full evalua-tion of the 3GPP LTE Release 10 & Beyond(LTE-Advanced) candidate is performed, con-firming that the proposal satisfies all the IMT-Advanced requirements.

ITU-R FRAMEWORK ANDEVALUATION PROCESS

The path toward IMT-Advanced officially start-ed in March 2008, when the Circular Letter wassent out by the ITU-R to invite submissions ofIMT-Advanced technology proposals. The ITU-R schedule spans over the 2008–2011 timeframeand is shown in Fig. 1, as in [3].

The radio interface development process iscovered in several steps, the first one represent-ed by the issuance of the Circular Letter (step1), after which step 2 copes with the develop-ment of candidate RITs and sets of RITs(SRITs). Step 3 represents the submission/recep-tion of the RIT and SRIT proposals (andacknowledgment of receipt) to Working Party5D (WP5D), the group within ITU-R responsi-ble for IMT systems. Step 4 indicates the phasein which evaluation of candidate RITs or SRITsby evaluation groups is carried out. Steps 5, 6,and 7 refer to the review and coordination ofoutside evaluation activities, the review to assesscompliance with minimum requirements, and,finally, the consideration of evaluation results,consensus building, and decision. Step 8 refers tothe development of radio interface recommen-dation(s).

The timing of these phases can partially over-lap, as is clear from the above schedule, and notall the phases are treated within ITU-R. In par-ticular, step 4 is external to ITU-R. Organiza-tions willing to become an IEG have been invitedto register with ITU-R.

In November 2008 the European EurekaCeltic project WINNER+ registered as an IEGat ITU-R. WINNER+ has been very active inthe IMT-Advanced process since its early stages.WINNER+ has participated in both rounds ofworkshops organized by the IMT-Advanced pro-ponents in 2009 and 2010, and the relevant ITU-R WP5D meetings, by submitting severalcontributions and sharing the adopted workmethod, intended work plan, and calibrationassumptions and results. A dedicated websitewas activated by WINNER+ [4] to share theupdated calibration data status in real time withall the other IEGs. The calibration methodologyproposed by WINNER+ has represented a basicguideline for all the IEGs. The alignment ofsuch results across different evaluation groupshas been verified, which is beneficial for therobustness of the entire ITU-R process. A corre-spondence group was also initiated on the ITU-R website to address questions to the proponentsand exchange comments among the differentevaluation groups. WINNER+ has had a highlevel of communication with others through thistool.

The WINNER+ project, in its 30-month life-time, has produced consistent research work [5]on optimization of the radio interface conceptsfor IMT-A systems, also thanks to the heritageof activities carried out in the former EuropeanUnion Framework Program 6 projects WINNERI and WINNER II. In particular, WINNER IIstrongly influenced the channel model definitionfor IMT-A [6]. Based on expertise in IMT-Aradio technology concepts and link- and system-level simulation tools, the WINNER+ Evalua-tion Group has considered the 3GPP LTERelease 10 & Beyond (LTE-Advanced) SRITproposal consisting of a time-division duplexing(TDD) RIT and a frequency-division duplexing(FDD) RIT [7]. The WINNER+ group has eval-uated all minimum requirements for IMT-A sys-tems by means of analytical, inspection, andsimulation activities in order to perform a fullevaluation of the LTE-Advanced candidate tech-nology.

For simulation purposes, in order to guaran-tee the reliability of the results, evaluated char-acteristics have been assessed by a plurality ofpartners. During the course of the work, greatemphasis has been placed on reflecting realistic

Figure 1. The ITU-R schedule for the IMT-Advanced process mapped to ITU-R WP5D meetings.

# 1 (Feb 2008)

Geneva

Step 1 and 2

Step 4

Step 3

Step 8

Step 5, 6 and 7

# 2 (Jun 2008)

Dubai

# 3 (Oct 2008)

Seoul

# 4 (Feb 2009)

Geneva

# 5 (Jun 2009)

Geneva

# 6 (Oct 2009) Dresden

# 7 (Feb 2010)

Torino

# 8 (Jun 2010) Da Nang

2009 2010

# 9 # 10

2008 2011 The timing of these

phases can partially

overlap, as it is clear

from the above

schedule, and not all

the phases are

treated within ITU-R.

In particular, step 4

is external to the

ITU-R. Organizations

willing to become an

IEG have been

invited to register

with ITU-R.



behavior of the system under consideration, bymodeling non-ideal aspects (including, e.g.,effects of channel estimation errors, CQI mea-surement errors, and feedback delay as well as acorrect modeling of the overhead in the system).Simulators of different partner organizationshave been calibrated in order to provide consis-tent results. The adopted calibration approach,detailed calibration results, and the require-ments assessment are provided later.

PERFORMANCE CRITERIA ANDEVALUATION SCENARIOS

According to the evaluation process of ITU-R,IMT-A candidate proposals need to fulfill a setof 13 requirements related to technical perfor-mance for IMT-A radio interface(s) [8]. Therequirements ensure that candidate systems fitinto the framework of IMT systems.

It is to be checked by IEGs through inspec-tion of the proposal whether the candidate sys-tem supports scalable bandwidths in the IMT-Aspectrum, a wide range of services, and intersys-tem handover with at least one IMT-2000 sys-tem.

Furthermore, candidate systems should bedesigned to reach certain performance require-ments under best case conditions. Calculationsshould prove that peak spectral efficiencyrequirements can be reached, and that userplane and control plane latency as well ashandover interruption times meet the require-ments.

A third set of requirements refers to theefficient use of the radio spectrum under nor-mal operating conditions. Link- and system-level simulations need to demonstrate high cellspectral efficiency while ensuring basic servicefor cell edge users. A high number of simulta-neous voice calls must be supported, and thesystem should operate at user speeds of up to300 km/h.

For these simulations the ITU-R givesdetailed guidelines for evaluation of RITs forIMT-A [9] to ensure comparable simulationresults across evaluation groups. According to[10] minimum requirements need to be fulfilledin three of four specific test environments thatreflect future use cases of IMT-A systems. Eachenvironment is associated with a deploymentscenario that specifies the simulation setup (e.g.,intersite distance, carrier frequency, maximumtransmit powers, channel model).

In particular, the deployment scenariosdefined in [9] are:

Indoor hotspot (InH): Small isolated cells atoffices or hotspot areas; targets high userthroughput or user density for pedestrian users.Two base stations operating at 3.4 GHz withomnidirectional antenna setup are mounted onthe ceiling of a long hall with adjacent offices(cell coverage area 3000 m2).

Urban microcell (UMi): High traffic and userdensity for city centers and dense urban areas.Outdoor and outdoor-to-indoor propagationcharacteristics for pedestrian users are assumed.Continuous hexagonal deployment is used with 3sectors/cell and below rooftop antenna mount-

ing. Base stations operate at 2.5 GHz and havean intersite distance of 200 m (cell coverage area0.035 km2).

Urban macrocell (UMa): Targets ubiquitouscoverage for urban areas. A similar hexagonaldeployment is used with larger intersite dis-tance of 500 m and antennas mounted clearlyabove the rooftop. Non-line-of-sight orobstructed propagation conditions are commonfor this scenario. Only vehicular users at mod-erate speed are assumed, suffering from anadditional outdoor to in-car penetration loss.Base stations operate at 2 GHz (cell coveragearea 0.22 km2).

Rural macrocell (RMa): Similar to UMa, buttargets larger cells with support for high-speedvehicular users. Base stations have an intersitedistance of 1732 m and operate at 800 MHz,which is more suitable for large cells (cell cover-age area 2.59 km2).

Suburban macrocell (SMa): This is an option-al scenario for the same test environment as ofthe UMa scenario. The key difference is anincreased intersite distance of 1299 m, and a mixof indoor and high-speed vehicular users (cellcoverage area 1.46 km2).

During the evaluation phase, the Indian eval-uation group TCOE India proposed in [10] anadditional optional scenario reflecting an impor-tant use case to serve rural areas. It can be char-acterized by:

Rural Indian open area: This is a large-cellcoverage scenario. Some parameters of the sce-nario may take several values (e.g., the carrierfrequency, terminal antennas height, and intersite distance). The intersite distance is 30–50 kmcorresponding to typical distance between vil-lages in India. In this scenario terminals are infixed positions with rooftop directional antennas.Base stations operate at 312–2300 MHz (cellcoverage area is up to 1962 km2).

WORKING METHOD OF THEWINNER+ EVALUATION GROUP

ASSESSMENT OF THE3GPP TECHNOLOGY CANDIDATE

In 2008 the 3GPP held two 3GPP IMT-Advanced Workshops. The goal of these work-shops was to investigate what were the mainchanges that could be brought forward toenhance the evolved universal terrestrial radioaccess radio interface as well as the evolved uni-versal terrestrial radio access in the context ofIMT-A.

In particular, the LTE-Advanced StudyItem was initialized in order to study the evo-lution of LTE, based on new performance tar-gets. This initiative has been collectingoperators’ and manufacturers’ views in order todevelop and test innovative concepts that willsatisfy the needs of the next-generation com-munications. The resulting technical report waspublished in June 2008 and a contribution wassent to ITU-R covering the work in 3GPP radioaccess network (RAN) on LTE-Advancedtoward IMT-A. Finally, the 3GPP has con-tributed to ITU-R toward IMT-A via its pro-

Candidate systems

should be designed

to reach certain

performance require-

ments under best-

case conditions.

Calculations should

prove that peak

spectral efficiency

requirements can be

reached and that

user-plane, and con-

trol-plane latency as

well as handover

interruption times

are meeting the

requirements.



posal “3GPP LTE Release 10 & Beyond (LTE-Advanced)” [7].

The new technical features of LTE-Advancedare defined in [11]. The main technical featuresare as follows.

Support of Wider Bandwidth — Carrieraggregation, where two or more component car-riers, each with a bandwidth up to 20 MHz, areaggregated, is considered for LTE-Advanced inorder to support downlink transmission band-widths larger than 20 MHz (e.g., 100 MHz).

Extended Multi-Antenna Configurations —Extension of LTE downlink spatial multiplexingis considered. LTE-Advanced supports spatialmultiplexing of up to eight layers for the down-link direction and up to four layers for the uplinkdirection. Enhanced Multi-User MIMO (MU-MIMO) transmission is supported in LTE-Advanced.

Coordinated Multiple Point Transmissionand Reception — Coordinated multi-point(CoMP) transmission/reception is considered forLTE-Advanced as a tool to improve the cover-age of high data rates, the cell-edge throughputand/or to increase system throughput. DownlinkCoMP transmission implies dynamic coordina-tion among multiple geographically separatedtransmission points. The 3GPP currently consid-ers the following two categories: Joint Processingand Coordinated Scheduling/Coordinated Beam-forming. Downlink CoMP transmission shouldinclude the possibility of coordination betweendifferent cells. Two implementations of CoMPcan be considered: inter-site CoMP and intra-site CoMP. Initially the focus of CoMP will beon intra-site schemes. In fact for Release 10,there will be no new standardized interface com-munication for support of inter-site CoMP,therefore no additional features are specified tosupport downlink CoMP. Uplink CoMP recep-tion is expected to have very limited impact onthe specifications. Uplink CoMP reception caninvolve joint reception of the transmitted signalat multiple reception points and/or coordinatedscheduling decisions among cells to controlinterference.

Relaying Functionality — Relaying is consid-ered for LTE-Advanced as a tool to improve thecoverage of high data rates, group mobility, tem-porary network deployment, the cell-edgethroughput, and/or to provide coverage in newareas. Relay nodes are placed throughout themacro-cell layout, hence modifying the referencelayout specified in [9]. Moreover the channelmodel to be used to model relay backhaulingtransmission link was not defined in [9]. Forthese reasons relay nodes have not been consid-ered as advanced feature to be used when assess-ing IMT-Advanced requirements.

The evaluation guidelines published by ITU-R in [9] are helpful for IMT-A systems evalua-tion but evaluating Beyond Release 10 systems isstill challenging since there is a need for specify-ing reference scenarios and missing parametersfor new features like e.g., CoMP or multilayerednetworks.

SPLITTING THE WORK: ANALYTICAL,INSPECTION, AND SIMULATION APPROACHES

In its Guidelines for evaluation of radio inter-face technologies for IMT-Advanced [9] theITU-R defined the characteristics for evaluatingIMT-A candidate proposals. The characteristicscan be classified based on the three differentmethods for evaluation:• Analytical• Inspection• Simulation (link-level or system-level)

Analytical evaluation comprises all character-istics that can be calculated. It is performed forthe characteristics of peak spectral efficiency,control and user plane latency, as well as intra-and interfrequency handover interruption time.Inspection is a non-numerical check by the IEGthat certain requirements are fulfilled and cer-tain capabilities are provided. The characteristicsbandwidth, intersystem handover, deployment inat least one of the identified IMT bands, channelbandwidth scalability, and support for a widerange of services are evaluated by inspection.

Numerical characteristics that are too compli-cated to be calculated are evaluated by simula-tive methods. These characteristics are cellspectral efficiency, cell edge user spectral effi-ciency, mobility, and VoIP capacity. The simula-tions results should respect the guidelines andthe deployment scenarios detailed in [9].

PREPARING THE WORK:CALIBRATION OF THE SIMULATORS

In the WINNER+ project the evaluations havebeen performed by several partners using differ-ent simulation tools. To ensure that all toolsyield coherent results, key components were cali-brated among partners. Specifically, the channelmodel implementation, which is technologyagnostic, and a basic setup of the baseline LTERelease 8 communication system were alignedamong partners. The calibration process wasimplemented using a stepwise approach withthree steps: channel model large-scale parame-ters calibration, channel model small-scaleparameters calibration, and baseline system cali-bration. Such calibration work provided highreliability to the WINNER+ IEG main evalua-tion work that was focused on the full assess-ment of the 3GPP LTE Release 10 & Beyond(LTE-Advanced) proposal.

The channel model proposed by ITU-R in [9]is far from being simple to implement. This iswhy the WINNER+ IEG addressed a channelmodel implementation calibration from thebeginning. The channel model calibration pro-cess was divided in two steps: large-scale andsmall-scale parameters calibration.

Large-scale calibration (LSC) is focused onthe calibration of the channel model implemen-tation without multipath effects (i.e., only withlarge-scale fading). The metrics used in this cali-bration are the path gain and wideband signal-to-interference-plus-noise ratio (SINR). Thepath gain is defined as the average signal attenu-ation between a user terminal and its servingbase station. The measure includes distanceattenuation, shadowing, and antenna gains (both

Uplink CoMP

reception is expected

to have very limited

impact on the

specifications. Uplink

CoMP reception can

involve joint

reception of the

transmitted signal at

multiple reception

points and/or

coordinated

scheduling decisions

among cells to

control interference.



at the base station and at the user terminal),while the effects from fast fading are excluded.The downlink wideband SINR, sometimes alsocalled the geometry, is the average powerreceived from the serving cell in relation to theaverage interference power received from allother cells plus noise. In addition to the evalua-tion principles and assumptions in [9] and thechannel model clarifications that followed, addi-tional assumptions concerning the cell selectionmechanism, feeder loss, and base station anten-na tilt have been used to derive the path gainand wideband SINR distributions. Exact valuesare included in [12].

Small-scale calibration (SSC) is focused onthe calibration of the multipath part of the chan-nel model. Given that the channel model is astochastic geometric model, the stochastic distri-butions of several geometric characteristics arecalibrated. These characteristics include thedelay spread, and the departure and arrivalangular spread at the base station and user ter-minal, respectively (also known as angle ofdeparture [AoD] and angle of arrival [AoA]).The root mean square delay spread and circularangular spread at the base station and user ter-minal are calculated for a large number of radiolinks, and in the calibrations the corresponding

distributions are compared. Mathematical defini-tions of these spread measures are included in[12]. The calibrations are performed separatelyfor line of sight (LoS), non line of sight (NLoS),and outdoor-to-indoor (OtoI) propagation con-ditions.

As an example of the calibration data collect-ed in this phase, we provide curves obtained inthe UMi deployment scenario in Fig. 2. Resultsof several partners are included and also theaveraged curves of the group. It can be conclud-ed that the calibration is achieved. The completecalibration data obtained by WINNER+ is avail-able in the WINNER+ IMT-A evaluation webpage [4].

WINNER+ has focused on evaluating the3GPP LTE Release 10 & Beyond (LTE-Advanced) proposal, and in order to prepare thesystem-level evaluations, a simulator calibrationfor the baseline configuration was performed inthe third step of the calibration process. The ref-erence baseline configuration is illustrated inFig. 3, and the detailed simulation parameterscan be found in [11].

Harmonization of simulators was done bycomparing uplink and downlink spectral efficien-cies (both cell and cell edge) for a baselinesetup. Implementations of all major parts of an

Figure 2. Channel model calibration (steps 1 and 2 of the calibration process) with examples of wideband SINR (left) and angle of depar-ture (right) distributions in the UMi NLoS scenario.

Partner #2

Wideband SINR (dB)

Cross checking and decision on calibration

-5

20

CD

F (%

]

10

0

30

40

50

60

70

80

90

100

-10 0 5 10 15 20

AOD (˚)

20

CD

F (%

)

10

0

30

40

50

60

70

80

90

100

0 40 80 120

Org 1Org 2Org 4Org 5Org 7Org 8Average

LSCsimulation

SSCsimulation

Partner #1

Wideband SINRpath gain

Delay spreadAoD spreadAoA spread

LSCsimulation

SSCsimulation

Partner #N

Wideband SINRpath gain

Delay spreadAoD spreadAoA spread

Org 1Org 2Org 4Org 5Org 7Org 8Average



LTE compliant protocol stack such as hybridautomatic repeat request (H-ARQ) retransmis-sions, channel status feedback loop, power con-trol, scheduling, and receiver setup wereincluded. For non-standardized algorithms base-line assumptions were made. By comparing thenormalized downlink and uplink user throughput(user spectral efficiency) distributions in Fig. 4,it can be seen that a good alignment betweenWINNER+ partners was achieved.

The presented information and benchmarkdata has been derived for all IMT-A deploymentscenarios, and shared with the other IEGs dur-ing the evaluation period in order to foster therequired coordination and unification of results.

LTE-ADVANCEDTECHNOLOGY CANDIDATE RESULTS

This section gives an introduction to a subset ofevaluation characteristics addressed by the WIN-NER+ IEG for the 3GPP LTE Release 10 &Beyond (LTE-Advanced) proposal assessment.The peak spectral efficiency is presented as anexample of the analytical method. This is fol-lowed up by simulation results based on theaforementioned calibration outcome.

Analytical Results — The peak spectral effi-ciency (PSE) is defined in [8]. It is basically thehighest theoretical data rate normalized bybandwidth assignable to a single mobile stationassuming error-free conditions. The WINNER+IEG evaluated PSE for LTE-Advanced FDD

and TDD modes in uplink and downlink. Inaddition to evaluation configuration parametersprovided in [9] with up to four Rx and four Txantennas at the base station and up to four Rxand two Tx antennas at the mobile station, con-figurations with up to eight antennas were alsoinvestigated for informative purposes.

From a mathematical point of view PSE cal-culation is not demanding. It is simply the num-ber of data bits that can be transmitted dividedby the bandwidth and the time needed for trans-mission.

But LTE-Advanced, as does any other mobileradio system, needs overhead that does not con-tribute to the data rate. Reference and synchro-nization signals as well as broadcast channelsand control signaling with channels carrying dif-ferent indicators and control information formsuch overhead. Depending on the mode and thedirection of transmission, different overheadtypes have to be taken into account. In TDDmode the guard period (GP) that separatesdownlink and uplink transmission in the timedomain adds additional overhead.

For the PSE calculation one may additionallydistinguish between different overhead typesthat add to the data rate or not. This topic wasraised during a workshop organized by 3GPP forall IEGs at the end of 2009 and finally clarifiedby ITU-R in a liaison statement in 2010. A fur-ther topic was the handling of the GP durationin TDD mode and its influence on the time nor-malization for PSE calculation.

The WINNER+ IEG provided multiple PSE

Figure 3. Baseline system calibration scenario and parameters. TDMA: time-division multiple access;FDMA: frequency-division multiple access.

ISD

57-sector UMi scenario

10 users per sector

Simulation parameter

Duplex method

Traffic

Downlink scheduling

Uplink scheduling

Uplink power control

...

Value

FDD

Full buffer

Round-robin (TDMA)

Round-robin (FDMA)P0 = 106.0 dBmAlpha = 1.0

...

Figure 4. Baseline system calibration (step 3 of the calibration process) for the UMi scenario.

Normalized user throughput (b/s/Hz]

Downlink

0.1

30

CD

F (%

)

20100

405060708090

100

0.0 0.2 0.3 0.4 0.5Normalized user throughput (b/s/Hz)

Uplink

0.1

30C

DF

(%)

20100

405060708090

100

0.0 0.2 0.3 0.4 0.5

Org. 1Org. 2Org. 4Org. 5Org. 6Org. 7

Org. 1Org. 5Org. 6



calculations for LTE-Advanced, and all of themclearly fulfilled the IMT-A requirements. Theresults for four-layer spatial multiplexing aresummarized in Table 1.

As it is clearly beyond the scope of this arti-cle to go into technical details, the interestedreader is referred to the final evaluation report[12] where the calculation is explained in detail.

Simulation Results — Simulations have beenderived by the organizations, and results arecompared to the ITU-R requirements. Theassessment is done in different ITU-R environ-ments, and for FDD and TDD RITs. The ITU-R guidelines impose that for the downlink, thenumber of antennas to be used should be higheror equal to n = 4 for the transmitter and m = 2for the receiver. However, for the uplink, onlythe receiver should use at least m = 2 antennas.The use of different transmitting schemesallowed by LTE-Advanced and the constraintgiven by the antenna number lead to differentsimulation results. The following definitions forthe transmission schemes hold:• SIMO: The transmitter uses one antenna

and the receiver m antennas. This schemeis called 1 × m single-input multiple-output.

• BF: The transmitter uses n and the receiverm antennas. The transmitter exploits the nantennas to orientate the transmittingpower of the transmitted data stream to thereceiver favorite direction. The scheme iscalled n × m beamforming.

• SU-MIMO: The transmitter uses n and thereceiver m antennas. The transmitter usesall n antennas to transmit for only onereceiver one or several data streams. Thisschemes is called n × m single-user multi-ple-input mutliple-output.

• MU-MIMO: Several receivers having mantennas share the n transmitting antennasto be served on the same time-frequencyresources. This scheme is called multi-userMIMO.Table 2 summarizes the main results in UMi

and UMa environments for FDD RITs. The resultspresented in this table for cell spectral efficiencyand cell edge spectral efficiency are averaged overresults coming from different organizations, evalu-ated using the same transmission scheme. We notethat different LTE-Advanced transmission schemespermit the requirement achievement for uplinkand downlink. The UMi and UMa deploymentscenarios are the most challenging since there is aneed to use MU-MIMO to achieve the downlinkrequirements. However, InH and RMa require-ments are met using SU-MIMO configuration.Uplink requirements are less demanding thandownlink requirements since they can be achievedwith SIMO configurations.

For the mobility assessment, the traffic chan-nel link data rate and support for mobility class-es are addressed. It is also shown that theirrequirements are also achieved for the consid-ered environments. Finally, the voice over IP(VoIP) capacity is assessed, and it is shown thatthe required number of active users per sectorper megahertz is achieved by the LTE-Advancedtechnology.

In general, the addressed requirements areachieved by simulations in all environments forFDD and TDD RITs. A complete set of assess-ment results for all ITU-R deployment scenariosderived by WINNER+ IEG is described in [12].The obtained results have confirmed that the3GPP LTE Release 10 & Beyond (LTE-Advanced) proposal satisfies all IMT-A require-ments.

CONCLUSIONSThe WINNER+ project responded to the ITU-R call to form an IEG and created its own eval-uation group. The evaluation effort has differentflavors ranging from careful study of the propo-nent proposal (inspection) through calculation(analytical) to link- and system-level simulations(simulation). Evaluations by simulations werepreceded by calibration. The stepwise calibra-tion exercise appeared to be a complex anddemanding task. During this step, communica-tion among independent evaluation groups wasrelevant. Making the results of the WINNER+IEG publicly available has enabled discussionsand the possibility to compare results amongother IEGs. Furthermore, WINNER+ gave ahint of one possible approach to coping withcalibration.

WINNER+ IEG also promptly reacted onproposed scenarios suggested by other IEGs, asin the case of the rural Indian open area addi-tional test scenario. The WINNER+ IEGresponse can be an example of an agile approachto the evaluation activity.

The WINNER+ evaluation group completedits assessment of the 3GPP LTE-Advanced pro-posal and submitted its final evaluation report toITU-R WP5D in June 2010. The main conclu-sion drawn from the results is that the 3GPPLTE Release 10 & Beyond (LTE-Advanced)proposal satisfies all the IMT-Advanced require-ments and thus qualifies as an IMT-Advancedsystem.

There is an expectation that further LTE evo-lution beyond Release 10 will provide even bet-ter performance since multiple featuresconsidered in further releases, such as relayingand coordinated multipoint (CoMP) transmis-sion and reception, were not part of the evaluat-ed proposal.

Table 1. Requirements and analytical PSE results for FDD and TDD RIT.

PSE in b/s/Hz ITU-R requirement FDD RIT assessment TDD RIT assessment

Downlink 15 16.3 15.8

Uplink 6.75 8.4 7.9

There is an expecta-

tion that further LTE

evolution beyond

Release 10 will

provide even better

performance since

multiple features

considered in further

releases e.g.,

Relaying, Coordinat-

ed Multipoint

Transmission and

reception (CoMP)

were not a part of

evaluated proposal.



ACKNOWLEDGMENT

This work has been performed in the framework ofthe CELTIC project CP5-026 WINNER+. Theauthors would like to acknowledge the contribu-tions of their colleagues in the WINNER+ consor-tium. The authors wish to thank colleagues fromEricsson, Per Skillermark and Johnan Nyström, fortheir effort in leading the simulations part of theWINNER+ evaluation group. The work of DavidMartín-Sacristán was supported by an FPU grantof the Spanish Ministry of Education.

REFERENCES[1] Doc IMT-ADV/1-E, “Background on IMT-Advanced.”[2] ITU-R Circular Letter 5/LCCE/2 and Addendums, “Invita-

tion for Submission of Proposals for Candidate RadioInterface Technologies for the Terrestrial Componentsof the Radio Interface(s) for IMT-Advanced and Invita-tion to Participate in their Subsequent Evaluation.”

[3] ITU-R Doc IMT-ADV/2 Rev1, “Submission and EvaluationProcess and Consensus Building.”

[4] WINNER+: A European ITU-R Evaluation Group;http://projects.celtic-initiative.org/winner+/WIN-NER+%20Evaluation%20Group.html.

[5] WINNER+ D4.2, “Final Conclusions on End-to-End Per-formance and Sensitivity Analysis,” June 2010;http : / /projects.celticinitiative.org/winner+/WIN-NER+%20Deliverables/D4.2_v1.0.pdf.

[6] P. Kyösti et al., “WINNER II Channel Models,” IST-WIN-NER D1.1.2 v. 1.1, Sept. 2007;https://www.ist-winner.org/WINNER2/Deliverables/D1.1.2v1.1.pdf

[7] Doc IMT-ADV/8, “Acknowledgment of Candidate Sub-mission from 3GPP Proponent (3GPP Organization Part-ners of ARIB, ATIS, CCSA, ETSI, TTA, AND TTC) underStep 3 of the IMT-Advanced Process (3GPP Technolo-gy),” Oct. 2009.

[8] ITU-R M.2134, “Requirements Related to Technical Per-formance for IMT-Advanced Radio Interface(s),” Nov.2008.

[9] ITU-R M.2135, “Guidelines for Evaluation of Radio Inter-face Technologies for IMT-Advanced,” Nov. 2008.

[10] ITU-R IMT-ADV/16, “Evaluation of IMT-Advanced Can-didate Technology Submissions in Documents IMT-ADV/4 and IMT-ADV/8 by TCOE India,” 2010.

[11] 3GPP TR 36.814, “Evolved Universal Terrestrial RadioAccess (E-UTRA); Further Advancements for E-UTRAPhysical Layer Aspects,” v. 9.0.0, Mar. 2010.

[12] Doc. IMT-ADV/22, “Evaluation of IMT-Advanced Candi-date Technology Submissions in Documents IMT-ADV/6,IMT-ADV/8, and IMT-ADV/9 by WINNER+ EvaluationGroup”;http://www.itu.int/ITU-R/index.asp?category=study-

groups&rlink=rsg5-imt-advanced&lang=en.

ADDITIONAL READING[1] ITU-R M.2133, “Requirements, Evaluation Criteria, and

Submission Templates for the Development of IMT-Advanced.”

UMi UMa

Requirement Assessment results Requirement Assessment results

Cell spectral efficiency in DL (b/s/Hz/cell) 2.6 2.88*(4 × 2 MU-MIMO) 2.2 2.38*

(4 × 2 MU-MIMO)

Cell spectral efficiency in UL (b/s/Hz/cell) 1.8

2.07*(1 × 4 SIMO)

2.41*(2 × 4 BF)2.59*(2 × 4 SU-MIMO)

1.4

1.60*(1 × 4 SIMO)

2.94 *(2 × 4 BF)1.97 *(2 × 4 SU-MIMO)

Cell edge spectral efficiency in DL (b/s/Hz/cell) 0.075 0.089*(4 × 2 MU-MIMO) 0.06 0.067*

(4 × 2 MU-MIMO)

Cell edge spectral efficiency in UL (b/s/Hz/cell) 0.05

0.082*(1 × 4 SIMO)

0.124*(2 × 4 BF)0.127*(2 × 4 SU-MIMO)

0.03

0.073*(1 × 4 SIMO)

0.092*(2 × 4 BF)0.091*(2 × 4 SU-MIMO)

MobilityTraffic channel link data rates in UL (b/s/Hz) 0.75 1.27*

(1 × 4 SU-MIMO) 0.55 1.36*(1 × 4 SU-MIMO)

Mobility classes supported

Stationary,pedestrian,

Vehicular(up to 30 km/h)

YesStationary,pedestrian,vehicular

Yes

VoIP capacity (active users/sector/MHz) 40 83* 40 66*

*Mean value of all contributing organizations for the given antenna configurations. Note that the mean value does not represent theperformance of one particular system setup. Values in bold (maximum values in case of multiple antenna configurations) are taken asmain results. UL: uplink; DL: downlink.



BIOGRAPHIESKRYSTIAN SAFJAN ([email protected]) graduated fromthe Wroclaw University of Technology, Poland, Faculty ofElectronics, Specialization of Telecommunications/DigitalSignal Processing in 2004. After graduation he joinedNokia Siemens Networks Sp z o.o., Poland, where he isinvolved in research and development of B3G systemswithin the Department of Radio System Technology. Hismain research interests are concentrated on advancedradio resource management methods.

VALERIA D’AMICO received an M.Sc. degree in electronicsengineering from the University of Catania, Italy. After aninternship in ST Microelectronics, she joined MarconiMobile. Since 2001 she is with Telecom Italia where shehas been involved in several activities targeting future-gen-eration communications, contributing to the Italian-fundedFIRB PRIMO project and the Eureka Celtic WINNER+ pro-ject. Currently she is involved in the EU ARTIST4G project,leading the work package on interference avoidance.

DANIEL BÜLTMANN received his Diploma (Dipl.-Ing.) in electri-cal engineering from RWTH Aachen University in 2004.Since January 2005 he has been employed as a researchassistant with the Research Group ComNets, RWTH AachenUniversity, where he is working toward his Ph.D. degree.The focus of his research is on LTE-Advanced radio resourcemanagement. He was involved in the WINNER II and WIN-NER+ projects.

DAVID MARTIN-SACRISTAN received his M.S. degree in telecom-munications engineering from the Polytechnic University ofValencia (UPV) in 2006. Nowadays, he is a Ph.D. student inthe Institute of Telecommunications and Multimedia Appli-cations (iTEAM), UPV. His research interests are focused onbeyond 3G networks including modeling and simulation,resource management, and link adaptation.

AHMED SAADANI received his engineering degree fromTunisia Polytechnic School in 1999, his Master’s degree in2000, and his Ph.D. degree in digital communications in2003 from the Ecole National Supérieure des Télécommuni-cations (ENST), Paris, France. He then joined Orange Labs,Issy les Moulineaux, France, as a research engineer, wherehe worked on advanced receivers and MIMO schemes for3G/3G+ systems. His current research interests are in coop-erative communications, relaying, and distributed MIMOfor 4G systems.

HENDRIK SCHÖNEICH received his Dipl.-Ing. degree in 2001 fora diploma thesis on co-channel interference cancellation inthe GSM system. He joined the Information and CodingTheory Laboratory (ICT) in 2001, where he worked as aresearch assistant on different research topics with a focuson interleave-division multiple access (IDMA). He receivedhis Dr.-Ing. degree for a thesis on adaptive IDMA in mobileradio communication systems. Since 2006 he has beenwith Qualcomm CDMA Technologies GmbH Nuremberg.His research interests include iterative interference cancella-tion, turbo equalization and decoding, semi-blind channelestimation, and related resource allocation strategies.



1 A full list of results fromall partners is available athttp://www.easy-c.com.

2 See ict-artist4g.eu.

INTRODUCTION

High spectral efficiency (i.e., high aggregatedcell data rate per unit of spectrum) is especial-ly important for data networks. Mobile datatraffic has recently surged due to the availabili-ty of affordable data dongles, notebooks, tabletcomputers with third-generation (3G) radiomodules, and smartphones with web-orienteduser interfaces. Vodafone, for example, hasobserved 70 percent growth of data trafficwithin one year for their European mobile net-works. So far, 3G networks could support thetraffic growth. However, eventually, more effi-cient wireless technology and novel deploy-ment concepts l ike small cells andheterogeneous networks are needed to providethe required capacity.

Ubiquitous user experience is key for the enduser to have a guaranteed minimum service

quality corresponding to a minimum data rate.Denser network deployments address this issuecaused by low link budget at the cell edge. How-ever, this goes along with larger areas where thetransmission is limited by interference.

Long Term Evolution (LTE) and mobileWiMAX use multiple-input multiple-output(MIMO)-orthogonal frequency-division multi-plexing (OFDM) and achieve improved spectralefficiency within one cell. However, inter-cellinterference is still preventing these technolo-gies from coming close to the theoretical ratesfor multi-cell networks. There are two funda-mental ways to deal with inter-cell interference:Coordination of base stations to avoid interfer-ence and constructive exploitation of interfer-ence through coherent base station cooperation.Conceptually, we extend single-cell MIMO tech-niques, such as multi-user (MU-MIMO), tomultiple cells.

This article shows results from the EASY-C1

project, which focused on coordinated multi-point (COMP) from 2007 to 2010 and set up twomultisite testbeds for LTE-based COMP inDresden and Berlin. ARTIST4G2 and otherforthcoming projects will continue to use theseplatforms.

COMP is a main element on the LTEroadmap beyond Release 9. In LTE Release 11,some simpler COMP concepts may appear, butit is generally expected that advanced COMPconcepts will take longer to be mature enoughfor commercial use.

The main scope of this article is to outlinethe basic COMP concepts, and highlight thepotentials and technical challenges when intro-ducing them in future mobile networks. More-over, we sketch practical COMP schemes foruplink and downlink, assess their performance inlarge-scale network simulations, and use field tri-als in urban areas to demonstrate the maturityof COMP.

ABSTRACT

Coordinated multipoint or cooperativeMIMO is one of the promising concepts toimprove cell edge user data rate and spectralefficiency beyond what is possible with MIMO-OFDM in the first versions of LTE or WiMAX.Interference can be exploited or mitigated bycooperation between sectors or different sites.Significant gains can be shown for both theuplink and downlink. A range of technicalchallenges were identified and partiallyaddressed, such as backhaul traffic, synchro-nization and feedback design. This article alsoshows the principal feasibility of COMP in twofield testbeds with multiple sites and differentbackhaul solutions between the sites. Theseactivities have been carried out by a powerfulconsortium consisting of universities, chip man-ufacturers, equipment vendors, and networkoperators.


Ralf Irmer, Vodafone

Heinz Droste, Deutsche Telekom

Patrick Marsch, Michael Grieger, and Gerhard Fettweis, Technische Universität Dresden

Stefan Brueck, Qualcomm CDMA Technologies GmbH

Hans-Peter Mayer, Alcatel-Lucent Bell Labs

Lars Thiele and Volker Jungnickel, Fraunhofer Heinrich-Hertz-Institut

Coordinated Multipoint: Concepts,Performance, and Field Trial Results

IRMER LAYOUT 1/19/11 3:33 PM Page 102


COORDINATION AND COOPERATIONIN MOBILE NETWORKS

One key element of mobile radio networks isspatial reuse (i.e., the reuse of resource elementssuch as timeslots or frequency bands) in a geo-graphical distance, where the signal strength isreduced due to path loss, shadowing, and so on.Historically, this was achieved using networkplanning with certain frequency reuse patterns,which have, however, the drawback of poorresource utilization. 3G and 4G technologies areusing full frequency reuse, which in turn leads tointerference between the cells.

In [1, 2] network coordination has been pre-sented as an approach to mitigate intercell inter-ference and hence improve spectral efficiency.Figure 1 shows the cooperation architecture forCOMP. The same spectrum resources are usedin all sectors, leading to interference for termi-nals (user equipment [UE] in Third GenerationPartnership Project [3GPP] terminology) at theedge between the cells, where signals from mul-tiple base stations are received with similar sig-nal power in the downlink. Multiple sectors ofone base station (eNB in 3GPP LTE terminolo-gy) can cooperate in intrasite COMP, whereasintersite COMP involves multiple eNBs.

The sectors at one site can be different self-sustained units, or different remote radio headslinked via fiber to a central baseband unit. TheeNBs may be interconnected by the logical X2interface. Physically, this could be a direct fastfiber link, or a multi-hop connection involvingdifferent backhaul technologies.

The cooperation techniques aim to avoid orexploit interference in order to improve the cell-edge and average data rates. COMP can beapplied both in the uplink and downlink. Allschemes come with the cost of increaseddemand on backhaul (high capacity and lowlatency), higher complexity, increased synchro-nization requirements, more channel estimationeffort, more overhead, and so on. The aim ofthis article is to highlight the potentials ofCOMP and its technical challenges to beaddressed for introducing it in next-generationmobile networks.

EVALUATION BY SIMULATION ANDFIELD TRIALS

Different approaches to COMP can be analyzedusing system-level simulations with hexagonalcells and evaluation methodologies customary inthe 3GPP, Next Generation Mobile Networks(NGMN), and International TelecommunicationUnion (ITU). Unless otherwise specified, theintersite distance in all computer simulationshas been set to 500 m, a terminal speed of 3km/h is assumed, and the system bandwidth is10 MHz.

The results of such simulations will be pre-sented in this article. However, it is not enoughto evaluate the feasibility of an approach solelybased on simulations. Field trials are essential tofind out the critical technical issues, and theyencourage an end-to-end view. The EASY-C

project has set up two outdoor testbeds withslightly different underlying technology andfocus, as shown in Table 1; see also [3–5].

UPLINK COORDINATED MULTIPOINT

OVERVIEWTheoretical work has shown that uplink (UL)COMP offers the potential to increase through-put significantly [1, 2], in particular at the celledge, which leads to enhanced fairness overall.Modeling some practical aspects such as a rea-sonably constrained backhaul infrastructureand imperfect channel knowledge, UL COMPpromises average cell throughput gains on theorder of 80 percent, and roughly a threefoldcell edge throughput improvement [6]. Thechannel information is available in the networkwithout resource-consuming feedback transmis-sions in the uplink. Also, the terminals need nomodifications in order to support UL COMP.Therefore, base station cooperation may beeasier to implement than in the downlink (DL).Only the interface between base station sites(X2) needs to be defined. In case of jointdetection in the UL, higher X2 capacity isneeded than for joint transmission in the DL.Although the UL capacity is not the bottleneckin today’s networks, guaranteeing a minimumdata rate, especially for cell edge users, isimproving user experience, and UL COMPmay be used to carry control traffic necessaryto implement DL COMP.

In general, the UL COMP schemes can beclassified as:

Interference-aware detection: Here, no coop-eration between base stations is necessary;instead, base stations also estimate the links tointerfering terminals and take spatially coloredinterference into account when calculatingreceive filters (interference rejection combining).

Joint multicell scheduling, interference pre-

Figure 1. Base station cooperation: intersite and intrasite COMP.

X2 interface (eNB-eNB)

Cell / sector Mobile

terminal (UE)

eNB

eNB eNB

eNB

Inter-Site COMP

Cell edge

Intra-Site COMP



diction, or multicell link adaptation, requiringthe exchange of channel information and/orscheduling decisions over the X2 interfacebetween base stations [7].

Joint multicell signal processing. Here,degrees of freedom exist in the way that decod-ing of terminals may take place in a decentral-ized or centralized way, and to which extentreceived signals are preprocessed before infor-mation exchange among base stations. In gener-al, there is a trade-off between using backhaulefficiently by a maximum extent of preprocessing(e.g., as in distributed interference subtraction,DIS, where decoded data is exchanged), butobtaining less CoMP gain, or using a large back-haul capacity (as in the case of the distributedantenna system, DAS, where quantized receivesignals are exchanged) and obtaining a betterperformance.

SELECTED SIMULATION RESULTSIn the following section selected UL cooperationschemes are introduced. During performanceevaluation it is distinguished between gains ofintrasite and intersite cooperation, where inter-site cooperation needs X2 backhaul capacity.

Uplink Interference Prediction — The basicidea of UL interference prediction [7] is to per-form link adaptation based on predicted signal-to-interference-plus-noise ratio (SINR) valuesthat are likely to occur during the associateddata transmissions. Prediction is enabled byexchange of resource allocation informationwithin a cluster of cooperating cells. In addition,the UL receivers provide channel state informa-tion related not only to their associated termi-nals, but also to the strongest terminals ofneighboring cells. Due to interference predic-tion, more appropriate link adaptation can berealized, and hence the performance can beimproved. The exchange of resource allocationinformation between two cells causes only mod-erate backhaul traffic in the range of 8 Mb/s.Whereas performance gains with intrasite coop-

eration prove to be rather low, we observe up to25 percent gain in spectral efficiency and 29 per-cent gain with respect to baseline cell edgethroughput if intersite cooperation including upto six interfering cells is simulated. The predic-tion accuracy degrades if the channel state infor-mation gets outdated. Therefore, the X2 latencyshould not exceed 1 ms, even at low terminalspeed.

Uplink Joint Detection — Uplink joint detec-tion means that signals received at different sec-tors are jointly processed [8]. Hence, virtualMIMO antenna arrays may spread out over dif-ferent users as well as different base station sec-tors at the network side. Most of theinformation exchange between cooperating cellsis caused by sharing the quantized basebandsamples received in each cell. Channel stateinformation and resource allocation tables areshared in the cooperation cluster as well. Firstestimates reveal that even with consideration ofless than half the cooperation cluster size asdescribed above for interference prediction, thecell-to-cell X2 traffic would exceed 300 Mb/s for10 MHz system bandwidth. This high amount ofbackhaul traffic motivates the investigation ofintrasite joint detection. In case of intersite jointdetection including up to three sectors per ter-minal, gains in spectral efficiency and cell edgethroughput account for 35 and 52 percent,respectively (2). Sticking with intra-site jointdetection, the improvements drop only moder-ately to 25 percent on average and 24 percent atthe cell edge (3).

Combining high throughput and low latencyas required by joint detection will cause a costburden for the backhaul, specifically the X2interface. Therefore, a combination of intrasitejoint detection (no X2 needed) and intersiteinterference predictions (low throughputdemand) has been considered. This even outper-forms the throughput-demanding intersite jointdetection, as shown in Fig. 2. However, the bur-den of low-latency X2 remains.

Table 1. COMP testbeds developed within the EASY-C project.

Dresden testbed Berlin testbed

Environment Dense urban

Trial setup 10 sites with up to a total of 28 sectors 4 sites with up to 10 sectors

Frequency 2.68 GHz DL, 2.53 GHz UL

Baseline technologyOFDMA in DL and UL, scalable bandwidth 5–20MHz, transmissions limited to a maximum of 40resource blocks (PRBs) in UL and 10 PRBs in DL.

DL: 2 × 2 MIMO-OFDMA, UL: 1 × 2 SC-FDMA,scalable bandwidth 1.5–20 MHz, full bandwidthcan be used in both up- and downlink

ProcessingReal-time DL transmission. For uplink COMPoffline processing. Scheduling is investigated inquasi-realtime.

Real-time PHY, adaptive MIMO multiple access andnetwork layer. PHY is extended for DL CoMP.

Backhaul and interconnects 5.4/5.8 GHz microwave with a net data rate of100 Mb/s and 1 ms delay

1 Gb/s Ethernet over optical fiber and free-space-optical links.

Testbed scope UL and DL MU-MIMO COMP, relaying, practicalissues

DL MU-MIMO, COMP, relaying, real-time demossuch as high-definition mobile video conference



SELECTED FIELD TRIAL RESULTSJoint decentralized and centralized detection ofterminals was evaluated in the Dresden testbed[9]. Two terminals with one transmit antennaeach transmitted continuous sequences of modu-lation and coding schemes, which were receivedby two base stations with one receive antennaeach (KATHREIN 80010541). The scenarioresembled a symmetric cell edge scenario, butthe terminals were moved such that interferenceconditions changed continuously. The receivesignals were recorded so that different coopera-tion schemes could be applied and evaluatedoffline.

The result plot in Fig. 3 shows the averagerates that could be achieved with different coop-eration strategies vs. the backhaul required.Here, square and round markers are used to dis-tinguish both UE types. We can see that in anLTE Release 8 system, where each UE unit isdecoded only by the serving base station, anaverage rate of about 1.5 b/channel use is possi-ble for UE 1 (square marker). This can beimproved to about 2.2 b/channel use simply if aflexible (i.e., transmission time interval [TTI]-wise) assignment of UE to eNBs is enabled, withthe option of local decoding with successiveinterference cancellation (SIC). A further rateimprovement of UE 1 is possible if DIS isenabled, where one UE unit is decoded first atone eNB, and the decoded data are then for-warded to the other eNBs for interference sub-traction, requiring a smaller extent of backhaul.This scheme turns out to reduce the outageprobability significantly. The remaining pointsshow the performance of a DAS, where theeNBs exchange quantized received signals, witheither 6 or 12 bits per antenna both for I and Qsignal dimensions. As compared to LTE Release8, in this scenario full DAS-based cooperationcan improve the average throughput by about 70

percent, but the backhaul required is more thantwo orders of magnitude larger than for decen-tralized concepts (DIS). Further measurementshave shown that DIS schemes become evenmore valuable in asymmetric scenarios, such thatan adaptive usage of centralized and decentral-ized cooperation schemes depending on theinterference situation appears promising.

The presented results provide evidence of thepotential benefits of using CoMP in specific sce-narios. Figure 4 shows the COMP gains in alarge-scale setup in the EASY-C testbed indowntown Dresden with 12 eNBs on five sites.The spacing between the sites is 350–600 m, withan antenna height of 15–35 m. Two UE units are

Figure 2. Performance of selected uplink COMP schemes: 1) inter-site interference prediction, 2) inter-site joint detection, 3) intra-sitejoint detection, 4) combining inter-site interference prediction with intra-site joint detection.

Cell edge throughput gain (%)

Spec

tral

eff

eici

ency

gai

n (%

)

Cel

l-to-

cell

(X2)

thr

ough

put

requ

irem

ent

(Mb/

s)

30

25

00 40 50 60 70

30

35

40

45

50

55

LTEA scheme

2

340

1

8

43

8

50

0

100

150

200

250

300

350

2

4

1

3

Figure 3. Achieved rates vs. required backhaul for different uplink cooperationschemes, as measured in field trials.

Average backhaul (b/channel use)

LTE Rel. 8

DIS

BS1

UE1 UE2

BS2

DAS (6-bit)

DAS (12-bit)

Flexible assignment

No backhaul required here

100 10-1

0.5

Ave

rage

rat

e (b

/cha

nnel

use

)

0

1.0

1.5

2.0

2.5

3.0

3.5

4.0

101 102



carried on a measurement bus on a 7.5 km lengthroute, as depicted in Fig. 4, which passes throughdifferent kinds of surroundings, including anunderpass, apartment buildings, a train station,and open spaces like parking areas. Convention-al non-cooperative decoding is compared tocooperative joint decoding. Through coopera-tion, average spectral efficiency gains of about20 percent were achieved. In certain areas, how-ever, gains above 100 percent were observed.Furthermore, the variance of achievable rates atdifferent UE positions was reduced, correspond-ing to fairer rate distribution throughout themeasurement area.

CHALLENGESFrom the experience of implementing and test-ing UL COMP, the following key challengeshave become apparent.

Clustering: Suitable clusters of cooperatingbase stations have to be found, which can bedone in a static way or dynamically, as discussedbelow.

Synchronization: Cooperating base stationshave to be synchronized in frequency such thatintercarrier interference is avoided, and in timein order to avoid both intersymbol and intercar-rier interference [10]. The maximum distance ofcooperating base stations is limited since differ-ent propagation delays of different terminalsmay conflict with the guard interval. This aspectmay be compensated through a more complexequalization.

Channel estimation: A large number of eNBsin the COMP cluster in the UL will require alarger number of orthogonal UL pilot sequences.

At some cluster sizes, the COMP gains are out-weighed by capacity losses due to additionalpilot effort.

Complexity: The above mentioned field trialshave been performed using orthogonal frequen-cy-division multiple access (OFDMA) in the UL,as this enables a subcarrier and symbol-wiseMIMO equalization and detection in the fre-quency domain. If single-carrier (SC)-FDMAwas used as in LTE Release 8, equalizationwould be more complex.

Backhaul: It can be a severe issue if central-ized decoding is applied. Hence, adaptive decen-tralized/centralized cooperation appears to be aninteresting option. Furthermore, source codingschemes appear interesting for backhaul com-pression.

DOWNLINKCOORDINATED MULTIPOINT

OVERVIEWBase station cooperation in the DL can alsoimprove average throughput and, more impor-tant, cell edge throughput [2]. 3GPP distinguish-es between the following categories of DLCOMP [11].

Coordinated scheduling/beamforming: Userdata is only available in one sector, the so-calledserving cell, but user scheduling and beamform-ing decisions are made with coordination amongthe sectors.

Joint processing COMP: User data to betransmitted to one terminal is available in multi-ple sectors of the network. A subclass of jointprocessing is joint transmission, where the datachannel to one terminal is simultaneously trans-mitted from multiple sectors.

Both coordinated scheduling/beamformingand joint transmission have been investigatedwithin the EASY-C project.

SELECTED SIMULATION RESULTSCoordinated beam selection [12] and co-scheduling are part of the investigated COMPschemes. Co-scheduling draws its gains frominterference avoidance and is less complex thanDL joint transmission. One approach whichincludes beamforming per cell is presentedhere. Synchronization of the cells is needed;however, there is no strict requirement onphase stability as known from coherent tech-niques. Multicell co-coordinated beamforminghas been assessed in system-level simulationstaking into account the latency for inter-NodeBcommunication.

The method is based on an extended precod-ing matrix index (PMI): the terminals measureand report the PMIs for their own cells (bestcompanion) and additionally the PMIs for theneighboring cells causing the strongest interfer-ence (worst companion) plus the channel qualityinformation for the case that these worst inter-ferers are not used.

The multicell scheduler is based on a dis-tributed approach, with overlapping clusters ofseven neighboring cells each. The scheduling iscoordinated within the clusters. The followingresults are given for four closely spaced antennas

Figure 4. Uplink COMP gains in EASY-C testbed in downtown Dresden.


at the base station and two UE antennas at 20MHz system bandwidth.

The simulations show significant gains forcoordinated DL scheduling, in particular formobiles at the cell edge. Additionally, thegains were evaluated for different radio chan-nels and different latencies for communicationbetween the sites. As can be observed in Fig.5, 1 ms latency/hop has only a moderateimpact on the gains. Even with highly time-variant channels such as urban macrocell(UMa) at 30 km/h, co-scheduling still providesa sensible improvement. Assuming 6 ms laten-cy per hop, the gains are still preserved forUE velocities up to 3 km/h.

The aggregated additional traffic on the back-haul sums up to approximately 5 Mb/s for 20MHz spectrum; as a result this technique is alsoeconomically attractive.

SELECTED FIELD TRIAL RESULTSSince joint transmission is regarded as the mostchallenging CoMP technique from the imple-mentation perspective, it has been implementedin both testbeds to investigate the feasibility ofcoherent transmission for intra- and intersiteCOMP [13, 14]. Significant throughput gainshave been demonstrated for specific interferencescenarios. The same techniques have also beenassessed in wide-area system-level simulations tostudy more complex scenarios. The followingenabling features were essential for the trials:• Sufficient timing and frequency synchro-

nization accuracy: In the trials GPS wasused, although network-based approachessuch as IEEE 1588v2 could also providesufficient accuracy.

• Low-phase-noise radio frequency (RF)oscillators were used.

• Cell-specific reference signals.• Time-stamped CSI feedback.• Synchronous exchange of data and channel

state information (CSI) between eNBs overthe X2 interface.

• Distributed precoding and the provision ofprecoded pilots.An example of the DL COMP experiments

conducted in the Berlin testbed is shown in Fig.7a. A distributed implementation of joint trans-mission has been demonstrated with synchro-nized base stations and cell-specific pilots.Terminals estimate the multicell channel andfeed the CSI back to their serving cells. Base sta-tions exchange CSI as well as data and indepen-dently perform pre-coding with the goal tomaximize the desired signals whilst minimizingmutual interference.

Quantization and compression of the CSI areimportant topics, but outside the scope of thistrial setup. CSI is fed back from the terminalssafely using UL resources at a data rate of 4.6Mb/s. Feedback interval and precoding delay are10 and 20 ms, respectively. The X2 interfacebetween base stations is realized using a 1 Gb/sEthernet connection over cooper, fiber, or free-space optics, depending on the setup. The bidi-rectional load is 300 Mb/s realized with 0.5 mslatency.

Measurements were taken in the laboratory[5] and over the air in both indoor and outdoor

environments (Fig. 6, bottom left). It wasobserved that the interference situation experi-enced at a terminal is indeed critical at the celledge if both base station signals are receivedequally strong on average at full frequencyreuse. Signal and interference links fade inde-pendently, and sometimes the signal is strongerthan the interference, while after a very shortdistance the opposite can be true. This is theorigin of the high outage probabilities observedat the cell edge in the interference-limited case(Fig. 6, bottom right). Once DL COMP isswitched on, significantly higher data rates canbe realized in both cells simultaneously, due tothe mutual interference cancellation. Moreover,the outage probability is remarkably reduced.Our experiments have shown that COMP gainsare significant for simple interference scenarios,and that the implementation challenges can beovercome.

In reality, non-cooperating cells would sur-round the cluster of cooperative cells, leading toa remaining interference floor not yet present inour trials. The presence of such external inter-ference has been studied in wide-area system-level simulations using basically the same COMPconcept also tested in the field. Note that theactive set is found so that in each cell two usersare randomly placed, and each user gets onlyone stream. In all cells, only those user setsrequesting the same cooperation cluster areinvestigated. In Fig. 6 (top right) we observe thatthere is no gain from using explicit CSI feedbackin the serving cell, exploited for multi-user DLbeamforming. Performance is equivalent to afixed grid-of-beams as in LTE Release 8 if theterminal estimates in addition the surroundinginterferers coherently, applies interference rejec-tion combining, and provides implicit frequency-selective feedback on interference-aware PMIand CQI, and the base station applies score-based scheduling [4]. Explicit CSI feedback isuseful for CoMP. With increasing cluster size,the interference floor is reduced and the perfor-mance enhanced accordingly, at the cost of addi-tional effort for overhead and backhaul. Formore details, see [5].


Figure 5. Downlink co-scheduling: spectral efficiency vs. cell edge throughputfor ITU UMa and SCME radio channels and different backhaul latencies.

Spectral efficiency (b/Hz/sector)

DL spectral efficiency vs. cell edge throughput

1.90

3.00E+055%-il

e C

DF

DL

UE

thro

ughp

ut (

b/s)

4.00E+05

5.00E+05

6.00E+05

7.00E+05

8.00E+05

9.00E+05

1.00E+06

2.00E+052.00 2.10 2.20

alpha 3.0alpha 2.0

alpha 1.0

alpha 0.5

2.30 2.40 2.50 2.60 2.70

3GPP_SCME_1ms3GPP_SCME_6ms3GPP_SCME_baselineITU_UMa_1msITU_UMa_6msITU_UMa_baseline



CHALLENGES

Our results indicate that the complexity of DLCOMP can be managed in real-world scenariosand that significant gain can be realized by form-ing small cooperation clusters in large-scale net-works. However, solutions for the following areneeded before it can be integrated in next-gener-ation mobile networks:• Reduced cost of base station synchroniza-

tion and low-phase-noise transmitters• Efficient feedback compression• Reduced feedback delay• Efficient channel prediction at the precoder• Flexible formation of cooperation clusters• Handling of outer interference within the

cluster• Efficient multi-user selection• Flexible networking behind COMP• Integration of COMP into higher layers

CLUSTERING OF CELLSAs demonstrated in previous sections, COMPhas the capability to enhance spectral efficien-cy and cell edge throughput significantly. How-ever, COMP requires additional signalingoverhead on the air interface and over thebackhaul in case of intersite cooperation.

Therefore, in practice only a limited number ofbase stations can cooperate in order to keepthe overhead manageable. The cooperating cellclusters should be set up adaptively based onRF channel measurements and UE positions inorder to exploit the advantages of COMP effi-ciently at limited complexity. A key require-ment for any adaptive cluster algorithm is thatit fits into the architecture of the radio accessand/or the core network of LTE. The 3GPPstandard already offers a framework for self-organizing networks (SONs) to support auto-matic configuration and optimization of thenetwork. Within EASY-C an adaptive mobile-station-aware clustering concept has beendesigned that can be integrated with smallstandard changes to the existing network archi-tecture and the SON concept of LTE.

In order to evaluate the performance of theadaptive clustering concept, system-level simu-lations were run employing a hexagonal net-work layout shown in Fig. 7a. The scenario wasconfigured with 19 3-sector sites of 500 m inter-site distance. The 3GPP UMa spatial channelmodel (SCM) at 2 GHz was used. The shadowfading standard deviation was set to 2 dB. Onehundred UE unitss were placed at random loca-tions within each of four hotspot areas. Figure

Figure 6. Top left: Distributed implementation of Joint transmission COMP. Top right: Performance and backhaul traffic vs. cluster sizeobtained from system-level simulations. Bottom left: Intra- and inter-site test scenarios in Berlin [5]. Bottom right: Measured throughputwith full frequency reuse in a two-cell scenario w/o external interference relative to the case of isolated cells.

Mar

chst

raβe Cluster size. i.e., number of sectors involved in JP CoMP

non-CoMP00

Med

ian

spec

tral

eff

icie

ncy

(b/s

/Hz/

sect

or)

Back

haul

tra

ffic

(pa

yloa

d on

ly)

(b/k

Hz)

5

10

150

100

50

15

2 3 4 5 6 7 8 9 10

Rate relative to isolated cell rate

00

CD

F

0.91

0.7

0.5

0.3

0.1

0.9 10.80.70.60.50.40.30.20.1

Interference limited, IntersiteInterference limited, IntrasiteCoMP IntersiteCoMP Intrasite

Channelfeedback

MT1

300m

N

Hardenbergstr.

Technical University

TUB(43 m)

TLabs(84 m)

540 m

Ernst ReuterPlaza

Strasse des17. Juni

Campus

485

m

480 m

HHI Einsteinuƒer

BS1

H11 H22

H12 H21

Localprecoder

DataMT1

DataMT2

MT2

BS2

Channelfeedback

Localprecoder

Inter-BS link (X2)

Channel feedbackexchange

Intersite sector

Intrasite sector

Intersite point

Intrasite point

K – 1traffic = rate ⋅ (1+ code_rate)

w.r.t. CoMP LTE mapping

LTE 1x1, round robinLTE 2x2, round robinLTE 2x2, score-basedCoMP, LTE L2S mapping



7b shows the result of the designed clusteringalgorithm, which was configured to obtain theoptimal solution for a disjoint set of clusterswith up to three sectors. The colors representthe different clusters. The clustering algorithmtook only long-term average received powermeasurements from UE into account in casethey were higher than –120 dBm. It is apparentfrom the figure that this concept managed toform clusters around the UE hotspots andavoided clusters in regions where not needed.The mean geometry gain due to adaptive clus-tering was about 6 dB for this scenario com-pared to LTE Release 8.

BACKHAUL FOR COMPARCHITECTURE AND TECHNOLOGIES

COMP approaches need to exchange directinformation between cells, with different require-ments of necessary backhaul throughput andlatency. Intra-site COMP can be realized with-out any impact on backhaul. In the case ofdeployment of remote radio units connected to acentralized baseband processing unit via Ether-net or fiber links, COMP backhaul requirementsshould also be no obstacle.

For connectivity between sites, the logical X2interface could be used. This could either be adirect physical link or a multihop link, depend-ing on the network’s backhaul architecture. Thedelay depends on the network topology, networknode processing delay and line delay (usuallyspeed of light). Gigabit Ethernet speeds of up 10Mb/s and delays of 0.1–20 μs with additionaldelays due to switching equipment. Other suit-able candidates are conventional and millimeter-wave microwave, with speeds up to of 800 Mb/sor 10 Gb/s, respectively, and delays as low as 150μs/hop.

LATENCY REQUIREMENTS

COMP has to be integrated with the hybridautomatic repeat request (HARQ) process; thus,the backhaul latency will put some limits on this,suggesting a maximum latency of 1 ms withoutLTE standard modification.

Another impact of backhaul latency is thatthe exchanged channel information is outdated.For example, a minor performance degradationwas estimated for coordinated scheduling con-sidering a X2 latency of 6 ms. In [15] a DLCOMP capacity gain reduction of 20 percent isestimated for joint transmission with 5 ms back-haul latency at 3 km/h.

CAPACITY REQUIREMENTSCOMP schemes require the exchange of channelstate information, control data, user data, andreceived signals, in a preprocessed or quantizedformat.

As shown earlier and in [16, 17], the backhaulrequirements vary strongly from a few megabitsper second up to 4 Gb/s for different COMPapproaches, considering a 10 MHz LTE X2 link.This also depends on the cluster size. Earlier weshowed an example of how backhaul can bereduced significantly even without major perfor-mance losses.

To conclude, state-of-the-art backhaul tech-nology can support COMP in principle. Howev-er, the cost of additional backhaul and accesscapacity gains has to be balanced in a networkdeployment.

CONCLUSIONS AND OUTLOOKThis article has shown that coordination of cellsin wide-area systems is not only beneficial foraverage spectral efficiency and cell edge data

Figure 7. Cell layout with UE positions and selected clusters: a) no clustering; b) adaptive clustering.

X-Distance (m)

Original CoMP cluster layout and UE positions

-1000

-1000

Y-D

ista

nce

(m)

-500

0

500

1000

-500 0 500 1000X-Distance (m)

Adapted CoMP Cluster Layout and UE Positions

-1000

-1000

Y-D

ista

nce

(m)

-500

0

500

1000

-500 0 500 1000

01

2

34

5

2122

23

2425

26

2728

291011

122223

24

67

8910

113637

38

3940

41 4243

44 4546

47

4849

50

1112

13

1415

161819

201516

17

1213

14

01

2

34

5

2122

23

2425

26

2728

291011

122223

24

67

8910

113637

38

3940

41 4243

44 4546

47

4849

50

1112

13

1415

161819

201516

17

1213

14



rates, but can also be implemented. COMP wasdemonstrated for uplink and downlink in twotestbeds in urban areas. COMP schemes for theUL range from joint multicell scheduling tomore complex joint detection, and can be cen-tralized or decentralized. In the DL the schemesrange from less complex coordinated schedulingto more challenging joint processing approaches.

From the technical as well as economic pointsof view, intrasite cooperation will be much easierto realize. However, intersite cooperation will beneeded in order to exhaust the full interferencereduction potential of base station cooperation.The combination of joint processing at one sitewith joint scheduling between the sites is ofgreat interest as it provides promising gains withlimited backhaul.

The following challenges needs to beaddressed in order to benefit from the promisingCOMP gains:• Backhaul with low latency and high band-

width. Today’s backhaul technologies cansupport COMP, but more effort is neededto reduce the amount of data exchangedbetween the sites.

• Clustering and multisite scheduling.• Channel estimation and efficient feedback

(for DL COMP).• Synchronization between sites is feasible

today, but the cell area where COMP canbe applied may be limited by the length ofthe cyclic prefix.

• Combination of UL and DL COMP andtheir integration into the LTE standard.This article, and the EASY-C project, have

already given some answers on COMP. Ongoingefforts to address the challenges in the researchcommunity — such as the ARTIST4G projectand 3GPP standardization — are important togain more insight into achievable spectral effi-ciency gains and the complexity of differentapproaches.

ACKNOWLEDGMENTThe authors acknowledge the excellent coopera-tion of all project partners within the EASY-Cproject and the support of the German FederalMinistry of Education and Research (BMBF).

REFERENCES[1] P. Marsch, S. Khattak, and G. Fettweis, “A Framework

for Determining Realistic Capacity Bounds for Distribut-ed Antenna Systems,” Proc. IEEE Info. Theory Wksp.‘06, Chengdu, China, Oct. 22–26, 2006.

[2] K. M. Karakayli, G. J. Foschini, and R. A. Valenzuela.“Network Coordination for Spectrally Efficient Commu-nications in Cellular Systems,” IEEE Wireless Commun.,vol. 13, no. 4, Aug. 2006, pp. 56–61.

[3] P. Marsch and G. Fettweis, “On Multi-Cell CooperativeTransmission in Backhaul-Constrained Cellular Systems,”Annales des Télécommun., vol. 63, no. 5–6, May 2008.

[4] V. Jungnickel et al., “Interference Aware Scheduling inthe Multiuser MIMO-OFDM Downlink,” IEEE Commun.Mag., vol. 47, no. 6, June 2009.

[5] V. Jungnickel et al., “Field Trials using CoordinatedMulti-Point Transmission in the Downlink,” 3rd IEEEInt’l. Wksp. Wireless Distrib. Net., IEEE PIMRC, Sept.2010.

[6] P. Marsch, Coordinated Multi-Point under a ConstrainedBackhaul and Imperfect Channel Knowledge, Ph.D. the-sis.

[7] A. Müller and P. Frank, “Cooperative Interference Pre-diction for Enhanced Link Adaptation in the 3GPP LTEUplink,” IEEE VTC–Spring, 2010.

[8] A. Müller and P. Frank, “Performance of the LTE Uplinkwith Intra-Site Joint Detection and Joint Link Adapta-tion,” IEEE VTC–Spring, 2010.

[9] M. Grieger et al., “Field Trial Results for a CoordinatedMulti-Point (CoMP) Uplink in Cellular Systems,” Proc.ITG/IEEE Wksp. Smart Antennas ‘10, Bremen, Germany,Feb. 23–24, 2010.

[10] V. Kotzsch and G. Fettweis. “Interference Analysis inTime and Frequency Asynchronous Network MIMOOFDM Systems,” IEEE WCNC ‘10, Sydney, Australia,Apr. 18–21, 2010.

[11] 3GPP TR 36.814, “Further Advancements for E-UTRAPhysical Layer Aspects,” Release 9, v. 9.0.0, Mar. 2010.

[12] J. Giese and M. A. Awais, “Performance Upper Boundsfor Coordinated Beam Selection in LTE-Advanced,”Proc. ITG/IEEE Wksp. Smart Antennas ‘10, Bremen, Ger-many, Feb. 23–24, 2010.

[13] G. Fettweis et al., “Field Trial Results for LTE-AdvancedConcepts,” Proc. IEEE ICASSP ‘10, Dallas, TX, Mar.14–19, 2010.

[14] L. Thiele, V. Jungnickel, and T. Haustein, “InterferenceManagement for Future Cellular OFDMA Systems UsingCoordinated Multi-Point Transmission,” IEICE Trans.Commun., Special Issue on Wireless Distributed Net-works, Dec. 2010.

[15] S. Brueck et al., “Centralized Scheduling for Joint-Transmission Coordinated Multi-Point in LTE-Advanced,”Proc. ITG/IEEE Wksp. Smart Antennas ‘10, Bremen, Ger-many, Feb. 23–24, 2010.

[16] C. Hoymann, L. Falconetti, and R. Gupta, “DistributedUplink Signal Processing of Cooperating Base Stationsbased on IQ Sample Exchange,” Proc. IEEE ICC ‘09,Dresden, Germany, June 14–18, 2009.

[17] L. Falconetti, C. Hoymann, and R. Gupta, “DistributedUplink Macro Diversity for Cooperating Base Stations,”Proc. IEEE ICC ‘09, Dresden, Germany, June 14–18,2009.

ADDITIONAL READING[1] R. Irmer et al., “Multisite Field Trial for LTE and

Advanced Concepts,” IEEE Commun. Mag., vol. 47, no.2, Feb. 2009, pp. 92–98.

BIOGRAPHIESRALF IRMER [SM] ([email protected]) received hisDipl-Ing. and Dr.-Ing. degrees from Technische UniversitätDresden in 2000 and 2005, respectively. He joined Voda-fone Group R&D in 2005, where he leads the WirelessAccess Group, which is responsible for evolution of LTE,WiFi, and other technologies, and defining Vodafone’sfuture network architecture. Before, he worked for fiveyears as a research associate at TU Dresden. He holds sev-eral patents, and has published more than 30 conferenceand journal publications. He had a leading role in severalresearch projects, including WIGWAM, WINNER, and EASY-C. He is a member of VDE and IET.

HEINZ DROSTE ([email protected]) received his Dipl.-Ing degree 1991 from the Open University, Hagen. Sincethen he has been working for Deutsche Telekom at a vari-ety of mobile communication related R&D projects. Anten-nas and radio wave propagation belong to his knowledgefield as well as system-level simulation and radio networkplanning. More recently he extended his expertise to thefield of techno-economical evaluations. In the frameworkof EASY-C he coordinates the partner activities in WorkingGroup 1, “Algorithm and Concepts.”

PATRICK MARSCH ([email protected]) received hisDipl.-Ing. and Dr.-Ing. degrees from Technische UniversitätDresden in 2004 and 2010, respectively, after completingan apprenticeship at Siemens AG and studying at the TUDresden and McGill University, Montréal, Canada. After aninternship with Philips Research East Asia in Shanghai, P.R.China, he joined the Vodafone Chair in 2005. He is thetechnical project coordinator of EASY-C, and is currentlyheading a research group on the analysis and optimizationof cellular systems.

MICHAEL GRIEGER received his Dipl.-Ing. from DHBW Stuttgartin 2005 and his M.Sc. from the Technische UniversitätDresden in March 2009. In 2008, funded by the HerbertQuandt/ALTANA Foundation, he studied at CTU, Prague.During his Master’s thesis, he conducted research in Prof.John Cioffi’s group at Stanford University on multicell sig-

COMP schemes for

the uplink range

from joint multi-cell

scheduling to more

complex joint

detection, and can

be centralized or

decentralized. In the

downlink, the

schemes range from

less complex

coordinated

scheduling to more

challenging joint

processing

approaches.



nal processing, which continues to be his major researchfocus today. An aspect of his research is the comparison ofinformation theoretic results to those of the “real world”using field trials.

GERHARD FETTWEIS [F] ([email protected]) earnedhis Dipl.-Ing. (1986) and Ph.D. (1990) degrees from AachenUniversity of Technology (RWTH), Germany. From 1990 to1991 he was a visiting scientist at the IBM AlmadenResearch Center, San Jose, California, working on signalprocessing for disk drives. From 1991 to 1994 he was withTCSI Inc., Berkeley, California, responsible for signal proces-sor development. Since 1994 he holds the Vodafone Chairat TU Dresden. He is coordinating the research projectEASY-C.

HANS-PETER MAYER ([email protected])received his Ph.D. degree in physics from the University ofTübingen in 1987. He joined Alcatel-Lucent and worked onhigh-speed optoelectronic and WDM components until1995. From 1996 to 1999, he has been responsible forearly UMTS system studies, followed by the realization offirst UMTS and HSPA trial systems. Within Bell Labs, he iscurrently responsible for the Advanced MAC departmentwith a focus on projects related to LTE-Advanced.

STEFAN BRUECK ([email protected]) studied mathe-matics and electrical engineering at the University of Tech-nology Darmstadt, Germany, and Trinity College Dublin,Ireland. He received his Dipl.-Math. and Dr.-Ing. degrees in

1994 and 1999, respectively. From 1999 to 2008 he wasworking for Lucent Technologies and Alcatel-Lucent in BellLabs and UMTS Systems Engineering, where he was respon-sible for the MAC layer design of the HSPA base station. InMay 2008 he joined Qualcomm Germany and currentlyleads the Radio Systems R&D activities in the CorporateR&D Centre Nuremberg. He is involved in several researchprojects on LTE-Advanced.

LARS THIELE [S‘05] received his Dipl.-Ing. (M.S.) degree inelectrical engineering from TU Berlin in 2005. Currently heis working towards his Dr.-Ing. (Ph.D.) degree at theFraunhofer Heinrich Hertz Institute (HHI), Berlin. He hascontributed to receiver and transmitter optimization underlimited feedback, performance analysis for MIMO trans-mission in cellular ODFM systems, and fair resource alloca-tion. He has authored and co-authored about 40conference and journal papers in the area of mobile com-munications.

VOLKER JUNGNICKEL [M‘99] ([email protected]) received a Dr.rer. nat. (Ph.D.) degree in physics from Humboldt Univer-sität zu Berlin in 1995. He worked on semiconductor quan-tum dots and laser medicine and joined HHI in 1997. He isa lecturer at TU Berlin and head of the cellular radio teamat HHI. He has contributed to high-speed indoor wirelessinfrared links, 1 Gb/s MIMO-OFDM radio transmission, andinitial field trials for LTE and LTE-Advanced. He hasauthored and co-authored more than 100 conference andjournal papers on communications engineering.



INTRODUCTION

The 3GPP Long Term Evolution (LTE) stan-dard is continuing to evolve to add more capa-bilities and enable transmission and reception ofeven higher data rates [1]. One of the key fea-tures to be introduced in LTE-Advanced(Release 10 and beyond) is multiple-inputs mul-tiple-outputs (MIMO) technique for uplink(UL). MIMO helps achieve the LTE-Advancedrequirements, which include UL peak spectrumefficiency of 15 b/s/Hz and UL average spectrumefficiency of 2.0 b/s/Hz/cell [2]. With four trans-mit antennas and four receive antennas, spatialmultiplexing for data channel achieves the ULpeak spectrum efficiency. Given the availabilityof multiple antennas, transmit diversity for con-trol channel provides robust signaling andimproves cell coverage. In addition, multi-userMIMO (MU-MIMO) helps attain the UL aver-age spectrum efficiency.

In this article, the evolution of UL-MIMOschemes is described. Precoded spatial multi-plexing for the UL data channel, the physicaluplink shared channel (PUSCH), is explained,emphasizing the cubic metric (CM) preservingcodebook design and its precoding gain. Trans-mit diversity for the UL control channel, thephysical uplink control channel (PUCCH), isalso described, together with its orthogonalresource allocation scheme. Additionally, theUL-MIMO aspects of reference signals anddownlink (DL) control signaling are discussed,particularly the impact of interlayer interferenceon channel estimation and a new reference sig-

nal multiplexing. To maximize the benefit ofMIMO, a baseband receiver capable of untan-gling the multiple spatially-multiplexed signalsreceived simultaneously is desired. In LTE-UL,single-carrier frequency-division multiple access(SC-FDMA) introduces intersymbol interference(ISI) in dispersive channels. Thus, a desiredMIMO receiver should address both ISI andspatial-multiplexing interference. We present anumber of receiver options, including linearminimum mean square estimation (MMSE)equalization, successive interference cancellation(SIC), and turbo equalization receivers. Linksimulation results are provided to demonstratethe performance of UL data channel and alsocompare performance among these differentreceivers.

SINGLE-CARRIER TRANSMISSIONThe UL multiple access in LTE is characterizedby SC-FDMA, where discrete Fourier transform(DFT) is followed by orthogonal frequency-divi-sion multiplexing (OFDM) [3, 4]. Since DFTspreads each time domain (TD) modulationsymbol over all the subcarriers assigned, itenables single-carrier transmission, as opposedto OFDM.

The main objective of single-carrier transmis-sion in the UL is to relax the requirement ofuser equipment (UE) power backoff [5]. Sincethe power amplifier of UE is nonlinear in gener-al, the output power tends to saturate aroundthe peak power. Therefore, in order to avoidnonlinear distortion, the power amplifier needsto operate far below the saturation point. Thisinevitably limits the maximum transmit power ofUE, thereby resulting in cell coverage limitation.The amount of power backoff depends on howmuch the amplitude of transmitted signal fluctu-ates around the average value, which is mea-sured by peak-to-average-power-ratio (PAPR) orCM. Since the required power backoff is practi-cally determined by adjacent channel interfer-ence caused by nonlinear power amplification(dominated by the cubic term), CM is a moreaccurate measure for power backoff require-ments. The more the amplitude of transmittedsignal fluctuates, the more power backoff the

ABSTRACT

The evolution of LTE uplink transmissiontoward MIMO has recently been agreed in3GPP, including the support of up to four-layertransmission using precoded spatial multiplexingas well as transmit diversity techniques. In thisarticle, an overview of these uplink MIMOschemes is presented, along with their impact onreference signals and DL control signaling.Receivers suitable for uplink MIMO are pre-sented, and their link performances are com-pared.


Chester Sungchung Park, Y.-P. Eric Wang, George Jöngren, and David Hammarwall, Ericsson Research

Evolution of Uplink MIMO forLTE-Advanced

PARK LAYOUT 1/19/11 3:34 PM Page 112


power amplifier demands and the lower power itcan radiate. Since an SC-FDMA signal resem-bles a conventional single-carrier signal, SC-FDMA has a lower CM than OFDM, therebyleading to less power backoff and thus improvedcoverage.

MIMO TECHNIQUES

UL DATA CHANNEL:PRECODED SPATIAL MULTIPLEXING

A block diagram of a transmitter and receiverfor the UL data channel is presented in Fig. 1a,which consists of a turbo encoder/decoder, ascrambler/descrambler, a modulation mapper/demodulator, a layer mapper/codeword mapper,a DFT spreader/despreader, a precoded spatialmultiplexer/frequency domain (FD) equalizer, aresource mapper/demapper, and inverse fastFourier transform (IFFT)/FFT. Analogous tothe transmitter for the Release 8 DL data chan-nel, the physical downlink shared channel(PDSCH), up to two codewords can be transmit-

ted in one subframe with separate control ofmodulation and coding scheme (MCS) andhybrid automatic repeat request (HARQ) func-tionality. One or two codewords are mapped toup to four independent data streams (referred toas layers hereafter) by the same codeword-to-layer mapping as for DL-MIMO [6]. DFTspreading is applied to each layer separately, andprecoded spatial multiplexing is applied in fre-quency domain. Specifically, the precoder mayspread each layer across multiple transmit anten-nas, as exemplified in Fig. 1b. The number oflayers (R) (referred to as rank hereafter) is equalto or smaller than the number of transmit anten-nas (NT). Finally, IFFT is performed on a per-antenna basis. The receiver simply reverses theabove operation, as shown in Fig. 1a.

The basic principles of the precoded spatialmultiplexing scheme for the UL data channelare similar to those for DL-MIMO [6]. The pre-coding operation is mathematically expressed asa left-multiplication of a DFT-spread layer signalvector (R × 1) by a precoding matrix (NT × R),which is chosen from a predefined set of matri-

Figure 1. Precoded spatial multiplexing for UL data channel: a) transmitter and receiver; b) illustration of precoding operation (2 trans-mit antennas, 1 layer).

IFFT

Antenna portsLayers

IFFT

Resourcemapper

Resourcemapper

DFTspreader

Precoder

DFTspreader

Modulationmapper

Layermapper

Scrambler

Equivalent channel(SNR)Precoder

Constructive combining

Destructive combining

=

Equivalent system

w1

h1

h2

w1 w2h2

w2h2

w2h2

w2h2

w1h1

w1h1

w1h1

w1h1

w1h1+w2h2

w1h1+w2h2

w1h1+w2h2

w1h1+w2h2

w2

Precoder

MIMO with precoder

Encoder

Codewords

Antenna portsLayersCodewords

w2

w1h1+w2h2

ModulationmapperScramblerEncoder

FFT

(a)

(b)

FFT

Resourcedemapper

Resourcedemapper

IDFTdespreader

FDequalizer

IDFTdespreader

Demodulator

Codewordmapper

DescramblerDecoder

DemodulatorDescramblerDecoder

+1

+1

= w1

w2

+1

+j

= w1

w2

+1

-1

=

:

:

:

: w1

w2

+1

-j



ces, a so-called codebook exemplified in Fig. 2.(Note that the rth column vector of the precod-ing matrix represents the antenna spreadingweight of the rth layer.) The main purpose ofthis kind of precoding is to match the precodingmatrix with the channel properties to increasethe received signal power and also to someextent reduce inter-layer interference, therebyimproving the signal-to-interference-plus-noise-ratio (SINR) of each layer. Assuming that theeNodeB is equipped with NR receive antennas,the capacity-achieving precoding, referred to aseigen-beamforming, precodes each layer withone of min(NR, NT) eigenvectors of the channelcorrelation. However, in practice, the selectionof precoder needs to be signaled from eNodeBto UE; thus, the number of precoding matrices(i.e., codebook size) is limited by the availableamount of DL control signaling. The use of afinite-sized codebook results in a reduced pre-coding gain. Additionally, as shown in Fig. 2, thealphabet for the nonzero elements of precodingmatrices is confined to quaternary phase shiftkeying (QPSK) (±1, ±j). This eases the selec-tion of precoding matrix, since it greatly simpli-fies matrix multiplication. Moreover, theconstant-modulus property of the QPSK alpha-bet enables all the power amplifiers to transmitwith equal power level. This balanced powerhelps maximize the power amplifier efficiency,regardless of which precoding matrix is used. Adrawback of this alphabet constraint is furtherreduction of the precoding gain.

A major difference from the precoder code-book used for DL-MIMO is that the UL code-book is carefully designed to not increase theCM, compared to single-antenna transmission.(In contrast, the Householder codebook of DL-MIMO linearly combines multiple layers [6] andthus inevitably increases CM.) A design con-straint of the CM-preserving codebook is thateach row should have at most one non-zero ele-ment. In other words, each transmit antenna isnot allowed to convey more than one layer; thus,no linear combination is allowed. This CM-pre-serving constraint gives rise to further reductionof the precoding gain. One of the objectives ofthe UL precoder design is to optimize the pre-coding gain under the CM-preserving constraint.

It is well known that the codebook maximiz-ing the minimum chordal distance makes sensefrom an information theoretic capacity point ofview [7]. However, in practice, UE antennas arelikely to be correlated; thus, UE antenna config-uration and indexing, shown in Fig. 2, also needto be considered, taking into account antennaspacing and polarization. Since each layer isassigned to a disjoint group of UE antennas,antenna grouping with lower interlayer correla-tion is preferred, as it leads to lower interlayerinterference. This is also consistent with eigen-beamforming in that eigenvectors of channelcorrelation tend to have elements with largeramplitude in one antenna group than in theother antenna groups. For example, in Fig. 2antenna grouping of (1,2) + (3,4) is preferred to(1,3) + (2,4), since two closely located or co-polarized antennas are grouped together in theformer grouping.

Under the above constraints, it easily follows

that the full-rank codebooks consist of the iden-tity matrices alone. For non-full-rank codebooks,it is desirable to maximize the precoding gainwhile preserving the CM. The design of 2-TXcodebooks is relatively simple. The rank-1 code-book consists of four constant modulus vectorsand two antenna selection vectors. The inclusionof constant modulus vectors enables the signalsfrom multiple UE antennas to be combined con-structively at the receiving eNodeB by introduc-ing a relative phase shift of 0°, 90°, 180°, or 270°between UE antennas. In the example shown inFig. 1b, a phase shift of 270° maximizes the sig-nal-to-noise ratio (SNR) of received signal dueto constructive combining, whereas a phase shiftof 90° minimizes it due to destructive combining.On the other hand, the inclusion of antennaselection vectors reduces transmit power and isintended for UE power saving. For example, ifUE antennas experience significant gain imbal-ance, one of the antenna selection vector can beused to switch off the low-gain antenna.

The design of 4-TX codebooks is more com-plicated. As presented in Fig. 2, the rank-1 code-book consists of 16 constant modulus vectors forconstructive combining and 8 antenna selectionvectors for UE power saving. For constant mod-ulus vectors, four relative phase shifts are con-sidered between the first two antennas. For eachof these relative phase shifts, only two relativephase shifts are considered between the last twoantennas. Additional relative phase shifts areapplied between the first and third antennas,and between the second and fourth antennas,accounting for imperfect antenna polarization.The rank-2 codebook covers every possibleantenna grouping: a total of three antennagroupings for two layers, with a larger portioninvolved with antenna groupings of (1,2) + (3,4)(i.e., the antenna grouping with lower intergroupcorrelation) than the other two antenna group-ings of (1,3) + (2,4) and (1,4) + (2,3). For therank-3 codebook, the first layer is assigned withtwo TX antennas, whereas layers 2 and 3 areeach assigned one antenna. Every antennagrouping (in total, six groupings) for the firstlayer is covered, along with an intragroup phaseshift of 0° or 180°. It should be noted that nointergroup phase shift is considered in the rank-2 and rank-3 codebooks, since it does not con-tribute to layer separation.

Our simulation results show that compared toeigen-beamforming, the codebook for the ULdata channel experiences up to 2 dB loss for the4 × 4 spatial channel model (SCM) of [8]. Onthe other hand, it tends to perform as well as theHouseholder codebook used in DL-MIMO,unless UE antenna correlation is extremely high.Note that the Householder codebook has a simi-lar codebook size, while its alphabet is 8PSK andthe CM is not preserved. Therefore, it can beconcluded that the loss of precoding gain (com-pared to eigen-beamforming) largely comes fromthe codebook size constraint, rather than thealphabet restriction and CM preserving con-straint.

The precoder is defined by rank and precod-ing matrix, which are selected by an eNodeBbased on uplink channel measurements and con-veyed to UE through the rank indicator (RI)

A major difference

from the precoder

codebook used for

DL-MIMO is that the

UL codebook is

carefully designed to

not increase the CM

compared to

single-antenna

transmission.

A design constraint

of CM-preserving

codebook is that

each row should

have at most one

non-zero element.



and precoding matrix indicator (PMI) fields inan uplink grant via the DL control channel, thephysical downlink control channel (PDCCH).The rank and precoding matrix may be selectedto maximize the throughput, together with MCS.In general, the higher SINR the eNodeB mea-sures in link quality, the higher rank the eNodeBselects. Link quality measurements may be car-ried out through the use of so-called soundingreference signals (SRSs). SRSs are explained inmore detail in the next section.

UL CONTROL CHANNEL: TRANSMIT DIVERSITYMultiple transmit antennas are used to providediversity gain for the UL control channel. A 2-TX transmit diversity (TxD) scheme for hybridautomatic repeat request (HARQ) and schedul-ing request (format 1/1a/1b) transmits the samecontrol information from different UE antennasby using different orthogonal resources (thusproviding no multiplexing gain). This is referredto as space orthogonal-resource transmit diversi-

ty (SORTD), and is illustrated in Fig. 3. For theRelease 8 UL control channel, the orthogonalresources are available through either the cyclicshift of length-12 phase-rotated sequences ororthogonal code covering of length-4 orthogonalsequences for control information [6]. Thedemodulation reference signal (DM-RS) for theUL control channel also uses the orthogonalresources through the cyclic shifts of length-12phase-rotated sequences or orthogonal code cov-ering of length-3 orthogonal sequences. (Thedetails of DM-RS will be given later.) Thus, inprinciple, a total of 36 orthogonal resources areavailable for the UL control channel in eachsubframe. Assuming perfect orthogonality, thesignals from multiple transmit antennas arereceived separately and combined constructively.Figure 3 shows that a 2-TX TxD scheme togeth-er with maximum ratio combining providestwofold transmit diversity. In the case of fourtransmit antennas, the UE from an eNodeBpoint of view transmits using two orthogonal

Figure 2. CM-preserving codebook and antenna configuration (4 transmit antennas).

Uniform Linear Array(ULA)

2 pairs of cross-pol. antennas

1

1

111-1

2

Rank-1Codebook

Index 0 to7 1

11jj

21

11-11

21

11-j-j

21

1j1j

21

1jj1

21

1j

-1-j

21

1j-j-1

2

1

1-111

2

Index 8 to15 1

1-1j-j

21

1-1-1-1

21

1-1-j-j

21

1-j1-j

21

1-jj

-12

1

1-j-1j

21

1-j-j1

2

1

1010

2

Index 16 to23 1

10-10

21

10j0

21

10-j0

21

0101

21

010-1

21

010j

21

010-j

2

1

1100

001-j

2

Rank-2Codebook

Index 0 to7 1

1100

001j

21

1-j00

0011

21

1-j00

001-1

21

1-100

001-j

21

1-100

001j

21

1j00

0011

21

1j00

001

-1 2

1

1010

0101

21

1010

010-1

21

10-10

0101

21

10-10

010-1

21

1001

0110

21

1001

01-10

21

100-1

0110

21

100-1

01-10

2

Index 8 to15

1

1100

0001

0010

2

Rank-3Codebook

Index 0 to5

1

1000

0010

0100

2

Rank-4Codebook

Index 0 0001

1

1-100

0001

0010

21

1010

0001

0100

21

10-10

0001

0100

21

1001

0010

0100

21

100-1

0010

0100

2

1

0110

0001

1000

21

01-10

0001

1000

21

0101

0010

1000

21

010-1

0010

1000

21

0011

0100

1000

21

001-1

0100

1000

2

Index 6 to11

2 3 4

2 pairs of ULA

1 1 32 3 4 2 4



resources (cyclic shift and orthogonal code cov-ering), and it is up to the UE implementation ifand how the transmission is spread over the fourantennas. One possibility is to let the third andfourth antennas use the same orthogonalresources allocated to the first and second anten-nas, respectively, and transmit the same informa-tion. Another possibility is to only transmit ontwo of the four UE antennas.

The same TxD scheme is also applied for thechannel quality information report (format2/2a/2b).

REFERENCE SIGNALS

DM-RS AND CHANNEL ESTIMATIONThe reception of the UL data/control channelgenerally requires channel estimation, which isfacilitated by DM-RS in Release 8 [6]. DM-RSis defined in the frequency domain by the cell-specific base sequence and its time-domain cyclicshift. The ISI introduced by SC-FDMA is aresampled version of the dispersive channel by afactor of N/M when N out of M subcarriers areassigned (N ≤ M). Consequently, different cyclicshifts provide orthogonal resources, as long asthe resampled delay spread is smaller than theminimum spacing of cyclic shifts. However, theintroduction of UL-MIMO requires an increase

in the number of orthogonal resources to facili-tate the estimation of multiple spatial multiplex-ing channels. This reduces the spacing betweenthe cyclic shifts, and as a result, the orthogonali-ty of DM-RS is no longer guaranteed in highlydispersive channels.

The same precoding is applied to both theUL data channel and DM-RS to provide theprecoding gain for channel estimation. There-fore, the required number of orthogonalresources for each UE is equal to the number oflayers, but it may be smaller than the number oftransmit antennas. For demodulation purposes,the channels seen by all layers need to be esti-mated, and each channel can be estimated sepa-rately thanks to the orthogonality of DM-RS.This enables low-complexity implementation ofMIMO channel estimation. Specifically, as shownin Fig. 4a, in slightly dispersive channels such asthe Extended Pedestrian A (EPA) channel [9],after multiplying the received signal with a con-jugate of DM-RS (e.g., DM-RS of the firstlayer), it is possible to separate all the channels,since the resampled delay spread of each chan-nel is confined to a pair of two consecutive cyclicshifts. However, in highly dispersive channelssuch as the Extended Typical Urban (ETU)channel [9], per-layer channel estimation tendsto experience significant loss, since the resam-

Figure 3. Transmit diversity for UL control channel — HARQ and scheduling request (2 transmit antennas).

r0 ... r11 : length-12 phase-rotated sequence

12 subcarriers(180 kHz)

1 slot (0.5 ms)

OCC: orthogonal code coveringCS: cyclic shift

Effective channel (SNR)

Orthogonalresource mapper

(CS1, OCC1)

h1

h1*

Maximum ratiocombining

h2

w1

CSn OCCn(DM-RS)

OCCn(UCI)

h2* h2 |h1|2+|h2|2

h1

w0

w0

r0


(CS2, OCC2)


(CS1, OCC1)


(CS2, OCC2)

w0

r11

w1 w2 w3

w2w1w0

w1 w2

w2 w3

w0 ... w3 : length-4 orthogonal sequence

w0 ... w2 : length-3 orthogonal sequenceOCCn(UCI)



pled delay spread exceeds the minimum spacingof cyclic shifts, thereby causing interlayer inter-ference, as illustrated in Fig. 4b. Furthermore,interlayer interference becomes more detrimen-tal in case of narrower user bandwidth (i.e.,when N/M becomes smaller), since the interpola-tion filter (i.e., sinc filter) involved in the resam-pling operation decays more slowly, as illustratedin Figs. 4c and 4d.

In order to improve inter-DMRS orthogonal-ity, orthogonal code covering the two slots in asubframe is additionally used. By spreading DM-RS across two slots within a subframe with twoorthogonal codes (+1, +1) or (+1, –1), DM-RSorthogonality between layers can be maintainedeven in highly dispersive channels. Since theorthogonal code covering orthogonality assumesthat the channel remains constant between twoslots, it is inherently vulnerable to high-mobilityscenarios. However, considering that UL-MIMO(especially with rank-4 transmissions) is mostlytargeted for the low-mobility scenario, orthogo-nal code covering remains attractive as an alter-native DM-RS multiplexing scheme. Another

advantage of orthogonal code covering is that itenables MU-MIMO pairing of UE units withdifferent user bandwidths. (Note that cyclic-shift-ed DM-RS cannot guarantee the orthogonalityin case of different user bandwidth.)

SRS AND PRECODER SELECTIONSince DM-RS is precoded together with the ULdata channel, it is impossible to select the pre-coder (i.e., the rank and precoding matrix) basedon DM-RS, since only a subset of the spatialchannel dimensions is observed. Instead, non-precoded SRS transmission is the conventionalway for the eNodeB to efficiently select anappropriate precoder for the UL data channel.SRS is antenna-specific, as opposed to layer-spe-cific in the case of DM-RS; thus, the requirednumber of orthogonal resources is the same asthe number of transmit antennas.

There exists much similarity between DM-RSand SRS. SRS is derived from the same cell-spe-cific base sequence as DM-RS, and SRS orthog-onality is provided through different cyclic shifts[6]. The difference from DM-RS is that two

Figure 4. Per-layer channel estimation (4 layers, system bandwidth: 20 MHz): a) EPA channel, 5 MHz; b) ETU channel, 5 MHz; c) EPAchannel, 1.4 MHz; d) ETU channel, 1.4 MHz.

Subcarrier index

(a)

EPA, 25RBs

500

2

0

Am

plit

ude

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

100 150 200 250

First layerSecond layerThird layerFourth layer

Subcarrier index

(b)

ETU, 25RBs

500

1

0

Am

plit

ude

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

100 150 200 250

Subcarrier index

(c)

EPA, 6RBs

100

2.5

0

Am

plit

ude

2

1.5

1

0.5

20 30 40 50 60 70Subcarrier index

(d)

ETU, 6RBs

10 0

2

0

Am

plit

ude

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

20 30 40 50 60 70





additional orthogonal resources are available inan FDMA fashion by using every other subcarri-er [6].

IMPACT ON DL CONTROL SIGNALINGUL scheduling grant is conveyed in the DL con-trol channel from eNodeB to UE [6], and itenables dynamic resource allocation for both thedata and control channels on a per-subframebasis. The introduction of UL-MIMO requiresadditional fields of DL control channel payloadto inform UE of the relevant resource alloca-tion, as follows:• PMI — RI: The 2-TX and 4-TX codebooks

consist of 7 and 53 precoding matrices,respectively.

• MCS — redundancy version (RV): MCSand RV are jointly signaled, as in theRelease 8 UL data channel [10]. Up to twocodewords can be transmitted in one sub-frame with per-codeword MCS and HARQcontrol. Thus, an additional MCS-RV forthe second codeword is necessary.

• New data indicator (NDI): Per-codewordHARQ control is assumed so that an addi-tional NDI is required for the second code-wordIt is worth mentioning that DM-RS resource

allocation (i.e., the cyclic shift and orthogonalcode covering) should be signaled without addi-tional fields of the DL control channel payloadfor both SU-MIMO and MU-MIMO. It is possi-ble to derive the cyclic shifts for multiple layersfrom one cyclic shift field of DL control channeland semi-statically signaled cyclic shift offsets.Orthogonal code covering is also possible to bederived from a single dynamically-signaled cyclicshift, if orthogonal code covering index is direct-ly mapped from the cyclic shift field.

It can be concluded that additional DL con-trol channel signaling required by UL-MIMOwith the above approach amounts to approxi-mately 9 and 12 bits for 2-TX and 4-TX cases,respectively, which justifies the need for new DLcontrol formats for spatial multiplexing modes.Since the number of blind decodings is deter-mined by the number of DL control formats,adding additional DL control formats results inan increase in the computational complexity ofDL control channel reception of UE. Thus, it isdesirable to limit the number of new DL controlformats (e.g., by sharing the payload size) and anew DL control format (format 4) is assigned inorder to support UL-MIMO.

HARQ acknowledgment (ACK)/negativeACK (NACK) is signaled in the physical HARQindicator channel (PHICH) from eNodeB to UE[6]. In order to support per-codeword HARQ,two orthogonal resources need to be allocated toeach user.

RECEIVERS FOR UL DATA CHANNEL

MMSE EQUALIZERA widely used simple solution is a linear MMSEequalizer, which produces an MMSE estimatefor each TD symbol. Due to the linearity andunitary properties of the DFT and IDFT, anequivalent equalizer is to obtain an FD MMSE

estimate for each FD symbol and then take theIDFT to get the TD MMSE symbol estimates.The MMSE estimate for each FD symbol isobtained by linearly combining FD received sig-nals collected from multiple receive antennasusing MMSE combining weights,

W(k) = (H(k)HH(k) + RU(k))–1H(k), (1)

where k is the subcarrier index, H(k) is a vectorrepresenting the frequency response for thelayer signal of interest (one element per receiveantenna), and RU(k) is the impairment covari-ance matrix capturing the spatial correlationbetween the impairment components. This sim-ple MMSE equalizer achieves performance veryclose to the theoretical capacity in a SIMO chan-nel. However, in a MIMO channel, linear MMSEequalizer performance is far from the capacitydue to the presence of spatial-multiplexing inter-ference.

SIC RECEIVERA SIC receiver as considered in [11] is a goodcandidate for MIMO reception. It was shownthat SIC receivers with per-layer MCS control infact achieve MIMO capacity in non-dispersivechannels [11]. The operation of SIC is simple.The receiver first detects the signal sent by thefirst layer, and after successful detection it can-cels the interference contributed by the detectedsignal before it detects the other layers. The pro-cess is repeated until all the layers are detected.Typically, each layer is cancelled only when thecyclic redundancy check (CRC) bits check. Withperfect per-layer MCS control, the transmissionrate for each layer is chosen to satisfy sufficientlylow block error rate (BLER); thus, SIC can mostlikely cancel interference contributed by previ-ously detected layers.

A SIC receiver consists of a number of linearMMSE equalizers as well. The role of theseMMSE equalizers is to suppress the residualinterference. The cancellation of other layers isreflected in the impairment correlation, and thusin the MMSE combining weights in Eq. 1. Thecancellation of the detected layers allows theMMSE equalizer to use its available degree offreedom to better suppress other dominantinterfering signals, resulting in a reduced inter-ference level and a higher SINR.

However, there are some limitations with SICreceivers in practice. Firstly, SIC does notaddress the ISI problem. Though frequency-domain MMSE equalization is effective in sup-pressing ISI, there is an opportunity to achievesmall performance improvement by better can-celling ISI [12]. Moreover, Turbo encoding andCRC checking is not necessarily performed on aper-layer basis as per-codeword (as opposed toper-layer) MCS control and CRC checking isused for UL-MIMO. For example, for the rank-4 transmission, the first and second layers aremapped to the same codeword and share thesame CRC bits. Hence, these two layers need tobe decoded jointly, rather than individually.Thus, prior to the decoding stage, the secondlayer receiver cannot cancel the interferencecontributed by the first layer, resulting in highintra-codeword interference.


With perfect

per-layer MCS

control, the

transmission rate for

each layer is chosen

to satisfy sufficiently

low block error rate

(BLER); thus, SIC can

most likely cancel

interference

contributed by

previously detected

layers.



TURBO EQUALIZATION RECEIVER

Ultimately, an optimal receiver is to form jointhypotheses on all the overlapping symbols. How-ever, this is practically impossible due to the useof higher order modulation and a large numberof overlapping symbols due to SC-FDMA andMIMO. A practically attractive solution, interms of performance and complexity trade-offs,is soft-cancellation-based MMSE turbo equaliza-tion. Such a receiver was first proposed forGSM/EDGE [13] and recently considered forLTE-UL [14].

The receiver architecture of the turbo equal-ization receiver is similar to SIC in the sensethat each codeword is successively detected,regenerated, and then cancelled from thereceived signal. However, unlike SIC, the cancel-lation of a detected layer takes place no matterwhether or not the CRC checks. Furthermore,each layer may be detected multiple times in around-robin manner. There are three basic ele-ments in a soft-cancellation-based MMSE turboequalization receiver.

Turbo Operation — This refers to the informa-tion exchange between several detection blocksin the receiver. For each codeword, informationexchange occurs between the FD equalizer/demodulator and the turbo decoder. In essence,both the equalizer/demodulator and turbodecoder are estimating the probabilistic informa-tion (usually log-likelihood ratio [LLR]) abouteach encoded bit. The sign of the estimatedprobabilistic information indicates whether a bitis likely to be 0 or 1, while the magnitude indi-cates the likelihood or reliability level. First, theequalizer and demodulator utilize the modula-tion constellation structure to derive an estimateof the probabilistic information from thereceived signal. Based on this probabilistic infor-mation, the decoder utilizes the forward errorcorrection (FEC) code structure to furtherimprove the probabilistic estimate about theencoded bits. This involves a simple modificationto the turbo decoder to generate LLR’s for theparity bits, in addition to those for the systematicbits, giving rise to a so-called soft-input soft-out-put (SISO) turbo decoder [13]. Such probabilis-tic information from the decoder is providedback to the demodulator, as a form of a prioriinformation, for the demodulator to further

improve the probabilistic estimate. Estimatesabout received bits become more accurate asmore information exchanges take place. Further-more, not only does the equalizer benefit fromhaving the probabilistic information from thedecoder that is estimating the same set of encod-ed bits, it also benefits from accessing probabilis-tic information from the decoders that areestimating other sets of encoded bits (e.g.,encoded bits mapped to the other layers.) Thisprobabilistic information can be used to formsoft symbols for the cancellation of both inter-layer interference and ISI. (see below)

Soft Symbol Subtraction — The way theequalizer utilizes the probabilistic informationfrom the decoder is to use this information toget MMSE symbol estimates of an interferingsymbol. From [15], such an estimate is given inthe form of conditional mean. Assuming bitsmapped to a TD constellation symbol are mutu-ally independent, the symbol probability is sim-ply the product of bit probabilities. Averagingthe product of symbol probability and the corre-sponding modulation value over the entire mod-ulation constellation yields the MMSE symbolestimate. After obtaining the MMSE estimatesof TD symbols, MMSE estimates of FD symbolsare obtained from taking DFT, as illustrated inthe layer signal regenerator in Fig. 5. Theseinterfering symbols, after channel filtering, aresubtracted from FD received signals.

Soft Decision Feedback MMSE Equalization— In fact, not only interfering symbols fromother layers are cancelled, soft symbols from thelayer of interest are also removed. This featureis referred to as soft decision-feedback equalizer(DFE) [13], as in this process, the ISI compo-nent due to time dispersion is removed. Thisproblem can be formulated as an MMSE-DFEproblem. An MMSE-DFE structure is showninside the codeword detector in Fig. 5. Theobjective of MMSE-DFE is to produce anMMSE symbol estimate to maximize post-equal-ization SINR, based on the knowledge of chan-nel response and a priori information of thesymbols that are under estimation. ISI is mitigat-ed through feedback filtering (FBF) using theregenerated layer and cancellation of the FBFoutput from the feed-forward filter (FFF) out-put. The FFF and FBF are jointly designed

Figure 5. Two key components in a turbo equalization receiver: a codeword detector that includes an MMSE-DFE receiver and a layerregenerator.

DFTSoft symbolmodulator

Demodulator

a prioriinformation

FDreceivedsignal

MMSE-DFE

Codeword detector

IDFT−

Feedforwardfilter LLR

Layer regeneratorRegenerated

layer

Scrambler

Regeneratedlayer Feedback

filter Descrambler SISOdecoder

LLR



based on the MMSE criterion, and the exactexpressions were derived in [13].

PERFORMANCE EVALUATIONTo illustrate the performance of UL-MIMO inan LTE system, we simulated link performancein the Extended Vehicular A (EVA) channel [9].We consider a scenario where 25 resource blocks(5 MHz) are allocated to a user, and full-ranktransmission is assumed. For each SNR, we eval-uate the average throughput performance over500 subframes. We assume that there is noantenna correlation.

Performance of 2 × 2 MIMO is shown in Fig.6a for the MMSE, SIC, and turbo equalizationreceivers. Here the set of MCSs defined for theRelease 8 UL data channel [10] is used for eachlayer and the highest MCS in this case has a trans-mission rate of around 18.75 Mb/s per layer. As in[14], an MCS is chosen based on long-term aver-age throughput at each SNR point, and thus eachcurve in Fig. 6 corresponds to the envelope of thethroughput curves for all the 29 MCSs defined in[10]. We see that turbo equalization offers signifi-cant performance gain over both MMSE and SIC,in detail, 4–5 dB gain over MMSE and 2–3 dBgain over SIC at a medium to high SNR. Further-more, all the receivers suffer similar degradationwhen practical channel estimation is considered.With antenna correlation, similar performancegain of the turbo equalization receiver over thelinear MMSE equalizer was observed in [14].

Performance of 4 × 4 MIMO is shown in Fig.6b. The gain of the turbo equalization receiverover SIC increases, compared to the 2 × 2 case.The 4 × 4 case has two layers mapped to onecodeword, and thus the larger gain here is due tobetter cancellation of intra-codeword interference.

CONCLUSIONAn overview of UL-MIMO for LTE-Advanced ispresented in this article. The evolution of theRelease 8 UL transmission scheme toward

MIMO has been agreed on in 3GPP, as detailedhere, which includes precoded spatial multiplex-ing, transmit diversity, and modifications on ref-erence signals and DL control signaling tosupport UL-MIMO. The performance of a num-ber of receivers for UL data channel is evaluatedusing link simulations, which show that signifi-cant performance gain is achievable when anadvanced receiver such as the turbo equalizationreceiver is used at the eNodeB.

ACKNOWLEDGMENTThe authors would like to thank the anonymousreviewers for their helpful comments and sugges-tions.

REFERENCES[1] E. Dahlman et al., 3G Evolution: HSPA and LTE for

Mobile Broadband, 2nd ed., Academic Press, 2008.[2] 3GPP TR 36.913, “Requirements for Further Advance-

ments for E-UTRA,” v. 8.0.1, Mar. 2009.[3] U. Sorger, I. De Broeck, and M. Schnell, “Interleaved

FDMA — A New Spread-Spectrum Multiple-AccessScheme,” Proc. IEEE ICC ‘98, Atlanta, GA, June 1998,pp. 1013–17.

[4] H. G. Myung, J. Lim, and D. J. Goodman, “Single CarrierFDMA for Uplink Wireless Transmission,” IEEE Vehic.Tech. Mag., vol. 1, no. 3, Sept. 2006, pp. 30–38.

[5] M. Noune and A. Nix, “Frequency-Domain Precoding forSingle Carrier Frequency-Division Multiple Access,” IEEECommun. Mag., vol. 48, no. 5, June 2009, pp. 68–74.

[6] 3GPP TS 36.211, “Evolved Universal Terrestrial RadioAccess (E-UTRA); Physical Channels and Modulation,” v.8.9.0, Dec. 2009.

[7] D. J. Love and R. W. Heath, Jr., “Limited Feedback Uni-tary Precoding for Spatial Multiplexing Systems,” IEEETrans. Info. Theory, vol. 51, no. 8, Aug. 2005, pp.2967–76.

[8] 3GPP TR 25.814, “Physical Layer Aspects for EvolvedUTRA,” v. 7.1.0, Sept. 2006.

[9] 3GPP TS 36.104, “Evolved Universal Terrestrial RadioAccess (E-UTRA); Base Station (BS) Radio Transmissionand Reception,” v. 8.10.0, June 2010.

[10] 3GPP TS 36.213, “Evolved Universal Terrestrial RadioAccess (E-UTRA); Physical Layer Procedure,” v. 8.8.0,Sept. 2009.

[11] M. K. Varanasi and T. Guess, “Optimum Decision Feed-back Multiuser Equalization with Successive DecodingAchieves the Total Capacity of the Gaussian Multiple-Access Channel,” Proc. Asilomar Conf. Signals, Sys.,Comp., Monterey, CA, Nov. 1997, pp. 1405–9.

Figure 6. Performance of UL-MIMO (EVA channel, 5 MHz): a) 2 transmit antennas, 2 receive antennas; b) 4 transmit antennas and 4receive antennas.

Es/No (dB)

(a)

100

5

0

Ave

rage

thr

ough

put

(Mb/

s)

10

15

20

25

30

35

40

20 30 40

MMSE (id)SIC (id)Turbo (id)MMSE (est)SIC (est)Turbo (est)

Es/No (dB)

(b)

10 0

10

0

Ave

rage

thr

ough

put

(Mb/

s)

20

30

40

50

60

70

80

20 30 40

MMSE (id)SIC (id)Turbo (id)MMSE (est)SIC (est)Turbo (est)



[12] G. Berardinelli et al., “Improving SC-FDMA Perfor-mance by Turbo Equalization in UTRA LTE Uplink,”Proc. IEEE VTC, Singapore, May 2008, pp. 2557–61.

[13] C. Laot, R. Le Bidan, and D. Leroux, “Turbo Equaliza-tion: Adaptive Equalization and Channel DecodingJointly Optimized,” IEEE JSAC, vol. 19, no. 9, Sept.2001, pp. 965–74.

[14] G. Berardinelli et al., “Turbo Receiver for Single UserMIMO LTE-A Uplink,” Proc. IEEE VTC, Barcelona, Spain,Apr. 2009, pp. 26–29.

[15] H. Stark and J. W. Woods, Probability and RandomProcesses with Application to Signal Processing, 3rded., Prentice Hall, 2002.

BIOGRAPHIESCHESTER SUNGCHUNG PARK received a Ph.D. degree in electri-cal engineering from the Korea Advanced Institute of Sci-ence and Technology (KAIST), Daejeon, in 2006. Since 2007he has been with Ericsson Research, USA, working on digi-tal baseband and front-end algorithms for LTE. His researchinterests include algorithm and hardware architecturedesign for MIMO-OFDM, error correction codes, and soft-ware-defined radios. From 2006 to 2007 he was with Sam-sung Electronics Inc., Giheung, Korea.

Y.-P. ERIC WANG received a Ph.D. degree from the Universityof Michigan, Ann Arbor, in 1995. Since then he has been amember of Ericsson Research, USA. His research interestsinclude coding, modulation, synchronization, multipleinputs-multiple outputs, channel equalization, and interfer-ence cancellation and suppression. He was an AssociateEditor for IEEE Transactions on Vehicular Technology from2003 to 2007.

GEORGE JÖNGREN received M.Sc. and Ph.D. degrees in electri-cal engineering from the Royal Institute of Technology(KTH), Stockholm, Sweden, in 1998 and 2003, respectively.He was a postdoctoral researcher at KTH for one yearbefore joining Qualcomm CDMA Technologies GmbH,Nuremberg, Germany. Since September 2005 he has beenwith Ericsson Research, Sweden, working on smart antennaresearch and implementation, as well as 3GPP standardiza-tion. He was elected Teacher of the Year in Electrical Engi-neering at KTH in 1997.

DAVID HAMMARWALL received a Ph.D. degree in telecommu-nications in 2007 from KTH, where he was awarded an“Excellent Graduate Student Position” from the President’soffice. In 2007 he joined Ericsson Research in Stockholm,where he is active in research and standardization of LTE.His research interests include wireless communications,receiver algorithms, resource optimization, beamforming,and scheduling.


INTRODUCTION

With the emergence of numerous smart mobiledevices such as handheld smart phones and net-books, the data usage on mobile networks isgrowing exponentially. As shown in Fig. 1, it isexpected that the mobile data traffic generatedwill grow more than 50 times compared to theend of 2009 by 2015 and even 500 times by 2020corresponding to about 57 Gbytes per month peraverage subscriber. Sixty Gbytes per month isgenerated by an asymmetric digital subscriberline (ADSL) subscriber at a constant bit rate of2 Mb/s based on the average online time of 136min in 2009. Thus, this expectation means thatmobile users in 2020 will be waiting for a similaruser experience as that of current wireline users.It is anticipated that all Internet applicationsused via fixed Internet access should also be sup-ported on the mobile access platform. Actually,some applications (e.g., social networks) mayeven be accessed more frequently via mobileaccess than fixed access. At the same time, it ispredicted that by 2020 more than 50 billion

devices (compared to 4.9 billion devices at theend of 2009) will be connected with mobilebroadband connections, and every one of us willbe surrounded by an average of 10 devices. Suchexplosive growth of wireless devices also drives anew shift of applications from current man-to-machine communications to future machine-to-machine (M2M) communications. This newparadigm of M2M applications also contributesto the increasing demand for mobile data appli-cations in the future.

The rapidly increasing mobile traffic puts ahuge demand on network capacity and quality ofservice (QoS) in mobile networks. The wirelessnetwork is evolving from the current third-gener-ation (3G) technologies to various fourth-gener-ation (4G) systems such as International MobileTelecommunications (IMT)-Advanced systems.Although IMT-Advanced technologies (e.g.,Long Term Evolution Advanced [LTE-A] in theThird Generation Partnership Project [3GPP])have achieved remarkable capacity increase incomparison with current 3G systems, they stillcannot satisfy the explosive increase of mobiledata traffic projected for 2020. For instance,assuming an average bit rate of 1 Mb/s duringthe busy hour (BH) per user, this implies ademand of average area capacity of 25 Gb/s/km2

in dense urban regions with typical user densityof 25,000 users/km2. To achieve this capacity,about 230 MHz bandwidth is required even ifthe cell average throughput of 3.7 b/s/Hz/cell isachieved as in LTE-A with 200 m intersite dis-tance (ISD) [1]. It is clear that we need a morefundamental breakthrough to increase the wire-less capacity in urban areas beyond LTE-A tech-nology. In this article we discuss various technicalchallenges involved as well as potential advancedtechnologies to achieve the aggressive target of25 Gb/s/km2 area throughput: interference miti-gation techniques, cooperative multiple-inputmultiple-output (MIMO) techniques, and cross-layer self-organizing networks (SONs). In thefollowing we discuss the advantages and techni-cal issues associated with urban small cell deploy-ment and the associated key performance target.Specifically, the urban small cell environmenthas the potential to provide more spatial degrees


ABSTRACT

In this article we present a survey on thetechnical challenges of future radio access net-works beyond LTE-Advanced, which could offervery high average area throughput to support ahuge demand for data traffic and high user den-sity with energy-efficient operation. We highlightvarious potential enabling technologies andarchitectures to support the aggressive goal ofaverage area throughput 25 Gb/s/km2 in beyondIMT-Advanced systems. Specifically, we discussthe challenges and solutions from the control-ling/processing perspective, the radio resourcemanagement perspective, and the physical layerperspective for dense urban cell deployment.Using various advanced technologies such asinterference mitigation techniques, MIMO, andcooperative communications as well as cross-layer self-organizing networks, we show thatfuture urban wireless networks could potentiallyoffer high-quality mobile services and offer anexperience similar to the wired Internet.


Sheng Liu and Jianjun Wu, Huawei Technologies

Chung Ha Koh and Vincent K. N. Lau, Hong Kong University of Science and Technology

A 25 Gb/s(/km2) Urban Wireless NetworkBeyond IMT-Advanced

KOH LAYOUT 1/19/11 3:35 PM Page 122

of freedom in the network, which facilitates effi-cient resource reuse, user plane/control planeseparation, and dynamic spatial path selections.We propose two radio access network (RAN)architectures, the cloud RAN and self-organiza-tion RAN, for reducing operation and controlcosts. We also elaborate on various advancedphysical layer techniques such as interferencemitigation techniques and cooperative MIMOcommunications, which contribute to fulfillingthe high demand in future wireless networks.

URBAN DENSE SMALL CELL DEPLOYMENTThere are various advantages associated withsmall cell urban deployment from the networkcapacity and energy efficiency perspectives. Forinstance, small cell deployment allows more effi-cient spatial reuse, which contributes toincreased network capacity. In addition, thedeployment of small cells also has benefits interms of energy efficiency. Energy-efficientdesign for green radio has become a trend forboth handsets and network infrastructure tolower total cost of ownership (TCO) for mobileoperators and CO2 emissions [2]. One exampleis the Green Radio project in Mobile VCE [3],which plans to achieve the goal of 100-foldreduction in power consumption over currentwireless communication networks. Moving theaccess network closer to the user is a keyapproach to reduce the transmit power required.

Although urban small cell deployment hasthe potential to achieve higher system capacity,there are various crucial technical challengesthat must be overcome.

Backhaul and Installation Cost — When thecell size shrinks, the number of radio accesspoints increases tremendously, which leads tohuge increases in backhaul cost and real estatecost of installing the access points. Therefore,cost-efficient backhauling schemes and easy-to-install access points must be considered.

Interference Management — For dense smallcell deployment, intercell interference willbecome more severe than in conventional macro-cellular systems since the cell edge areas and thenumber of interference sources might becomelarger. In macrocell networks MIMO-basedintercell interference mitigation techniques, suchas coordinated multipoint transmission (CoMP),have been proven to be very effective for han-dling interference [4]. However, such schemesmay not be practically feasible in dense smallcell networks due to the increased number ofinterference sources and radio access points.

Mobility Management — A small cell networkimplies frequent handover, which gives rise toheavy signal loading in mobility managementand an increase in the probability of droppedcalls.

Resource Scheduling — The radio propaga-tion environment in a small cell network is verydifferent from that in a macrocell network. Forexample, in a macrocell network the base stationlocations are usually carefully planned, and usersare partitioned into cell center and cell edge

users. However, in a dense small cell networkthe locations of the access points may not be ascarefully planned as in the macrocell scenario,and there is no distinct user partitioning sincethe cell radius is very small, and therefore thedifferences in signal strengths among users maynot be very large. Additionally, in a traditionalmacrocell network there are usually two to fouradjacent cells that dominate the intercell inter-ference, and the impact of remote cells can beignored due to signal attenuation over longpropagation distances. However, in small cellnetworks, interference may be coming not onlyfrom the first tier neighboring cells but also fromthe other cells. As a result, radio resources(power, frequency, time, and space) schedulingand optimization become much more complicat-ed, and conventional centralized solutions maynot be viable in such complex networks.

High Operational Expenditures — The num-ber of nodes in dense small cell networks is sig-nificantly larger than in conventional macrocellnetworks. As a result, the cost of site mainte-nance will be very high if each node does notsupport self-organizing operations such as self-configuration, self-optimization, and self-healing.

Figure 2 illustrates a hierarchical architectureof future dense small cell network, consisting ofdense small cells and macrocells. Specifically,network equipment within the service range of amacrocell base station includes a large numberof small cell access points (SAPs), which arecontrolled by the AP managers. The SAPs mightform a wireless backhaul by connecting withneighboring SAPs via the wireless medium andrelaying traffic for each other. On the otherhand, the SAPs might have centralized controlsignaling according to the control managementpolicy of the capacity enhancement technologies.While the small cell concept has been around fora long time for 2G, 3G, or LTE-A systems, thenotion of small cells in the current systems isquite different from the dense small cell networkwe discussed above. For instance, the notion ofsmall cells in current systems is expected to playa complementary role rather than a major role

Figure 1. Growth of transferred data in Western Europe.


20202019

Average subscriber transferred data. (Gbytes/month):56.9

2018201720162015201420132012201120102009

39.927.718.812.58.15.163.222.001.090.520.220.09

0.09

Cellular modem

10.00

Tran

sfer

red

data

per

ave

rage

sub

scri

ber

(bas

is: a

ll su

bscr

iber

s)(G

byte

s/m

onth

)

0.00

20.00

30.00

40.00

50.00

60.00

17.011.07.04.42.71.610.960.580.320.170.06Handset data +VoIP



in providing capacity and coverage. Hence, thephysical layer signal processing, resource man-agement algorithms, and architecture of currentsystems are not optimized for dense small celldeployment, and these existing technologies can-not deal with the above challenges. In this articlewe elaborate on various advanced enabling tech-nologies to support the vision of dense small cellnetworks.

TARGET PERFORMANCETable 1 summarizes the major target perfor-mances of dense small cell networks. We aim tosupport an average user data rate of up to 1Mb/s, which is similar to the current ADSL-likeuser experience. For a typical user density of25,000 user/km2, this implies an average areacapacity of 25 Gb/s/km2. In addition, we considervarious per-cell performance targets.1 For exam-ple, the target average and peak downlink spec-trum efficiency are about 5.5 b/s/Hz and 45b/s/Hz, respectively. These per-cell targets repre-sent about 50 percent improvement over theLTE-A systems in micro- or picocellular envi-ronments. While these target performances arerather challenging, they could potentially beachievable via various advanced enabling tech-nologies on network architecture, cross-layer

self-organizing networks, as well as advancedinterference mitigation techniques in the physi-cal layer.

RADIO ACCESSNETWORK ARCHITECTURE

The future mobile network is a heterogeneousnetwork that supports macros, picos, femtos,relays, and distributed antenna system (DAS)in the same spectrum. Such a network enableshuge system capacity by allowing vertical cov-erage and optimal usage of local and wide areacells. For indoor applications, femto is an effi-cient scheme to enhance indoor coverage andthroughput. For urban and dense small cellenvironments, it is envisioned that dense smallcells will play a major role in supporting highsystem capacity. The dense small cells shouldbe supported by the control management andmaintenance operations in the RAN architec-ture. A key factor is the availability of abun-dant physical f iber, wavelength-divisionmultiplexed passive optical network (WDM-PON), or ultra-wide band microwave, based onwhich the following two RAN architectures areconsidered.

Centralized processing: For smaller-scalenetworks or when low-cost backhaul transmis-sion resources are widely available, the radiofrequency (RF) processing can be distributed atremote radio units (RRUs) such as SAPs, whilethe baseband and radio resource management(RRM) processing can be centralized at thenetwork controller. Based on this centralizedarchitecture, the system capacity can be maxi-mized with network MIMO signal processingand global RRM. Performance-wise, the cen-tralized RAN architecture is one of the mostefficient ways to overcome the interference andresource management issues in small cell

Figure 2. Small cell network architecture.

AP manager AP manager

User density: 25000 user/km2

Macro-BTS

Smallcell APs SAPs

Network manager

Table 1. System performance targets.

Metric Potential target

Average spectrum efficiency (b/s/Hz) DL: 5.5 b/s/Hz, UL: 3.7 b/s/Hz

Peak spectral efficiency (b/s/Hz) DL: 45 b/s/Hz, UL: 25 b/s/Hz

Peak data rate (b/s/cell) DL: 4.5 Gb/s/cell, UL: 2.5 Gb/s/cell

Average areal capacity (b/s/km2) 25 Gb/s/km2

1 The per-cell metrics(average cell throughput,average spectrum efficien-cy, peak spectral efficien-cy, etc.) are widely usedfor characterizing the sys-tem performance ofmacrocell systems. How-ever, these metrics focusmore on the performancegain achieved by the phys-ical layer signal processingand radio resource controltechniques, and less onthe evolution of networkarchitecture.



deployment. However, the issue is scalabilitydue to the computation complexity and signal-ing overhead involved. Depending on the scaleof the network, cost reduction and load sharingmay be accomplished by cloud computing tech-nology [5].

Distributed processing: For large-scale net-works or when backhaul transmission is expen-sive, distributing some RAN computation andprocessing at the SAPs is preferred. In this casethe RAN consists of dense small cells with com-pact micro/pico base stations, which take care ofthe baseband processing and maybe part of theRRM control, whereas the network controllertakes care of registrations and maintenance ofthe network. The advantage of this architectureis reduced computational and signaling loadingin the network controller, and as a result, thisarchitecture is more scalable. However, the dis-tributive processing requirement poses greatchallenges on the robustness and effectiveness ofthe control algorithms.

In the following we illustrate two implemen-tation examples of centralized and distributiveRANs, the cloud RAN and distributed SON net-works, respectively.

CENTRALIZED CLOUD RANFigure 3a illustrates a typical implementation ofthe cloud RAN [6]. A cloud RAN is a radioaccess network installing many small RRUs andcentralized processing units (CPUs) based onthe software defined radio (SDR) multiprotocolplatform. It uses virtualized baseband processingby combining all radios and computing resourcescheduling. As Fig. 3a shows, the CPUs are con-nected to the RRUs (which correspond to SAPsin urban wireless network) via WDM-PON andcontrol the RRUs based on all the knowledge ofcentralized processing.

The digitalized I/Q radio signal has very highspeed up to several gigabits per second or high-er when MIMO RRUs are used and the band-width is more than 20 MHz. Thus, opticaltransmission is usually necessary for this archi-tecture. In the near future, it is in generalagreed that a WDM-based access network willenable next-generation optical broadbandaccess. In contrast to time-division multiplexing(TDM)-based passive optical network (PON)that offers only tens of megabits per second,WDM-PON will enable the delivery of muchhigher capacity services to subscribers sinceeach optical network unit (ONU) will be servedby a dedicated wavelength channel to communi-cate with the central office or optical line termi-nal (OLT). As a result, the deployment ofcentralized architecture for outdoor applicationswill become popular.

When the signals of distributed antennas areconnected together, joint signal processing andcooperative RRM become easier and more flex-ible. With the rapid development of multicoreprocessors, a cloud-computing-based platformwill be feasible to carry out all physical layerand medium access control (MAC) layer pro-cessing. Because the cell size is small, signalsfrom hundreds or even thousands of cells can becentralized without long-distance optical trans-mission.

DISTRIBUTED SELF-ORGANIZATION RAN

Figure 3b illustrates an example implementationof a distributed SON [5]. The SON is a solutionfor simpler operations and better maintenanceof networks, which gives the network elementsself-organizing functions to allow the system tooperate and configure with less human interven-tion. Distributed SON allows the relative lowernetwork entities to have SON functions, whileon the other hand centralized SON allows onlyupper-level network entities to operate SONfunctions. In urban wireless networks establishedby distributed SON, each SAP operating SONfunctions collects the information about environ-ment changes (e.g., the installation of a newSAP and neighboring SAP actions) and makesself-optimization decisions for mobility robust-ness, energy savings, coverage optimization, andso on. Moreover, the SAPs exchange informa-tion through the interface to help reconfigura-tion of neighboring SAPs.

It can be a key enabler to manage and oper-ate several layers for interworking and givingfaultless service with lower cost. This servicewould be useful in particular for a dense smallcell network comprising a large number of SAPs.The main SON functions include self-configura-tion, self-optimization, and self-healing. Follow-

Figure 3. RAN architectures: a) centralized cloud; b) distributed self-organized.

Small-cellcluster

Small-cellcluster

RRU RRU

RS

WDM-PONCentralizedprocessing unit

(CPU)

(a)

(b)

SON

SON SON

SON

SAPs

Interfacebetween SAPs

CentralizedOAM

Network manager

BTS manager

OAM(Operation andmanagement)

OAM


ing are detailed descriptions of SON functionsand requirements with considerations of urbansmall cells.

Self-Configuration — SAPs should be auto-matically configured to provide wireless servicewhen connecting with a core network. After theinstallation of a new SAP, the automatic config-uration process contains the connection to acore network entity, authentication, and recogni-tion of neighbor SAPs. The initial parameter set-ting for reliable initial service without interferingwith other SAPs and macro-base transceiver sta-tion (BTS) is also an important feature of self-configuration.

Self-Optimization — In order to maximize net-work performance and keep the system’s relia-bility, optimization considers urban environmentcharacteristics of scattered channel, user mobili-ty, and high density. Self-optimization encapsu-lates the procedures of the monitoring mode fordetecting variance, updating the neighbor list,reconfiguring system parameters, and exchang-ing configured information; thus, it should becarefully operated to handle a small load ofwork.

Interaction between Entities — Self-opti-mization of SAPs needs sensing and detectionprocedures of neighbor environments. Thismight lead to establishing interactions betweenSAPs and between an SAP and upper networkequipment, especially when end equipment hasits own SON function. The interface and controlsignaling are significantly necessary to supportthe SON function.

Fault Management — Failure detection andlocalization belongs to self-healing procedures. Ifa failure happens in SON procedures such asnegative effects and false setting of parameters,

SAP and network entities should contain self-disabling capability and self-healing procedures.

RADIO RESOURCE MANAGEMENT INDENSE SMALL CELL NETWORKS

Efficient radio resource management (RRM)schemes for urban wireless networks play animportant role in utilizing the limited radio spec-trum resources. General RRM involves strate-gies and algorithms for controlling parameterssuch as transmit power, data/control channelallocation, and load balancing. In this section wefocus on RRM approaches for controlling adense population environment. In our urbanmodel, typical user density is 25,000 user/km2.For a dense urban environment, it is envisionedthat an efficient distributed approach to resourceallocation will have significant influence onachieving high system capacity.

In this section we describe the advanced fea-tures of RRM adapted to the urban wirelessenvironment. It deals with a plane separationscheme using the hierarchical cell structure ofurban wireless networks. We then introduce traf-fic distribution and caching approaches for effi-cient resource usage that work with high userdensity.

USER PLANE/CONTROL PLANESEPARATED HIERARCHICAL CELL STRUCTURE

Figure 4 shows the proposed separate userplane/control plane hierarchical cell structure,where a macro-BTS and a number of SAPs sharethe same spectrum to form two-tier coverage.Signaling channels as well as traffic channels fordelay-sensitive services of small data volume(e.g., voice over IP [VoIP] and gaming) areoffered by the macro-BTS; meanwhile, otherdata traffic is transmitted by the channels estab-lished by the SAP.

As illustrated in Fig. 4, UE1 has traffic chan-nels offered by SAP1 and SAP2, and a signalingchannel with the macro-BTS. When UE1 moveswithin the coverage of the macro-BTS, only traf-fic channels set up with the SAP may bechanged, but the signaling channel with themacrocell is always connected. Since the signal-ing link is free of frequent handover, signalingload for mobility management can be alleviatedand call drop of real-time services avoided.Thanks to orthogonal frequency-division multi-ple access (OFDMA), if orthogonal time-fre-quency resource blocks are allocated to themacro-BTS andrespectively, the macro and SAPscan share the same spectrum almost withoutinterference.

Moreover, this two-tiered structure has anotheradvantage, i.e., the macro-BTS can provide wire-less backhaul for the small cells. If in-band relay-ing is employed, the SAP actually acts as a relaynode. Since the antenna height of the macro-BTSis high enough, there may be a line-of-sight (LOS)path between the macro and an SAP, which makesthe use of point-to-point microwave links possible.Together with smart traffic distribution and localcontent caching, this structure will greatly reducethe cost of backhauling.


Figure 4. Illustration of user plane/control plane separated hierarchical cellstructure.

SAP 2

SAP 3

UE2

UE1

SAP 1

Microwave or self-backhaul



SMART TRAFFIC DISTRIBUTION ANDLOCAL CONTENT CACHING

Note that mobile data traffic is differentiable. Itis a fact that a lot of traffic is low value-addeddata for Internet services that occupies mostmobile network resources but treats the networkmerely as a pipeline, while only high value-added data such as IP Multimedia System (IMS)services, VoIP, mobile gaming, and mobileTV/music require QoS guarantees and core net-work services.

In addition, most areas where massive mobiledata occurs are convenient to Internet access viaa fixed access network. Usually the backhaul linkto connect RANs and mobile core networks ismore expensive than the links offered bymetropolitan area networks because the back-haul link conveys not only user traffic but alsothe control signaling, so it must be low-latencyand high-security, and even provide special func-tions like timing and frequency reference. Addi-tionally, the backhaul link conveys video service,and web browsing accounts for a big portion ofmobile data traffic. The contents of such servicescan be prefetched or selectively stored close tothe access point so that it can be accessed with-out being redirected from the application serversthroughout the core network.

Hence, we can integrate the deep packetinspection (DPI) function in the base station, sosmart traffic distribution and local contentscaching can be implemented at the RAN side.By offloading the low value-added traffic tofixed access networks, the bandwidth for back-hauling can be effectively reduced. Only a smallpart of data traffic relying on QoS guaranteeand core network services and control signalingpass through the backhaul link to the core net-work. Moreover, contents including web objects,downloadable objects (media files, software, anddocuments), real-time media streams, and so onare fetched from local caches, which implies thatthe load of backhaul can be further alleviated. Infact, smart traffic distribution and local contentcaching also has the benefit of improving QoSand user experience due to reduced delay.

INTERFERENCE MITIGATIONTECHNIQUES IN

DENSE SMALL CELL NETWORKS

In outdoor environments, interference manage-ment and effective transmission schemes are sig-nificant to enhance the capacity, and they shouldbe adapted to the characteristics of urban chan-nels and deployment environments. The desiredpolicies and algorithms for urban outdoor wire-less networks, which are different from indoorand macrocellular environments, should considerthe following aspects.

Usage of a complex propagation environ-ment: In contrast to a macro base station whoseantenna is tower-mounted, the antenna height ofa pico base station is usually 5~10 m. In urbanregions such an antenna height implies numer-ous complicated scatters and shades in the prop-agation paths, so the coverage area of each

picocell becomes very irregular, and they overlapeach other. This results in a complex propaga-tion environment but also brings rich spatialdegrees of freedom, which implies potentialcapacity that can be exploited.

Necessity of global optimization: Due to thecomplex propagation environment as explainedabove, the system model actually becomes a par-tial interference channel for most users. Thus,interference management cannot be handled ineach cell separately, but should be consideredfrom the viewpoint of global optimization.

Interference characteristics: In a dense smallcell network, the separation between outer celland inner cell is not clear due to the small cellradius. Moreover, intercell interference is gener-ated in universe to/from not only a one-tierneighboring cell, but also two- or three-tier cells.

In this section we represent an advancedapproach of dynamic spatial selection adaptingto the urban wireless environment. We thenintroduce various cooperative transmissionapproaches that can overcome limitations ofLTE-A technologies in urban small cell deploy-ment.

DYNAMIC SPATIAL PATH/BEAM SELECTION IN ADENSE SMALL CELL NETWORK

A simple scheme is to avoid interference throughdynamic spatial path selection. As illustrated inFig. 5, UE1 will select path 1 because interfer-ence signals from SAP1 to UE2 are weak enoughthat the interference of SAP1 to UE2 can beignored. Similarly, UE2 selects path 2 under thesame principle. It is actually an altruistic algo-rithm and only feasible when there are manycandidate paths to be selected. In order to applythis algorithm, each UE just reports its worstpath (i.e., from which it hears almost nothing).

In fact, physical-layer-based interference pro-cessing, which highly depends on the accurateestimation of real-time channel parameters vary-ing every frame, usually needs interlinks amongSAPs. However, MAC layer processing onlyrelies on long-term channel average, and thusthe intersite links for cooperation are not neces-sary. Since beamforming is an efficient way toenrich spatial degrees of freedom, coordinatedbeamforming, already adopted in LTE-A [1],can be further used together with the selectionof spatial path (i.e., joint spatial path and beamselection) for interference mitigation.

COOPERATIVE TRANSMISSION INDENSE SMALL CELL NETWORKS

Small cell size brings more spatial degrees offreedom, which facilitates cooperative transmis-sion of multiple network nodes. When the desti-nation is far from the source node, cooperativetransmission has the benefit of capacity by therelaying function of some other nodes, so infor-mation flow is transferred via them instead ofover the direct air link.

One novel technique in LTE-Advanced forcooperation is CoMP, which directly improvescell-edge user throughput by avoiding or elimi-nating intercell interference; however, due tocomplexity it is not a cost-efficient approach to

In a dense small cell

network, the

separation between

outer cell and inner

cell is not clear due

to the small cell

radius. Moreover,

intercell interference

is generated in

universe to/from not

only a one-tier

neighboring cell,

but also two- or

three-tier cells.



combat interference in a dense small cell net-work. Cooperative techniques have the followingadvanced approaches for enhancing capacity andreducing cost overhead in urban wireless net-works.

Cooperative Relaying — Although relayingalso consumes radio resources, it can sti l limprove system capacity due to interferencelocalization. Based on cooperation among mul-tiple relay nodes or relay nodes and the sourcenode, cooperative relaying can further improvesystem throughput and reliabil ity due toexploiting spatial degrees of freedom. Further-more, relay also helps reduce terminal powerconsumption, which is significant to greenradio.

Mobile Terminal Cooperation — At the sametime, the number of subscriptions is increasingdramatically with the rapid development ofmobile computing and the advent of M2M ser-vice. As the mobile terminal becomes more pow-erful, sophisticated cooperative communicationcan be carried out by the terminals. The mobileterminal either acts as a relay node just like arelay station or establishes a direct air link withother terminals to enable device-to-device (D2D)transmission. Thus, we can achieve performanceimprovement in terms of throughput and relia-bility without increased infrastructure cost.Despite the fact that the capacity gain of mobilecooperation diminishes fast with increased dis-tance, mobile cooperation still plays a big role inenabling M2M services in future mobile net-works since sensors must have a very low trans-mit power limit, but handsets or cellular modemsare distributed everywhere and close to thesesensors.

Other Approaches — MAC layer scheduling,including distributed frequency reuse (DFR) anddistributed power control, is also essential forcooperative radio resource management for thedistributed architecture. In a small cell networkthere is no distinct partition of cell center andcell edge users, which means that fractional fre-quency reuse (FFR), succeeding in the macrocellnetwork, may not appropriate for the densesmall cell network. Hence, radio resourcesincluding time, frequency, beam, power, and spa-tial path cannot be scheduled independently byeach cell but preferably in a cooperative way.

PHYSICAL LAYER ENHANCEMENTThere are various advanced physical layer tech-niques introduced in LTE-A systems to enhancethe physical layer data rate for macrocell scenar-ios, such as carrier aggregation and enhancedMIMO. Carrier aggregation supports both con-tiguous and non-contiguous spectra and asym-metric bandwidth for frequency-division duplex(FDD) with maximum 100 MHz aggregatedspectrum in LTE-A systems. However, the effi-ciency of bandwidth aggregation is limited by theavailable spectrum and maximum site powerconstraint. On the other hand, with a maximumof eight antennas equipped, enhanced MIMO inLTE-A systems supports up to 8-stream spatialmultiplexing in both the uplink and downlink.However, in the small cell network, the antennaheight of a pico base station is usually 5~10 m,and there is very limited space for antennamounting, which implies that the number ofantennas for each pico base station is difficult toincrease. In addition, there is a diminishingreturn on spatial multiplexing gain in high orderMIMO when the overhead of pilot preamblesand the associated signaling are taken intoaccount. As a result, we cannot merely rely onthe existing techniques in LTE-A; careful designof physical layer techniques to exploit the uniquecharacteristics of a dense small cell network isneeded to meet the aggressive area capacity infuture networks. In addition to MIMO and carri-er aggregation, the following physical layerenhancement techniques can be effective toboost the spectral efficiency in dense small cellnetworks.

High-order modulation: Exploiting the inter-ference mitigation techniques and small propa-gation loss in a small cell network, mobile usersmay be able to operate at very high signal-to-interference-plus-noise ratio (SINR); hence, wecould exploit very high-order modulation (128-quadrature amplitude modulation [QAM] and256-QAM) to achieve high peak rate and aver-age throughput.

Filter-bank-based multicarrier: In addition,filter-bank-based multicarrier (FBMC) [7] maybe a better choice of multiple access techniquein dense small cell networks instead of OFDMAor SC-FDMA in LTE. The main advantages ofFBMC include very low out-of-band frequencyleakage and cyclic prefix (CP)-free operation.Note that in the user plane/control plane sepa-rated hierarchical cell structure, the macro- andsmall cells share the same spectrum, and thedelay spreading of signals from the small cell is

Figure 5. Illustration of dynamic spatial path selection.

SAP 1

SAP 2

Path 1

Path 2

UE1

UE2



usually small, but that from the macro is rela-tively large. Clearly, the CP length should meetthe delay spreading of signals from the macro-cell, which is a redundancy for the small cell andthus causes a loss in spectral efficiency. Adopt-ing FBMC can overcome this problem at a costof somewhat increased processing complexity.

CONCLUSIONSBased on the above discussions, it is seen thatthe future radio access network should offer veryhigh average areal throughput to support a vastamount of mobile data traffic and user density,and should be energy-efficient. Although wire-less technologies seem to encounter a bottleneckin fundamental limitations, we can still move for-ward with advanced radio access network archi-tecture and novel techniques for small cellnetworks.

REFERENCES[1] T. Abe, “3GPP Self-Evaluation Methodology and Results

— Assumptions” 3GPP TSG-RAN1 tech. rep., NTT DoCo-Mo, 2009.

[2] R. Irmer and S. Chia, “Signal Processing Challenges forFuture Wireless Communications,” Proc. ICASSP, 2009,pp. 3625–28.

[3] Mobile VCE, “CORE-5 Research Area: Green Radio”;http://www.mobilevce.com/infosheets/GreenRadio.pdf.

[4] T. Nakamura, “Proposal for Candidate Radio InterfaceTechnologies for IMT-Advanced Based on LTE Release10 and Beyond (LTE-Advanced),” ITU-R WP 5D 3rdWksp. IMT-Advanced, Oct. 15, 2009.

[5] 3GPP TR 32.821, “Study of Self-Organizing Networks(SON) Related Operations, Administration and Mainte-nance (OAM) for Home Node B (HNB) version 9.0.02009-06,” 3GPP TSG-RAN, 2009.

[6] China Mobile Research Institute, “C-RAN — The RoadTowards Green RAN,” White Paper, v. 1. 0. 0, Apr. 2010.

[7] T. Ihalainen et al., “Filter Bank Based Multi-Mode Multi-ple Access Scheme for Wireless Uplink,” EUSIPCO ‘09,Aug. 2009, pp. 1354–58.

BIOGRAPHIESSHENG LIU ([email protected]) received B.S., M.S., andPh.D. degrees in electrical engineering from the Universityof Electronic Science and Technology of China (UESTC, in1992, 1995, and 1998, respectively. From 2001 to 2005 heserved as a senior architect in UTStarcom Shenzhen R&D

Center, where he worked on RAN architecture and RRMalgorithms. Since 2005 he has been with Huawei Technolo-gies Corporation, where he has led several standardresearch projects including HSPA+, UMB, and 802.16m.Currently he is the system architect of the NG-Wireless Pro-gram in Huawei. His research interests include MIMO,interference alignment, cloud RAN, heterogeneous net-work, and cooperative communications. He is the inventorof 16 U.S. and 50+ China granted patents or patent appli-cations.

JIANJUN WU ([email protected]) graduated fromSouthwest Jiaotong University in April 2001. He joinedHuawei as a wireless engineer in 2001. From 2001 till 2003he was engaged in the development of NodeB for WCDMAsystems, and during this time he also designed and devel-oped smart antenna for WCDMA and CDMA2000 systems.From 2003 to 2004 he worked on the Huawei B3G projectas a system engineer, responsible for system design. SinceJuly 2005 he has led Huawei’s WIMAX Research project,and was responsible for the standard research within theIEEE and WiMAX Forum. Since May 2007 he has beenresponsible for system and architecture evolution research.

CHUNG HA KOH ([email protected]) received her B.S., M.S.,and Ph.D. degrees from the Department of Electrical andElectronic Engineering of Yonsei Univerisy, Seoul, Korea, in2004, 2006, and 2010, respectively. Since May 2010 shehas been with Hong Kong University of Science and Tech-nology (HKUST) as a research associate. Her currentresearch interests are resource allocation, self-organizingnetworks, cross layer optimization, delay optimal systems,and femtocell protocols.

VINCENT K. N. LAU ([email protected]) received his B.Eng.(Distinction 1st Hons — ranked 2nd) from the Departmentof Electrical and Electronics Engineering, University ofHong Kong in 1992. He joined the Hong Kong Telecom(PCCW) after graduation for three years as system engi-neer, responsible for transmission systems design. Heobtained the Sir Edward Youde Memorial Fellowship,Rotoract Scholarship, and Croucher Foundation in 1995and went to the University of Cambridge for a Ph.D. inmobile communications. He completed his Ph.D. degree intwo years and joined Bell Labs — Lucent Technologies,New Jersey, in 1997 as a member of technical staff. He hasbeen working on various advanced wireless technologiessuch as IS95, 3G1X, and UMTS, as well as wideband CDMAbase station ASIC design and p 3G technologies, such asMIMO and HSDPA. He joined the Department of ECE,HKUST as an associate professor in August 2004 and waspromoted to professor in July 2010. He has also been thetechnology advisor and consultant for a number of compa-nies such as ZTE and Huawei, ASTRI, leading several R&Dprojects on B3G, WiMAX, and cognitive radio. He is thefounder and co-director of the Huawei–HKUST InnovationLab.

Although wireless

technologies seem to

encounter a

bottleneck in

fundamental limita-

tions, we can still

move forward with

advanced radio

access network

architecture and

novel techniques for

small cell networks.



SYNCHRONIZATION OVER ETHERNET AND IP INNEXT-GENERATION NETWORKS

etwork synchronization deals with the distributionof time and frequency over a network of clocks,

including clocks spread over a wide area. The goal is toalign the time and frequency scales of all the clocks byusing the communications capacity of links between nodes.

A synchronization network is the facility that imple-ments network synchronization. The basic elements of asynchronization network are nodes (autonomous and slaveclocks) and communication links interconnecting them.Since the 1970s and ’80s, most telecommunications opera-tors have set up synchronization networks to synchronizetheir switching and transmission equipment.

Over this time, network synchronization has been gain-ing increasing importance in telecommunications. As amatter of fact, the quality of many services offered by net-work operators to their customers depends on networksynchronization performance.

Since the introduction of early digital switching systems,network synchronization was needed to avoid slips in cir-cuit-switched voice and data networks. The deploymentsynchronous digital hierarchy/synchronous optical network(SDH/SONET) networks imposed new and more complexrequirements on the quality of synchronization systems. Tostudy those new problems, international standard bodiesestablished specific work groups, which culminated in the’90s with the release of a new series of InternationalTelecommunication Union — Telecommunication Stan-dardization Sector (ITU-T) Recommendations on synchro-nization of digital networks (G.810, G.811, G.812, andG.813), as well as their counterparts released by theAlliance for Telecommunications Industry Solutions(ATIS) and Telcordia (e.g., GR-1244) in the United Statesand by the European Telecommunications Standards Insti-tute (ETSI) in Europe.

More recently, it has been recognized that the importanceof network synchronization goes even further: asynchronoustransfer mode (ATM) and cellular mobile telephone net-works (Global System for Mobility [GSM], Global Packet

Radio Services [GPRS], Universal Mobile Telecommunica-tions Services [UMTS]) are two striking examples where theavailability of network synchronization references has beenproven to significantly affect the quality of service.

Traditionally, synchronization has been distributed totelecommunications network nodes using circuit-switchedlinks in time-division multiplexing (TDM). In particular,E1 and DS1 circuits have been most commonly used overEuropean and North American standard plesiochronousdigital hierarchy (PDH) systems, respectively.

The recent migration of network operators to the pack-et-switched next-generation network (NGN) once againposes newer and even more difficult problems for networksynchronization. Today, as fixed and mobile operatorsmigrate to NGN infrastructures based on IP packet switch-ing, Ethernet transport is becoming increasingly common.This trend is driven by the prospect of lower operationcosts and the convergence of fixed and mobile services.However, migrating trunk lines to IP/Ethernet transportposes significant technical challenges, especially for circuitemulation and synchronization of network elements.

Therefore, the network evolution toward IP packetswitching has led to increased interest on the part of com-munications engineers in synchronization distributionusing packet-based methods. After a few years of decliningresearch, considerable new investigation activity on net-work synchronization has restarted in both industry andacademia.

International standard bodies have also resumed signifi-cant levels of activity on this subject. Since 2004 the ITU-Thas been developing a new set of Recommendations,specifically for synchronization on packet-switched net-works, beginning with ITU-T RecommendationG.8261/Y.1361, “Timing and Synchronization Aspects inPacket Networks.” In 2002 IEEE released a new “Stan-dard for a Precision Clock Synchronization Protocol forNetworked Measurement and Control Systems” (IEEE1588, revised in 2008).

N

GUEST EDITORIAL

Stefano Bregni Ravi Subrahmanyan

LYT-GUEST EDIT-Bregni 1/20/11 12:09 PM Page 130


At this point, it is worth pointing out that the tradition-al model, in which synchronization distribution is engi-neered carefully for optimal performance and survivability,may give way to scenarios in which there is greater expec-tation of automatic, self-configured operation while stillmaintaining adequate synchronization quality. This modelis similar to that of Ethernet “plug and play,” in whichEthernet equipment may be connected into a networkwithout significant, or any, a priori configuration, yetexpected to come up and work satisfactorily. As NGN syn-chronization is transported increasingly via packet net-works, there are indications that such an expectation willarise for synchronization as well.

This consideration widens the scope of interest in syn-chronization beyond specialists, reaching the wider audi-ence of telecommunications engineers in general. Anexample is the distribution of synchronization to next-gen-eration wireless base stations, which are connected to thecore network only via packet-switched networks, but stillrequire highly accurate synchronization to meet standardquality of service expectations.

Toward this end, this special issue aims to introducereaders to some notable changes in network synchroniza-tion technology, which have recently arisen from the evolu-tion to NGN. Two articles (Ferrant et al. and Garner et al.)review the current state of standardization activity in theITU-T and IEEE. Two articles (Cosart and Shenoi) dealwith aspects of characterization and measurement of syn-chronization performance in packet networks. Finally, onearticle (Ouellette et al.) describes the use of IEEE 1588 fortime synchronization.

In further detail, the article by J.-L. Ferrant and S.Ruffini summarizes the work done by ITU-T Q13/15 overthe last six years to standardize the transport of timingover packet networks, including a summary of the relevantdocuments published by the ITU-T. It also provides insightinto the future work in ITU-T Q13/15 on the transport oftiming in packet networks.

The article by G. M. Garner and H. Ryu presents theAudio/Video Bridging (AVB) project in the IEEE 802.1working group, focused on the transport of time-sensitivetraffic over IEEE 802 bridged networks. The IEEE802.1AS is the AVB standard that will specify require-ments to allow for transport of precise timing and synchro-nization in AVB networks. This article provides a tutorialon IEEE 802.1AS and also new simulation results for tim-ing performance.

The article by L. Cosart describes techniques for perfor-mance data measurement and analysis of NGN packet net-work synchronization. It introduces some of the new metricsthat are used for performance evaluation of packet timing.

The article by K. Shenoi presents metrics and analyticalmethods suitable for specifying timing requirements inNGN packet networks. It provides a brief overview of tim-ing fundamentals, followed by an explanation of how pack-et-based methods transfer timing. Two groups of metrics,the TDEV and MTIE families, are discussed.

Finally, the article by M. Ouellette, K. Ji, S. Liu, and H.Li describes the use of IEEE 1588 and boundary clocks fortime distribution in telecommunications networks. This

technology is primarily used to serve the radio interfacesynchronization requirements of mobile systems such asWiMAX and LTE, and to avoid the dependence on GPSsystems deployed in base stations. It also presents somepreliminary field trial results, which indicate that it is pos-sible to transfer accurate phase/time across a telecom net-work for meeting the requirements of mobile systems.

This issue does not include an update on the topic ofsynchronous Ethernet. Recent activity in the standardsbodies has recognized that no single method is likely toachieve acceptable results for both time and frequency dis-tribution, and a combination of methods will be required.Research activity is now focused on using synchronousEthernet to transfer frequency, and then using a differentprotocol as an overlay to distribute time. (IEEE 1588 isone example of a protocol that may be used in combina-tion with synchronous Ethernet, although this has yet to beproven to work well enough). Ferrant et al. (IEEE Com-munications Magazine, 2008) recently provided a review ofsynchronous Ethernet. Another article will be published ina forthcoming issue of IEEE Communications Magazineproviding a further update on this topic, including aspectsof how synchronous Ethernet may be used together withother protocols for time distribution.

BIOGRAPHIESSTEFANO BREGNI [M’93, SM’99] ([email protected]) is an associate profes-sor at Politecnico di Milano, where he teaches telecommunications networksand transmission networks. In 1990 he graduated in telecommunicationsengineering at Politecnico di Milano. Beginning in 1991, he worked on SDHand network synchronization issues, with special regard to clock stabilitymeasurement, first with SIRTI S.p.A (1991–1993) and then with CEFRIEL(1994–1999). In 1999 he joined Politecnico di Milano as a tenured assistantprofessor. Since 2004 he has been a Distinguished Lecturer of the IEEE Com-munications Society, where he holds or has held the following official posi-tions: Member at Large on the Board of Governors (2010–2012), Director ofEducation (2008–2011), Chair of the Transmission, Access and Optical Sys-tems (TAOS) Technical Committee (2008–2010; Vice-Chair 2002–2003,2006–2007; Secretary 2004–2005) and Member at Large of theGLOBECOM/ICC Technical Content (GITC) Committee (2007–2010). He is orhas been Technical Program Vice-Chair of IEEE GLOBECOM 2012, SymposiaChair of GLOBECOM 2009, and Symposium Chair for eight other ICC andGLOBECOM conferences. He is Editor of the IEEE ComSoc Global Communi-cations Newsletter and Associate Editor of IEEE Communications Surveysand Tutorials. He was tutorial lecturer for four IEEE ICC and GLOBECOMconferences. He has served on ETSI and ITU-T committees on digital networksynchronization. He is an author of about 80 papers, mostly in IEEE confer-ences and journals, and of the book Synchronization of Digital Telecommu-nications Networks (Wiley, 2002). His current research interests focus mainlyon traffic modeling and optical networks.

RAVI SUBRAHMANYAN [M’89, SM’97] ([email protected]) is a seniordesign engineer with Immedia Semiconductor, Andover, Massachusetts,where he is involved in the architecture and design of high-performancevideo products focusing on next-generation multimedia delivery. He waspreviously a systems & applications engineering manager at National Semi-conductor Corp. He was also with AMCC in Andover, where he wasinvolved in the design of communications ICs and multicore PowerPC CPUs.He has participated in ITU-T and ATIS on topics related to timing and syn-chronization in communications networks, most recently working on syn-chronization over packet networks. His interests are in high speed andcustom design (currently as applied to communications integrated circuitsand video codecs), signal processing, timing recovery, and communicationsnetwork architectures. He received M.S. and Ph.D. degrees in electricalengineering from Duke University, and a B.Tech. degree (also in electricalengineering) from the Indian Institute of Technology, Bombay. He has 50publications including conference presentations and papers in refereedjournals. He also has 20 issued or pending patents. He has served on vari-ous conference committees, and has been involved with ComSoc’s TAOS TCsince 2008, where he currently serves as Vice-Chair. He is also active on theorganizing committee of the Workshop on Synchronization in Telecommu-nications Networks (WSTN).

GUEST EDITORIAL

LYT-GUEST EDIT-Bregni 1/20/11 12:09 PM Page 131

INTRODUCTION

In 2001, all time-division multiplexing (TDM)hierarchies — plesiochronous digital hierarchy(PDH), synchronous digital hierarchy (SDH),and optical transport networks (OTNs) — werefully standardized.

For the synchronization aspect, the jitter andwander of PDH interfaces were specified inG.823 and G.824; the reference clock of digitalnetworks was specified in G.811; and G.810 pro-vided the definitions related to synchronizationin TDM networks.

SDH was defined as the new synchronizationnetwork, with the definition of slave clocks inG.812 (synchronization supply unit [SSU]) andin G.813 (SDH equipment clock [SEC]). The jit-ter and wander of STM-N interfaces was speci-fied in G.825, and the SDH synchronizationlayer was specified in G.781.

The last TDM hierarchy, OTN, has beendesigned as an asynchronous network, and thereference timing signal should be carried by theSDH clients. The only requirement assigned toOTN was that STM-N signals should be trans-ported by OTN without degradation of their tim-ing characteristics; at this time almost nobodythought that the transport of synchronization bysignals other than STM-N could raise any issuesin the future.

All these documents specified the transport of anetwork reference frequency through TDM net-works; the transport of time was not specified. Theexistence of Network Time Protocol (NTP) wasconsidered good enough to transport time with an

accuracy of 0.5 s for time stamping of events; therewas no need to distribute more accurate time ref-erence over the network.

The main requirement for client synchroniza-tion at that time came from mobile networksusing frequency-division duplex (FDD) mode;the transport of a reference frequency with anaccuracy of 50 parts per billion (ppb) could bedone via SDH synchronization networks and tra-ditional 2048/1544 kb/s interfaces.

The evolution of transport and access net-works emphasizing data led to the definition ofnew synchronization requirements. In order toreduce the cost of mobile backhauling networks,the transport network between the mobileswitches and the base stations could be based onEthernet and IP links.

The first approach in the migration fromTDM to packet networks was based on timingrecovery as carried by the circuit emulationservices (CES). The use of an adaptive clockrecovery technique in the CES timing recov-ery process, however, raised some concernson the quality of the recovered timing, espe-cially in case of highly loaded packet net-works.

In order to increase the quality of the timingcarried in packet networks, some operators thenpushed in the International TelecommunicationUnion — Telecommunication StandardizationSector (ITU-T) for the transport of a frequencyreference over the physical layer of Ethernetlinks. ITU-T SG15 Question 13 developed thisconcept and defined synchronous Ethernet,based on previous work to define the transportof timing through SDH networks.

Some mobile network techniques requiredthe transport of accurate time through net-works; therefore, in parallel with the work inITU-T, the IEEE worked on a new version ofits Precise Time Protocol (PTP), IEEE 1588-2008 in order to transport time informationwith a high accuracy through wide area net-works (WANs). ITU-T decided to use this timeprotocol to transport accurate time and fre-quency over telecom networks, and Q13 hasdefined a series of recommendations addressingthe transport of frequency and time as summa-rized in Fig. 1.


ABSTRACT

This article summarizes the work done byITU-T Q13/15 over the last six years to stan-dardize the transport of timing over packet net-works. It provides a summary of the publisheddocuments in this area from ITU-T while pro-viding some of the background that went intoeach document including the specification ofsynchronous Ethernet and IEEE 1588 telecomprofiles. Finally, it provides insight into thefuture work on the transport of timing in packetnetworks in ITU-T Q13/15.

SYNCHRONIZATION OVERETHERNET AND IP NETWORKS

Jean-Loup Ferrant, Calnex Solutions

Stefano Ruffini, Ericsson

Evolution of the Standards forPacket Network Synchronization

FERRANT LAYOUT 1/19/11 3:27 PM Page 132

SYNCHRONIZATION ASPECTS INPACKET NETWORKS: ITU-T G.8261,

G.8262, G.8264Starting in 2004, the transport of TDM throughCES was the first work item addressed by ITUon the synchronization aspects in packet net-works. At that time, the focus was on CES andits ability to properly interwork with legacyTDM networks. Moreover, the CES require-ments include synchronization since many appli-cations, including base stations, need a referencefrequency via their traffic interface. The synchro-nization reference is required in order to gener-ate a radio signal at the output of the basestation with both long-term frequency accuracyand short-term stability (measured over a veryshort period, in the sub-millisecond range) with-in 50 ppb.

The work on the CES requirements was com-pleted with the approval of the first revision ofG.8261, which contained the following itemsrelated to CES:• A network model, combining a TDM net-

work and CES, and allowing the definitionof the wander budget that can be allocatedto the CES part in different network sce-nario cases.

• A guideline (included in Appendix VI,G.8261) that could provide assistance in theacquisition of performance characteristics.In particular, a testing method was definedto evaluate the performance of CES sys-tems in the presence of different types ofnetwork loads generating packet delay vari-

ation (PDV). Note: It was not possible todirectly specify PDV patterns representingthe stress in real networks due to the lackof network measurements at that time. Thisguideline is only informative and is not anormative part of G.8261.

• A model of CES equipment including pack-etizer and depacketizer interworking func-tions. These functions are applicable to thedifferent modes of CES: network-syn-chronous solutions, and differential andadaptive methods.The most common clock recovery mode used

with CES is the adaptive mode (also known asadaptive clock recovery, ACR) as normally a ref-erence network clock is not available at theedges of the packet network. Depending on theclock characteristics of the base station, the per-formance of the CES might not be good enoughto ensure the delivery of 50 ppb accuracy in anynetwork load condition; therefore, operatorsasked ITU to work on a more robust solution.

To address the problems with adaptive meth-ods, ITU-T proposed to transport a referencefrequency within the physical layer of Ethernetsignals, and it was immediately agreed on. Thisapproach was identified with the new term syn-chronous Ethernet.

ITU decided to provide full interworkingbetween SDH and synchronous Ethernet net-works, defining hybrid equipment with bothtypes of interfaces at the boundary betweenthese networks. At the network level, interoper-ability means that the G.803 synchronization ref-erence chain could be done with SDH networkentities (NEs) or synchronous Ethernet equip-


Figure 1. Current structure of ITU-T Q13/15 Recommendations for synchronization in packet networks.

G.8261 Frequency: G.826x Time/Phase:

G.827x

Definitions / terminology

Basics

Clock

Methods

Profiles

Networkrequirements

SyncE NetworkJitter-Wanderincluded in G.8261

G.8261.1(networkPDV_frequency)

G.8271

G.8262(SyncE clock)

G.8263 (Packet clock)

G.8264 (SyncE-architecture)

G.8265(Packet-architecture

for frequency)

G.8265.1 (PTPprofileFrequency)

Agreed Document status: Ongoing

G.8265.2(PTPprofileFrequency 2)

G.8271.1(networkPDV_time/phase)

G.8271.2 may be needed in future

G.8272 PRTC

G.8273 BC TC?

G.8275 (Packet-architecture

for time)

G.8275.1 (PTPprofileTime/phase)

G.8275.2(PTPprofileTime/phase 2)

G.8260 (Sync definition)

To address the

problems with

adaptive methods,

ITU-T proposed to

transport a reference

frequency within the

physical layer of

Ethernet signals and

it was immediately

agreed. This

approach was

identified with the

new term Syn-

chronous Ethernet.


ment, and the restoration of the synchronizationnetwork operates in the same way, independentof the type of equipment. The network architec-ture for synchronous Ethernet was defined inAnnex A of G.8261.

The compatibility of SDH NEs and syn-chronous Ethernet equipment in the synchro-nization reference chain led to the conclusionthat both types of equipment must be built witha clock with identical main parameters (e.g.,noise generation, noise transfer, and noise toler-ance, phase discontinuity, and transient respons-es). This enabled the approval of synchronousEthernet in a reasonable period of time, sincethere was no need to perform all the networksimulations that were done to validate SDH. Butthis also led to the decision to specify twooptions for the synchronous equipment clock inG.8262, in the same way as had been done inG.813 for SDH, although a convergent solutionwould have been preferable.

The identical behavior of SDH NEs and syn-chronous Ethernet equipment in case of syn-chronization network protection led to thefollowing consequences:• The holdover characteristics of G.8262 and

G.813 are identical.• The restoration principles of synchronous

Ethernet networks are identical to thosedefined for SDH: the protection of a chainof 20 G.8262 clocks must not generate morethan 1 μs of phase error, and the use ofsynchronization status message (SSM) ismandatory to propagate the quality level(QL) information between directly linkedequipment. In this respect a fundamentalcharacteristic of the restoration via SSM ofthe physical-layer-based synchronization isthat the timing information is propagatedequipment by equipment, and the SSM isnever passed transparently through thesenetwork elements.While it was relatively simple to reuse the

SDH clock specification (G.813) to specify thesynchronous Ethernet clock specification(G.8262), this was not the case for the transportof SSM.

In SDH, the SSM code is regularly transmit-

ted in a dedicated slot of each frame (125 μs);this periodicity was short enough to provide afast restoration mechanism and protect againstbit errors on the SSM message. For Ethernet, anew transmission of SSM over packets had to becreated.

The SSM propagation method imposes thatthe SSM generated by one NE of the chain issent to the next NE of the chain, and only to thisone; the SSM must never be transparently passedthrough an NE.

According to G.813 and G.781, the round-trippropagation of a message through a chain of 20NEs must not exceed 15 s; otherwise, a betterholdover characteristic would be required forG.8262. In addition, the restoration must not beblocked or delayed by lost packets.

In order to fulfill all these requirements, ithas been decided to define in G.8264 an Ether-net synchronization messaging channel (ESMC)able to carry two types of messages; a heartbeatmessage sent periodically, typically 1 s, carryingthe QL value, and an event message sent once incase of change of the QL. This QL value is codedin the SSM transported by the ESMC.

These messages are transported via an IEEE802.3 Organization Specific Slow Protocol(OSSP) provided by IEEE to ITU.

Another emerging aspect that has beenaddressed during the initial development of thepacket timing standards has been the definitionof a generalized packet timing distribution wherethe timing could be carried by specific protocolssuch as NTP [1] or PTP [2, IEEE 1588]. Also, inthis case the actual clock recovery technique isbased on adaptive methods.

The use of these techniques, similar as whenusing CES with adaptive clock recovery, requireda specific investigation of the performanceaspects as described in the following section. Inaddition, the use of IEEE 1588 packets requiredthe development of a specific telecom profile.

PACKET TIMING PERFORMANCEASPECTS: ITU-T G.8260,

G.8261.1, G.8263

The distribution of timing via packets has raisedsome new concerns due to the important rela-tionship between the performance that can beachieved by these methods (in particular if basedon adaptive methods) and the packet networkcharacteristics.

In particular in these cases the key parameteris the PDV introduced by the network.

Discussions on these aspects were first dis-cussed in ITU-T when the adaptive clock recov-ery method was introduced several years ago.The adaptive clock recovery method supportssynchronization of CES (e.g., to recover theclock of constant bit rate [CBR] services carriedover asynchronous transfer mode (ATM) adap-tation layer 1 (AAL1) in ATM networks [3]). Inaddition, the European TelecommunicationsStandards Institute (ETSI) published a relateddocument (“TR101685, Timing and Synchroniza-tion of Aspects of Asynchronous Transfer Mode[ATM] Networks” [4]). This report provided


Figure 2. Measured packet delay (copy of the left hand side of FigureI.2/G.8260).

PDV phase; samples: 596,919; initial phase offset: 270.020 μs

0.000 hours

10.36 hours

1.00 hours/div

1.17ms

90.0μs/div

180μs



some initial hints on the performance character-istics of clock recovery when using adaptivemethods. Further work was required in order tobetter understand the behavior of real packetnetworks and the effect on clock recovery.

After these initial studies, significant improve-ments were made in clock recovery technology,based on a better understanding of packet net-work behavior. The characterization of packetnetworks via PDV statistical distributions gavethe group some understanding of the impair-ment that could be present.

However, the most significant progress inthese studies was made only when data from realnetwork were made available. Figures 2 and 3provide an example of the data presented duringthe last Study Period in Q13/15 meetings andrecently included in the recommendations as anexample of PDV measured in a packet network[5, G.8260].

In particular, Fig. 2 provides an example ofthe packet delay variation measured over severalhours and Fig. 3 provides the corresponding his-togram.

One important aspect that was addressedduring the related studies is the introduction ofthe concept of a network clock carried by packet-based methods (e.g., using NTP or PTP pack-ets). It was recognized that from a performancepoint of view this case is analogous to the CESclock recovery using adaptive method (i.e.,ACR).

Indeed, the basic principle in the packet-based clock is to compare the time of arrival of apacket as calculated by the local clock (slaveclock) with the expected arrival time of thepacket generated by a master. The method inthis case is to use time stamps to transmit thetimes of departure and arrival.

The comparison of local time of arrival withthe content of the time stamp as generated bythe master corresponds to a measurement offsetin a one-way packet transfer and is analogous tothe phase error measurements obtained in CESadaptive clock recovery methods where theexpected arrival time is defined by the periodici-ty of the packets.

The current status of the discussions istwofold:• Characterization and modeling of the pack-

et clock (packet-based equipment clock,PEC), which is planned to be included inthe G.8263 draft

• The definition of PDV metrics and relatednetwork limits (G.8260 and G.8261.1,respectively)The work on the characterization of a packet

clock is quite complex due to the fact that multi-ple implementations are indeed possible. A logi-cal model for the packet clock was finally agreedon and included in the G.8263 draft. Some otherimportant aspects will need to be further dis-cussed before G.8263 is finalized: bandwidth,holdover, and PDV tolerance.

In particular, the PDV tolerance and themetrics used to describe it are among the mostcontroversial points in this discussion.

This work is also known by the generic termPDV metrics. In particular, this term indicates amethod to measure the significant characteristics

of the delay variation in the network and buildthe actual requirements in terms of PDV. Inparticular, the goal is to formulate packet-basedstability quantities (metrics) that will provide ameans of estimating the physical-based stabilityquantities for the packet clock output.

The initial results of the PDV metrics discus-sion are included in the ITU-T G.8260, recentlyconsented by SG15.

The related network limits are planned to beincluded in G.8261.1. An initial draft has beenprepared, but it may take some more time beforeit is completed.

THE FIRST IEEE 1588 TELECOMPROFILE: ITU-T G.8265,

G.8265.1

The use of packet-based protocols to distribute anetwork clock was initially described in G.8261(SG8261)).

As described in this Recommendation (clause7), different protocols (e.g., NTP, PTP) might beused to distribute a frequency synchronizationreference signal end to end (i.e., without supportfrom the network), leading to similar perfor-mances.

Remember, the initial focus of the new proto-cols was to distribute frequency only in order tosupport the interworking of packet and TDMnetworks, and applications that require a syn-chronization reference (CES and wireless).

The next step is to address the needs of appli-cations also requiring accurate time synchroniza-tion references (e.g., in the sub-microsecondrange).

To meet these requirements ongoing stan-dardization activities are focusing on PTP (IEEE1588), Transparent Clocks and Boundary Clocks.

Although the IEEE 1588 standard was finallyreleased in 2008 [2], it was not sufficient to beused for the deployment of the related synchro-nization solutions in Telecom.

One main aspect is that the IEEE 1588-2008specification in reality provides a list of options(e.g., mapping over Ethernet or over IP/UDP,master selection process, use of unicast or multi-

Figure 3. PDV histogram (copy of the right hand side of Figure I.2/G.8260).

180.00μs

Bins = 2048

PDV phase; samples: 596,919; Iinitial phase offset: 270.020 μs

Measurements = 596,919 1.17.00ms

100

1

10

1 k

10 k



cast mode, etc.) and each application has todefine its own specific setting. The list of thesedetails is called a “profile” by IEEE 1588. Thismeans that ITU-T had to define a telecom pro-file (or profiles) in order to use IEEE 1588.

A second major aspect is that compliancewith IEEE 1588 does not imply any performanceguarantee. The actual performance that can beachieved depends on aspects such as the net-work architecture, support or not of boundaryclock/transparent clock in the network, quality ofthe clocks in the various types of equipment, andso on.

In particular, when there is no timing support(e.g., the transport nodes do not support theboundary clock, which filters the timing itreceives before propagating it), the slave clockmust filter any PDV that is introduced by thenetwork.

The work on the IEEE 1588 telecom profileand related performance aspects was initiated in2008 with focus on the frequency synchroniza-tion needs. The initial plan was to include thespecification of the telecom profiles as a seriesof documents under the G.8264 umbrella.

To improve the structure of the Recommen-dations, it was then agreed that the work on theIEEE 1588 profile should be addressed in Rec-ommendations clearly separated from the docu-

ments dealing with synchronous Ethernet(G.8264 is mostly known as the “SyncE SSMRecommendation”).

A second decision was also made to separatebetween the general aspects (e.g., architecture)that might be applicable to different packet pro-tocols (e.g., including NTP) and the actual tele-com profile.

This approach was also found useful in orderto allow the definition of several (frequency syn-chronization) profiles in the future.

The architecture recommendation is nowcalled G.8265 [5]. The profile(s) are specified byG.8265.1 [5].

A decision was taken to address initially thecase of frequency synchronization distribution end-to-end without timing support from the network(see the general architecture in Fig. 5), consid-ered as the simplest one.

According to this scenario a Packet Masterclock distributes timing packets towards the con-nected slaves over an IEEE 1588-unaware packetnetwork (i.e., according to ITU-T terminology“without timing support”).

Despite the relatively simple environment,the definition of this Telecom Profile requiredlengthy and careful discussions.

The IEEE 1588 protocol was originallydesigned for use in a LAN, almost in a plug-and-play approach (see IEEE 1588-2002). A secondversion was then released as to add some fea-tures that should have made it also more suit-able for use in telecom (and other applications).One main addition was the possibility to use theunicast mode (as opposed to the default multi-cast mode used in the initial IEEE 1588 applica-tions).

Nevertheless, despite the IEEE 1588-2008revision, several concerns still remained forusing the PTP protocol in a telecom environ-ment; they were carefully addressed during sev-eral Q13 meetings based on a large amount ofcontributions.

One key aspect is that the common practice

Figure 5. General packet network timing Architecture (copy of Figure 1/G.8265).

Fout+δ1

Packet master clock

Packet network

1 Note: the reference may be from a PRC directly, GPS or via a synchronization network

Packet slave clock

Fout+δ3

Packet slave clock

Fout+δ2

Packet slave clock

Packet timing signals

Reference1

Fi

Figure 4. Packet-based methods (copy of Figure 3/G.8261).

Time stampmaster

Time stamp processing Packet switched

network

PRCreference

Recovered reference timing signal

Time stamp



for the Telecom operators is to have a full con-trol on the operation of the network: for instancethe master priority is statically defined and themaster selection is decided by the slaves.

On the other hand, the basic principle of theIEEE 1588 redundancy is to provide some auto-matic planning of the network at start up as well assupport automatic restorations after failures (basedon the Best Master Clock Algorithm [BMCA]). Inparticular, according to the IEEE 1588 defaultapproach, the master could decide whether or notit is the “grandmaster for the network.”

The discussion on the BMCA was probablythe most controversial one. A final agreementresulted in the scheme shown in Fig. 6, whereseveral instances of PTP slave are required as toallow the Packet Slave to be connected to sever-al Masters at the same time, supporting theappropriate redundancy requirements.

In order to simplify the discussion it was alsoagreed to focus on a full unicast approach.

The first telecom profile was finally agreedon at the June 2010 SG15 meeting and is includ-ed in G.8265.1 [5, G.8265.1]. Future frequencysynchronization profiles (e.g., supporting amixed multicast/unicast environment) might beaddressed in future versions of the G.8265.1.

TIME AND PHASE SYNCHRONIZATIONAND FUTURE WORK

Time synchronization has traditionally beenrequired to support functions such as billing andalarming. In this case the requirements are onthe order of tens or hundreds of milliseconds.

An effective and convenient distribution ofthe time synchronization reference (time of day[ToD]) calls for a hierarchical time synchroniza-tion network and a protocol that can read a serv-er clock, transmit the data to one or more clients,and adjust each client clock. For this purpose inIP networks NTP [1] has been chosen.

More stringent time synchronization require-ments are related to the support of the correctgeneration of the signals on the radio interface.In the past this was mainly required for instancein case of code-division multiple access (CDMA)technology (the time synchronization require-ment is for ± 3 μs with respect to the CDMAsystem time, which in its turn is traceable andsynchronous to UTC).

In this case the traditional approach has beento deploy a GPS receiver in every Base Station.

The deployment of time-division synchronousCDMA (TD-SCDMA) technology, as well as theforeseen wider use of time-division duplex(TDD) technology in the future, is increasingthe need to deliver accurate time or phase syn-chronization in the networks. Particularly in thecase of LTE TDD, Third Generation Partner-ship Project (3GPP) TS 36.133 specifies 3 μs or10 μs maximum time difference between basestations for small and large cells, respectively. Incase of TD-SCDMA, 3GPP TR 25.836 specifies3 μs maximum time difference between BaseStations. Additional examples are related to thesupport of functions such as multimedia broad-cast/multicast service over a single frequencynetwork (MBSFN) or coordinated multipointtransmission (also known as network multiple-input multiple-output [MIMO]), also with

Figure 6. Model of a telecom slave required to define the alternate BMCA (copy of Figure 3/G.8265.1).

QL

PTSFTime-

stamps

Timestampsfrom SOOCs

ENABLE_REQUESTING_SYNC_DEL_RESP

Timestamps forfrequencyrecovery

ENABLE_REQUESTING_UNICAST-ANNOUNCE


Selector

GM#1, priority_GM#1GM#2, priority_GM#2GM#N, priority_GM#N

List of grandmasters

Requestsync / del_resp

GMselection Selected

grandmaster TelecomSlave

Control and processing block

PTSF fromSOOCs

PTSFprocessing

QL fromSOOCs

QLprocessing

ENABLE_REQUESTING_UNICAST_ANNOUNCE

Requestannounce

QLPTSF

Time-stamps ENABLE_REQUESTING_

UNICAST-ANNOUNCE


SOOCinstantiation

Managementinformation

SOOCinstantiation

SOOCinstantiation

Grandmaster1PTP domain = x

Grandmaster2PTP domain = x

GrandmasterNPTP domain = x

QL

PTSFTime-

stampsENABLE_REQUESTING_UNICAST-ANNOUNCE


Network

Slave-onlyOCinstance2

Slave-onlyOCinstanceN

Slave-onlyOCinstance1



requirements on the order of microseconds orsub-microseconds.

This situation is driving the definition ofalternative solutions to the use of GPS or, moregenerally, a global navigation satellite system(GNSS). In fact sometimes the use of a GPSreceiver might not be feasible or effective. Note,however, that the use of GNSS will remainimportant in the future: it will provide the refer-ence to the packet masters, and allow a mixedarchitecture where some nodes may directlyreceive time reference from the GNSS, and oth-ers will get the reference through the networksfrom these nodes.

A significant number of activities are plannedaddressing various aspects of the distribution oftime over packet networks and in general overany relevant transport technology (OTN, GPON,VDSL2, etc.).

In this respect Q13 has defined a new seriesof Recommendations, G.827x, which will coverall relevant aspects: network requirements,architecture, PTP profile, and clocks.

CONCLUSIONS

Synchronization networks have adapted to mergewith the widespread use of packet networks asneeded for different types of networks and appli-cations. Synchronous Ethernet is being deployedand is able to interwork with SDH synchroniza-tion networks. For IEEE 1588, the new telecomprofile for the transport of frequency is now con-sented, and other frequency profiles might bespecified in the future.

The main area of study for ITU-T Q13/15 isnow the transport of time and phase throughpacket networks, as required by some mobiletechnologies (e.g., LTE TDD).

A new area of work is to define an architec-ture where the transport of time and phase couldbe supported by networks where a reference fre-quency is available, in order to get a more flexi-ble time synchronization solution.

New issues which did not exist for the trans-port of frequency have to be considered in caseof time synchronization, such as the asymmetryof the two directions.

Many aspects need to be solved in the nextfew years, and several recommendations (G.827xseries) will have to be completed. Q13 has shownto be a very active group and expects to over-come these new challenges like it has addressedthe previous technical issues.

REFERENCES[1] RFC 5905, “NTP Network Time Protocol Version 4: Pro-

tocol and Algorithms Specification.”[2] IEEE 1588-2008, “Standard for a Precision Clock Syn-

chronization Protocol for Networked Measurement andControl Systems.”

[3] I-ETS 300 353, “Broadband Integrated Services DigitalNetwork (B-ISDN); Asynchronous Transfer Mode (ATM);Adaptation Layer (AAL) Specification — Type 1.”

[4] ETSI TR 101 685, “Timing and Synchronization Aspectsof Asynchronous Transfer Mode (ATM) Networks.”

[5] ITU-G.G826x series: G.8260, “Definitions and Terminolo-gy for Synchronization in Packet Networks”; G.8261,“Timing and Synchronization Aspects in Packet Net-works”; G.8262, “Timing Characteristics of a Syn-chronous Ethernet Equipment Slave Clock (EEC)”;G.8263, “Timing Characteristics of Packet-Based Equip-ment Clocks (PEC) and Packet-Based Service Clocks(PSC)” (only draft available); G.8264, “Distribution ofTiming Information through Packet Networks”; G.8265,“Architecture and Requirements for Packet-Based Fre-quency Delivery”; G.8265.1, “Precision Time ProtocolTelecom Profile for Frequency Synchronization.”

ADDITIONAL READING[1] ITU-G.81x series: G.810, “Definitions and Terminology

for Synchronization Networks”; G.811, “Timing Charac-teristics of Primary Reference Clocks”; G.812, “TimingRequirements of Slave Clocks Suitable for use as NodeClocks in Synchronization Networks”; G.813, “TimingCharacteristics of SDH Equipment Slave Clocks (SEC).”

[2] ITU-G.82x series: G.823, “The Control of Jitter and Wan-der within Digital Networks which are Based on the2048 kb/s Hierarchy”; G.824, “The Control of Jitter andWander within Digital Networks which are Based onthe 1544 kb/s Hierarchy”; G.825, “The Control of Jitterand Wander within Digital Networks which are Basedon the Synchronous Digital Hierarchy (SDH).”

BIOGRAPHIESJEAN-LOUP FERRANT ([email protected]), grad-uated from INPG Grenoble (France), joined Alcatel in 1975and worked on analog systems, PCM, and digital cross-connects. He has been working on SDH synchronizationsince 1990 and on SDH and OTN standardization for morethan 15 years in ETSI TM1, TM3, and ITU-T SG13 andSG15. He has been rapporteur of SG15 Q13 on networksynchronization since 2001. He was one of the Alcatel-Lucent experts on synchronization in transport networksuntil he retired in March 2009. He is still rapporteur ofSG15 Q13, sponsored by Calnex Solutions.

STEFANO RUFFINI ([email protected]) joined Erics-son in 1993 and has been working on synchronizationaspects for about 15 years (currently as Expert R&D andmember of the Research & Innovation Team, Ericsson). Hehas represented Ericsson in various standardization organi-zations (including ETSI, ITU, 3GPP, and IETF) and is current-ly actively contributing to ITU-T SG15 Q13 (serving asassociate rapporteur and editor) and to other relevantstandardization bodies. He is one of the Ericsson expertsinvolved in the definition of the equipment and networksynchronization solutions.

A significant number

of activities are

planned addressing

various aspects on

the distribution of

time over packet

networks and in

general over any

relevant transport

technology (OTN,

GPON, VDSL2, etc.).

In this respect Q13

has defined a new

series of recommen-

dations, G.827x



INTRODUCTION

The Audio/Video Bridging (AVB) project in theIEEE 802.1 working group is focused on thetransport of time-sensitive traffic over IEEE 802bridged networks. The initial emphasis was onconsumer audio/video (A/V) applications. As theproject developed, the focus broadened toinclude professional A/V, industrial automation,and automotive applications. Additional AVBapplications are expected to include wirelesscommunications and smart grid. All of theseapplications have stringent timing requirements.Current bridged networks do not have mecha-nisms that enable meeting these requirementsunder general traffic conditions.

Goals of the AVB project, driven mainly bythe initial emphasis on consumer applicationsbut also useful for the later applications, are thatthe bridges be low-cost, and that AVB systemsrequire minimal configuration and are as nearplug-and-play as possible. Regarding the former,the philosophy has been to make the bridges asinexpensive as possible and place any high costin the respective end devices. This ensures thatthe cost of applications with stringent require-ments is borne only by those applications.Regarding the latter, the result has been to spec-ify as few options as possible.

The carrying of time-sensitive traffic requiresthree main functions. First, precise timing and

synchronization is needed so that individual traf-fic streams will meet their respective jitter, wan-der, and time synchronization requirements onegress. Second, a mechanism for applications toreserve the necessary network resources is need-ed. Finally, bridge forwarding and queueingmechanisms are needed so that latency require-ments are met. These functions are provided bythree AVB standards: IEEE 802.1AS (precisetiming and synchronization) [1], IEEE 802.1Qat-2010 (“Virtual Bridged Local Area Networks,Amendment 14: Stream Reservation Protocol[SRP]”), and IEEE 802.1Qav-2009 (“VirtualBridged Local Area Networks, Amendment 12:Forwarding and Queueing Enhancements forTime-Sensitive Streams”). A fourth AVB stan-dard, IEEE 802.1BA [2], specifies “AVB pro-files” (i.e., the parameters and options of theother three standards needed to transport trafficstreams of each respective AVB application).

The initial versions of the three main AVBstandards are either completed or nearly com-pleted. IEEE 802.1Qat and IEEE 802.1Qav arepublished. IEEE 802.1AS has completed the ini-tial sponsor ballot and one recirculation, and iscurrently in a second recirculation; completion isexpected in early 2011.

The AVB networks focus on IEEE 802 tech-nologies. In the current versions of the AVBstandards, these include full-duplex IEEE 802.3(Ethernet) operating at rates of 100 Mb/s orhigher, 802.11 (WiFi) operating at rates of 100Mb/s or higher (which means that 802.11 trans-port must use IEEE 802.11n), and 802.3 Ether-net passive optical network (EPON). However,an informative annex of IEEE 802.1AS describestransport over a coordinated shared network(CSN). Examples of such networks, alsodescribed in the Annex, are those based on theMultimedia over Coax Alliance (MoCA) stan-dard and International TelecommunicationUnion — Telecommunication StandardizationSector (ITU-T) Recommendation G.hn.

This article focuses on synchronization ofAVB networks using IEEE 802.1AS. It is intend-ed as a tutorial on AVB synchronization. AVBsynchronization has been described previously[3–6]. However, [3–5] were prepared well beforecompletion of IEEE 802.1AS and are not up todate, and [6] was a presentation without anaccompanying paper or tutorial. In addition, this

ABSTRACT

The Audio/Video Bridging project in theIEEE 802.1 working group is focused on thetransport of time-sensitive traffic over IEEE 802bridged networks. Current bridged networks donot have mechanisms that enable meeting theserequirements under general traffic conditions.IEEE 802.1AS is the AVB standard that willspecify requirements to allow for transport ofprecise timing and synchronization in AVB net-works. It is based on IEEE 1588-2008, includes aPTP profile that is applicable to full-duplexIEEE 802.3 transport, and adds specificationsfor timing transport over IEEE 802.11, IEEE802.3 EPON, and CSN media. This article pro-vides a tutorial on IEEE 802.1AS that updatesearlier descriptions, and new simulation resultsfor timing performance.


Geoffrey M. Garner and Hyunsurk (Eric) Ryu, Samsung Advanced Institute of Technology

Synchronization of Audio/VideoBridging Networks Using IEEE 802.1AS

GARNER LAYOUT 1/19/11 3:28 PM Page 140


article contains new simulation results for AVBnetwork timing performance that correspond toup-to-date default parameters specified in802.1AS.

The article is organized as follows. The nextsection gives an overview of IEEE 802.1AS. Wethen describe how synchronization is transportedusing IEEE 802.1AS. We then describe how bestmaster selection is done using IEEE 802.1AS.We then describe the Precision Time Protocol(PTP) profile that is part of IEEE 802.1AS andthe performance requirements for 802.1AS time-aware systems. We then describe the new simu-lation results for 802.1AS timing performance.The final section presents conclusions.

OVERVIEW OF IEEE 802.1ASA bridge or end station that meets the require-ments of IEEE 802.1AS, and therefore is able totransport synchronization, is referred to as atime-aware bridge or end station, respectively.Since IEEE 802.1AS operates over a variety ofmedia, the 802.1AS architecture divides a time-aware system (system refers generically to bridgeor end station) into media-independent andmedia-dependent entities. All the entities arelocated in a layer above the IEEE 802.1 mediumaccess control (MAC), MAC relay, and linklayer control (LLC) sublayers.

IEEE 802.1AS relies on the transfer of timestamps using mechanisms that are media depen-dent. For the case where the medium is full-duplex Ethernet, 802.1AS uses a subset of IEEE1588-2008 [7]. Specifically, IEEE 802.1ASincludes an IEEE 802-specific profile of IEEE1588-2008. For 802.11 links, 802.1AS uses timingfacilities, developed initially for location deter-mination, defined in IEEE 802.11v [8]. ForEPON links, 802.1AS uses the timing facilitiesdefined in the IEEE 802.3 multipoint controlprotocol (MPCP). For CSNs, it is possible to useinherent timing facilities or the PTP profile.

The protocol defined by IEEE 1588 isreferred to as PTP, and the IEEE 1588 profilesare referred to as PTP profiles. By analogy, theprotocol defined by IEEE 802.1AS is referred toas the generalized PTP (gPTP); the gPTP includestransport of synchronization over all media (i.e.,not only those where transport is part of thePTP profile). The links that connect ports oftime-aware systems are at least logically point-to-point; that is, gPTP information sent by agPTP port is received by a specific gPTP port atthe other end of the logical link. Whether thelinks are also physically point-to-point is media-dependent. The specific subset of IEEE 1588used in the PTP profile contained in IEEE802.1AS, and the differences between PTP andgPTP, are described in more detail later.

IEEE 802.1AS establishes a synchronizationhierarchy within an AVB network using an algo-rithm that is very similar to the default best mas-ter clock algorithm (BMCA) of [7]. Thisalgorithm is part of the media-independentlayer. It operates autonomously, and results inthe selection of one time-aware system as thegrandmaster and all ports of all the time-awaresystems having master, slave, or passive roles.Each time-aware system, except for the grand-

master, receives synchronization information onits single slave port and transmits synchroniza-tion information on any master ports. The syn-chronization hierarchy forms a synchronizationspanning tree, and ports are placed in the passiverole to break loops. The transport of synchro-nization is hop by hop; each time-aware systemexcept the grandmaster uses incoming synchro-nization information on its slave port to synchro-nize to the grandmaster. Each time-aware systemtransmits synchronization information on anymaster ports.

IEEE 802.1AS requires that all bridges andend stations be time-aware. The enforcement ofthis requirement is media-dependent. A full-duplex Ethernet port uses the IEEE 1588 peerdelay mechanism to determine whether the sys-tem at the other end of the link is time-aware.This is done by detecting whether the system atthe other end of the link responds to peer delaymessages and, if so, whether the measured linkdelay exceeds a specified threshold. An IEEE802.11 link determines that gPTP cannot run onthe link based on information provided by the802.11v protocol. An EPON can always rungPTP because the MPCP timing facilities arealways present. A CSN that uses its inherenttiming facilities can always run gPTP; alterna-tively, if the CSN uses the PTP profile, the IEEE1588 peer delay mechanism is used to determinewhether the system at the other end of the linkis time-aware.

TRANSPORT OF SYNCHRONIZATIONIN IEEE 802.1AS

Synchronization transport by an 802.1AS time-aware system is functionally equivalent to syn-chronization transport by an IEEE 1588boundary clock (or ordinary clock) that uses thepeer delay mechanism. It may be shown that thetransport is also functionally equivalent to trans-port by an IEEE 1588 peer-to-peer transparentclock.

The time synchronization model in IEEE802.1AS assumes that a time-aware system has afree-running local clock it uses to time-stamp thedeparture and arrival of various time synchro-nization messages. There must be a single com-mon local clock for all time-stamping in thenode, but otherwise the clock is free-running(i.e., there is no requirement to physically adjustthe frequency of this clock, although it is notprohibited). Each node uses the arrival anddeparture time stamps for the various messagesand time synchronization information carried bythe messages from upstream nodes to determinegrandmaster time that corresponds to anydesired local clock time

A time-aware system is required to measurepropagation delay on each logical link attachedto each gPTP port. This measurement is media-dependent. For full-duplex Ethernet, the mea-surement uses the IEEE 1588 peer delaymechanism. The peer delay measurement isillustrated in Fig. 1 (adapted from [1, 7]). Theport that wishes to measure propagation delay,termed the peer delay initiator, sends a Pde-lay_Req message and time-stamps the depar-

Since IEEE 802.1AS

operates over a

variety of media, the

802.1AS architecture

divides a time-aware

system (system refers

generically to bridge

or end station) into

media-independent

and media-depen-

dent entities. All the

entities are located in

a layer above the

IEEE 802.1 MAC,

MAC relay, and LLC

sublayers.



ture of that message (time t1). The messagearrives at the node at the other end of the link,termed the peer delay responder, which time-stamps the arrival of the message (time t2). Thepeer delay responder sends a Pdelay_Respmessage to the peer delay initiator at a latertime, and time-stamps the departure (time t3).The Pdelay_Resp message also conveys thetime t2 to the initiator. The initiator time-stampsthe arrival of the Pdelay_Resp message (timet4). Finally, the responder conveys the time t3 tothe initiator in a Pdelay_Resp_Follow_Upmessage, which is not time-stamped. At the con-clusion of this message exchange, the peer delayinitiator can compute the propagation delayunder the assumption that the link is symmetric(i.e., the delay is the same in both directions).The propagation delay is equal to half the differ-ence between the interval t4 – t1 and the intervalt3 – t2, but with both intervals referred to a com-mon time base (since, in general, the frequenciesof the local clocks at the time-aware systems willbe different). The referencing of both intervals

to a common time base is done by multiplyingthe interval t3 – t2 by the measured ratio of theinitiator to responder local clock frequency.1

A peer delay initiator uses the departure andarrival times of the successive Pdelay_Respmessages to measure the ratio of its local clockfrequency to that of the peer delay responder.This rate ratio is used in the link delay computa-tion, as described above, and in computing syn-chronized (i.e., grandmaster) time, as describedshortly. IEEE 802.1AS does not prescribe thespecific algorithm for the rate ratio measure-ment; any algorithm is allowed provided that themeasurement can be made to within ±0.1 ppm.An example is given where rate ratio is comput-ed as the ratio of the interval between thearrivals of successive Pdelay_Resp messages tothe interval between the departures of the samemessages.

Every full-duplex Ethernet port of everytime-aware system periodically initiates a peerdelay measurement. As indicated above, theresult of this measurement is used to determineif gPTP can be run on the link. In addition, thelink delay and rate ratio will be continuouslyknown to all ports of all links that can run gPTP,which allows faster reconfiguration if there is agrandmaster or network topology change. Notethat while the link delay is expected to be rela-tively static for full-duplex Ethernet links, thefrequencies of the local clocks of the endpointtime-aware systems, and therefore the rate ratio,may vary over time (e.g., due to temperaturechanges).

A time-aware system synchronizes to thegrandmaster using time synchronization informa-tion received on its slave port. Here, the termsynchronize means that the time-aware system isable to compute the grandmaster time corre-sponding to any desired local clock time. If thetime-aware system can do this, it can provide thegrandmaster time (i.e., the network synchronizedtime) whenever desired. There is no requirementthat the local clock frequency be physicallyadjusted to match the grandmaster frequency(although, as indicated earlier, this is not prohib-ited).

Time synchronization is performed as follows.Periodically, a time-aware system that is not thegrandmaster receives time synchronization infor-mation on its slave port. This information con-sists of a grandmaster time and a correspondinglocal clock time. The period for the sending ofthis information by a master port is termed theSync interval, following IEEE 1588 terminolo-gy. The format of this information and the mes-sages used to convey the information aremedia-dependent. For example, in full-duplexEthernet media the PTP messages Sync andFollow_Up [7] are used, and the correspon-dence between grandmaster and local clock timeis obtained using the timestamp informationfrom the upstream time-aware system carried inthe Follow_Up message and the timestamp ofthe arrival of the Sync message. The process isdescribed in more detail below. In IEEE 802.11media, the information is conveyed by a singleIEEE 802.11v Timing Measurement ActionFrame. IEEE 802.1AS causes this frame to betransmitted and obtains the information from a

Figure 1. Link delay measurement using peer delay mechanism.

Time stampsknown by peerdelay intiator

t4

t1

t2

t1

t1,t2,t4

t1,t2,t3,t4

t3

t4 – t1 t3 – t2

Pdelay_Req

Pdelay_Respt2

Peer delay initiator

Peer delay responder

Pdelay_Resp_Follow_Upt3

Figure 2. Transport of time synchronization information over full-duplex Ether-net media.

Time-aware system A

master port Slave port

Sync

Master port

Time-aware system B Time-aware system C slave port

Follow_Up (preciseOriginTimestamp,correctionFieldA, rateRatioA)

Sync tr,C

ts,B

ts,A

tr,B

Follow_Up (preciseOriginTimestamp,correctionFieldB, rateRatioB)

1 An error remains due toany difference between theinitiator and grandmasterfrequency; however, theeffect of this error can beshown to be negligible [1].



received frame via service interface primitivesthat are part of the 802.11v MAC layer manage-ment entity (MLME). In EPON media, the cor-respondence is conveyed using anorganization-specific slow protocol (OSSP) mes-sage. IEEE 802.1AS causes this message to besent and obtains information from a receivedmessage via service interface primitives. TheEPON MPCP counter is, in effect, the localclock, and the setting of this counter is donewithin the EPON MPCP layer (i.e., outside ofIEEE 802.1AS). The transport of synchroniza-tion information for IEEE 802.11 and IEEE802.3 EPON media is described in more detail in[1, clauses 12 and 13].

Figure 2 (taken from [1]) illustrates the syn-chronization process for the case of full-duplexEthernet media. The figure shows three time-aware systems, labeled A, B, and C. A masterport of A sends a PTP Sync message to B, andtime-stamps the message ts,A relative to the localclock of A. The Sync message is received andtime-stamped on the slave port of B at localtime tr,B. At a later time, A sends a PTP Fol-low_Up message, which contains (among otherinformation) a preciseOriginTimestamp, a cor-rectionField, and the measured cumulative rat-eRatio of the frequency of the local clock of Arelative to the grandmaster frequency (this latterfield is contained in a standard organizationTLV). The computation of these values willbecome clear after the set of computations doneby time-aware system B is described. The pre-ciseOriginTimestamp value is the same valuethat was sent by the grandmaster when this par-ticular synchronization information originated.The correctionField value is chosen so that thesum of the preciseOriginTimestamp and correc-tionField is the grandmaster time that corre-sponds to local time ts,A. At time ts,B, B sendsand time-stamps a Sync message. At a latertime, B sends a Follow_Up message. The fields

of the Follow_Up message are set as follows:• The preciseOriginTimestamp is set equal to

the preciseOriginTimestamp of the mostrecently received Sync and Follow_Upmessage on the slave port,

• The cumulative rateRatio is set equal to thecumulative rateRatio of the most recentlyreceived Sync and Follow_Up messageon the slave port multiplied by the currentneighbor rate ratio measured by the slaveport, and

• The correctionField is set equal to the sumof the correctionField of the most recentlyreceived Sync and Follow_Up messageon the slave port, the link delay measuredby the slave port, and the time interval ts,B– tr,B multiplied by the newly computedcumulative rateRatio.The sum of the preciseOriginTimestamp and

correctionField of the Follow_Up message sentby B is the grandmaster time that corresponds tothe local time ts,B when the Sync message issent. The time interval ts,B – tr,B multiplied bythe cumulative rateRatio is the PTP residencetime, and the fact that a time-aware systemalters only the correctionField and not the pre-ciseOriginTimestamp is analogous to way inwhich an IEEE 1588 peer-to-peer transparentclock transports synchronization. However, gPTPcould have specified that the full grandmastertime (except for any sub-nanosecond portion, asin PTP this must be carried in the correctionfield) be carried in the preciseOriginTimestampfield as would be done by an IEEE 1588 bound-ary clock. In fact, the boundary clock and peer-to-peer transparent clock are functionallyequivalent in the manner in which they transportsynchronization (since they differ only in howthe grandmaster time is distributed between thepreciseOriginTimestamp and correctionField).The key difference between the boundary clockand peer-to-peer transparent clock is that the

Table 1. Default values and required supported ranges for PTP attributes.

Attribute Default value Required supported values

gPTP domain number 0 There is a single gPTP domain, with domain number 0

logAnnounceInterval 0 0, 127*

logSyncInterval –3 –3, 127*

logPdelayReqInterval 0 0, 127*

announceReceiptTimeout 3 3

priority1246 (network infrastructure time-aware systems)250 (portable time-aware systems)248 (other time-aware systems)

246 (network infrastructure time-aware systems)250 (portable time-aware systems)248 (other time-aware systems)

priority2 248 248

Observation interval foroffsetScaledLogVariance 0.125 s 0.125 s

*When the value of logAnnounceInterval, logSyncInterval, or logPdelayReqInterval is 127, the port does not send Announce, Sync, orPdelay_Req messages, respectively.



former invokes the BMCA and PTP statemachine, and the latter does not; for this reason,the gPTP time-aware system is equivalent to aPTP boundary clock. This topic is discussed inmore detail in [9].

BEST MASTER SELECTION INIEEE 802.1AS

The description of best master selection in thissection is a summary of the description in [4]. Amore detailed description is contained in [4, ref-erences therein, especially 11–14].

The 802.1AS BMCA is very similar to thedefault IEEE 1588 BMCA. Within an IEEE802.1AS network, all time-aware systems arerequired to invoke the BMCA, and best masterselection information is conveyed using PTPAnnounce messages on all media. As indicatedabove, the BMCA is part of the media-indepen-dent layer.

The gPTP BMCA differs from the defaultPTP BMCA as follows:• In gPTP, a time-aware system need not be

grandmaster-capable, regardless of thenumber of ports, while in PTP it is only anordinary clock (i.e., a single-port clock) thatneed not be grandmaster-capable (i.e., canbe slave-only).

• In gPTP, there is no foreign master qualifi-cation as in PTP; all Announce messagesreceived on a port are used immediately.

• In gPTP, a port whose role is determined tobe master becomes master immediately;there is no pre-master state as in PTP.The gPTP BMCA is expressed using a subset

of the formalism for Rapid Spanning Tree Proto-col (RSTP) (see IEEE 802.1Q-2005, “IEEEStandard for Local and Metropolitan Area Net-works, Virtual Bridged Local Area Networks”).This is possible because both the BMCA andRSTP create spanning trees. The root of thespanning tree created by the BMCA is the grand-master, unless no time-aware system in the net-work is grandmaster-capable. The IEEE 1588attributes priority1, clockClass, clockAccuracy,offsetScaledLogVariance, priority2, and clockI-dentity are concatenated, as unsigned integers inthat order, into the overall attribute systemIden-tity. The first part of the IEEE 1588 dataset com-parison algorithm [7, Fig. 27] is expressed interms of a comparison of systemIdentities. Sixdifferent, but related, priority vectors are defined,which are set and compared in four interactingstate machines; these machines also set the role(i.e., PTP state) of each port. The operation ofthese state machines is equivalent to the datasetcomparison and state decision algorithms of [7].

PTP PROFILE CONTAINED INIEEE 802.1AS AND DIFFERENCESBETWEEN PTP AND IEEE 802.1AS

IEEE 802.1AS includes a PTP profile, which isused for transport over full-duplex, IEEE 802.3links. The PTP profile specifies attribute valuesand the subset of IEEE 1588 options used inIEEE 802.1AS. However, IEEE 802.1AS alsocontains specifications beyond PTP, such as thespecifications for transport over IEEE 802.11and 802.3 EPON, as well as those related to

Table 2. Parameters for simulation cases.

Parameter Value

Number of time-aware systems, including grandmaster 8 (7 hops)

Link medium Full-duplex IEEE 802.3 (Ethernet)

Sync interval 0.125 s

Peer delay interval 1.0 s

Free-running, local clock tolerance ±100 ppm

Residence time, Pdelay turnaround time 1 ms (cases 1 and 2), 10 ms (case 3), 50 ms (case 4)

Link propagation delay 500 ns

Phase measurement granularity of local oscillator 40 ns

Local clock wander generation Not modeled (case 1); FFM at level of TDEV mask of [1, Fig. B-1];see performance requirement 4 in article (cases 2–4)

Endpoint filter bandwidth 1 Hz, 10 mHz, 1 mHz (for each of cases 1–4)

Endpoint filter gain peaking 0.1 dB

Simulation time 10 010 s

Maximum time step 0.001 s



time-aware system performance.The default values and required supported

values for time-aware system attributes are givenin Table 1. Note that the required supported val-ues are minimum requirements for an AVB net-work (i.e., all AVB applications can assume thatthese values are supported). However, a particu-lar AVB application may require that additionalvalues be supported.

The PTP options used in this PTP profile are:1. The BMCA is an alternate BMCA that dif-

fers from the default BMCA of [7] asdescribed above.

2. The management mechanism uses a SimpleNetwork Management Protocol (SNMP)Management Information Base (MIB).

3. The path delay mechanism is the peer delaymechanism.

4. The transport mechanism is full-duplex Eth-ernet.

5. A time-aware system is a PTP ordinaryclock or boundary clock, depending onwhether it has one or more than one gPTPports, respectively. All time-aware systemsare two-step clocks (as defined in IEEE1588-2008).

6 Each gPTP port of a time-aware systemmeasures the frequency offset of its neigh-bor at the other end of the attached linkrelative to itself. The frequency offset, rela-tive to the grandmaster, is accumulated in astandard organization TLV that is attachedto the Follow_Up message.

7 The PTP path trace feature is required.8 A standard organization TLV is defined

that allows a port of a time-aware system torequest that its neighbor slow down orspeed up the rate at which it sendsSync /Follow_Up , peer delay, and/orAnnounce messages.

9 The acceptable master table feature is usedwith IEEE 802.3 EPON links to ensure thatthe optical line terminal (OLT) is masterand optical network units (ONUs) are slaves.In addition to containing the above PTP pro-

file, IEEE 802.1AS specifies performancerequirements for time-aware systems:1 The fractional frequency offset of the local

clock, relative to the TAI frequency, shallbe within ±100 ppm.

2 The local clock frequency shall be 25 MHzor greater (i.e., the time measurement gran-ularity is no worse than 40 ns).

3 The jitter generation of the local clock shallnot exceed 2 ns peak-to-peak, measuredthrough a 10 Hz first-order high-pass filter(and having low-pass characteristic speci-fied in [1]).

4 The wander generation of the local clockshall be within the time deviation (TDEV)mask of [1, Fig. B-1].

5 The residence time (i.e., the time intervalbetween the sending of a Sync message ona master port and receipt of the most recentSync message on the slave port) shall notexceed 10 ms.

6 The Pdelay turnaround time (i.e., the timeinterval between the receipt of a Pde-lay_Req message and the sending of thecorresponding Pdelay_Resp message

shall not exceed 10 ms.7 The error inherent in any scheme used to

measure neighbor rate ratio shall notexceed 0.1 ppm.

SIMULATION RESULTS FORTIMING PERFORMANCE

References [5, 6] present simulation results for802.1AS timing performance. The results aregiven in the form of maximum time intervalerror (MTIE; see ITU-T Rec. G.810, “Defini-tions and Terminology for Synchronization Net-works”) for transport over one and sevem hops,respectively. An eight-node network was simulat-ed because one objective for AVB is that endsystems may be separated by up to seven hops.However, these simulations did not considerlocal clock wander generation (see requirement4 in the previous section). In addition, sincethose simulations were done, the residence timeand Pdelay turnaround time requirements (seerequirements 5 and 6 of the previous section)were increased from 1 to 10 ms. New simula-tions were performed, which include local clock

Figure 3. MTIE results for cases 1-6, node (time-aware system) 2.

Observation Interval (s)

Comparison of jitter/wander accumulation MTIE at time-aware system (node) 2 1 Hz, 10 mHz, and 1 mHz endpoint filter bandwidths 1, 10, 50 ms residence time and Pdelay turnaround time (with clock wander generation) 1 ms residence time and Pdelay turnaround time (without clock wander generation) Sync Interval = 0.125 s Pdelay Interval = 1.0 s

1e-3

1e-4

MTI

E (n

s)

1e-5

1e-3

1e-2

1e-1

1e+0 1e+1 1e+2

1e+3

1e+4

1e+5

1e+6

1e+7 1e+8

1e+9 1e+10

1e-4 1e-2 1e-1 1e+0 1e+1 1e+2 1e+3 1e+4 1e+5

1 Hz, 1 ms, no clock wander generation1 Hz, 1 ms, with clock wander generation1 Hz, 10 ms, with clock wander generation1 Hz, 50 ms, with clock wander generation10 mHz, 1 ms, no clock wander generation10 mHz, 1 ms, with clock wander generation10 mHz, 10 ms, with clock wander generation10 mHz, 50 ms, with clock wander generation1 mHz, 1 ms, no clock wander generation1 mHz, 1 ms, with clock wander generation1 mHz, 10 ms, with clock wander generation1 mHz, 50 ms, with clock wander generationUncompressed SDTV (SDI Signal)Uncompressed HDTV (SDI Signal)MPEG-2, after network transportMPEG-2, no network transportDigital audio, consumer interfacesDigital audio, professional interfacesFemtocell

1 mHz

1 Hz

10 mHz



wander generation and reflect the increased resi-dence and Pdelay turnaround times. The simula-tor and model used in [5, 6] were used here, butwith the addition of a noise model that gener-ates wander at the level of the TDEV mask ofperformance requirement 4 of the previous sec-tion. This mask has a flicker frequency modula-tion (FFM) characteristic, and its values at 0.05 sand 10 s are 0.25 ns and 50 ns, respectively (andit is specified only for times between 0.05 s and10 s; see [1] for details). The noise is simulatedusing the technique of [10]. The simulator isdescribed in [5].

Parameters for the simulation cases are givenin Table 2. Case 1 is from [5, 6], and is repeatedhere for comparison with the new cases. Thenew cases include:• Addition of local clock wander generation

at the level of the 802.1AS requirement• Addition of clock wander generation and

increase of residence and Pdelayturnaround times to 10 ms (the 802.1ASrequirement)

• Addition of wander generation and increaseof residence and Pdelay turnaround timesto 50 ms (i.e., exceeding the 802.1ASrequirement by a factor of 5)

As in [5, 6], a single run was made for eachendpoint filter bandwidth, for each case. Thefree-running clock frequency offset for eachtime-aware system was initialized randomly,from a uniform distribution over the frequencytolerance range.

MTIE results for nodes (time-aware systems)2 and 8 are given in Figs. 3 and 4, respectively,and are compared with MTIE requirements(masks) derived from jitter, frequency offset,and frequency drift requirements for the respec-tive audio/video and femtocell technologies.MTIE is peak-to-peak phase variation, as a func-tion of observation interval; see [5, 6, referencestherein] for more detail, including derivation ofthe MTIE masks.

The results show that the addition of clockwander generation, as well as the increase of res-idence and Pdelay turnaround times, have littleeffect on the resulting MTIE. In fact, increasingthe residence and Pdelay turnaround times to 50ms has little impact. The results indicate that a 1mHz endpoint filter bandwidth enables all therequirements to be met. Increasing the band-width to 10 mHz enables all the requirementsexcept those for uncompressed SDTV to be met,and increasing the bandwidth to 1 Hz enablesonly the audio requirements to be met. Notethat [5, 6] also gave results that indicated that a10 Hz endpoint filter bandwidth will allow onlythe professional audio requirements to be met.While the new cases were not run with a 10 Hzendpoint filter, it is expected that this resultwould be the same.

CONCLUSIONSThis article has provided a tutorial on IEEE802.1AS, and has presented new simulationresults based on requirements in the 802.1ASsponsor ballot draft.

IEEE 802.1AS is based on a subset of IEEE1588-2008. It includes a PTP profile that is appli-cable to full-duplex IEEE 802.3 transport, andadds specifications for timing transport overIEEE 802.11, IEEE 802.3 EPON, and CSNmedia. The requirements of IEEE 802.1AS werechosen to provide for low-cost bridges, while stillallowing application performance requirementsto be met. Applications with more stringent tim-ing requirements will use narrower-bandwidthendpoint filters. IEEE 802.1AS has very fewuser-configurable options, consistent with thegoal that AVB networks be plug-and-play.

The new simulations were performed to con-firm the level of support that can be providedfor various end applications. The results of simu-lations show that all application requirementsconsidered in 802.1AS can be met with the useof a 1 mHz filter bandwidth. With a 10 mHzbandwidth, all requirements are met with theexception of uncompressed SDTV. With a 1 Hzbandwidth, requirements for audio are met.

REFERENCES[1] IEEE P802.1AS/D7.0, “Draft Standard for Local and

Metropolitan Area Networks — Timing and Synchro-nization for Time-Sensitive Applications in BridgedLocal Area Networks,” Mar. 23, 2010.

[2] IEEE P802.1BA/D2.0, “Draft Standard for Local andMetropolitan Area Networks-Audio Video BridgingFigure 4. MTIE results for cases 1-6, node (time-aware system) 8.

Observation Interval (s)

Comparison of jitter/wander accumulation MTIE at time-aware system (node) 81 Hz, 10 mHz, and 1 mHz endpoint filter bandwidths1, 10, 50 ms residence time and Pdelay turnaround time (with clock wander generation)1 ms residence time and Pdelay turnaround time (without clock wander generation)Sync Interval = 0.125 sPdelay Interval = 1.0 s

1e-3

MTI

E (n

s)

1e-41e-3

1e-2

1e-1

1e+0

1e+1

1e+2

1e+3

1e+4

1e+5

1e+6

1e+7

1e+8

1e+9

1e+10

1e-4 1e-4-2 1e-1 1e+0 1e+1 1e+2 1e+3 1e+4 1e+5

1 Hz

10 mHz

1 mHz

1 Hz, 1 ms, no clock wander generation1 Hz, 1 ms, with clock wander generation1 Hz, 10 ms, with clock wander generation1 Hz, 50 ms, with clock wander generation10 mHz, 1 ms, no clock wander generation10 mHz, 1 ms, with clock wander generation10 mHz, 10 ms, with clock wander generation10 mHz, 50 ms, with clock wander generation1 mHz, 1 ms, no clock wander generation1 mHz, 1 ms, with clock wander generation1 mHz, 10 ms, with clock wander generation1 mHz, 50 ms, with clock wander generationUncompressed SDTV (SDI Signal)Uncompressed HDTV (SDI Signal)MPEG-2, after network transportMPEG-2, no network transportDigital audio, consumer interfacesDigital audio, professional interfacesFemtocell



(AVB) Systems,” Aug. 13, 2010.[3] G. M. Garner et al., “IEEE 802.1 AVB and its Application

in Carrier-Grade Ethernet,” IEEE Commun. Mag., Dec.,2007, pp. 126–34.

[4] M. D. Johas Teener and G. M. Garner, “Overview andTiming Performance of IEEE 802.1AS,” Proc. IEEE ISPCS‘08, Ann Arbor, MI, Sept. 22–26, 2008, pp. 49–53.

[5] G. M. Garner, A. Gelter, and M. D. Johas Teener, “NewSimulation and Test Results for IEEE 802.1AS TimingPerformance,” Proc. IEEE ISPCS ‘09, Brescia, Italy, Oct.12–16, 2009, pp. 109–15.

[6] G. M. Garner, “Synchronization of Audio/Video BridgingNetworks using IEEE 802.1AS,” NIST-ATIS-Telcordia Wksp.Synchronization Telecommun. Sys., Mar. 9–11, 2010.

[7] IEEE Std. 1588-2008,” IEEE Standard for a PrecisionClock Synchronization Protocol for Networked Measure-ment and Control Systems,” Revision of IEEE Std 1588-2002, IEEE Instrumentation Measurement Society, July24, 2008.

[8] IEEE P802.11v/D14.0, “Draft Standard for InformationTechnology — Telecommunications and InformationExchange between Systems — Local and MetropolitanArea Networks- Specific Requirements — Part 11: Wire-less LAN Medium Access Control (MAC) and PhysicalLayer (PHY) Specifications, Amendment 8: IEEE 802.11Wireless Network Management,” Aug. 2010.

[9] G. M. Garner, M. Ouellette, and M. D. Johas Teener,“Using an IEEE 802.1AS Network as a Distributed IEEE1588 Boundary, Ordinary, or Transparent Clock,” Proc.IEEE ISPCS ‘10, Portsmouth, NH, Sept. 29–Oct. 1, 2010,pp. 109–15.

[10] J. A. Barnes and C. A. Greenhall, “Large Sample Simu-lation of Flicker Noise,” 19th Annual Precise Time and

Time Interval Apps. Planning Meeting, Dec. 1987.

BIOGRAPHIESGEOFFREY M. GARNER ([email protected]) graduatedfrom MIT (S.B., 1976, S.M., 1978, Ph.D., 1985). Since 1993his work has focused on network timing, jitter, and syn-chronization; network performance and quality of service;standards development; and simulation. He has been aconsultant since 2003; current projects include work on theIEEE Audio/Video Bridging standard for precise timingtransport, IEEE 802.1AS (for Samsung Electronics), andsimulation of timing performance for new OTN clients (forHuawei Technologies). He is editor of IEEE P802.1AS, and isa member of the IEEE Registration Authority Committee.Prior to 2003, he was a Distinguished Member of TechnicalStaff in Bell Labs Lucent Technologies.

HYUNSURK (ERIC) RYU ([email protected]) graduatedfrom POSTECH (B.S., 1992, M.S., 1994, Ph.D., 1998). Hehas been a member of technical staff in Samsung AdvancedInstitute of Technology, Samsung Electronics, after receiv-ing his Ph. D. He has been a project leader of the multiplenetworking technologies R&D projects and was a contribu-tor to the related standardization in ITU-T SG15, IEEE802.1, and IEEE 802.3. In addition, he has been activelyinvolved in the development of the new IEEE 802.1 AudioVideo Bridging standard from the beginning. Recently hehas expanded his research interests to bio-mimic comput-ing, communication, and cognitive applications.



INTRODUCTION

Traditionally, timing measurements involve sam-pling signal edges of an oscillator or networkequipment timing signal. Phase (phase devia-tion) in this context is usually represented inunits of time, in which case it is also referred toas time interval error (TIE). It is computed bycomparing time-stamped edges to a mathemati-cally derived ideal signal, and frequency is calcu-lated by counting signal edges between each pairof time stamps and taking a ratio of the countover the sample duration.

Other useful quantities can be derived fromthese basic calculations, including Allan devia-tion (ADEV), modified Allan deviation(MDEV), phase power spectral density (PPSD),time deviation (TDEV), and maximum timeinterval error (MTIE) [1–3]. The latter two areof particular importance to telecom: MTIE,which describes the maximum phase swings overa time window, and TDEV, which assesses noiseprocesses, have both been used to set networkand equipment synchronization limits in thestandards bodies.

It should be emphasized that these traditionaltiming measurements are still important for thecharacterization of packet network timing. Pack-et network equipment such as a packet slavedevice produces signals that may be character-ized in the same way as those described above,and the definitive way of assessing performanceof such a device is to measure that output timingsignal.

There is also, however, a great need for study-ing the timing characteristics of the packet net-work itself; this is required both for

understanding the behavior of packet networkclocks and for designing these devices optimally.This involves a measurement of a different kind.Rather than timing signal edges, packets aretime-stamped as they traverse two nodes in anetwork. The packets could just as well originateor terminate at one of these nodes. Just like fortiming signal edges, a primary reference isrequired in the ideal case for timing packets. Inaddition, a common timescale at the two nodesis required in the ideal case. This can be provid-ed by a global navigation satellite system (GNSS)system such as GPS.

In contrast, for the traditional synchroniza-tion measurement, only a primary frequency ref-erence is required as frequency stability andaccuracy, not absolute time, are the quantitiesunder study. To summarize, the traditional syn-chronization measurement is a single-node mea-surement requiring precision time-stamping ofthe signal edges with a single primary frequencyreference, while the packet timing measurementis a two-node measurement requiring precisiontime-stamping of the packets, ideally with prima-ry time references at both nodes.

Regarding the study of packet timing, thereare two aspects of the measurement result thatare of interest: the nominal time it takes to tran-sit the network and the variation in that transittime over time. The former quantity is referredto as latency, while the latter is called packetdelay variation (PDV). The focus of the metricsdescribed below is on the characterization andunderstanding of PDV, which is also the quanti-ty responsible for providing the greatest chal-lenge for packet timing equipment.

PACKET TIMING PROBEPacket timing measurements involve the produc-tion or selection of special probe packets withinthe overall mix of traffic packets in the network.These probe packets could be IEEE 1588 Preci-sion Time Protocol (PTP) packets or NetworkTime Protocol (NTP) packets, for example. Aseries of these probe packets are time-stampedat two nodes in the network, most generally inboth the forward and reverse directions.

There are two approaches to the design of apacket timing probe, one resulting in a passive

ABSTRACT

As the transport of data across the networkrelies increasingly on Ethernet/IP methods andless on the TDM infrastructure, the need forpacket methods of synchronization transport aris-es. Evaluation of these new packet methods of fre-quency and time transport requires newapproaches to timing measurement and analysis.This article describes these new packet measure-ment techniques and introduces some of the newmetrics being used for packet timing data analysis.


Lee Cosart, Symmetricom, Inc.

NGN Packet Network SynchronizationMeasurement and Analysis

COSART LAYOUT 1/19/11 3:26 PM Page 148


probe and the other resulting in an active probe[4, 5]. The passive probe relies on other networkequipment to produce the packet stream andserves only to time-stamp the packets as theypass through the probe via a network tap or mir-rored port. The active probe establishes a proto-col with another paired device as well astime-stamping the probe packets. In the casewhere an active probe is used to measure packetdelay in both directions, each of the two paireddevices serves to originate probe packets in onedirection and terminate probe packets in theother direction.

Examples of both probe types are shown inFig. 1. In this case, IEEE 1588 PTP packets areused as the probe packets. Note that the setupfor the passive probe is more complicated, witha slave device needed to establish the packetprotocol and an Ethernet tap needed to send acopy of the packet stream going into and outfrom the slave into the passive probe. The activeprobe both sets up the session and time-stampsthe packets, and thus requires neither additionalpiece of equipment. The IEEE 1588 grandmas-ter supplies PTP packets and time-stamps pack-ets in both directions, just as the probe does. Allgrandmaster time stamps can be transported tothe probe, so it can serve as the central measure-ment data collection point.

PACKET TIMINGMEASUREMENT DATA

It is instructive to look at a sequence of rawpacket time stamp pairs and describe how pack-et delay samples are derived from them. Theprobe packets that traverse a network understudy are emitted from a source at some speci-fied rate. In the example shown in Fig. 2, therate is 64 Hz. Each individual packet is time-stamped at two nodes in the network. If bothforward and reverse directions are being studied,there are two probe packet sequences. These aredenoted in Fig. 2 as F and R time stamp pairs.

The F and R time stamps are used to con-struct forward and reverse packet delay sequencesby computing the differences between timestamps in the pair. To take one example, the firstforward packet delay value is the differencebetween the time stamps in the line F,00167,1223305830.490552012 – 1223305830.488078908= 2.473 ⋅ 10–3 s.

The placement of the individual packet delaycalculations is set by the originating time stamp,in this case in Unix time (UTC seconds since Jan-uary 1, 1970). In the example shown in Fig. 2, thefirst time stamp for both the forward and reversesequence occurs at 2009/10/06 15:10:30, which hasbeen assigned to a time of 0.0000. The next sam-ple occurs approximately 1/64 s (0.015625 s) laterfor both forward and reverse sequences.

Thus, two packet delay sequences are pro-duced, forward and reverse, and these are mea-surements of one-way packet delay. Generally,these two sets of data are analyzed separately,particularly when packet-based frequency trans-port is the goal. When time transport via a two-way packet timing protocol is of interest, bothdirections must be considered together. Metrics

applicable to both of these situations are dis-cussed in turn below.

PACKET DELAY STATISTICSUnlike the phase (TIE) data from a traditionalsynchronization measurement, where currentphase values are generally correlated to neigh-boring preceding and subsequent phase valuesbecause of oscillator stability, the data in a pack-et delay sequence can vary considerably frompoint-to-point. Often a relatively short packetdelay measurement will exhibit similar peak-to-peak behavior as a much longer measurement.This is true for other statistics such as mean andstandard deviation as well.

Given these typical characteristics of packetdelay sequence data (PDV phase), it is often use-ful to construct a histogram with the collection ofpacket delay samples. In other situations, statisticsmight vary over the duration of a packet delaymeasurement, in which case dynamically trackinga statistic such as minimum, mean, or standarddeviation could provide insight into the data.

Figure 1. Passive and active packet timing probes using IEEE 1588 PTP pack-ets as probe packets.

GPS

GPS

Ethernet tap

PDV measurement andanalysis software

Network

1588 Grandmaster

1588 slave

Active probe

PDV measurement and analysis software

Passive probe

GPS

Figure 2. Packet time stamp pairs, and the corresponding forward and reversepacket delay.

Forward Reverse

R,00162; 1223305830.478035356; 1223305830.474701511 F,00167; 1223305830.488078908; 1223305830.490552012 R,00163; 1223305830.492882604; 1223305830.489969511 F,00168; 1223305830.503473436; 1223305830.505803244 R,00164; 1223305830.508647148; 1223305830.505821031 F, 00169; 1223305830.519029300; 1223305830.521302172 R,00165; 1223305830.524413852; 1223305830.521446071 F,00170; 1223305830.534542972; 1223305830.536801164 R,00166; 1223305830.540181132; 1223305830.537115991 F,00171; 1223305830.550229692; 1223305830.552551628

#Start: 2009/10/06 15:10:30 0.0000, 2.473E-3 0.0155, 2.330E-3 0.0312, 2.273E-3 0.0467, 2.258E-3 0.0623, 2.322E-3

#Start: 2009/10/06 15:10:30 0.0000, 3.334E-3 0.0153, 2.913E-3 0.0311, 2.826E-3 0.0467, 2.968E-3 0.0624, 3.065E-3

Packet delay sequence

Packet Timestamps



PACKET DELAY DISTRIBUTION ANDSUMMARY STATISTICS

An example of a histogram formed from a packetdelay sequence (PDV histogram) is shown in Fig.3. This figure shows the raw packet delay sequencefrom the six-day measurement taken on a produc-tion Ethernet network spanning hundreds of kilo-meters (Fig. 3b) and the corresponding histogram(Fig. 3a). Statistics based on this distribution areshown to the right of the histogram plot. Themean is 4.55 ms, the standard deviation is 13.26μs, and the peak-to-peak is 145 μs.

The shape of the distribution is an importantaspect of this approach to the analysis. Of partic-ular note is the asymmetry; this histogram isessentially a one-sided distribution, that is, thetail exists only to the right since packet transitdelay is limited to some minimum value and areconcentrated there. There is a much greaterprobability that packets experience a shortertransit delay than a longer one. Such a one-sideddistribution is fairly common for a one-waypacket delay measurement. Sometimes particularnetwork equipment — such as firmware-basedenterprise routers — will alter this, as will highlevels of load, particularly when a networkbecomes congested. In that case, very few pack-ets if any are able to traverse the network in theminimum possible time, and the distribution willexhibit greater symmetry.

DYNAMIC PACKET DELAY STATISTICS

Packet networks are, of course, dynamic bynature, and stationarity cannot be assumed ingeneral, particularly in the long term. While con-ditions in the packet delay sequence (Fig. 3b)appear to be fairly constant throughout the sixdays, tracking standard deviation over 100 sintervals shows variations in traffic load reveal-ing clear, repeating 24-h cycles (Fig. 3c). Thecycles peak at approximately 4 p.m. local timeand bottom out at approximately 8 a.m. localtime.

FREQUENCY TRANSPORT (ONE-WAY)PACKET DELAY METRICS

When a histogram is constructed from a packetdelay sequence, any temporal characteristics pre-sent are hidden since the time a sample occurredis not preserved; the sample value is renderedinto a bin count somewhere in the histogram.Tracking a statistic as described above begins toreveal temporal characteristics, but one canimagine that some of the stability analysis pro-vided by such calculations for TIE data asADEV, MDEV, MTIE, and TDEV or relatedcalculations could be useful for packet delaysequence analysis.

Systematic noise processes appear in simple

Figure 3. a) Histogram of a six-day packet delay sequence; b) plot of the packet delay sequence (PDVphase) over six days; c) plot tracking standard deviation of the PDV phase over the six days.

12.0 hours/div

4.5 ms 4.7 ms

PDVhistogram

PDVphase

PDVphasestddev

μ = 4.55 msσ = 13.26 μsρ-ρ = 145 μs

τ = 100 s

Symmetricom TimeMonitor Analyzer; Production Network; 2009/03/27; 13:53:33

100

1

(a)

(b)

(c)

1000

10

4.7 ms

4.5 ms

0.0days

6.4days

22 μs

2 μs

The shape of the dis-

tribution is an impor-

tant aspect of this

approach to the

analysis. Of particular

note is the asymme-

try; this histogram is

essentially a one-

sided distribution,

that is, the tail exists

only to the right

since packet transit

delay is limited to

some minimum

value and are con-

centrated there.



and complex networks because of both the tech-nologies employed, and the characteristics anddesign of the network devices such as switches,routers, access equipment (e.g., digital subscriberline [DSL] and gigabit passive optical network[GPON]), and transport equipment (e.g.,microwave systems). If a packet clock is to recov-er frequency optimally in such a situation, anunderstanding of these temporal characteristicsis important.

Just like packet slave clock algorithms mustselect and reject packets to optimize timing per-formance, analysis algorithms themselves canprovide greater insight through packet selection.Modification to two stability metrics used forTIE analysis, TDEV and maximum average timeinterval error (MATIE, also known as ZTIE),has proved fruitful for packet timing data analy-sis [6]. These are parallel to the standard TDEVand MTIE metrics used for TIE data analysis;TDEV focuses on systematic noise, and MTIE/MATIE focus on frequency offsets. For packettiming data, large short-term movement reducesMTIE to an essentially single-dimensional quan-tity, a flat line or nearly flat line at the overallpeak-to-peak phase; this is the rationale forchoosing the averaging in MATIE for packettiming data. A normalized version of MATIEcalled maximum average frequency error(MAFE) is derived from MATIE by dividingeach MATIE(τ) value by the applicable τ.

Two approaches have been taken to incorpo-rating packet selection into these calculations,preprocessed packet selection and integrated packetselection. A number of packet selection algo-rithms are possible. Three are described in detailbelow in the discussion of preprocessed and inte-grated packet selection: min, percentile, andband. All three can be applied using either thepreprocessed or integrated approach. Thesepacket selection methods are related to theclient/slave algorithms themselves. In order for apacket slave to optimally recover time and fre-quency, packet selection is essential. In this way,calculations with packet selection provide insightinto how a packet slave algorithm might performunder the studied network conditions.

PREPROCESSED PACKET SELECTIONWith preprocessed packet selection, a packetselection algorithm is applied to the packet delaysequence (PDV phase), and the standard TDEV,MATIE, or MAFE calculation is run with themodified packet delay sequence. If x is the origi-nal packet delay sequence, and x′ is the modifiedpacket delay sequence, the idea is that TDEV,MATIE, or MAFE operates on the x′ sequencerather than original x sequence.

There are several important parameters oroptions that must be chosen when applying pre-processed packet selection. One is the selectiontime window duration. Another is whether ornot to overlap the time windows. As an example,a 100 s time window might be chosen with non-overlapping windows. If the measurement dura-tion is 50,000 s, the sequence would be dividedinto 500 segments of 100 s, with each segmentproducing a single value based on packet selec-tion. If overlapping windows are chosen instead,more points are generated in the x′ sequence. If

the window advances point by point, there arenearly as many points in the x′ sequence as therewere in the original x sequence.

As indicated above, a number of packet selec-tion algorithms are possible. A fairly simple one,min, effective for sequences with a good popula-tion of minimum delay packets, is minimumselection. Drawing on the 100 s time windowexample above, and assuming a 64 Hz probepacket rate, the minimum packet delay would beselected from 6400 samples in a 100 s window,with each selected minimum over 100 s produc-ing a single sample in the x′ sequence.

A similar selection algorithm, percentile, takesa number of packets at or near the minimum,and averages them together to produce an x′sequence sample. For example, the minimum 10packet delay samples could be found in a 100 swindow and averaged together.

Generalizing the percentile selection methodfurther, the band selection method first sorts allthe data in a time window from minimum tomaximum, then selects a cluster of points atsome chosen range, say between the 20th and30th percentiles, and averages them together toproduce a sample in the x′ sequence.

INTEGRATED PACKET SELECTIONTDEV and MATIE (and by extension MAFE)all contain averaging as a component of the cal-culation. Integrated packet selection replaces theaveraging with a selection process such as themin, percentile, and band described above. As aresult a new self-contained metric is formed suchas minTDEV, percentileTDEV, bandTDEV,minMAFE, percentileMAFE, or bandMAFEThese metrics and packet selection procedureshave been discussed in the Alliance for Telecom-munications Industry Solutions (ATIS) andInternational Telecommunication Union (ITU),and are included in recently published docu-ments [6, 7]. The metrics based on the bandselection process are the most general; in fact,the other selection processes and the standardcalculation are all special cases of the band met-ric. As an example, consider the TDEV group.Standard TDEV is bandTDEV with indexingbased on the 0 and 100 percentiles. The min-TDEV metric is bandTDEV with both indiceschosen at the 0 percentile. The percentileTDEVcalculation is bandTDEV with the lower indexbased on the 0 percentile.

To clarify this, definitions of TDEV, minT-DEV, and bandTDEV are shown in Eqs. 1, 2,and 3, which also serve to clarify the min andband selection methods. For the bandTDEV cal-culation as defined below, the x′ sequence repre-sents the sorted phase sequence from minimumto maximum over the range i ≤ j ≤ i + n – 1. Theindices a and b are set based on two selectedpercentile levels, A and B. The averaging is thenapplied to the x′ variable indexed by a and b.The number of averaged points m is related to aand b: m = b – a + 1.

(1)TDEVn

xn

xn

xi ni

n

i ni

n

ii

nτ( ) = − +

⎡

⎣+

=+

= =∑ ∑ ∑

1

6

12

1 12

1 1 1⎢⎢

⎤

⎦⎥

2

Systematic noise pro-

cesses appear in sim-

ple and complex

networks both

because of the tech-

nologies employed

and because of the

characteristics and

design of the net-

work devices such as

switches, routers,

access equipment

(DSL and GPON for

example), and trans-

port equipment

(microwave systems

for example).



(2)

(3)

Again, relating the other TDEV calculationsto bandTDEV:1)TDEV is bandTDEV (0.0 to 1.0)2)minTDEV is bandTDEV (0.0 to 0.0)3)percentileTDEV is bandTDEV (0.0 to B)

with B between 0.0 and 1.0The effectiveness of packet selection is illus-

trated in the two measurement examples shownin Fig. 4. The first measurement (Fig. 4a) wastaken on a laboratory network composed of fivecarrier class switches with multiple stream trafficgeneration composed of small, medium, andlarge packets: 64 bytes, 576 bytes, and 1518 bytes,respectively. In this example, where 80 percent ofthe packet delay data is at or very near the mini-mum, minTDEV noise levels are below TDEVnoise levels for all values of integration time τ.Clearly a packet clock algorithm based on mini-mum packet selection would likely have greatadvantages over one that utilizes all data withoutany packet selection in this situation.

Similarly, in a measurement taken on a net-work of cascaded routers (Fig. 4b), where thepacket delay data is concentrated in a bandabove the minimum, advantages can be achievedby focusing the TDEV analysis on that band.This is accomplished using the bandTDEV cal-culation with appropriate choice of lower andupper indices based on chosen lower and upperpercentiles. In this case, the 40th and 60th per-centiles were chosen. In so doing, the bandT-DEV calculation shows lower noise than thestandard TDEV calculation throughout most ofthe range of integration time τ.

TIME TRANSPORT (TWO-WAY)PACKET DELAY METRICS

Packet-network-based time transport requiresmore than the assessment of variations in packetdelay required for frequency transport. Whilepacket network frequency transport can benefitfrom a two-way protocol, a stream in a singledirection could be used. Time transport, by con-trast, requires a two-way protocol in order toestimate one-way packet delay using a measure-ment of a round-trip. In such a situation, asym-metry between the upstream and downstreampaths has a direct impact on the ability to accu-rately transport time. Thus, certainly one of thegoals of two-way transport metrics is to assessasymmetry.

The measurement setup for two-way packettiming demands that both forward and reversepacket flows are measured at the same time andthat there is a common, stable time reference atboth measurement points. The one-way mea-surement for the frequency transport metricsrequires only that the two clocks run at the samerate since packet delay variation is the criticalquantity. Further, for frequency transport, one-way packet flows can be studied independently.

TWO-WAY PACKET TIMINGMEASUREMENT DATA

The first step toward performing two-way packettiming analysis is the construction of a two-waydata set from a simultaneously measured pair ofupstream and downstream one-way packet delaysequences. The procedure for doing so is out-lined in the top and middle parts of Fig. 5.

The forward and reverse packet delaysequences shown in Fig. 5 are themselves derivedfrom four time stamps, two time stamp pairs,such as those shown in the top part of Fig. 2.

bandTDEV x i n x i n x ibm bm bmτ( ) ( ) ( ) ( )[ ]= + − + +1

62 2

2

where ⋅ = ′( ) +=

∑x im

xband mean j ij a

b

_1

min min min minTDEV x i n x i n x iτ( ) ( ) ( ) ( )[ ]= + − + +1

62 2

2

where for ⋅ = ≤ ≤ +( ) ⎢⎣ ⎥⎦x i x i j i njmin min −−( )1

Figure 4. a) PDV histogram followed by TDEV compared to minTDEV fordata with a well-populated minimum; b) TDEV compared to bandTDEV fordata concentrated above the minimum.

40 μsec 84 μsec

100

1

10k

1.0 s

10 μs

1 ns10 s 100 s 1 ks 10 ks

Symmetricom TimeMonitor Analyzer, Xli 1588 PDV Phase; 2006/10/09; 20:59:41

Symmetricom TimeMonitor Analyzer, TP5000 Fwd PDV Phase; 2008/10/17; 01:30:27

0.0 hours 2.0 hours30 μs

90 μs

1.0 s 10 s0.1 μs

10 μs

1 μs

100 s 1 ks

TDEV

PDVHistogram

minTDEV

TDEV

bandTDEV

PDV



Based on the time when individual samplesoccur, forward and reverse packet delaysequence samples are combined into sampleswith three components — time, forward packetdelay, and reverse packet delay — for each sam-ple. Organizing the data in such a way makes itconvenient for performing two-way packet tim-ing calculations to do such things as investigateasymmetry in packet transport.

TWO-WAY PACKET TIMINGBASIC CALCULATIONS

There are two fundamental quantities for two-way packet timing analysis, round-trip and offset.These form the basis for the metrics involvingpacket selection discussed below in the next sec-tion. The round-trip calculation sums the for-ward and reverse packet delay, and the offsetcalculation takes the difference of the two.

It is useful to consider these from the per-spective of one-way delay, since one-way packetdelay is often estimated as half of the round-trippacket delay for devices that can only measureround-trip delay. The bias produced by a packetalgorithm is related to half the full offset sincethe client/slave device is dividing its locally madefull round-trip measurements by two. Forinstance, if the actual forward and reverse one-way delays in a network were 99 μs and 101 μs, apacket slave would estimate a one-way delay of100 μs, which would result in a 1 μs bias for eachdirection, half of the 2 μs difference.

This kind of normalization is accomplished bydividing the full round-trip and full offset bytwo. These will be referred to as normalizedround-trip and normalized offset. The importanceof this normalization for the offset calculation isillustrated by the measurement example shownin Fig. 6 where a normalized packet offset pre-dicts the 2 μs bias in the slave device.

TWO-WAY PACKET TIMING CALCULATIONSWITH PACKET SELECTION

Just as one-way packet timing analysis can bene-fit from packet selection, so can two-way analy-sis. Figure 5 shows construction of a new set oftwo-way data by applying min packet selection toboth the forward and reverse packet delaysequences. In this case, for the purpose of illus-tration, a time window is set to three samples,and non-overlapping windows are used. Theresulting new data set is thus a third the size ofthe original data set.

Normalized offset and normalized round-tripcan be calculated using a min packet selectionversion of two-way data producing new metricsminRoundtrip and minOffset. These are then thebasis for other metrics.

One such metric is produced by plottingminOffset against minRoundtrip as a scatter plotto form minimum time dispersion (minTDISP).Such a relation does not produce a plot withunique values of minRoundtrip mapping to sin-gle values of minOffset — that is, this is not aplot of minOffset as a function of minRoundtrip.This is why this data is best plotted as a scatterplot rather than connecting the dots.

As is the case for all calculations based on

minOffset, a time window must be chosen asinput parameter. The calculation could berepeated for different time windows each ofwhich would produce a different minTDISPscatter plot.

The ideal situation for minTDISP is a clusterof data sitting on the x-axis with little dispersioneither above or below the x-axis (little variationin minOffset away from zero) and little disper-sion along the x-axis (limited variation in min-Roundtrip).

The two-way packet measurement in Fig. 6,taken on an Ethernet network used for wirelessbackhaul, is represented as a minTDISP plot(Fig. 6a). The apex of the minTDISP data con-verges to a minOffset of –2 μs. Thus, a clearasymmetry between forward and reverse chan-nels is seen in this packet network measurement.A measurement of a slave clock operating underthese asymmetrical conditions (Fig. 6b) not sur-prisingly shows the same 2 μs error when recov-ering time.

Other two-way metrics have been proposedand are being studied in industry and withinstandards bodies. Since asymmetry is of particu-lar interest, in many cases these are based onminOffset. Examples of these are minOffsetmean, minOffset standard deviation, and minOff-set percentile plotted as a function of time win-dow.

In contrast to minTDISP, where a particularpacket selection time window is chosen inadvance of performing the calculation, theseminOffset mean/standard deviation/percentilecalculations start with a small time window toproduce a single numerical result, and thenincrease the time window to produce additionalnumerical values, finally plotting these resultsagainst time window value.

The values themselves are a chosen statisticof a minOffset sequence such as mean, standard

Figure 5. Constructing a two-way data set from forward and reverse packetdelay sequences and then f′ and r′ from f and r (the two-way data set) with a3-sample time window.

#Start: 2010/03/06 17:15:30 0.0000, 1.47E-6 0.1000, 1.54E-6 0.2000, 1.23E-6 0.3000, 1.40E-6 0.4000, 1.47E-6 0.5000, 1.51E-6

#Start: 2010/03/06 17:15:30 0.0000, 1.47E-6, 1.11E-6 0.1000, 1.54E-6, 1.09E-6 0.2000, 1.23E-6, 1.12E-6 0.3000, 1.40E-6, 1.13E-6 0.4000, 1.47E-6, 1.22E-6 0.5000, 1.51E-6, 1.05E-6

Time(s) f(μs) r(μs) f’(μs) r’(μs)0.0 1.47 1.110.1 1.54 1.09 1.23 1.090.2 1.23 1.120.3 1.40 1.130.4 1.47 1.22 1.40 1.050.5 1.51 1.05

Two-way data set

Constructing f’ and r’from f and r with a 3-sample time window

Minimum search sequence

Forward packet delay sequence #Start: 2010/03/06 17:15:30 0.0000, 1.11E-6 0.1000, 1.09E-6 0.2000, 1.12E-6 0.3000, 1.13E-6 0.4000, 1.22E-6 0.5000, 1.05E-6

Reverse packet delay sequence



deviation, or 95th percentile. Extensions to thesecan be imagined as integrating other packetselection techniques such as percentile or band;another possible extension would be to plotsome other statistic such as one based on theAllan deviation family of metrics against timewindow.

CONCLUSIONSAs this article has discussed, a number of newmetrics are currently being studied for the analy-sis of packet delay variation data. Just as was thecase for traditional synchronization measure-ments, these metrics serve several purposes.First, they are tools for gaining insight into thebehavior of the timing characteristics of packetnetworks. Second, some of them, perhaps inconjunction with other related new metrics, canform the basis for setting packet network limits,much has been done with MTIE and TDEV fortraditional synchronization measurements. Atthis stage, standards committees have beenstudying and defining the new packet metrics.

The next step will be to use these metrics todefine limits based on application requirementsfor the synchronization signals produced by thepacket client/slave device and understanding ofthe essential characteristics of the packet algo-rithms that convert packet timing into synchro-nization signals. It is worth pointing out thatthese synchronization signals themselves can bemeasured in the traditional way with networklimits such as MTIE and TDEV masks with thechoice of mask based on the particular applica-tion requirements. For example, if an applicationdemands better than 15 parts per billion fre-quency offset, the measurement of the packetslave synchronization output signal can be ana-lyzed for this.

As discussed above, these new metrics incor-

porating packet selection provide insight intopacket algorithms since the algorithms them-selves rely on packet selection to produce opti-mal timing performance. The need forsynchronization in packet networks arises as thetelecommunications infrastructure migrates fromcircuit-switched networks to packet-switched net-works. Applications relying on precise time andfrequency drive the requirement for packet-based synchronization and the necessity forpacket network timing analysis.

REFERENCES[1] S. Bregni, Synchronization of Digital Telecommunica-

tions Networks, Wiley, 2002.[2] K. Shenoi, Synchronization and Timing in Telecommuni-

cations, BookSurge Publishing, 2009.[3] ITU-T Rec. G.810 “Definitions and Terminology for Syn-

chronization Networks,” 1996.[4] L. Cosart, “Precision Packet Delay Measurements Using

IEEE 1588v2,” ISPCS ‘07, Vienna, Austria, Oct. 2007.[5] L. Cosart, “Packet Network Timing Measurement and

Analysis Using an IEEE 1588 Probe and New Metrics,”ISPCS ‘09, Brescia, Italy, Oct. 2009.

[6] ATIS-0900003.2010 Technical Report “Metrics Charac-terizing Packet-Based Network Synchronization,” 2010.

[7] ITU-T Rec. G.8260 “Definitions and Terminology forSynchronization in Packet Networks: Appendix I,” 2010.

ADDITIONAL READING[1] M. Weiss, “Time Domain Representation of Oscillator

Performance,” 32nd Annual Time & Frequency Metrolo-gy Seminar, Boulder, CO, June 2007.

BIOGRAPHYLEE COSART ([email protected]) is a senior technol-ogist with Symmetricom, Inc. A graduate of Stanford Uni-versity, he worked as an R&D engineer at Hewlett-Packard/Agilent prior to joining Symmetricom in 1999. His R&Dactivities have included measurement algorithm develop-ment and mathematical analysis for a variety of testequipment for which he holds several patents. He serveson the ATIS and ITU-T committees responsible for networksynchronization standardization as chair, contributor, andeditor.

Figure 6. a) Ethernet wireless backhaul asymmetry; b) IEEE 1588 slave 1PPS under these asymmetrical network conditions.

Symmetricom TimeMonitor Analyzer; Ethernet Wireless Backhaul; 2009/04/28; 11:37:01

MinTDISP

270.0 μs265.6 μs

-2.0 μs

(a)

(b)

0.5 μs/div

-6.0 μs

1588Slave1 PPS

vs. GPS

22.7 hours0.0 hours

2.0 μs

0.5 μs/div

-1.0 μs

The need for syn-

chronization in pack-

et networks arises as

the telecommunica-

tions infrastructure

migrates from circuit-

switched networks

to packet-switched

networks. Applica-

tions relying on pre-

cise time and

frequency drive the

requirement for

packet-based syn-

chronization and the

necessity for packet

network timing

analysis.

COSART LAYOUT 1/21/11 10:09 AM Page 154


BACKGROUND

The principal requirement for synchronizationand timing in telecommunications networks is tosupport real-time services such as voice andvideo communications, particularly of an interac-tive or conversational nature. For example, forsignals such as speech, voice-band (modem/fax),and video, the information signal (analog) at thesource end is converted into digital format at aparticular sampling rate. At the destination thereis a conversion back to analog, and if the conver-sion rates are not equal, the quality of experi-ence (QoE) degrades.

In traditional circuit-switched telecommunica-tions networks the need for network-wide syn-chronization arises to ensure propercommunication of information (bits) itself, andarises from the nature of the methods of multi-plexing and switching employed. In time-divisionmultiplexing (TDM) networks the transmittedsignals themselves are often suitable for carryingtiming information [1]. This enables the creationof a synchronization network with each nodeprovided timing information traceable to a pri-mary reference clock (PRC) [1, 2]. In traditionalnetworks the service requirements are piggy-backed on transport requirements and therebyavailable by default.

In next-generation networks (NGN) that arebased on packet switching and statistical multi-

plexing principles, network-wide synchronizationis not required to keep the transmission pipesfrom operating in a bit-error-free mode. Howev-er, the need for synchronization remains for theservices provided, particularly in the case of real-time signals and the circuit emulation of con-stant bit rate data streams. Whereas goodsynchronization will improve the functioning of apacket network as a whole, good timing is essen-tial at all points in the network at which serviceis delivered or where a format conversion isrequired for converting between packet-switchedand circuit-switched formats. Equipment per-forming such C2P functions is referred to as aninterworking function (IWF) or gateway, and syn-chronization is required to reliably support real-time applications such as circuit emulationservices (CES), voice over IP, video over IP,IPTV, and so on.

The notion of a synchronization network canbe achieved in packet networks by embeddingtiming information in the physical layer, asexemplified by Synchronous Ethernet. Alterna-tively, timing information can also be carried ata higher layer as exemplified by methods basedon Precision Time Protocol (PTP) [3] or Net-work Time Protocol (NTP) [4].

Quantifying timing requirements requires thedevelopment of suitable metrics and analyticalmethods. These and related performance aspectsare the subject of this article. In the ensuing sec-tions a brief overview of timing fundamentals isprovided, followed by an explanation of howpacket-based methods transfer timing. Twogroups of metrics, the TDEV and MTIE families,are discussed to clarify how they quantify thedeparture of a timing signal from ideal.

TIMING FUNDAMENTALSITU-T Rec. G.810 [5] defines a clock as “anequipment that provides a timing signal” andfurther explains that in the context of telecom-munications networks a clock can be viewed as asignal generator that provides the appropriatesignals to other devices in the network, effective-ly synchronizing the network. This definition canbe extended to include the subsystem within anetwork element that governs the temporalbehavior of other functions of the network ele-

ABSTRACT

Circuit-switched networks based on time-divi-sion multiplexing require synchronization todeliver information, whereas packet-switchednetworks can deliver information in an asyn-chronous environment. However, all real-timeservices require that synchronization and timinginformation be delivered over the network. Per-formance of timing distribution is quantifiedusing particular metrics and adherence torequirements determined by using masks. Thetraditional metrics, TDEV and MTIE, haveextensions to packet-switched networks foraddressing the corruption of timing informationby packet delay variation. The principles of met-rics and masks and these extensions are present-ed here.


Kishan Shenoi, Shenoi Consulting

Performance Aspects of Timing inNext-Generation Networks

SHENOI LAYOUT 1/19/11 3:37 PM Page 156


ment, especially related to output signals andfunctions such as analog-to-digital and digital-to-analog conversions. Implicit in the notion of atiming signal is the information that can be uti-lized to generate a clock signal, most often pic-torially viewed as a pulse train where a risingedge of the waveform identifies the instant ofinterest.

CLOCK NOISE AND PACKET DELAY VARIATIONThe notion of clock noise, or time error, is illus-trated in Fig. 1a. When the principal item ofinterest is the frequency of the clock, the noise inthe clock output is obtained as the time (phase)difference between the timing edge of the clockrelative to the corresponding edge of an ideal orreference clock that is known to be of muchhigher quality than the clock under test. Thesequence {x(n)} (an equivalent notation is {xn})represents the time error. In practice it is a mea-sured quantity and various metrics can be com-puted on the vector {x(n); n = 0, 1, 2, …, (N –1)}. When the principal characteristic of interestis time (sometimes referred to as wall-clock), thenotion of time error is simply the differencebetween wall-clock reading of clock under testand the reference clock. In such a case the timeinterval between samples of this time error com-putation need not necessarily be uniform but itis important that the wall-clock reading apply tothe same instant in time.

In packet-based methods the timing infor-mation is based on the time of arrival and timeof departure of packets. If Tn is the time ofdeparture of timing packet #n, and the time ofarrival at the destination is Sn, the transit time

of the packet is simply v(n) = (Sn – Tn). Thepacket delay variation (PDV) is defined as thedifference between the actual transit delay anda reference transit delay. There are other defi-nitions of PDV that are targeted to other appli-cations [6]. The specific choice of referencedelay is application-dependent, and commonchoices are delay of first packet of session, mini-mum delay, average delay, and so on. The PDVsequence is analogous to the time errorsequence and the same collection of metricscan be computed.

In traditional TDM architectures physicallayer methods are employed to deliver timinginformation. For example, the transmissionschemes in optical networks based on syn-chronous digital hierarchy (SDH) areisochronous with suitable signal features thatallow the receiver of the signal to extract a recov-ered clock that represents the characteristics,from a timing perspective, of the clock employedin the transmitter.

In packet-based methods, a packet timingpacket flow is established between the master(source of timing) and slave (recipient of tim-ing). In fact there could be packet flows in eitheror both directions. Timing information in thisscenario is comprised of the time-stamps associ-ated with the instant the packet leaves one clockand arrives at the other. Clock noise is intro-duced in the form of packet delay variation andthe effects of imprecision in the manner in whichtime-stamps representing the times of arrivaland departure are struck. Nevertheless, even inpacket-based methods, a clock output is generat-ed that can be of the form in Fig. 1a.

Figure 1. Illustrating the notion of clock noise (time error and packet delay variation).

Time-of-departure

Time-of-arrival

τ0

Sn

Tn

Reference(ideal)

Clockoutput

(a)

(b)

(n-1)

x(n-1) x(n)

v(n) v(n-1) v(n+1)

x(n+1)

(n) (n+1) Implicit in the notion

of a timing signal is

the information that

can be utilized to

generate a clock

signal, most often

pictorially viewed as

a pulse train where a

rising edge of the

waveform identifies

the instant of

interest.



CLOCK NOISE MODEL

The time error is modeled as a combination ofsystematic and random components andexpressed as:

x(t) = α + β ⋅ t + (0.5 ⋅ D ⋅ t2 + φ(t)) (1)

Systematic components include a constant, α,(α is the initial phase error); a possible linearfunction of time, β ⋅ t (β is the fractional frequen-cy offset); and in some cases a quadratic term,0.5 ⋅ D ⋅ t2 (D is the frequency drift). For quanti-fying the error in frequency alignment any con-stant phase error (i.e., α) can be ignored; fortime alignment (or phase alignment) the con-stant term should be (nominally) zero. A similarmodel can be postulated for delay in packet net-works. However, if we assume that the networkdoes not create or destroy packets, then on along-term basis β and D should be zero. Thisassumption is valid since packets used for timingpurposes can be sequence numbered, and conse-quently duplicate and/or lost packets can beidentified.

Clearly, the ideal value of time error (orpacket delay variation) is identically zero. Alltiming metrics quantify the non-zero behavior ofthe time error. Note that the time error is essen-tially a random process and, in practice, metricsare computed on the (measured) time errorsequence {x(n)}. Thus the computed metrics areakin to estimates of statistical expectations.

An implicit assumption is that the perfor-mance estimates based on data obtained fromone sufficiently long experiment is adequate tocharacterize the statistical behavior of the clockoutput or timing signal. This is definitely true forstatic conditions such as in the study of thebehavior of traditional (TDM) network equip-ment clocks. For dynamic conditions such asthose encountered in packet networks the packetdelay variation may not be a stationary process.Nevertheless, metrics are invaluable for purposesof analysis. However, the analytical thought pro-cesses involved are somewhat more complex thanin the static case and require a good understand-ing of the metric, the strengths and weaknesses,and areas of applicability. Additional explanationand observations regarding metrics for evaluatingtiming performance are available in [1, 7–9].

PACKET-BASED TIMING METHODSPhysical layer methods used in traditional TDMnetworks are well documented [1, 7]. Here theemphasis is placed on methods suitable forNGNs and are based on transfer of packetsbetween the two devices.

Precision Time Protocol (PTP) [3] and Net-work Time Protocol (NTP) [4] are two protocolsthat are being studied for use in packet-basedtiming methods for telecommunications applica-tions. From a time-transfer perspective the twomethods are identical in principle. Both proto-cols specify mechanisms for communicating tim-ing information comprising the time-of-arrivaland time-of-departure of designated packets andprovide specific formats for the time-stampsused to convey this timing related information.

Synchronization of slave to master necessi-tates that the notion of 1 second be the same (ornearly so) in both devices. This implies that thetwo are syntonized (aligned in frequency). Syn-tonization implies that the wall-clock valuesprogress at the same rate and thus the wall-clockdifference, or time offset, will be constant.Whereas there are other methods of ranging toobtain the time offset (ε), PTP and NTP devicesestimate ε as one-half the round-trip delay. Thatis, there is an implicit assumption that the delayis the same in the two directions. Any asymmetrywill result in an error in the offset estimate (εerr).Figure 2 indicates the calculation in the case of asingle exchange of packets. In practice a flow oftiming packets is set up between the devices inorder to address packet delay variation and localoscillator drift.

Delivery of timing information across a pack-et network is based on some fundamentalpremises:• Every path between source and destination

has the notion of a nominal delay. Thedelay can change from packet to packet andit is this packet delay variation (from nomi-nal) that introduces the clock noise thatimpacts clock recovery.

• Missing or duplicated packets can be detect-ed. Consequently, on a long-term basis atleast, the rate at which the packets leavethe source matches the rate at which pack-ets are (perceived) received at the destina-tion.

• The local timekeeping mechanism is basedon a local oscillator that is reasonably sta-ble. Generally speaking, a longer observa-tion interval provides a better estimate ofthe appropriate oscillator frequency correc-tion. This is true provided the local oscilla-tor remains constant over the interval. Anydrift in the oscillator (frequency) will cor-rupt the estimate.

• The network attempts to deliver each pack-et, individually, in a timely manner. Store-and-forward schemes can protect thecontent of packets but could disrupt thetiming information.

• For clock recovery purposes, it is permissi-ble to select, from the population of pack-ets, a subset that exhibit particularproperties. This selection process is part ofthe “secret sauce” of algorithms.A common misconception is that timing infor-

mation is carried solely by time-stamps embed-ded in the packet directly or by an associationmechanism. Actually time-stamps are just oneaspect, albeit a very important aspect, of timinginformation. The time-of-departure and time-of-arrival instants are equally important and timinginformation implies the association of time-stamps with these events. Any error, or lack ofprecision, in associating the time-stamp and rep-resentative event is not distinguishable from net-work-introduced packet delay variation.

THE CLOCK RECOVERY FUNCTIONIn order to better appreciate the role of metricsand masks, a brief explanation of the principlesof clock recovery is provided. In essence, clock

The time-of-depar-

ture and time-of-

arrival instants are

equally important

and timing informa-

tion implies the

association of time-

stamps with these

events. Any error,

or lack of precision,

in associating the

time-stamp and

representative event

is not distinguishable

from network-intro-

duced packet delay

variation.



recovery can be viewed as the process of regen-erating a clock that matches the source (master)by filtering out the accumulated clock noise inthe timing reference.

The general structure of the signal processingassociated with clock recovery is depicted in Fig.3. The slave generates a control signal for adjust-ing its local time-base by comparing its output tothe timing reference extracted from the signalsent by the master. In the TDM case, the refer-ence is obtained by extracting a recovered clockfrom the physical layer transmission medium; inthe packet-based methods the reference isextracted based on time-stamps in the packetand time-of arrival/departure measurementsbased on the local clock. This model of a lockedloop (e.g., phase-locked) is common to all clockrecovery schemes.

It is important to observe that in both theTDM and packet-based cases, the clock noise ofthe master and transit-time variation introducedby the transmission medium affect the slaveclock in much the same manner, in principle.Two important differences are the rate of trans-fer, typically very high in TDM, and scale oftime error, usually much higher in packet-basedmethods (see [7, 9] for more related informa-tion).

From a signal processing perspective, theloop presents a low-pass filter characteristicbetween the timing reference input and theclock output; between the local oscillator andoutput the effective transfer function is high-passin nature. The high-pass and low-pass character-istics have the same (nominal) cut-off frequency.The implication is that very narrow-band low-pass characteristics require high performanceoscillators. In actual packet-based clock recoveryalgorithms, sophisticated packet pre-processingtechniques are employed to reduce the power ofthe clock noise in the timing reference as seen

by the loop and thereby, albeit to a limitedextent, relax the oscillator performance require-ments (and therefore cost).

One consequence of this low-pass behavior isthat the clock noise in the output contains allthe low-Fourier-frequency components presentin the timing reference fed to the loop. Thisimplies that the tolerance limit of the loop andthe clock output requirements are almost thesame for large observation intervals (i.e., lowFourier frequency) — some allowance must beprovided for locally generated noise. The obser-vation interval that quantifies large is related tothe bandwidth (or time-constant) of the clockrecovery loop.

THE TDEV FAMILY OF METRICSTime Variance (TVAR) and its square-root,Time Deviation (TDEV) are metrics that havebeen used within the timing community for sev-eral decades [1, 7]. TVAR can be interpreted asa spectral decomposition of the clock noisepower. Whereas a power spectrum is a functionof Fourier frequency, f, TVAR is a function of atime variable τ, usually referred to as the obser-vation interval and f and τ are (proportional to)the reciprocal of the other. The value of τ deter-mines the center frequency and bandwidth of afilter and TVAR(τ) is the power observed whenthe given time-error signal is passed through thisfilter. Assuming the nominal sampling interval isτ0, then TVAR (and TDEV) are computed forvalues of τ that are integer multiples of τ0 (i.e., τ= n ⋅ τ0).

TDEV, and other metrics of the TDEV fami-ly, provide insight into the stability of the timingsignal (see [1, 7, 8] for additional discussion onclock stability metrics). These are developed onmeasured values of the time error and in packetnetwork scenarios are based on measurements of

Figure 2. Principles of timing transfer over packet networks.

2

(time-of-departure)

@T1

Time-base Time-base Wall-clock Wall-clock

T ττ = T + ε

τ3 = T3 + ε

(τ2 – T1) – (T4 – τ3)

Master Slave

Transit delay = ΔMS

Transit delay = ΔSM

Slave time offset = ε

(time-of-departure)

ε ≅ εerr=;

@T3

τ2 = T2 + ε

(time-of-arrival) @T2

(time-of-arrival)

@T4

2

ΔMS – ΔSM

In actual packet-

based clock recovery

algorithms,

sophisticated packet

preprocessing

techniques are

employed to reduce

the power of the

clock noise in the

timing reference as

seen by the loop and

thereby, albeit to a

limited extent, relax

the oscillator perfor-

mance requirements

(and therefore cost).



the transit delay of (timing) packets over thenetwork between the master and slave. Thesemetrics are, generically, xTDEV(τ = n ⋅ τ0),where τ0 is the (nominal) sampling interval ofmeasurement, typically the (nominal) packetinterval, and are computed over the measure-ment data set comprised on N values, {x(kτ0); k= 0, 1, …, (N – 1)}.

For a given value of observation interval τ =n ⋅ τ0, a representative value, Xm, is developedfor the n samples in each observation windowspanning the indices k = m through k = (m + n– 1). The x in xTDEV identifies the manner inwhich this representative value is obtained.xTDEV is a measure of the standard deviation of{Xm} after removing any constant and linearcontribution as is clear from the double differ-ence (Xm+2 – 2 ⋅ Xm+1 + Xm) in Eq. 2 (see [7]for additional information).

(2)

For regular TDEV, the representative value iscomputed as the average of the n samples in theobservation interval. Different choices for x thathave been proposed include:• min (minTDEV): The representative value,

Xm, is the minimum of the n samples.• band (bandTDEV): given two probabilities

(expressed as percentages), α and β with β

> α, Xm is the average of the population ofsamples in the β-percentile of least delay ,after removing the samples in the α-per-centile of least delay, within the n samplesin the mth observation window. percentileT-DEV is a special case of bandTDEV with α= 0.

• cluster (clusterTDEV): in this case Xm is theaverage of the samples that fit a particularrule, called the clustering rule. That is, clus-terTDEV is a class of TDEV defined by aselection process. Numerous choices for theclustering rule can be proposed.Generally speaking, xTDEV provides a mea-

sure of the stability of the timing informationderived from the packet transit delay sequence ifthe rule x is applied in the selection process toselect a representative sample over an observa-tion interval of duration τ = n ⋅ τ0.

THE MTIE FAMILY OF METRICSConsider the hypothetical situation where data isbeing written into a buffer under control ofclock A and is being read out of the buffer undercontrol of clock B. If the two clocks are perfectlysynchronized then over any interval of time(observation interval) the buffer fill level willremain nominally constant. Any relative wanderbetween the two clocks will be reflected in avariation of the fill level and consequently exces-sive wander will result in buffer overflow orunderflow. The peak-to-peak variation of the

xTDEV n

X X XM

m m mm

τ τ= ⋅( ) =

⎛⎝⎜

⎞⎠⎟⋅ ⋅ − ⋅ +[ ]

−+ +

=

0

2 121

62

1

2 00

3M −

∑

Figure 3. Clock recovery processing.

(Controlled)oscillator

Clockoutput

Clock recovery function

Σ

Network noise (PDV)

PP: preprocessing

(a)

C

Master

(b)

IN

A

B

PLL function

(Servo function)

Loopfilter

Timingreference

ΣLPF Clock output

HPF

Local oscillator noise

PP

The notion of a

mask, as associated

with a metric, is

essentially the limit

beyond which the

equipment is

deemed to be

non-conformant to a

particular standard or

application. That is,

the mask is an

expression of what is

good enough.



(relative) time error then provides guidance onthe sizing of buffers and/or the propensity fordata loss. Maximum Time Interval Error (MTIE)provides a quantitative measure that indicatesthe peak-to-peak behavior as a function of obser-vation interval. Additional description, includingthe mathematical formulas, is available in [1, 7,10].

With some abuse of notation, denote byMT(n) or MT(τ) the MTIE value for observationinterval τ = n ⋅ τ0. MT(τ) is the maximum peak-to-peak variation over all intervals of duration≤τ. First, it is clear that MT(τ) is a monotonicallynon-decreasing function of τ. Second, if clockerror is dominated by systematic components(frequency offset and/or frequency drift) thenfor large τ we can say that MT(τ) exhibits a lin-ear/quadratic behavior. One very common situa-tion is where the two clocks being compared arevery stable (≈ zero frequency drift) but may havea small frequency offset, and in this situation therelationship MT(τ) ≈ β ⋅ τ applies for large obser-vation intervals. If the two clocks are locked thenthere is no frequency offset and the MTIEbecomes constant for large τ.

In traditional clock measurements the timeerror sequence is generally smooth and the non-decreasing property of MTIE is not an impedi-ment to understanding the relative wanderbetween the clocks. In the case where the timeerror represents a packet delay variation, it isnot uncommon to see sharp and rapid changesin the time error. Since it is known a priori thatthere will be some filtering of the time error inthe (slave) clock recovery system, a variant ofMTIE called MATIE has been proposed [7]. Thetime error is filtered using a first-order rectangu-lar-window low-pass filter prior to analysis andthis is the source of the A (for average) in theacronym MATIE. The metric is generally used toascertain whether there is any systematic compo-nent in the time error (i.e., PDV). Presence of afrequency offset will result in a linear behaviorof MATIE for large τ and the slope will be the(fractional) frequency offset. The metric Maxi-mum Average Frequency Error (MAFE) wasdefined to point out this behavior and provide anumerical value for this offset (in fractional fre-quency units); MAFE and MATIE are essentiallyequivalent.

MASKSThe notion of a mask, as associated with a met-ric, is essentially the limit beyond which theequipment is deemed to be non-conformant to aparticular standard or application. That is, themask is an expression of what is good enough.For example, an MTIE mask will correspond toa predetermined line on the log-log plot used fordisplaying MTIE; the measured MTIE should liebelow this line for compliance.

It is common to break down the masks intotwo major areas of study. These two areas relateto:• Clock related masks: In both cases, i.e.,

TDM or packet network architectures,there is an implicit or explicit notion of aclock signal that is used to time processesor other signals. Clock related masks apply

to these signals and determine whether theclock output is fit for purpose regardless ofwhether the network is packet-switched orcircuit-switched.

• PDV related masks: Networks add packetdelay variation that is, practically speaking,clock noise since it impairs the ability of thesink to recreate a suitable timing signal thatmimics the source. PDV related maskstherefore pertain to the measurement of anetwork and ascertain the ability of the net-work to support packet-based time/timingtransfer. Although peculiar to packet net-works, PDV metrics used for quantifyingthe strength of the PDV often have a paral-lel with a TDM counterpart. The develop-ment of PDV metrics and masks is a newarea of study in the industry and standardsare still in development.There are several masks available in stan-

dards documents. These are usually targetedtowards specific equipment or applications. Herewe address MTIE masks because that is oftenthe more relevant mask in the field. The TDEVbehavior is also important but requires muchmore subject matter expertise in order to inter-pret the TDEV results.

In all telecommunications networks there isthe notion of good enough in terms of frequencyaccuracy. In traditional TDM networks the goalis to have all network elements aligned to a pri-mary reference clock (PRC) whose innate frac-tional frequency accuracy is better than 1 ×10–11. However, clock performance requirementsat the network edge can be tailored to the appli-cation or end-point equipment. The most com-mon example of this situation is the case ofwireless base-stations that derive their timingfrom the backhaul signal. The primary require-ment in this case is (fractional) frequency accu-racy of better than 50 × 10–9, when measured(hypothetically) at the radio interface. To allowfor some margin in the base-station circuitry, itis common to impose a requirement of X on thetiming recovered from the network (backhaul

Figure 4. MTIE masks for 1 × 10–9 (1ppb) and 15 × 10–9 (15ppb) allowancecombined with the G.824 traffic mask.

MTI

E

Symmetricom TimeMonitor AnalyzerMTIE; Fo = 10.00 MHz; Fs = 1.000 Hz; 2010/03/18 13:41:54Simulated phase; Samples: 1000000; WhitePM: 1.000 μs; FlickerPM: 10.00 ns;RWalkPM: 10.00 ps

1ms

1μs

100μs

10ms

10μs

100ms

10.00s

100.0s

1.000s

10.00ks

100.0ks

1.000ks

G.824 TrafG.824T 1PPBG.824T 15PPB



signal). X is typically in the range between 1 ×10–9 and 15 × 10–9.

It is straightforward to generate a suitableMTIE mask that merges the traditional andapplication specific requirements. Suppose thetraditional network limit is expressed by a maskdescribed by the function μT(τ). It is clear that afractional frequency offset of y corresponds toan MTIE curve that is a linear function of τ,namely MT(τ) = y ⋅ τ. Then the appropriateMTIE mask that merges the two limits is givenby M(τ) where

M(τ) = max{MT(τ) ; μT(τ)} (3)

Wireless base-stations often derive their tim-ing from the backhaul signal and it is common toinvoke the conventional masks that are used inTDM for expressing network limits. However,these masks often have a 1 × 10–11 asymptotewhich may be overly stringent for the wirelesscase. Figure 4 shows the conventional trafficmask from ITU-T Rec. G.824 (for DS1) [11]that is modified to account for 1 × 10–9 and 15 ×10–9 offset allowances. The MTIE computed fora synthetic sequence is shown in Fig. 4 solely forillustrative purposes. The synthetic sequenceincludes a frequency offset of 5 × 10–9.

The notion of a tolerance mask is the follow-ing. If the PDV across a network is measured, itshould be possible to determine whether a slaveclock will be able to generate a clock output thatmeets a certain (application-specific or otherwiseprescribed) mask. To determine whether theslave clock will meet an output MTIE mask,there will probably be a tolerance mask at theinput that is based on a metric of the MTIE fam-ily. Likewise, a slave clock output TDEV maskwill determine a tolerance mask for a metric ofthe TDEV family. These tolerance masks areimportant since they actually determine networklimits for the PDV.

CONCLUDING REMARKSThe intent of this article is to provide an intro-duction to performance criteria associated withdelivering timing over packet networks. The cor-responding body of knowledge for TDM net-works is quite mature as exemplified by thenumerous standards available governing clockbehavior so it is possible to leverage the parallelbetween clock noise in TDM architectures andclock noise in packet-based methods as mani-fested in the form of packet delay variation. Thedevelopment of suitable standards for packet-based methods in telecommunications is anongoing activity in the ITU-T and ATIS stan-dards bodies.

The principal metrics used in TDM are MTIEand TDEV. For application in packet networks

these need to be extended. The MTIE and TDEVfamilies of metrics described here represent thestate of understanding for quantifying clocknoise in packet-switched networks. There havebeen several contributions made that suggestthat these metrics have uses far beyond simpleclock noise characterization. For example, moni-toring packet flows and computing simple met-rics can provide network operators a real-timeview of network loading and can prove to beuseful information for network managementpurposes.

ACKNOWLEDGMENTSThe author would like to acknowledge the excel-lent comments and suggestions from the review-ers. These were very constructive and improvedthe content and flow of the article.

REFERENCES[1] S. Bregni, Synchronization of Digital Telecommunica-

tions Networks, Wiley, 2002.[2] ITU-T Rec. G.811, “Timing Characteristics of Primary

Reference Clocks,” 1997.[3] IEEE Std. 1588-2008, “IEEE Standard for a Precision

Clock Synchronization Protocol for Networked Measure-ment and Control Systems,” Nov. 2008.

[4] IETF RFC 5905, “Network Time Protocol (Version 4): Pro-tocol and Algorithms Specification,” June 2010.

[5] ITU-T Rec. G.810, “Definitions and Terminology for Syn-chronization Networks,” 1996.

[6] ITU-T Rec. Y.1540, “Internet Protocol Data Communica-tion Service — IP Packet Transfer and Availability Per-formance Parameters,” 2007.

[7] K. Shenoi, Synchronization and Timing in Telecommuni-cations, BookSurge Publishing, 2009.

[8] S. Bregni, “Clock Stability Characterization and Mea-surement in Telecommunications,” IEEE Trans. Instru-mentation Measurement, vol. 46, no. 6, Dec. 1997.

[9] R. Subrahmanyan, “Timing Recovery for IEEE 1588App l ications in Telecommunications,” IEEE Trans.Instrumentation Measurement, vol. 58, issue 6, June2009.

[10] S. Bregni, “Measurement of Maximum Time IntervalError for Telecommunications Clock Stability Characteri-zation,” IEEE Trans. Instrumentation Measurement, vol.45, no. 5, Dec. 1996.

[11] ITU-T Rec. G.824, “The Control of Jitter and Wanderwithin Digital Networks which are Based on the 1544kb/s Hierarchy,” 2000.

BIOGRAPHYKISHAN SHENOI [M’76, SM‘09] ([email protected])received his B.Tech. degree from the Indian Institute ofTechnology (IIT) Delhi in 1972, an M.S. from ColumbiaUniversity in 1973, and a Ph.D. from Stanford Universityin 1977, all in electrical engineering. He has been activein the telecommunications field since 1977 and since2006 has been the principal consultant at Shenoi Consult-ing in Saratoga, California. Prior to this, he worked atSymmetricom, DSC Communications, and ITT AdvancedTechnology Center. He is named on 40 U.S. patents andauthored several papers. He has published two books,Digital Signal Processing in Telecommunications (PrenticeHall, 1995) and Synchronization and Timing in Telecom-munication (BookSurge Publishing, 2009). He serves asco-chair of the Technical Committee for the annual NIST-sponsored workshop on Synchronization in Telecommuni-cations Systems (WSTS).

The MTIE and TDEV

families of metrics

described here

represent the state

of understanding for

quantifying clock

noise in packet-

switched networks.

There have been sev-

eral contributions

made that suggest

that these metrics

have uses far beyond

simple clock noise

characterization.



INTRODUCTION

The explosive growth of mobile data devices andapplications is driving the industry to develophigher-bandwidth and more efficient radio tech-nologies like WiMAX, time-division synchronouscode-division multiple access (TD-SCDMA),Long Term Evolution (LTE), and LTE-Advanced.Current mobile radio technologies such as GlobalSystem for Mobile Communications (GSM), Uni-versal Mobile Telecommunications System(UMTS), and CDMA2000 are used to providevoice mobility services, and although these tech-nologies have been augmented with data connec-tivity, they do not provide the same scale thelatest radio technologies mentioned above offer.

As mobile data traffic increases, operators arebeing forced to rethink their backhaul networkinfrastructure. In many cases the current back-haul is based on time-division multiplexing(TDM) and asynchronous transfer mode (ATM)

network technologies, and is seen as a bottleneckto offering higher mobile data connectivity andservices. For instance, E1/T1 interfaces and syn-chronous digital hierarchy/optical network(SDH/SONET) interfaces are typically used tointerconnect base stations with the core wirelessswitching centers. These interfaces have servedwell the demand for voice and data services, butare now seen as not having enough flexibility andcapacity to meet the predicted growth of mobiledata traffic. Several operators are now adoptingpacket transparent network (PTN) and opticaltransport network (OTN) as a way to mitigatethe growth in mobile data traffic. However, thereis one aspect for which PTN and OTN (whencompared to a TDM network) were not initiallydesigned, and that is the transfer of both fre-quency and phase/time signals across the network(as opposed to a TDM network, which was pri-marily designed for the transfer of accurate fre-quency). Mobile systems such as GSM andUMTS require a frequency reference at the basestation with a frequency error is ±0.05 parts permillion (ppm) [1]. This requirement has beenpractically met for many years using the transferof E1/T1 circuits through the backhaul network,where the E1/T1 circuits are traceable to a pri-mary reference clock (e.g., Cesium clock).

The latest radio technologies such as TD-SCDMA and LTE-Advanced are not just forcingthe need for the transfer of frequency, but alsofor the transfer of phase/time across a telecomnetwork. Current mobile systems such asCDMA2000 require both frequency and phase/time at the base station (in contrast to GSM/UMTS, which require only frequency).CDMA2000 is successfully served today by GPSreceivers deployed in every base station. TheGPS receiver, equipped with sophisticated algo-rithms and oscillators, is used to maintainmicrosecond-level accuracy traceable to GPSsystem time. For instance, CDMA2000 requiresthat base stations be synchronized to GPS sys-tem time with an accuracy of ±3 μs in normaloperating mode and no more than ±10 μs when

ABSTRACT

This article describes the use of IEEE 1588and boundary clocks for clock distribution(phase/time transfer) in telecom networks. Thetechnology is primarily used to serve the radiointerface synchronization requirements of mobilesystems such as WiMAX and LTE, and to reducethe deployment and dependence of GPS systemsin base stations. We discuss the most importantfunctions that are necessary for phase/timetransfer and present some initial field trialresults using a chain of cascaded boundary clocksand synchronous Ethernet links across a packetand optical transport network that spans tens ofkilometers and tens of network elements. Theresults indicate that it is possible to transferaccurate phase/time in a telecom network andmeet the requirements of mobile systems. Thearticle also discusses some of the challenges andhighlights the ongoing activities in standardiza-tion bodies so that IEEE 1588 can be used as atechnology in telecom networks.


Michel Ouellette, Kuiwen Ji, and Song Liu, Huawei Technologies Inc., Ltd.

Han Li, China Mobile Research Institute

Using IEEE 1588 and Boundary Clocksfor Clock Synchronization inTelecom Networks

OUELLETTE LAYOUT 1/19/11 3:38 PM Page 164


GPS signals become unavailable. The ±10 μs isrequired to be met for up to 8 h, although somevendors do advertise a longer autonomy periodas a product differentiator.

Although GPS has many merits, there are net-work operators seeking to minimize the use ofGPS within their network, especially at the basestation sites. This is driven by the fact that thenumber of base stations to be deployed is expect-ed to significantly increase in the coming years,as well as the need to use alternative methods totransfer time through the network. Other factsinclude that the base station coverage area isbecoming smaller, and some base stations will bedeployed indoors where there is difficulty inobtaining satellite signals. The cost and operationof installing receiver antenna, mounts, cabling,and so on must be taken into account.

These are some of the reasons the telecomindustry is actively looking at the IEEE 1588 proto-col [2] for transferring accurate phase/time (as wellas frequency) across packet and optical networks.Considerable work has been done by the industrysince the initial proposals [3, 4] for applying IEEE1588 to the telecom environment were made.

The next section of this article discusses theaspect of frequency transfer, while the followingsection presents the most important IEEE 1588functions necessary for the transfer of phase/timein telecom. We then discuss field trial results.Finally, we provide an overview of challengesand future work.

FREQUENCY TRANSFERThis section summarizes some of the latestdevelopment in the industry for the transfer offrequency (syntonization) over packet and opti-cal networks. International TelecommunicationUnion — Telecommunication StandardizationSector (ITU-T) Study Group (SG) 15/Question(Q) 13 has spent a considerable amount of timestudying the aspects of frequency over packetnetworks. ITU-T Recommendation G.8261 [5]provides general information related to packetnetwork timing solutions. One of the technolo-gies specified in ITU-T is called synchronousEthernet or SyncE [5–7] and was primarilydesigned for use in packet networks (Ethernetsystems). SyncE is a physical-layer node-by-nodefrequency transfer method that inherits a lot ofthe SDH/SONET properties such as equipmentclocks. SyncE also specifies a messaging channelto exchange information on the quality of theclock being distributed in the network (analo-gous to SSM in SDH/SONET). SyncE is knownto provide guaranteed frequency performancewell below what is required by mobile radiotechnologies (i.e., ±0.05 ppm). In normal oper-ating mode (when traceable to a primary refer-ence clock), SyncE can offer long-term frequencyperformance of about ±0.00001 ppm. It is alsopossible now to transport GE/10GE (andbeyond) SyncE signals directly over an asyn-chronous OTN network where the bit and timingintegrity are preserved as they traverse the end-to-end OTN network [8]. This is useful for oper-ators that want to transfer a frequency referenceacross the core and metro portion of their net-work or in a multi-operator environment.

Another technology for frequency transferdeveloped in ITU-T is based on the use of IEEE1588. The development of the first IEEE 1588profile [9, 10] by ITU-T specifies the functionsand interoperability for the transfer of frequency(not phase/time) between a Packet Master Clock(e.g., IEEE 1588 Grandmaster) and Packet SlaveClock (e.g., Base station). The profile can besaid to offer packet-layer end-to-end frequencytransfer as opposed to SyncE, which is physical-layer node-by-node frequency transfer. The pro-file does not specify performance aspects andassumes that only the Packet Master and PacketSlave (end-points of the network) provide syn-tonization functions (i.e., the network elementsdo not offer any IEEE 1588 assistance).

It is important to note that the original inten-tions behind IEEE 1588 were not to provideend-to-end frequency transfer but rather node-by-node time transfer. This article addresses thenecessary building blocks for providing IEEE1588 node-by-node phase/time transfer in tele-com environments.

TIME TRANSFER BASED ON IEEE1588

IEEE 1588 OVERVIEWThe use of IEEE 1588 for clock synchronizationis seeing a considerable uptake in various fields.Originally developed for the test and measure-ment community, it is not uncommon today tosee IEEE 1588 being applied to other fields suchas automation, power systems, military and tele-com. This section describes some of the buildingblocks necessary to support node-by-node timetransfer based on IEEE 1588, with the intent ofaddressing telecom and mobile backhaul needs.

ITU-T is currently engaged in developing asecond IEEE 1588 profile. The profile will speci-fy that all nodes be IEEE 1588 capable (e.g.,boundary clock), since this is seen as necessaryfor mitigating the effects of latency and delayvariation as well as developing and providingguaranteed performance bounds. Industrial-typeapplications, power-based systems, and con-sumer/professional studio type applications [11]rely on such an approach.

IEEE 1588 BEST-MASTER CLOCK ALGORITHMThe Best-Master Clock Algorithm (BMCA)

is used to create a master-slave hierarchy withina set of distributed clocks, with the goal of pro-ducing a tree-based topology. The BMCA is adistributed algorithm running in every IEEE1588 node, and is functionally similar to Span-ning Tree Protocol used to build the data span-ning tree of bridged Ethernet LANs. Themaster-slave hierarchy is composed at the top ofthe tree of a root (grandmaster clock) being themost accurate clock, branch/forking points form-ing the tree (boundary clocks), and leaves (ordi-nary slave clocks) being the least accurate clocks.The BMCA advertises the clock quality of IEEE1588 nodes via a message called the Announcemessage. These messages travel through the net-work and are used by an IEEE 1588 node tocompare its own clock quality to any receivedclock quality and to decide what kind of role

The use of IEEE1588

for clock synchro-

nization is seeing a

considerable uptake

in various fields.

Originally developed

for the test and

measurement

community, it is not

uncommon today to

see IEEE 1588 being

applied to other

fields such as

automation, power

systems, military

and telecom.


(i.e., root, branch/forking point, leaf) the nodewill take within the hierarchy. Figure 1 shows atree-based master-slave hierarchy of IEEE 1588clocks that provides a timing loop-free topology.The BMCA is also capable of handling networktopology changes (e.g., link or node failure) andto provide re-convergence of the master-slavehierarchy during such events. Alternativeapproaches to create a master-slave hierarchy ofclocks are based on manual configuration orpossibly based on link state protocols as theywould permit distributed computation of thetrees in a manner similar to the multicast treesproduced by IEEE 802.1aq Shortest Path Bridg-ing. These are being studied.

TIMESTAMPINGHardware timestamping is another importantfunction of IEEE 1588. The level of time accura-cy transferred across a network is related to theprecision at which timestamps can be measuredin every node along the master-slave hierarchy.Although 1588 does not strictly specify howtimestamps are captured, it does point out thatto achieve a high-level of accuracy timestampsshould be captured at the physical (or as close tothe medium as possible) in order to minimizeany delay variability such as protocol stacks.

There is ongoing work in IEEE802.3(IEEE802.3bf task force) to specify how theMAC/PHY can be used to provide timestamp indi-cations. The proposed model consists of the defini-tion of a time sync service interface (TSSI) that sitsbetween the medium access control (MAC) andphysical (PHY) layers, and an external time syncclient. Timestamp indications are provided to theclient every time a message timestamp point (SFD)crosses the reconciliation sublayer (RS). There arealso provisions made for PHY vendors to specifythe inbound and outbound latency values (thelatency between the RS layer down to the medi-um). Such values are necessary to minimize anydelay uncertainty and help construct a more accu-rate delay budget across the network, as explained

later. Although the work in IEEE 802.3bf primarilyaddresses the needs of the IEEE 802.1AS proto-col, it is also applicable to telecom.

IEEE 1588 SYNCHRONIZATION PROCESSThe IEEE 1588 protocol defines a set of period-ic message exchange used to interrogate thepropagation delay of the network and calculatethe time offset between master-slave ports thatare geographically separated across links/nodes.This is done by exchanging timestamps and isknown as the synchronization process or two-way time transfer.

Figure 2 shows the synchronization process.The master sends a Sync message to the slave attime t1, which is the timestamp based on its ownclock. The Sync message is received by the slaveat time t2, based on its own clock. The differencebetween t2 and t1 is the clock offset including anypropagation delay between the master and slaveport. The slave then sends a Delay_Requestmessage to the master, which is transmitted attime t3 and measured by the slave clock, andreceived by the master at time t4 and measured bythe master. The master then transmits aDelay_Response message to the slave with allappropriate timestamps. The slave can then deter-mine the master-slave clock offset and propaga-tion delay of the network, and use these values toalign its timebase to the master. This is donebased on a system of equations as shown in thefigure. However, the propagation delay of thenetwork is calculated based on the assumptionthat the two directions (forward M → S andreverse S → M) are symmetrical. If the network isnot symmetric, a time error will be produced. Themagnitude of the error will be proportional tohalf the difference in the delay between the for-ward and reverse directions. The standard definesthe message and synchronization exchange pro-cess but does not specify how to apply the correc-tion and adjust the clock’s timebase.

SYNTONIZATION PROCESSThe section above described the synchronizationprocess, but also of importance for the transferof phase/time is the syntonization process. Syn-tonization implies that the frequency of clocksused to maintain time must be well behaved dur-ing the exchange of messages, as frequency is aprerequisite for time transfer. If the difference infrequency between a master port and a slave portis faster or slower, an accumulation of time errorwill be produced. The IEEE 1588 standard isfairly silent on the syntonization process, butdescribes an approach that may be used to com-pute the frequency offset between master andslave ports based on the interarrival and interde-parture times of Sync messages. This allows aslave to measure the frequency offset of its localoscillator relative to the master and physicallyadjust the frequency of the local oscillator. IEEE1588 also discusses other approaches where phys-ical signals may be used to syntonize clocks, andone such example could be synchronous Ether-net, as explained earlier. In both cases, it is wellknown that jitter and wander can accumulate in areference chain of syntonized clocks [12]; to limitthis accumulation, proper parameterization ofphase locked loops (PLLs) must be specified.


Figure 1. BMCA master-slave hierarchy (clock tree).

Grandmaster clock(root)

M

M M

M

M M M

Boundary clock (branch point)

Boundary clock (branch point)

Ordinary slave clock (leaf)

Ordinary slave clock (leaf)

S

S

S S

S S P

M: Master port S: Slave port P: Passive port


SYNCHRONIZATION ANDSYNTONIZATION PLANES

One trend in the industry is to combine the useof IEEE 1588 for the time function (synchro-nization process) and synchronous Ethernet forthe frequency function (syntonization process),especially in the case where an operator hasalready deployed SyncE. These two processescan be seen as operating in their own planes andare to some extent independent in terms of theirprotocols, functions, protection mechanisms, andso on. Figure 3 shows the two planes; the bottomone is the SyncE frequency sync plane and thetop one is the IEEE 1588 time sync plane.

One item worth discussing is the interactionand coordination of both planes. Since both areindependent, one can imagine situations wherethe reference chains of network elements maydiffer to some extent and lead to an accumula-tion of time error that is well outside require-ments. For instance, it can be difficult to managetwo different planes so as to result in a congru-ence of the frequency and time reference chains,and Fig. 3 illustrates an example where the ref-erence chains in both chains do not exactly fol-low the same path through the network. It canalso be difficult today to guarantee that the con-vergence time of both planes would be the sameduring failure events. There could also be caseswhere loss of frequency produces a reconfigura-tion of the SyncE reference chain, but is notseen by the IEEE 1588 reference chain (i.e., thedefault BMC algorithm described above is notinvoked). It is important to note that all theseissues would be minimized if IEEE 1588 wouldbe used simultaneously for both the time andfrequency functions (without the use of SyncE).In such a case congruence of both planes wouldbe guaranteed, and since frequency and timewould be distributed node-by-node, it could beenvisaged that performance of frequency wouldbe at par when compared to SyncE. The deci-sion on how to operate the planes should ulti-mately be left to the network operator and willlikely be studied by ITU-T.

BOUNDARY CLOCKSEarlier we discussed the role of the BMCA inestablishing the master-slave hierarchy. It wasnoted that the root of the tree is known as thegrandmaster clock and that Boundary Clocks(BC) are used to logically segment the networkby creating branches in the synchronization tree.Typically BCs have multiple ports, and a sepa-rate copy of the protocol state machine anddataset must be maintained for each port. Thisinformation is used to derive the master-slavehierarchy, and only one of port in a IEEE 1588node can be in a slave state at any given time,while the others are in master or passive state.The BC timestamps incoming and outgoingIEEE 1588 messages, and uses the timestampinformation to synchronize its timebase to thegrandmaster (GM) timebase. BCs can includePLL filtering to limit accumulation of errorthrough a chain of cascaded nodes, although thisis not specified in IEEE 1588. A model anddescription of a boundary clock is presented in[13], and conceptually (excluding the timestamp

part) a BC is similar to the type of clock systemsfound in SDH and SyncE network elements.

ONE PULSE-PER-SECOND INTERFACEThe so-called One Pulse-Per-Second (1PPS) isan electrical signal used for precise timekeepingand time measurements. The rising edge of thesignal is used to precisely indicate the rollover ofthe Universal Time Coordinates (UTC) second,and its accuracy generally ranges from severalnanoseconds to a few microseconds. The 1PPSsignal can be found on several devices such astest and measurement equipment, GPS receivers,and current cellular base stations such asCDMA2000 and WiMAX. The 1PPS signal isalso used in various types of interfaces such asPTTI (military/defense) and NEMA-183 (naviga-tion), where the interface specifies not just theelectrical signal but also a data communication


Figure 2. Synchronization process.

Sync

Del_Req

Del_Resp

t1 + clock_offset

t2 (=t1 + clock_offset + prop_delay)

t3

(t1, t2, t3, t4)

Clock_offset = [ (t2 − t1) − (t4 − t3) ] / 2Propagation_delay = [ (t2 − t1) + (t4 − t3) ] / 2

t1

t4

M ← S

M → SMaster port Slave port

Figure 3. Synchronization and synchronization planes.

BC

BC

BC BC

1PPS 2MHz

GE

Wireless switching centers

Grandmaster (time + frequency)

SyncE frequency plane

IEEE 1588 time plane

NE1

NE2

NE4

NE4

NE1

NE2

NE3

NE3 Base stations

1588 time chain SyncE freq chain Service



channel. The data channel is typically a unidirec-tional serial interface that defines the character-istics of a protocol to transmit data from atransmitter to a receiver. The data channel cantransmit information beyond the rollover of asecond such as GPS time, day of year, satelliteID, frequency information, and estimated qualityof time. Grandmaster clocks, for instance, canreceive their time reference signal via a 1PPSinterface of a GPS receiver; likewise, boundaryclocks and ordinary clocks can recover a 1PPSsignal to transfer time to an application or fortest and measurement purposes.

TIME ERROR BUDGETIn order to transfer phase/time, it is importantthat the various sources of error be known and

quantified (e.g., through simulations). Many sys-tems make use of a budget to describe theseerrors and their magnitude; therefore, a timeerror budget of a hypothetical network referencemodel is essential. Such modeling is necessary toderive the proper network limits and ensure thatthe application’s requirements can be met. Forinstance, a mobile backhaul can be decomposedinto several high-level parts with clear demarca-tion points; the GM clock generating time, thepacket transport network used to transfer time,and the recipient recovering and using time.Each part must be assigned a portion of theapplication’s allowable time error budget, andeach part can then be further subdivided intoseveral subparts. For example, the GM’s timeerror budget might be related to the error pro-duced due to the GPS receiver, the length ofcables, and so on. The packet transport networktime error might be related to aspects such astimestamping error, delay difference betweenforward and reverse directions, PLL filtering,PHY latency, thermal variations, and clock syn-thesis. The recipient of time might also share thesame factors as the packet transport network, inaddition to internal and external aspects oftransferring time to the application.

LABORATORY AND FIELDTRIAL RESULTS

This section presents some of the initial laborato-ry results and field trials conducted in Chinaabout a year and a half ago. The field trial evalu-ated the transfer of phase/time to TD-SCDMAmobile base stations, based on the use of IEEE1588 boundary clocks embedded in packet trans-port network equipment, as well as the use ofsynchronous Ethernet to transfer frequency. It isthe authors’ belief that no other field trials of thisscale have been reported in the industry, exceptin some very recent contributions to ITU-T.

LABORATORY RESULTSOne laboratory test topology is shown in Fig. 4,which was used for benchmarking performance.

The GPS antenna signal is connected to theGM device. The jitter on the GPS signal is fil-tered using a Rubidium oscillator embedded intothe GM, in order to produce a highly stable 2MHz and 1PPS reference signal. Device undertest (DUT) #1 and DUT#2 were configured asPTP boundary clocks and used an OCXO oscilla-tor. DUT#1 was phase-aligned to the GM usingthe 1PPS interface. The link connecting DUT#1and DUT#2 is a synchronous Gigabit Ethernet(SyncE) link, and there was no background traf-fic. The time and frequency measurement equip-ment compare the recovered time output (1PPS)and recovered frequency output (2 MHz) ofDUT#2 to their respective reference signals.

Figure 5 shows the frequency (2 MHz) andtime (1PPS) performance. The top figure showsthat the TIE (time interval error with a samplingrate of 30 measurements/s) of SyncE variedbetween –2.5 and 2.0 ns over 24 h, and showsthat PRC frequency traceability was achieved.This is the expected behavior. The bottom figureshows that peak-to-peak jitter of the recovered

Figure 4. Laboratory testbed.

DUT#1 DUT#2

GE

GM

GPS

1PPS 1PPS

2MHz 2MHz

Referencesignals

Recoveredsignals

1PPS + 2MHz

BCBC

Time tester

Freq tester

Figure 5. SyncE performance (top figure) and 1PPS performance (bottom figure).



1PPS (top trace) compared to the reference1PPS (bottom trace) was about 2 ns, but theresult also shows a static time error of about 3ns between the recovered 1PPS and reference1PPS. Note: Due to the resolution of the figure,the legend in the bottom right corner does notappear properly; for reference, the x-axis scalewas 5 ns/division and 1 ns/subdivision. This staticoffset was due to the cables used in the test andmeasurement, as the length of the 1PPS cablesgoing to the test equipment was not taken intoaccount and compensated for.

FIELD TRIAL TOPOLOGY AND RESULTSThe initial field trial topology is shown in Fig. 6,and includes several of the characteristics andfunctions discussed earlier. The field trials wereconducted in a city in China.

The GPS devices provide a 1PPS referencesignals into the IEEE 1588 Grandmaster devices(GM#1 and GM#2). The network consists of ametro 10GE ring and an access GE ring. Thereare two time synchronization reference chainsconsisting of 15 switches each (Chain#1 andChain#2), which were used for protection anddiversity of time transfer. Each network switchwas configured as an IEEE 1588 BC clock fortime transfer and SyncE for frequency transferand were deployed in various central officesacross the city. Chain#1 and Chain#2 spannedapproximately 52 and 45 km, respectively. Back-ground traffic included live mobile calls as well as80 percent load (on and off) during the measure-ment period. The BMCA algorithm was used toestablish the master-slave hierarchy of both syn-chronization reference chains. The state of eachport is shown in the figure, and the ports shownas passive (dashed lines) eliminate timing loops.The time measurement equipment (time tester)was placed at the last network switch to measurethe cumulative time error of the reference chain,and the measurement was done by comparingthe recovered 1PPS signal of the last switch witha reference 1PPS also obtained from GPS.

Cumulative Time Error of Reference Chain#1 — Figure 7 shows the time error (recovered1PPS signal) of Chain#1 for a period of 15 h.The time error varied from approximately 79 to123 ns. The short-term noise of the 1PPS wasprimarily due to hardware timestamping error,SyncE frequency error, and time offset calcula-tion and adjustments. The long-term noise of the1PPS was primarily due to the oscillator andtime-PLL used in the switches. The result alsoshows a static time error of about 100 ns, whichis due to the fiber asymmetry along the 52 kmreference Chain #1. In this initial field trial, anyfiber asymmetry between switches was not com-pensated for. What is important to note is thatthe results were well under the performancerequirements of ±3 μs for TD-SCDMA mobilebase stations, and further improvements havebeen made since these initial field trials.

Cumulative Time Error during Rearrange-ment of the Reference Chain — To create aswitch of the reference chain (i.e., a switch fromreference Chain #1 to #2), the 1PPS signal atGM#1 was disconnected. The execution of the

BMC algorithm elected GM#2 as the new GM,and a new master-slave hierarchy was created withan appropriate change of port states. In this fieldtrial the last network switch changed its right-sideport to slave state (the dashed line became a thickline) and its left-side port to passive state (thethick line changed to a dashed line). As shown inFig. 8, the rearrangement of the reference chainwas created at about 350 s. A time error transientof about 60 ns was produced during the conver-gence of the BMC algorithm. This 60 ns was dueto a combination of holdover and time offset com-putation during the rearrangement period. Theresult also shows a static time error of about 80 ns,which was due to the fiber asymmetry along the 43km reference Chain #2.

ADDITIONAL FIELD TRIAL RESULTSIn another field trial, the topology of the networkwas significantly augmented in terms of numberof nodes as well as the use of live TD-SCDMAbase stations. The reader is referred to [14] formore details on the network topology. The basestation was the last device of the reference chainand was capable of recovering time by two meth-ods: via a fast Ethernet interface where the basestation supported IEEE 1588 or via a 1PPS input

Figure 6. Field trial topology.

10GE RAN

GE RAN

Recovered 1PPS

Reference 1PPS(from GPS)

Reference chain #1

52 km

Reference chain #2

43 km

GM#1 GM#2

1PPS 1PPS

M

M

M

M M

M

M M

MM

S

S

S

S

S

S S

S

P

P

GPS GPS

Time tester


interface provided by the collocated switch/BC.Another notable difference when compared tothe previous results was that fiber asymmetrybetween each switch was manually measured andcompensated for (in order to remove any statictime error as observed in Figs. 7 and 8).

Figure 9 shows the results of the recovered1PPS using method#1 (top graph) and method#2(bottom graph). The top figure showed thatmethod#1 (using IEEE 1588 to synchronize thebase station) generated better performance andresulted in a smaller static time error thanmethod#2 (using 1PPS to synchronize the basestation) in the bottom figure. However, the statictime error in the bottom graph was due, at thetime of trial, to the base station vendor not capa-ble of properly compensating for the cable lengthused to transfer the recovered 1PPS between theswitch and the base station. This generated anarbitrary static time error as shown in the bottomgraph. It is important to note that every cableused in time transfer (e.g., for transferring time toan application or for measurement purposes)must be compensated for. Indeed, both methods#1 and #2, if properly implemented, shoulddeliver the same level of performance.

Figure 10 shows the cumulative time errorresult for a period of approximately 16 h mea-sured at one base station vendor. The results

indicate that the peak-to-peak 1PPS perfor-mance was kept roughly within +80 ns and –50ns over the course of 16 h, which was well withinthe performance requirements of ±3 μs for TD-SCDMA mobile base stations. More analysis onthe behavior is currently being conducted, andincludes aspects related to short-term buffering,timestamping functions, and the oscillator usedin the base station.

Finally, some road tests were also conductedto verify if there was any impact to voice anddata services when base stations were recoveringphase/time using IEEE 1588 and others usingGPS. The road tests consisted of driving betweenbase stations and verifying that successful hand-offs and call completion could be made. In [14]the initial results showed that the handoff andcall completion ratios were not impacted by theuse of IEEE 1588. In addition, the average voicemean opinion score (MOS) was measured to be3.46 (out of 5), which is typical of mobile voicequality.

In summary, it is worth pointing out that thetechnology proposed here is based on node-by-node phase/time transfer, where each networknode supports boundary clocks. For this reason,packet delay variation (PDV) in the networkdoes not impact phase/time transfer. Since thereis no PDV, the control algorithm to recover fre-quency and time is less complex, since correc-tions are done in every network node betweenthe GM and slave. Note that the frequency inthis testbed was obtained via synchronous Ether-net and time via PTP, but it is believed that theresults would be similar if PTP was used in everynetwork node for frequency recovery rather thansynchronous Ethernet.

FUTURE WORKThere is still work to be done on applying IEEE1588 in the telecom environment, but as shownin this article, the use of node-by-node clock syn-chronization is a necessary requirement to pro-vide guaranteed performance. Topics includeevaluating the implications of deploying IEEE1588 functionalities in existing network infra-structure. Other important topics include study-ing the aspect of link/fiber asymmetry andmitigation techniques to minimize this impair-ment, since measuring and compensating forevery link/fiber in a network is practically infea-sible or cost-prohibitive. The allocation of a timeerror budget, the modeling and simulation ofboundary clocks, the protection aspects, andmapping and transfer of IEEE 1588 into theOTN are other topics worth addressing. Thesetopics will most probably be studied in greatdetail by ITU-T SG15.

CONCLUSIONIt is not uncommon to hear in the industry thatIEEE 1588 can achieve sub-microsecond-levelaccuracy. Unfortunately, this statement is notalways true; it depends primarily on how IEEE1588 is deployed and how the various functionswill be used within the network. The use ofnode-by-node time transfer (default mode ofoperation of IEEE 1588) using boundary clocks

Figure 7. Cumulative time error.


Time (s)Cursor A: 0 Cursor B: 55705 5000

90.00 Tim

e in

terv

al e

rror

(ns)

85.00

79.19

95.00

100.00

105.00

110.00

115.00

120.00 123.00

0 10000 15000 20000 25000 30000 35000 40000 45000 50000 55705

Figure 8. Reference chain rearrangement.

Cursor A: 791

Tim

e er

ror

(ns)

Time (s)

Chain #1time error

(about 100 nsstatic time error

due to fiber asymmetry)

Chain #2time error

(about 80 nsstatic time error

due to fiber asymmetry)

Rearrangement chain #1 to chain #2

Cursor B: 803

50

50.0

41.8

60.0

70.0

80.0

90.0

100.0 105.7

0 100 150 200 250 300 350 400 450 500 550 600 650 700 750 803



is a strong requirement in order to provide guar-anteed performance, when compared to end-to-end frequency and time transfer where there isno support for IEEE 1588 within the telecomnetwork elements [15]. Using node-by-node timetransfer (where each node is IEEE 1588 capa-ble) removes the dependence on packet delayvariation and minimizes the complexity and pro-prietary rights of the control algorithms requiredfor frequency and phase/time recovery.

This article has presented some of the mostimportant building blocks that are necessary forusing IEEE 1588 in a telecom and mobile environ-ment, as well as initial field trial results. The use ofIEEE 1588 and boundary clocks provides an alter-native solution for operators that want to minimizethe use of GPS within their network, especially assome of them will be deploying a large number ofbase stations (in the coming years) that require asource of accurate phase/time.

REFERENCES[1] 3GPP TS25.402 v. 7.5.0, “Synchronization in UTRAN

Stage 2,” section 4.2, Dec. 2007.[2] IEEE Std. 1588-2008, “IEEE Standard for a Precision

Clock Synchronization Protocol for Networked Measure-ment and Control Systems,” July 2008.

[3] G. Algie, “Proposal for IEEE 1588 use over Metro Ether-net Layer2 VPNs,” Proc. IEEE 1588 Conf. ‘03, Gaithers-burg, MD, 2003.

[4] G. Algie and M. Ouellette, “IEEE 1588 Time ServiceEnablers for Metro Ethernet Solutions,” Proc. NIST-ATIST1X1 Wksp. Synchronization Telecommun. Sys., Broom-field, CO, 2004.

[5] ITU-T G.8261, “Timing and Synchronization Aspects inPacket Network,” Apr. 2008.

[6] ITU-T G.8262, “Timing Characteristics of SynchronousEthernet Equipment Slave Clock (EEC),” Aug. 2007.

[7] ITU-T G.8264, “Distribution of Timing through PacketNetworks,” Oct. 2008.

[8] J. L. Ferrant et al., “OTN Timing Aspects,” IEEE Com-mun. Mag., vol. 48, no. 9, Sept. 2010, pp. 62–69.

[9] ITU-T G.8265.1, “ITU-T PTP Profile for Frequency Distri-bution without Timing Support from the Network (Uni-cast Mode),” Consented June 2010.

[10] J. L. Ferrant et al., “Development of the First IEEE 1588Telecom Profile to Address Mobile Backhaul Needs,” IEEECommun. Mag., vol. 48, no. 10, Oct. 2010, pp. 118–26.

[11] G. M. Garner, M. D. Tenner, and A. Gelter, “New Sim-ulation and Test Results for IEEE 802.1AS Timing Per-formance,” ISCPS ‘09, Brescia, Italy, Oct. 2009.

[12] S. Bregni, Synchronization of Digital Telecommunica-tions Networks, Wiley, 2002.

[13] J. C. Eidson, Measurement, Control and Communica-tion Using IEEE 1588, Springer-Verlag, 2006.

[14] M. Ouellette, “Examples of Time Transport,” Joint ITU-T/IEEE Wksp. Future Ethernet Transport, Geneva,Switzerland, May 2010; http://www.itu.int/ITU-T/work-sem/tfet/programme.html.

[15] R. Subrahmanyan, “Time Recovery for IEEE 1588 Applica-tions in Telecommunications,” IEEE Trans. InstrumentationMeasurement, vol. 58, no. 6, June 2009, pp. 1858–69.

BIOGRAPHIESMICHEL OUELLETTE [M] ([email protected]) is aproject manager and technical leader in Huawei’s IP Net-work Solutions and Clock Lab, where he focuses on mobilebackhaul networks and the development and analysis ofpacket network architectures/protocols for accurate fre-quency and phase/time transfer. He actively participates inITU-T and IEEE 802.3 standardization. Prior to joiningHuawei, he spent 12 years at Nortel focusing on ATM/TDMpseudowires, synchronous Ethernet, clock algorithms forbase stations, TCP/IP active queue management, and ATMswitching. He has been granted 10 patents, has publishedin more than 10 international journals, and has generatedmore than 50 ITU contributions. He received his B.A.Sc.and M.A.Sc. in electrical/computer engineering from theUniversity of Ottawa in 1995 and 1997, and l’EcoleNationale Superieure des Telecommunication.

KUIWEN JI is the technical leader for synchronization solu-tions and has worked in Huawei since 2001. He focuses onthe synchronization solutions for SDH/OTN/IP products. Healso attends and participates in ITU-T, IETF, and IEEE stan-dardization, and is a steering committee member of theWSTS workshop on synchronization.

LIU SONG is a system engineer in Huawei’s Clock Lab inShenzhen, China, where he focuses on the network prod-uct’s synchronization solutions and implementation. He is aparticipant of ITU-T and actively contributes to the devel-opment of the frequency profile and time profile based onIEEE 1588v2 protocol. He has five years work experience inthe telecom network and product synchronization fieldsince joining Huawei in 2004. He received a bachelor’sdegree from the University of China in 2002.

HAN LI graduated from Beijing University of Posts andTelecommunications (BUPT) and obtained his Ph.D. in2002. He has been working in China Mobile Research Insti-tute since 2004. He is currently the deputy director ofresearch and in charge of transport and access area. Hehas profound knowledge in OTN, PTN, PON, and time syn-chronization technology, and has published more than 50articles, applied for 20 patents, and delivered more than60 ITU-T contributions.

Figure 9. Time error performance via method #1 (top graph) and method #2(bottom graph).

Time (s)321

0

Tim

e er

ror

(ns)

of

1PPS

sig

nal

-5

510

15

20

25

30

35

40

125 156 187 218 249 280 311 342 373 404 435 466 497 528 559 59063 94

Time (s)32 1

110 Ti

me

erro

r (n

s) o

f 1P

PS s

igna

l

105

115

120

125

130

135

140

145

125 156 187 218 249 280 311 342 373 404 435 466 497 528 559 590 63 94

Figure 10. Time error (ns) measured at the base station over 16 hours.

Time (s)

2522

60

Tim

e er

ror

(ns)

of

1PPS

sig

nal 80

40

20

0

-20

-40

-60

5042

2

7562

10

082

1260

2 15

122

1764

2

2016

2

2268

2

2520

2

2772

2 30

242

3276

2 35

282

3780

2

4032

2

4284

2 45

362

4788

2

5040

2

5292

2 55

442



Company Page

Accumold ...................................................................................................S15

ADVA Optical Networking ......................................................................S Cover 4

Agilent Technologies.................................................................................3, 13

Anritsu........................................................................................................S1

Elsevier .......................................................................................................11

European Center for Information & Communications Technologies........17

GL Communications.................................................................................14

Huber + Suhner ........................................................................................S11

IEEE GreenCom.......................................................................................101

IEEE PIMRC 2011 ...................................................................................139

IWCE..........................................................................................................155

Luna Technologies.....................................................................................S13

Micrel .........................................................................................................S9

OFC/NFOEC 2011....................................................................................S Cover 3

RSOFT Design Group..............................................................................S Cover 2

Samsung .....................................................................................................Cover 4

Silicon Labs................................................................................................Cover 2

Stanford Research Systems ......................................................................1

Steepest Ascent .........................................................................................19

u2t Photonics .............................................................................................S7

Vectron International................................................................................5

Wiley-Blackwell .........................................................................................Cover 3

Xilinx ..........................................................................................................S3

Zarlink Semiconductor .............................................................................S5

S indicates Optical Supplement

ADVERTISING SALES OFFICESClosing date for space reservation:1st of the month prior to issue date

NATIONAL SALES OFFICEEric L. Levine

Advertising Sales Manager IEEE Communications Magazine

3 Park Avenue, 17th FloorNew York, NY 10016Tel: (212) 705-8920Fax: (212) 705-8999

Email: [email protected]

SOUTHERN CALIFORNIAPatrick Jagendorf

7202 S. Marina Pacifica DriveLong Beach, CA 90803

Tel: (562) 795-9134Fax: (562) 598-8242


NORTHERN CALIFORNIAGeorge Roman

4779 Luna Ridge CourtLas Vegas, NV 89129Tel: (702) 515-7247Fax: (702) 515-7248Cell: (702) 280-1158


SOUTHEASTScott Rickles

560 Jacaranda CourtAlpharetta, GA 30022Tel: (770) 664-4567Fax: (770) 740-1399


EUROPERachel DiSanto

Huson International MediaCambridge House, Gogmore Lane

Chertsey, Surrey, KT16 9APENGLAND

Tel: +44 1428608150Fax: +44 1 1932564998


ADVERTISERS’ INDEX

LYT-ADINDEX-February 1/21/11 10:06 AM Page 176

IEEE Communications Magazine â€¢ February 2011 Vol. 49, No. 2

Documents

Transcript of IEEE Communications Magazine â€¢ February 2011 Vol. 49, No. 2