ieee wireless magazine

A Publication of the IEEE Communications SocietyIn cooperation with IEEE Computer and VehicularTechnology Societies

®

WIRELESSIEEE

COMMUNICATIONS

February 2013, Vol. 20, No. 1• INTEGRATION OF LOCAL AREA AND WIDE AREA SYSTEMS• COORDINATION AND COOPERATION IN LARGE-SCALE NETWORKS• MULTICELL COOPERATION FOR LTE-A HETEROGENEOUS NETWORK• PRACTICAL CHALLENGES OF INTERFERENCE ALIGNMENT• CSI SHARING STRATEGIES FOR TRANSMITTER COOPERATION• MULTICELL COOPERATION WITH MULTIPLE RECEIVE ANTENNAS• INTERFERENCE MANAGEMENT IN 3GPP LTE-A NETWORKS

MULT ICELL COOPERAT ION

February Wireless 2013 Cover_Layout 1 2/21/13 4:37 PM

Read a sample chapter online and order your copy

LTE and the Evolution to 4G Wireless, Second Edition

www.agilent.com/fi nd/LTEBook2

Next generation wireless communications promise a world of possibilities with important design and measurement challenges. From basic concepts to advanced testing, the second edition of LTE and the Evolution to 4G Wireless is packed with critical insights. To better accelerate your drive to the forefront.

New in the Second Edition

More detail on physical and upper layer MIMO and advanced receiver testingLatest on Releases 11 and 12How to lower non-signal testing costs

The challenge of LTE-Advanced is huge.

So is our second edition of technical insights.

© Agilent Technologies, Inc. 2013 u.s. 1-800-829-4444 canada: 1-877-894-4414

Scan or visit http://qrs.ly/p1263t1 for a video interviewwith Moray Rumney

IENYCM3217indd 1IENYCM3217_402162 1/10/13 4:00 PM1/10/13 4:00 PM

http://qrs.ly/p1263t1

www.agilent.com/find/LTEBook2

1

WIRELESS IEEE

COMMUNICATIONS

10GUEST EDITORIAL

EKRAM HOSSAIN AND ZANDER (ZHONGDING) LEI

12FUTURE STEPS OF LTE-A: EVOLUTION TOWARD

INTEGRATION OF LOCAL AREA ANDWIDE AREA SYSTEMS

YOSHIHISA KISHIYAMA, ANASS BENJEBBOUR, TAKEHIRO NAKAMURA,AND HIROYUKI ISHII

19MULTICELL COOPERATION: EVOLUTION OFCOORDINATION AND COOPERATION IN

LARGE-SCALE NETWORKSHARALD BURCHARDT AND HARALD HAAS

27MULTICELL COOPERATION FOR LTE-ADVANCED

HETEROGENEOUS NETWORK SCENARIOSBEATRIZ SORET, HUA WANG, KLAUS I. PEDERSEN,

AND CLAUDIO ROSA

35THE PRACTICAL CHALLENGES OF

INTERFERENCE ALIGNMENTOMAR EL AYACH, STEVEN W. PETERS, AND ROBERT W. HEATH JR.

43CSI SHARING STRATEGIES FOR TRANSMITTER

COOPERATION IN WIRELESS NETWORKSPAUL DE KERRET AND DAVID GESBERT

50MULTICELL COOPERATIVE SYSTEMS WITH

MULTIPLE RECEIVE ANTENNASINSOO HWANG, CHAN-BYOUNG CHAE, JUNGWON LEE,

AND ROBERT W. HEATH, JR.

IEEE Wireless Communications • February 2013

FE B R U A RY 2013/VO L . 20, NO. 1

59INTERFERENCE MANAGEMENT THROUGH COMP IN

3GPP LTE-ADVANCED NETWORKSSHAOHUI SUN, QIUBIN GAO, YING PENG, YINGMIN WANG,

AND LINGYANG SONG

67HOW DO WE DESIGN COMP TO

ACHIEVE ITS PROMISED POTENTIAL?CHENYANG YANG, SHENGQIAN HAN, XUEYING HOU,

AND ANDREAS F. MOLISCH

75ON THE IMPACT OF NETWORK GEOMETRIC MODELS ONMULTICELL COOPERATIVE COMMUNICATION SYSTEMS

ANVAR TUKMANOV, ZHIGUO DING, SAID BOUSSAKTA, AND ABBAS JAMALIPOUR

82ON GREENING CELLULAR NETWORKS VIA

MULTICELL COOPERATIONTAO HAN AND NIRWAN ANSARI

90OPTIMAL TRADE-OFF BETWEEN POWER SAVING AND

QOS PROVISIONING FORMULTICELL COOPERATION NETWORKS

XI ZHANG AND PING WANG

MESSAGE FROM THE EDITOR-IN-CHIEF — 2INDUSTRY PERSPECTIVE — 3

BOOK REVIEWS — 6SCANNING THE LITERATURE — 8

Cover image: Getty Images

SPECIAL ISSUE

MULTICELL COOPERATION

LYT-TOC-February_LYT/TOC/OCT 2/22/13 11:30 AM Page 1


ME S S A G E F R O M T H E ED I T O R- I N -CH I E F

his issue features a feature topic on “Mul-ticell Cooperation,” which contains 11excellent papers collected by our Guest

Editors, Ekram Hossain and Zander(Zhongding) Lei. They worked extremely hardto edit this feature topic due to a fairly largenumber of submissions to the Call for Papers.Multicell cooperation is a newly emergingcommunication technology that can help tosignificantly improve the overall system capaci-ty of a cellular system through mitigating orcoordinating intercell interference. A multicellcooperation system works based largely oncooperative communications at both the basestation (BS) and mobile station (MS) levels,and it has become a viable technology toaddress various critical issues in throughputand coverage enhancement in fourth genera-tion mobile cellular systems. As a matter offact, multicell cooperation has been identified as one of the keytechnologies adopted in the LTE-A standard, which offers apeak data transmission rate up to 1 Gb/s. Therefore, the impor-tance and timeliness of this feature topic is evident. Due to theexcellent work done by the Guest Editors, this feature topicoffers you a collection of 11 articles, contributed by experts

from both academia and industry. We do hopethat you will enjoy reading them.

BIOGRAPHYHSIAO-HWA CHEN [S’89, M’91, SM’00, F’10]([email protected]) is currently a Distinguished Professorin the Department of Engineering Science, National ChengKung University, Taiwan. He obtained his B.Sc. and M.Sc.degrees from Zhejiang University, China, and a Ph.D. degreefrom the University of Oulu, Finland, in 1982, 1985, and1991, respectively. From 2001 to 2003, he served as thefounding director of the Institute of Communications Engi-neering at National Sun Yat-Sen University, Taiwan, whichwas the first communication research institute establishedin southern Taiwan. Due to his instrumental role and strongleadership, this institute has educated thousands oftelecommunication postgraduate degree holders for Tai-wan, contributing to the fast-growing local telecommunica-tion industry. He has authored or co-authored over 300technical papers in major international journals and confer-ences, six books, and more than 10 book chapters in the

areas of telecommunications, including the books Next Generation Wireless Sys-tems and Networks (Wiley, 2005) and The Next Generation CDMA Technologies(Wiley, 2007). He has been an active volunteer with IEEE in various technicalactivities for over 22 years. He has served as General Chair, TPC Chair, and Sym-posium Chair for many international conferences. Recently, he served as lead TPCChair for IEEE SmartGridComm 2012,Taiwan, November 2012, and is currentlyserving as Vice TPC Chair for IEEE WCNC 2013, Shanghai, China.

MULTICELL COOPERATION FOR THE 4G MOBILE CELLULAR SYSTEMS

Director of MagazinesSteve Gorshe, PMC-Sierra, Inc, USA

Editor-in-ChiefHsiao-Hwa Chen, Nat’l. Cheng Kung Univ., Taiwan

Associate Editor-in-ChiefDilip Krishnaswamy, Qualcomm Research Center

Senior AdvisorsHamid Ahmadi, Motorola, USA

Uday Desai, Indian Inst. Technology-Hyderabad, IndiaDavid Goodman, Polytechnic University, USA

Abbas Jamalipour, University of Sydney, AustraliaTero Ojanperä, Nokia, Finland

Thomas F. La Porta, Penn State University, USAMichele Zorzi, University di Padova, Italy

Advisory BoardDonald Cox, Stanford University, USA

Yuguang Fang, University of Florida, USAHequan Wu, Chinese Academy of Eng., China

Mahmoud Naghshineh, IBM Watson Research, USAKaveh Pahlavan, Worcester Polytechnic Inst., USA

Mahadev Satyanarayanan, CMU, USA IEEE Vehicular Technology Liaison

Theodore Rappaport, Univ. of Texas, Austin, USAIEEE Computer Society Liaison

Mike Liu, Ohio State University, USATechnical Editors

Abouzeid Alhussein, Rensselaer Polytechnic Inst., USAAbderrahim Benslimane, University of Avignon, France Xiuzheng Cheng, George Washington Univ., USASunghyun Choi, National Seoul University, KoreaEkram Hossain, University of Manitoba, Canada

Thomas Hou, Virginia Tech., USARose Qingyang Hu, Utah State University, USA

Minho Jo, Korea University, KoreaNei Kato, Tohoku University, Japan

Phone Lin, National Taiwan University, TaiwanYing-Dar Lin, National Chiao-Yung University, Taiwan

Stanley Kuang-Hao Liu, Nat’l. Cheng Kung Univ., TaiwanJavier Lopez, University of Malaga, SpainZhisheng Niu, Tsinghua University, China

Mohammad S. Obaidat, Monmouth University, USASymeon Papavassiliou, Natl. Tech. Univ. Athens, Greece

Vincent Poor, Princeton Univ., USAYi Qian, University of Nebraska – Lincoln, USA

Kui Ren, Illinois Institute of Tech., USAJohn Shea, University of Florida, USA

Sherman Shen, Univ. of Waterloo, CanadaChonggang Wang, InterDigital Commun., USA

Richard Wolff, Montana State Univ., USAQian Zhang, Hong Kong Univ. Science & Tech.,

Hong KongYanchao Zhang, Arizona State University, USA

Department EditorsBook Reviews

Satyajayant Misra, New Mexico State Univ., USAIndustrial Perspectives

Jianfeng Wang, Philips Research North America, USAScanning the Literature

Pan Li, Mississippi State University, USASpectrum Policy and Reg. Issues

Michael Marcus, Marcus Spectrum Solns., USA2013 Communications Society

Elected OfficersVijay K. Bhargava, President

Sergio Benedetto, President-ElectLeonard Cimini, VP–Technical Activities

Abbas Jamalipour VP–ConferencesNelson Fonseca, VP–Member Relations

Vincent Chan, VP–PublicationsAlex Gelman, VP-Standards Activities

Members-at-LargeClass of 2013

Gerhard Fettweis • Stefano GalliRobert Shapiro • Moe Win

Class of 2014Merrily Hartman • Angel Lozano

John S. Thompson • Chengshan XiaoClass of 2015

Nirwan Ansari • Stefano BregniHans-Martin Foisel • David G. Michelson

2013 IEEE OfficersPeter W. Staecker, President

J. Roberto B. de Marca, President-ElectMarko Delimar, Secretary

John T. Barr, TreasurerGordon W. Day, Past-President

E. James Prendergast, Executive DirectorDoug Zuckerman, Director, Division III

IEEE Wireless Communications (ISSN 1536-1284) ispublished bimonthly by The Institute of Electrical andElectronics Engineers, Inc. Headquarters address:IEEE, 3 Park Avenue, 17th Floor, New York, NY10016-5997; Tel: (212) 705-8900; Fax: (212) 705-8999;E-mail: [email protected]. Responsibility for thecontents rests upon authors of signed articles and notthe IEEE or its members. Unless otherwise specified,the IEEE neither endorses nor sanctions any positionsor actions espoused in IEEE Wireless Communications.

Annual subscription: Member subscription: $40 peryear; Non-member subscription: $250 per year. Singlecopy: $50.

Editorial correspondence: Manuscripts for considera-tion may be submitted to the Editor-in-Chief: Hsiao-HwaChen, Department of Engineering Science, NationalCheng Kung University, Tainan 70101, Taiwan. Electronicsubmissions may be sent to: [email protected].

Copyright and reprint permissions: Abstracting ispermitted with credit to the source. Libraries permitted tophotocopy beyond limits of U.S. Copyright law for pri-vate use of patrons: those post-1977 articles that carry acode on the bottom of first page provided the per copy feeindicated in the code is paid through the CopyrightClearance Center, 222 Rosewood Drive, Danvers, MA01923. For other copying, reprint, or republication per-mission, write to Director, Publishing Services, at IEEEHeadquarters. All rights reserved. Copyright © 2012 byThe Institute of Electrical and Electronics Engineers, Inc.

Postmaster: Send address changes to IEEE WirelessCommunications, IEEE, 445 Hoes Lane, Piscataway, NJ08855-1331; or E-mail to [email protected] in USA. Periodicals postage paid at New York,NY and at additional mailing offices. Canadian GST#40030962. Return undeliverable Canadian addresses to:Frontier, PO Box 1051, 1031 Helena Street, Fort Eire, ONL2A 6C7.

Subscriptions: Send orders, address changes to: IEEEService Center, 445 Hoes Lane, Piscataway, NJ 08855-1331; Tel: (908) 981-0060.

Advertising: Advertising is accepted at the discretion of thepublisher. Address correspondence to: AdvertisingManager, IEEE Wireless Communications, 3 Park Avenue,17th Floor, New York, NY 10016.

Joseph Milizzo, Assistant PublisherEric Levine, Associate Publisher

Susan Lange, Online Production ManagerJennifer Porcello, Production Specialist

Catherine Kemelmacher, Associate Editor

WIRELESS IEEE

COMMUNICATIONS

HSIAO-HWA CHEN

D

®

LYT-EDIT NOTE-February_LYT/EDIT NOTE/OCT 2/21/13 4:03 PM Page 2

mailto:[email protected]




3IEEE Wireless Communications • February 2013

IN D U S T RY PE R S P E C T I V E S

INTRODUCTION

More than any time ever before, today technology has asignificant impact on people’s lives. The proliferation ofslim, mobile, and portable devices such as notebooks,ultrabooks, tablets, and smartphones is a clear testamentto the importance of computing and communication inmodern society. In the particular case of wireless systems,this explosive growth has led to an insatiable need to con-stantly boost capacity so that innovative uses can beoffered to consumers. This vicious cycle has brought tomarket many innovative wireless solutions including theIEEE 802.11 family of standards, third generation (3G),and fourth generation (4G), to name a few.

When it comes to wireless systems with data ratesgreater than 1 Gb/s at a few meters, the most notableexamples include the IEEE 802.11ac and IEEE 802.11adamendments to the base IEEE 802.11 standard [2], withkey parameters summarized in Table 1. Several companieshave announced products implementing these technolo-gies, with a few of those products already available, orsoon to be available, to consumers.

The data rates offered by IEEE 802.11ac and IEEE802.11ad can meet the needs of many applications, withreplacement of wired digital interface (WDI) cablesarguably the most prominent new use of these technolo-gies. This includes replacing WDI cables used between anotebook/tablet/smartphone and certain display (e.g.,HDMI, DisplayPort) and IO (e.g., USB) devices, therebyenabling a fully wireless user experience [4].

THE EVOLVING LANDSCAPE IN WDIS

The vicious cycle of technology permeates both wired andwireless interfaces. Since replacement of WDI cables is theprimary use of this new wave of multi-gigabit per secondwireless systems, it is important to understand how promi-nent WDIs are expected to evolve over time. This is illus-trated in Table 2, which focuses primari ly on the

approximate maximum data rates currently supported byeach of the corresponding WDI. Future versions of theseWDIs are under development and promise to offer evenhigher data rates.

Supporting the data rates outlined in Table 2 over awireless system is nontrivial. While IEEE 802.11ac andIEEE 802.11ad are able to support degrees of these WDIs,when it comes to supporting, say, the high-end data ratesfor certain WDIs shown in Table 2, it becomes apparentthat there is sti l l a performance gap. This gap is onlyexpected to grow larger, given the rapid evolution in WDIspeeds – clearly, this is one of the major drivers of thevicious cycle behind multi-Gb/s wireless technologies. Assuch, we can conclude that next generation wireless sys-tems needs to support data rates in the order to tens ofGb/s to be able to keep up with the evolution of WDIspeeds.

THE WAY TOWARD TENS OF GIGABIT PER SECOND WIRELESS

So far today, there has been little discussion in the indus-try on ways to achieve tens of Gb/s wireless speeds withranges of a few meters. Despite that, there is growing con-sensus being formed that the only practical way of gettingthere is through the use of higher frequencies, with 60GHz being the starting point.

The amount of unlicensed spectrum in frequencies below6 GHz is very limited. Even though there are ongoingefforts to allocate additional unlicensed spectrum in the 5GHz band, the added capacity will be far short of allowingtens of gigabits per second rates to be offered at ranges of5–10 m. As such, the 60 GHz frequency band, with its vastbandwidth of around 7 GHz, is currently seen as the onlyviable solution to achieve those goals. Referring back toTable 1, this implies that reusing and extending IEEE802.11ad is the way toward tens of gigabits per second wire-less, since this technology operates in the 60 GHz frequencyband. Arguably, unlicensed bands higher than 60 GHz (e.g.,terahertz) could also be used to achieve tens of gigabits per

Table 1. Brief comparison between two > 1 Gb/s systems.

Parameter IEEE 802.11ac IEEE802.11ad

Approx. max.theoreticaldata rate(Gb/s)

Per device 3.5 (4¥ 4, 160MHz channel)

6.8

Aggregate 6.95

Range at max data rate (m) Few (< 10)

Channel bandwidth (MHz) 20/40/80/160 2160

Number of channels in the US Five 80 MHz Three

Frequency band (GHz) 5 60

Expected publication Feb/14 Dec/12

Table 2. Evolution of prominent WDIs.

WDI Version Approx. max. data rate(Gb/s)

DisplayPort1.0–1.1 8.6

1.2 17.3

HDMI1.0–1.2a 5

1.3–1.4a 10.2

USB2.0 0.48

3.0 5

Thunderbolt 1.0 10

THE PURSUIT OF TENS OF GIGABITS PER SECOND WIRELESS SYSTEMSCARLOS CORDEIRO, INTEL CORP.

LYT-INDUSTRY-Cordeiro_LYT/SCANNING/JUNE 2/21/13 3:25 PM Page 3

4

second wireless, but technologies operating at these fre-quencies are nowhere near commercialization.

REVIEW OF BEAMFORMING IN IEEE 802.11AD

An overview of the IEEE 802.11ad standard is given in [3].The unique feature of this standard is the fact that com-munication in 60 GHz requires the use of high gain direc-tional antennas in order to compensate for the large pathloss [1]. While directional communication addresses thelink budget issue, it introduces the need for finding the

direct ion of communication with a neighbor. This isaddressed by a novel beamforming (BF) protocol.

BF is the process used by a pair of stations to achievethe necessary link budget for subsequent communication.In IEEE 802.11ad, BF consists of two phases, as depictedin Fig. 1: sector level sweep (SLS) and beam refinementphase (BRP). SLS enables a pair of devices to establishthe initial (coarse) direction of communication. At the endof SLS, a pair of devices determines the best transmit sec-tor for communication with each other. BRP is used forreceive antenna training and to fine tune antenna settingsto improve the quality of directional communication, andhence obtain a multi-gigabit-per-second link.

One of the most important characteristics of the BFprotocol in IEEE 802.11ad is that it is capable of discover-ing a single radio frequency (RF) path for communicationbetween a pair of devices. Thus, this means that even if adevice possesses more than one antenna array, only one ofthose antenna arrays is used for communication betweenany transmitter-receiver device pair. This is depicted inFig. 2.

ACHIEVING TENS OF GIGABITS PER SECOND WIRELESSSimilar to the evolution of IEEE 802.11a/b/g toward

IEEE 802.11n, we believe that the next step in the evolu-tion of 60 GHz standards beyond IEEE 802.11ad wil linclude the introduction of multiple-input multiple-output(MIMO) and channel bonding capabilities. With MIMO,different signals are transmitted from different 60 GHzantenna arrays, as shown in Fig. 3. This results in animprovement in diversity and performance.

To illustrate the gains possible with MIMO in 60 GHz,we have simulated the scenario shown in Fig. 4, wherein



Figure 1. Beamforming protocol in IEEE 802.11ad.

Transmit sector sweep

Initiator sector sweep

Sector sweepfeedback

STA1

STA2

BRP-TXBRP-RX

BRP-RX

Sectorsweep

ACK

BRP-TX

BRP

Responder sector sweep

Sector level sweep

Transmit sector sweep

BRP with finalfeedback

Beam refinement

Figure 2. IEEE 802.11ad beamforming: single RF path .

S1

S1

S1

ReceiverTransmitter

Figure 3. Next generation 60 GHz beamforming: > 1 RF path.

Sn

S2

S1

ReceiverTransmitter

Figure 4. Simulation scenario.

Nullforming

Receiver device

Link 1

Nullforming

Beamforming

Tran

smitt

er d

evic

e

Reflector

Link 2

Each antennaarray is 2 × 8

5 m

10 cm

y (m

)


5

two devices are separated by 5 m, and each device isequipped with two antenna arrays each with 2 ¥ 8 ele-ments. The center of the two antenna arrays in a deviceare separated by 10 cm. A reflector with a reflection coef-ficient of 2 dB is placed next to the two devices to form areflected path between the two devices. The distance tothe reflection point (y) is varied from 1 to 3 m. The totaltransmit power of each antenna array is set to 10 dBm,and the noise figure and implementation loss are set to 15dB. Both beamforming and nullforming techniques areused to enable two s imultaneous data transmissionsbetween the two devices over the direct path (i.e., link 1)and the reflected path (i.e., link 2). The MCS is chosenfrom MCS 13 (693 Mb/s) to MCS 24 (6.76 Gb/s) of IEEE802.11ad based on the received signal-to-interference-plus-noise ratio (SINR). Figure 5 shows the results obtained,where an aggregate data rate of at least 11 Gb/s can beachieved with two simultaneous transmissions.

When coupled with channel bonding, the aggregatedata rate is essentially multiplied. Therefore, MIMO withchannel bonding can lead to data rates greater than 20Gb/s for two bonded channels and 30 Gb/s for three bond-ed channels. The use of higher-order MIMO schemes andlarger numbers of elements per antenna array than thoseshown in Fig. 4 can lead to even higher aggregate datarates.

CONCLUSIONS AND FUTURE DIRECTIONS

The next few years will experience a significant growth inthe availability of commercial wireless technologies capa-ble of multi-gigabit-per-second data rates. While this willprovide a major technological advancement compared tothe capacity of today’s short-range wireless systems basedon IEEE 802.11 [2], these speeds fall short of meeting therates of future WDIs. The use of MIMO and channelbonding at 60 GHz frequencies is expected to provide thenext major bump in wireless speeds that can lead to datarates on the order of tens of gigabits per second, thus serv-ing as the path toward supporting the envisioned evolutionin WDI speeds. That said, research is still in its infancywhen it comes to the use of MIMO in 60 GHz, and furtherwork is needed on developing this capability and provingits commercial feasibility.

ACKNOWLEDGMENTSThe author would like to thank Minyoung Park from Intelfor providing helpful contributions to this article.

REFERENCES[1] C. Cordeiro, “Evaluation of Medium Access Technologies for Next

Generation Millimeter-Wave WLAN and WPAN,” IEEE ICC, 2009.[2] IEEE 802.11-2012 Standard, http://www.ieee802.org/11/[3] C. Cordeiro et al., “IEEE 802.11ad: Introduction and Performance

Evaluation of the First Multi-Gb/s WiFi Technology,” ACM mmCom,2010.

[4] C. Hansen, “WiGig: Multi-Gb/s Wireless Communications in the 60GHz Band,” IEEE Wireless Comms, Dec. 2011.


Figure 5. Simulation results of MIMO with 60 GHz.

Distance to the reflection point, y (m)1

2

PHY

rate

(G

b/s)

0

4

6

8

10

12

14

16

1.5 2

Link2 data rate

Link1 data rate

Sum rate

2.5 3



http://www.ieee802.org/11/


BO O K RE V I E W S

EDITED BY SATYAJAYANT MISRA

Fundamentals of Wireless Sensor Net-works: Theory and Practice, Wal-tenegus Dargie and ChristianPoellabauer, Wiley Series on WirelessCommunications and Mobile Com-puting; Series Editors: Xuemin (Sher-man) Shen and Yi Pan; John Wileyand Sons, Ltd.; Edition 1, 2010;ISBN: 978-0-470-997-659; Hard-cover, 332 pages.Reviewer: Satyajayant Misra

Wireless sensor networks (WSNs)have been an area of sustained researchfor more than a decade. Although theirinitial application has been restricted toniche domains, such as in the militaryand for academic projects, recently,they are finding use in many fields, suchas civil and mechanical engineering,biology, ecology, medicine, and individ-ual civilian uses. Several recent effortshave also been launched to make wire-less sensor nodes easy to assemble,using commercial off-the-shelf compo-nents, and easy to program using freelyavailable open source software. Severalcommercial ventures have been startedto help users set up their own WSNs,gather data and store it in the cloud,and also perform sophisticated analysisin the cloud. Looking into the future,these spirited efforts will result inWSNs becoming ubiquitous in our day-to-day life. This exciting phase of WSNdevelopment will enable users, practi-tioners, enthusiasts, and researchersattempting to apply WSNs in theirdomains. This calls for availability ofliterature and texts that provide anunderstanding of the fundamentals ofWSNs from both the theoretical andthe practical perspectives. These textsshould not only be written with studentsand researchers in the area in mind, butalso aim to benefit a wide spectrum ofreaders. This book, titled Fundamentalsof Wireless Sensor Networks: Theory andPractice, is a good match for the needsof this wide spectrum of possible users,including students at the graduate andundergraduate levels. The book pro-vides an introduction to the fundamen-tal concepts and principles of WSNs,and a survey of protocols, algorithms,and technologies in subdomains, suchas the network protocol stack, middle-ware, and applications.

This book is divided into three

broad sections: Introduction, BasicArchitectural Framework, and Nodeand Network Management, containinga total of 12 chapters. The Introduc-tion section contains four chapters.Chapter 1 provides the motivation forthe use of WSNs by juxtaposing WSNsagainst traditional networks (e.g. ,wired networks). The chapter alsoprovides some insights on the natureof typical sensors that are attached tosensor nodes, and the classification forthese sensors. It also details the chal-lenges and constraints with WSNs, interms of the environment, wirelesschannel, and energy scarcity, amongothers. The second chapter providesdetails on several application scenar-ios for WSNs: structural health moni-toring, traffic monitoring, healthcare,volcano monitoring, underground min-ing, pipeline monitoring, and precisionagriculture. There are of course sever-al other applications that may be envi-sioned for WSN application, such ashabitat monitoring and ecological sur-veys of animals and their migrationpatterns. Chapter 3 presents the archi-tecture of a wireless sensor node,which includes the sensing subsystemconsisting of the analog-to-digital con-verter as the integral part; the proces-sor/microcontroller subsystem, whichuses different technologies, such asapplication-specific integrated circuitsand field programmable gate arrays;and communication interfaces, such asthe serial peripheral interface and theinter-integrated circuits. The chapteralso presents some of the prototypicalarchitectures, such as the Intel iMotenode architecture, the XYZ architec-ture, and the Hogthrob architecture.Chapter 4 presents different featuresand functional aspects of operatingsystems for a wireless sensor node,and also presents and compares fourOSs: TinyOS, SOS, Contiki, andLiteOS.

The second section of the book per-tains to basic architectural frameworks,

and provides a detailed discussion ofprotocols and algorithms used at dif-ferent network protocol layers. Chap-ter 5 presents a discussion about thephysical layer architectures and con-cepts. It starts by presenting the basiccomponents of a wireless sensor sys-tem, and then presents source encod-ing, channel encoding, differentmodulation techniques for communica-tion in the wireless channel, and signalpropagation. Chapter 6 starts with anoverview of medium access control(MAC) protocols: the paradigms ofcontention-based and contention-freeMAC; then presents MAC protocolsused in the ad hoc wireless domain,and the requirements of a MAC proto-col in WSNs. The chapter also pre-sents several contention-free,contention-based, and hybrid MACprotocols for WSNs. Chapter 7 pre-sents an overview of the network layer,the routing metrics, and also the dif-ferent routing paradigms, such asflooding and gossiping, data-centricrouting, proactive routing, on-demandrouting, hierarchical routing, location-based routing, QoS-based routing, andthe corresponding protocols.

The third section deals with nodeand network management, presentingseveral additional techniques and solu-tions for dealing with system and net-work dynamics and WSN-specificrequirements. Chapter 8 deals withpower management in WSNs: localpower management aspects and dynam-ic power management. Chapter 9 dealswith time synchronization primitivesfor the nodes in the WSN. It presentsthe basics for time synchronization inWSNs and the corresponding chal-lenges. The chapter further presentsseveral time synchronization protocols,such as lightweight tree-based synchro-nization, the timing-sync protocol forsensor networks, the flooding time syn-chronization protocol, and protocolsfor multihop time synchronization.Chapter 10 deals with localization —ranging techniques, different tech-niques for range-based localization,range-free localization, and event-basedlocalization. Chapter 11 deals withWSN security, that is, the security fun-damentals necessary for proper opera-tion of a WSN, the challenges tosecurity, and potential security attacks.The chapter also presents protocolsand mechanisms to improve securityand provide reusable primitives. Chap-ter 12 deals with sensor network pro-

Given its breadth of coverage ofthe topic of wireless sensor networksand its close following of the state

of the art, the book would serve notonly as a good textbook, but also

as a very good reference forresearchers and practitioners.

LYT-BOOK REVIEW-February_LYT/SCANNING/JUNE 2/21/13 3:23 PM Page 6

IEEE Wireless Communications • February 2013 7

gramming: the challenges to program-ming wireless sensor nodes; the node-centric programming constructs in theevent-driven programming paradigm;the idea of macroprogramming, wherea node is no longer the focus, but agroup of nodes or a whole network; theidea of dynamic reprogramming of sen-sor nodes post-deployment; and theavailable sensor network simulatorsand tools.

The book provides a holistic view ofWSNs from the perspectives of boththeory and practice. Given its breadthof coverage of the topic of wireless sen-sor networks and its close following ofthe state of the art, the book wouldserve not only as a good textbook, butalso as a very good reference forresearchers and practitioners. The bookalso has several interesting exercisequestions at the end of each chapter.The references provided at the end ofeach chapter are fairly representative ofthe state of the art.

Smart Grid Communications andNetworking, Edited by Ekram Hossain,Zhu Han, and H. Vincent Poor, Cam-bridge University Press, Edition 1, May2012, ISBN-13: 9781107014138,Hard cover, 481 pages.Reviewer: Satyajayant Misra

Due to its huge economic, environ-mental , and societal benefits , thesmart grid has become a critical agen-da worldwide. The smart grid is anenhanced power grid system, in whichgrid entit ies can cooperate toimprove, the response of the powergrid, reduce peak demand, and makepower generation, distribution, andconsumption economically. A majorenabler of this cooperation is efficientcommunication and informationexchange between the grid entities,the producers, distributors, policymakers, and the consumers. The com-munication and information exchangewill be facilitated by several wirelessand wire-line technologies, such asIEEE 802.11, ZigBee, 6LoWPAN,Ethernet, Power-Line Communica-tion, etc. The smart grid will provideinformation and intelligence to thepower grid to enable grid automation,active operation, and efficient demandresponse. This makes the need forresearch and development in commu-nication protocols for the smart gridan area needing immediate attentionand thus, research in the developmentof smart grid communications and

networking systems is gaining momen-tum. Smart Grid Communications andNetworking, edited by Ekram Hossainfrom the University of Manitoba,Canada, Zhu Han from the Universityof Houston, USA, and H. VincentPoor from the Princeton University,USA—a recently published book byCambridge University Press on smartgrid communications has beenreleased for enthusiasts, practitioners,and engineers at the right time.

The book contains twenty chapters,which are organized into six parts. Partone consists of four chapters coveringthe basics of communication architec-tures, models, and technologies, net-work control systems, demand-sidemanagement models, as well as thevehicle-to-grid systems in the futuresmart grid. Part two onwards; the bookdeals in greater detail with communica-tion, namely existing paradigms, tech-nologies, and security and privacy issueson the smart grid. Part two, consistingof four chapters, deals with the physicaldata communications, access, detection,and estimation techniques for the smartgrid systems. Chapter 5 presents com-munication and access technologiesfrom legacy systems, to technologiesthat will find application in smart gridsof the future. The emerging paradigmof machine-to-machine (M2M) commu-nications is reviewed in Chapter 6.Chapter 7 deals with the important areaof bad-data detection in the smart gridespecially in a distributed manner.Chapter 8 deals with distributed stateestimation with emphasis on a learning-based framework.

Part three of the book consists oftwo chapters and focuses on wirelessnetworks and performance evaluationof network architectures and protocolsfor the smart grid systems. Chapter 9presents networking technologies forwide-area power-systems measurement,monitoring, protection, and control.Chapter 10 presents factors to take intoaccount when using wireless technolo-gies (especially wireless sensors andactuators) for smart grid applications.Wireless sensor and actuator networks

will play a very important role for thegeneration systems, transmission anddistribution systems, as well as in theconsumers' premises in the smart gridenvironments. Part four of the book,consisting of four chapters, focuses onthe applications of the technologies forthese sensor networks, major designchallenges, as well as implementationand performance evaluation of sensornetworking protocols for the smart gridsystems. Chapter 11 deals with researchchallenges and potential application ofwireless sensor networks for smart grid.Chapter 12 presents sensor techniquesand network protocols customized tothe needs of the smart grid. Chapter 13presents potential methods, such aspervasive service-oriented network andcontent-aware intelligent control, fordevelopment/deployment of sensor andactuator networks in smart grid. Chap-ter 14 implementation and performanceevaluation of wireless sensor networksfor smart grid. Security is a significantissue in the smart grid research and ithas attracted substantial research inter-est from both academia and industry.Part five of this book is dedicated tosecurity issues in the smart grid. Fivechapters in this part deal with issuessuch as cyber-attack impact analysis,jamming-based attack and bad-datainjection attack, cross-layer design forsecurity provisioning, and intrusiondetection, and application-driven designapproach for secured smart grid appli-cations. Lastly, part six overviews sever-al smart grid field trials.

The articles in the book are well-organized and well-articulated, withaccess to relevant references at theend of each chapter. It will be a valu-able source of information for peoplenew to the area of smart grid commu-nications and networking and thesmart grid systems in general. Also,scientists and engineers alreadyinvolved in smart grid research willfind this book a very good referencesource. This is one of the few books onthe smart grid available today which isuseful to the researchers in the area ofcommunications, networking, and sig-nal processing who intend to work inthe smart grid area.

Even though this is an edited book,the book has a very good flow and thematerials are coherently presented. Allof the major aspects related to smartgrid communications and networkinghave been covered. This book may bealso useful to readers from the powerengineering community seeking to opti-mize the communication systems forsmart grid.

BO O K RE V I E W S

Even though this is an editedbook, the book has a very good

flow and the materials are coherently presented. All of the

major aspects related to smart gridcommunications and networking

have been covered.

LYT-BOOK REVIEW-February_LYT/SCANNING/JUNE 2/21/13 3:23 PM Page 7


SC A N N I N G T H E L I T E R AT U R E

EDITED BY PAN LI

Crowdsourcing to Smartphones: Incen-tive Mechanism Design for MobilePhone SensingDejun Yang, Guoliang Xue, Xi Fang,and Jian Tang, ACM MobiCom, Istan-bul, Turkey, August 2012

Mobile phone sensing is a newparadigm which takes advantage of thepervasive smartphones to collect andanalyze data beyond the scale of whatwas previously possible. In a mobilephone sensing system, the platformrecruits smartphone users to providesensing service. Existing mobile phonesensing applications and systems lackgood incentive mechanisms that canattract more user participation. Toaddress this issue, the authors designincentive mechanisms for mobile phonesensing. They consider two systemmodels: the platform-centric modelwhere the platform provides a rewardshared by participating users, and theuser-centric model where users havemore control over the payment theywill receive. For the platform-centricmodel, the authors design an incentivemechanism using a Stackelberg game,where the platform is the leader whilethe users are the followers. They showhow to compute the unique Stackel-berg Equilibrium, at which the utilityof the platform is maximized, and noneof the users can improve its utility byunilaterally deviating from its currentstrategy. For the user-centric model,the authors design an auction-basedincentive mechanism, which is compu-tationally efficient, individually ratio-nal, profitable, and truthful. Throughextensive simulations, they evaluate theperformance and validate the theoreti-cal properties of our incentive mecha-nisms.

A Picasso: Flexible RF and SpectrumSlicing Steven Hong, Jeff Mehlman, and SachinKatti, ACM SIGCOMM, Helsinki, Finland,August 2012

This paper presents the design,implementation and evaluation ofPicasso, a novel radio design that allowssimultaneous transmission and recep-tion on separate and arbitrary spectrumfragments using a single RF front endand antenna. Picasso leverages thiscapability to flexibly partition fragment-ed spectrum into multiple slices that

share the RF front end and antenna,yet operate concurrent and indepen-dent PHY/MAC protocols. The authorsshow how this capability provides ageneral and clean abstraction to exploitfragmented spectrum inWiFi networksand handle coexistence in dense deploy-ments. They prototype Picasso, anddemonstrate experimentally that aPicasso radio partitioned into fourslices, each concurrently operating fourstandard WiFi OFDM PHY and CSMAMAC stacks, can achieve the same sumthroughput as four physically separateradios individually configured to oper-ate on the spectrum fragments. Theauthors also demonstrate experimental-ly how Picasso’s slicing abstraction pro-vides a clean mechanism to enablemultiple diverse networks to coexistand achieve higher throughput, bettervideo quality and latency than the bestknown state of the art approaches.

Combined Relay Selection and Coop-erative Beamforming for Physical LayerSecurityJunsu Kim, Aissa Ikhlef, and Robert Schober,Journal of Communications and Networks,14(4), 364-373, August 2012

This paper proposes combined relayselection and cooperative beamformingschemes for physical layer security.Generally, high operational complexityis required for cooperative beamform-ing with multiple relays because of therequired information exchange and syn-chronization among the relays. On theother hand, while it is desirable toreduce the number of relays participat-ing in cooperative beamformingbecause of the associated complexityproblem, doing so may degrade thecoding gain of cooperative beamform-ing. Hence, the authors propose com-bined relay selection and cooperativebeamforming schemes, where only twoof the available relays are selected forbeamforming and data transmission.The proposed schemes introduce aselection gain which partially compen-sates for the decrease in coding gaindue to limiting the number of partici-pating relays to two. Both the caseswhere full and only partial channel stateinformation is available for relay selec-tion and cooperative beamforming areconsidered. Analytical and simulationresults for the proposed schemes showimproved secrecy capacities compared

to existing physical layer securityschemes employing cooperative relays.

Outage Probability and Outage-BasedRobust Beamforming for MIMO Inter-ference Channels with Imperfect Chan-nel State InformationJuho Park, Youngchul Sung, Donggun Kimand H. Vincent Poor, IEEE Transactions onWireless Communications, 11(10), 3561– 3573, Octorber 2012

In this paper, the outage probabilityand outage-based beam design for mul-tiple-input multiple-output (MIMO)interference channels are considered.First, closed-form expressions for theoutage probability in MIMO interfer-ence channels are derived under theassumption of Gaussian-distributedchannel state information (CSI) error,and the asymptotic behavior of the out-age probability as a function of severalsystem parameters is examined by usingthe Chernoff bound. It is shown thatthe outage probability decreases expo-nentially with respect to the quality ofCSI measured by the inverse of themean square error of CSI. Second,based on the derived outage probabilityexpressions, an iterative beam designalgorithm for maximizing the sum out-age rate is proposed. Numerical resultsshow that the proposed beam designalgorithm yields significantly better sumoutage rate performance than conven-tional algorithms such as interferencealignment developed under the assump-tion of perfect CSI.

DOF: A Local Wireless InformationPlaneSteven Hong and Sachin Katti, ACMSIGCOMM, Toronto, ON, Canada,August 2011

The ability to detect what unlicensedradios are operating in a neighborhood,their spectrum occupancies and thespatial directions their signals aretraversing is a fundamental primitiveneeded by many applications, rangingfrom smart radios to coexistence to net-work management to security. Thispaper presents DOF, a detector that ina single framework accurately estimatesall three parameters. DOF builds onthe insight that in most wireless proto-cols, there are hidden repeating pat-terns in the signals that can be used to

LYT-SCANNING-February_LYT/SCANNING/JUNE 2/21/13 3:22 PM Page 8


construct unique signatures, and accu-rately estimate signal types and theirspectral and spatial parameters. Thepaper shows via experimental evalua-tion in an indoor testbed that DOF isrobust and accurate, and achievesgreater than 85 percent accuracy evenwhen the Signal-to-Noise Ratios(SNRs) of the detected signals are aslow as 0 dB, and even when there aremultiple interfering signals present. Todemonstrate the benefits of DOF, theauthors design and implement a prelim-inary prototype of a smart radio thatoperates on top of DOF, and showexperimentally that it provides a 80 per-cent increase in throughput over Jello,the best known prior implementation,while causing less than 10 percent per-formance drop for co-existing WiFi andZigbee radios.

Interference Channel with ConstrainedPartial Group DecodingChen Gong, Ali Tajer, and XiaodongWang, IEEE Transactions on Communi-cations, vol. 59, no. 11, November 2011,pp. 3059–71

This paper proposes novel coding anddecoding methods for a fully connectedK-user Gaussian interference channel.Each transmitter encodes its informa-tion into multiple layers and transmitsthe superposition of those layers. Eachreceiver employs a constrained partialgroup decoder (CPGD) that decodes itsdesignated message along with a part ofthe interference. In particular, eachreceiver performs a twofold task by firstidentifying which interferers it shoulddecode and then determining which lay-ers of them should be decoded. Deter-mining the layers to be decoded anddecoding them are carried out in a suc-cessive manner, where in each step agroup of layers with a constraint on itsgroup size is identified and jointlydecoded while the remaining layers aretreated as Gaussian noise. The decodedlayers are then subtracted from thereceived signal and the same procedureis repeated for the remaining layers.The authors provide a distributed algo-rithm, tailored to the nature of theinterference channels, that determinesthe transmission rate at each transmit-ter based on some optimality measureand also finds the order of the layers tobe successively decoded at each receiv-er. They also consider practical designof a system that employs the quadra-ture amplitude modulations (QAM)and rateless codes. Numerical resultsare provided on the achievable sum-

rate under the ideal case of Gaussiansignaling with random codes as well ason the system throughput under practi-cal modulations and channel codes. Theresults show that the proposed multi-layer coding scheme with CPGD offerssignificant performance gain over thetraditional un-layered transmission withsingle-user decoding.

No Time to Countdown: MigratingBackoff to the Frequency DomainSouvik Sen, Romit Roy Choudhury, andSrihari Nelakuditi, ACM MobiCom, LasVegas, NV, USA, September 2011

Conventional WiFi networks performchannel contention in time domain.This is known to be wasteful becausethe channel is forced to remain idlewhile all contending nodes are backingoff for multiple time slots. This paperproposes to break away from conven-tion and recreate the backing off opera-tion in the frequency domain. The basicidea leverages the observation thatOFDM subcarriers can be treated asinteger numbers. Thus, instead of pick-ing a random backoff duration in time,a contending node can signal on a ran-domly chosen subcarrier. By employinga second antenna to listen to all thesubcarriers, each node can determinewhether its chosen integer (or subcarri-er) is the smallest among all others. Infact, each node can even determine therank of its chosen subcarrier, enablingthe feasibility of scheduled transmis-sions after every round of contention.The authors develop these ideas into aBack2F protocol that migrates WiFibackoff to the frequency domain.Experiments on a prototype of 10USRPs confirm feasibility, along withconsistent throughput gains over 802.11.Trace based simulations affirm scalabil-ity to larger, real-world network topolo-gies.

PACP: An Efficient PseudonymousAuthentication-Based Conditional Pri-vacy Protocol for VANETsDijiang Huang, Satyajayant Misra,Mayank Verma, and Guoliang Xue,IEEE Transactions on Intelligent Trans-portat ion Systems, vol. 12, no. 3,September 2011, pp. 736–46

This paper proposes a new privacypreservation scheme, named pseudony-mous authentication-based conditionalprivacy (PACP), which allows vehiclesin a vehicular ad hoc network(VANET) to use pseudonyms instead

of their true identity to obtain provablygood privacy. In the proposed scheme,vehicles interact with roadside units tohelp them generate pseudonyms foranonymous communication. In thescheme setup, the pseudonyms are onlyknown to the vehicles but have noother entities in the network. In addi-tion, the proposed scheme provides anefficient revocation mechanism thatallows vehicles to be identified andrevoked from the network if needed.Thus, the authors can provide condi-tional privacy to the vehicles in the sys-tem, that is, the vehicles will beanonymous in the network until theyare revoked, at which point, they ceaseto be anonymous.

Achieving MAC-Layer Fairness inCSMA/CA NetworksYing Jian, Ming Zhang, and Shigang Chen,IEEE/ACM Transactions on Network-ing, vol. 19, no. 5, October 2011, pp.1472–84

CSMA/CA networks, including IEEE802.11 networks, exhibit severe fairnessproblem in many scenarios, where somehosts obtain most of the channel’s band-width while others starve. Most existingsolutions require nodes to overheartransmissions made by contendingnodes and, based on the overheardinformation, adjust local rates toachieve fairness among all contendinglinks. Their underlying assumption isthat transmissions made by contendingnodes can be overheard. However, thisassumption holds only when the trans-mission range is equal to the interfer-ence range, which is not true in reality.As this paper reveals, the overhearing-based solutions, as well as severalnonoverhearing AIMD solutions, can-not achieve MAC-layer fairness in vari-ous settings. The authors propose anew rate control protocol, called Pro-portional Increase Synchronized multi-plicative Decrease (PISD). Withoutrelying on overhearing, it provides fair-ness in CSMA/CA networks, particular-ly IEEE 802.11 networks, by using onlylocal information and performing local-ized operations. It combines severalnovel rate control mechanisms, includ-ing synchronized multiplicativedecrease, proportional increase, andbackground transmission. The authorsprove that PISD converges and achieves(weighted) fairness. They further intro-duce Queue Spreading (QS) to achieveMAC-layer fairness when there aremultiple contention groups, in whichcase PISD will fail.

SC A N N I N G T H E L I T E R AT U R E

LYT-SCANNING-February_LYT/SCANNING/JUNE 2/21/13 3:22 PM Page 9

10 IEEE Wireless Communications • February 2013

t is well recognized that the conventional cellular networkscan no longer meet the user performance requirements in termsof throughput and coverage, which is due to the significant

increase in the number of mobile users along with their trafficdemands and the limited spectrum resource. This has stimulat-ed the search for new approaches to alleviate this problem. Recent-ly, cooperative communications at both the base station (BS)and mobile station (MS) levels has been proposed as a keytechnology to address this problem. Multicell cooperation is a newand emerging communication paradigm promising significantimprovement in system capacity by mitigating intercell interfer-ence. It has been shown that, with the aid of multicell coopera-tion, a significant improvement of cellular system performance canbe achieved, and the increase in throughput can be as large asan order of magnitude. Although the advantages of multicell coop-eration have been demonstrated, many problems related to theresearch and development of practical and effective schemesfor multicell cooperation exist. They have attracted consider-able research attention in both academia and industry. Thisspecial issue of IEEE Wireless Communicationsputs together somerecent results in the area of multicell cooperation for the nextgeneration cellular wireless networks. It includes 11 articles, a briefaccount for each of which is provided below.

In the first article, “Future Steps of LTE-A: Evolutiontoward Integration of Local Area and Wide Area” by Kishiya-ma et al., the authors focus on the importance of integratinglocal area and wide area in the future of LTE-Advanced (LTE-A) networks. The concept of a macro-assisted small cell isintroduced for this integration in a frequency-separateddeployment scenario. The authors also identify some potentialtechnologies such as 3D MIMO/beamforming, receiver inter-ference cancelation, and dynamic TDD, to enhance the spec-trum efficiency of higher frequency bands.

In the next article, “Multicell Cooperation: Evolution of Coop-eration and Cooperation in Large-Scale Cellular Networks” byBurchardt and Hass, the authors emphasize the importance of mul-ticell cooperation in large-scale cellular networks. They presentdifferent techniques to perform such cooperation includinguser-based cooperation, system-wide optimization, and large-scalemultiple-antenna systems. A recently proposed Pareto-based opti-mal OFDMA transmission scheduling method is introduced asan example in which both the concept of user-based coopera-tion and multiple-antenna systems are combined.

In the third article, “Multicell Cooperation for LTE-AdvancedHeterogeneous Networks Scenarios,” Soret et al. present twoscenarios in which simple multicell cooperation is used forLTE-A heterogeneous networks. In the co-channel deploy-ment scenario, the cross-tier interference is mitigated by usingenhanced intercell interference coordination, which alsoensures that the number of offloaded users is sufficient. In the otherscenario, different carriers are assigned to the deployed macro andsmall cells while using collaborative intersite carrier aggrega-tion to fully utilize the fragmented spectrum. In addition, the authorspoint out the importance of explicit terminal support as well as dis-tributed coordination among the base stations for both the sce-narios to perform optimally.

Interference alignment is emerging as a key transmissionstrategy to facilitate interference cancellation for communica-tion in interference channels. In the article “The PracticalChallenges of Interference Alignment,” El Ayach, Peters, andHeath, Jr. give an overview on interference alignment and itschallenges. These challenges include the system performancein realistic propagation environments, how to obtain the chan-nel state information at the transmitter, and the practicality ofinterference alignment in large-scale wireless networks.

Cooperation among multiple transmit antennas has been shownto be an effective interference management tool for communi-cations over interference channels. In the fifth article, “CSISharing Strategies in Wireless Networks,” de Karet and Ges-bert emphasize the cost of exchanging the channel state infor-mation (CSI) among the collaborating transmitters. Theypresent two scenarios, the interference alignment and networkMIMO scenarios, in which the need for full CSI sharing couldbe compromised to get more practical and scalable coopera-tion while maintaining the same performance. To reduce the amountof CSI sharing in the interference alignment scenario, theauthors use specific antenna configurations, whereas in thenetwork MIMO scenario they exploit the decay property of sig-nal power. Then they present the challenges of precodingdesign in the network MIMO channels.

The sixth article, “Multicell Cooperative Systems with Mul-tiple Receive Antennas” by Hwang et al., emphasizes theimportance of using multiple receive antennas in a multicellcooperative system. The authors review recent developments inmulticell advanced receiver-processing algorithms to mitigatethe interference. Next, they outline the main limitations to achieve

GUEST ED I T O R I A L

MULTICELL COOPERATION

I

EKRAM HOSSAIN ZANDER (ZHONGDING)LEI

LYT-GUEST EDIT-Hossain_Layout 1 2/21/13 4:39 PM Page 4


cooperation in multicell cooperative systems and provide anoverview of recent research efforts to overcome such limita-tions using multiple receive antennas.

Coordinated multipoint transmission/reception (CoMP), inwhich base stations cooperate to mitigate or eliminate intercellinterference, has been specified as one of the enabling tech-niques for LTE-Advanced networks (in Release 11). In thearticle “Interference Management through CoMP in 3GPPLTE-Advanced Networks,” Sun et al.discuss the performance ben-efits of CoMP in both downlink and uplink scenarios definedin the 3GPP LTE-Advanced standard. In downlink CoMP,they focus on four different techniques: transmission scheme (whichincludes dynamic point selection, dynamic point blanking, jointtransmission, and coordinated scheduling/beamforming), CSIreport, interference management, and reference signal design.Next, the authors explain briefly why it is easier to supportCoMP in the uplink than in the downlink. Then the efficiencyof CoMP is evaluated for four different scenarios defined inthe standard, and several unsolved problems as well as theirpossible solutions are also outlined.

In the next article, “How Do We Design CoMP to AchieveIts Promised Potential?,” Yang et al. first discuss the featuresof CoMP systems that are different from those of single-cell MIMOsystems. Then the challenges related to CSI acquisition are dis-cussed for both time-division duplexing (TDD) and frequency-division duplexing (FDD) systems. Also, the challenges thatarise due to the limited backhaul capacity for the cooperating basestations in a CoMP system are discussed, and possible approach-es to tackle these challenges are outlined.

In the ninth article, “On the Impact of Network GeometricModels on Multicell Cooperative Communications Systems,”Tukmanov et al. emphasize the trade-off between complexity andprecision in modeling multicell cooperative networks. Theauthors describe the Wyner model as a well-known deterministicapproach to model multicell cooperative networks, and then theypresent the stochastic geometry approach to model the random mul-ticell cooperative wireless networks. The authors give examplesof some common tools that can be used to model multicell coop-erative networks considering fixed base stations and random userlocations as well as random base stations and user locations.

An interesting overview on how to improve the energy effi-ciency in multicell cooperative cellular networks is given by Hanand Ansari in their article “On Greening Cellular Networks viaMulticell Cooperation.” The authors discuss how the number ofactive base stations can be reduced and thus energy saved byusing traffic-demand-aware multicell cooperation. Also, the authorsintroduce the concept of energy-aware multicell cooperation in whichthe power consumption of on-grid base stations is reduced by offload-ing traffic to off-grid base stations powered by renewable energy.To this end, the authors discuss how CoMP can be used toimprove the energy efficiency of cellular networks as well as thechallenges of energy-efficient multicell cooperation in next gen-eration cellular networks.

In the last article, “Optimal Trade-off between Power Saving andQoS Provisioning for Multicell Cooperation networks,” Zhang etal. develop a new scheme to achieve an optimal trade-off betweenthe power saving and QoS provisioning in multicell cooperation.To obtain the optimal power scheduling policy that minimizesthe power consumption at the base stations while guaranteeingQoS to the users, the authors formulate a stochastic optimizationscheme based on a semi-Markov decision process.

The articles included in this Special Issue consider differentaspects of multicell cooperative cellular wireless networkswhich will be very significant in the context of the evolvingLTE-Advanced (and beyond) networks. We hope that youenjoy reading this issue and find the articles useful. We wouldlike to thank all the authors for submitting their work and thereviewers for their time in reviewing the articles. Our special thanksto Dr. Hongyang Chen who first conceived the idea to organizethis Special Issue and has helped us throughout the process.Also, we acknowledge the support from Professor Hsiao Hwa Chenwho has guided us in this endeavor.

BIOGRAPHIESEKRAM HOSSAIN [S’98, M’01, SM’06] ([email protected]) is a professorin the Department of Electrical and Computer Engineering at the University ofManitoba, Winnipeg, Canada. He received his Ph.D. in electrical engineering fromthe University of Victoria, Canada, in 2001. His current research interestsinclude design, analysis, and optimization of wireless/mobile communicationsnetworks, cooperative and cognitive wireless networks, and green radio tech-nology. He has authored/edited several books in these areas (http://www.ee.uman-itoba.ca/~ekram). He serves as the Editor-in-Chief for IEEE Communications Surveysand Tutorials (for the term 2012–2013), and an Editor for IEEE Journal onSelected Areas in Communications - Cognitive Radio Series and IEEE Wireless Com-munications. Also, he serves on the IEEE Press Editorial Board (for the term2013–2015). Previously, he served as Area Editor for IEEE Transactions onWireless Communications in the area of Resource Management and MultipleAccess from 2009 to 2011 and an Editor for IEEE Transactions on MobileComputing from 2007 to 2012. He has won several research awards includ-ing the University of Manitoba Merit Award in 2010 (for Research and Schol-arly Activities), the 2011 IEEE Communications Society Fred Ellersick PrizePaper Award, and the IEEE Wireless Communications and Networking Confer-ence 2012 Best Paper Award. He is a Distinguished Lecturer of the IEEE Com-munications Society for the term 2012–2013. He is a registered ProfessionalEngineer in the province of Manitoba, Canada.

ZANDER (ZHONGDING) LEi [M’01, SM’04] is a senior scientist and laboratory head,Institute for Infocomm Research, Singapore. He received his B.S. (Honor, 1stupper) and Ph.D. degrees from Beijing Jiaotong University in 1991 and 1997,and his M.Eng. degree from Tong Ji University (Shanghai Railway Institute) in1994. From 1997 to 1999, he was a post-doctoral fellow at the National Uni-versity of Singapore. Since 1999 he has been with the Institute for Infocomm ResearchSingapore, leading research teams working on next generation cellular andlocal area networks. He was a technical leader in the IEEE 802.22 standardworking group that published the first IEEE standard for cognitive radio sys-tems. His research interests include next generation/emerging wireless commu-nication systems/technologies, cooperative communications, and cognitiveradio, among others. He has 18 patents granted and over 90 technical paperspublished. He is a co-recipient of the Institution of Engineering of SingaporePrestigious Engineering Achievement Awards in 2005 and 2007. He has beenthe PHY Technical Editor for the IEEE 802.22 standard since 2006. He is an Edi-tor for IEEE Communications Letters and IEEE Communications Magazine, aLead Guest Editor for a EURSIP WCN Special Issue on Multicell Cooperation,and a Guest Editor for a EURSIP WCN Special Issue on Cognitive Radio.

GUEST ED I T O R I A L

LYT-GUEST EDIT-Hossain_Layout 1 2/21/13 4:39 PM Page 5


http://www.ee.umanitoba.ca/~ekram

IEEE Wireless Communications • February 201312 1536-1284/13/$25.00 © 2013 IEEE

MU LT I C E L L CO O P E R AT I O N

INTRODUCTIONThe Third Generation Partnership Project(3GPP) standardized the radio interface specifi-cations for the next generation mobile communi-cations system, called Long-Term Evolution(LTE) [1, 2]. The initial specifications for LTE(i.e., Release 8) were finalized in 2008, and com-mercial LTE service in Japan was launched inDecember 2010. Standards development forLTE is continued toward establishing anenhanced LTE radio interface called LTE-Advanced (LTE-A; LTE Release 10 andbeyond), and now specifications for LTE Release

11 are in their finalization phase. In order to continue to ensure the sustain-

ability of 3GPP radio access technologies overthe coming decade, 3GPP standardization willneed to identify and provide new solutions thatcan respond to future challenges. Toward thisend, the 3GPP initiated discussions on furthersteps in the evolution of LTE toward the future(i.e., Release 12 and onward) [3]. The generalconsensus in 3GPP so far is that the furtherstandardization of LTE will continue to followstep-by-step evolutionary phases in the form ofLTE Releases 12, 13, and so on. We refer tothese new steps in the LTE evolution as LTE-B,LTE-C, and so on to indicate the next steps ofLTE-A. Here, LTE-B refers to the next set ofreleases, LTE Releases 12 and 13 [3].

This article provides our views on futurerequirements, the evolution concept, and candi-date solutions for the next steps after LTE-A.The goal is to establish a long-term evolutionconcept and identify the candidate technologiesrequired to satisfy the future requirements of thenext decade. The envisioned evolution conceptconsists of two key aspects:• The efficient integration of wide and local

area enhancements• The efficient utilization of higher and wider

frequency bands for small cells along withexisting lower frequency bands for macrocellsTo concretize this evolution concept, poten-

tial spectrum-efficiency-enhancing candidatetechnologies are identified for both the widearea and the local area, on both the transmitterand receiver sides. Furthermore, we introduce amacro-assisted small cell, called the Phantomcell, as a key solution to efficiently support theintegration of the local area with the wide area.A macro-assisted small cell, the Phantom cellindeed establishes a new form of multicell coop-eration between macrocells and small cellsrequired for further network densification andspectrum extension into higher frequency bands.In LTE-A, technologies for multicell coopera-tion between macrocells and small cells are stan-dardized such as coordinated multipoint (CoMP)

YOSHIHISA KISHIYAMA, ANASS BENJEBBOUR, AND TAKEHIRO NAKAMURA, NTT DOCOMO, INC.HIROYUKI ISHII, DOCOMO INNOVATIONS, INC.

ABSTRACTIn the future steps of 3GPP LTE-Advanced,

we will need to ensure the sustainability of 3GPPradio access technologies in order to respond tothe anticipated challenging requirements of thefuture. To this end, this article presents ourviews on the evolution concept and candidatetechnologies for future steps of LTE-A takinginto account the ever increasing importance oflocal area (small cells) and the need for furtherspectrum extension to higher frequency bands.In our evolution concept, we emphasize the inte-gration of the local area with the wide area as anew form of cooperation between conventionalmacrocells in lower frequency bands and smallcells deployed in higher frequency bands. Fur-thermore, we identify the potential key technolo-gies for further spectrum efficiencyenhancements for both the local area and thewide area, and for the efficient integration of thelocal and wide areas while assuming frequencyseparation between macrocells and small cells.In particular, multicell cooperation based onmacrocell assistance of small cells (referred to asPhantom cell) is introduced, and the relatedbenefits are discussed. Finally, in order todemonstrate the potential capacity gains of ourevolution concept, system-level simulation resultsare provided for using outdoor small cells inhigher/wider frequency bands.

FUTURE STEPS OF LTE-A:EVOLUTION TOWARD INTEGRATION OF

LOCAL AREA AND WIDE AREA SYSTEMS

KISHIYAMA LAYOUT_Layout 1 2/21/13 3:17 PM Page 12


transmission/reception and enhanced intercellinterference coordination (eICIC) [4]. SuchLTE-A technologies, however, were developedmainly assuming single-frequency operation formacrocells and small cells with co-channel inter-cell interference between them. In our evolutionconcept, the frequency-separated deploymentbetween the wide area and the local area is iden-tified as an important key scenario toward theefficient utilization of higher frequency bands.For such a frequency-separated scenario, thereare no issues related to macro-to-small intercellinterference, but other new issues related to cov-erage, mobility, and energy/battery savings arise.We first explain these new issues, and then dis-cuss the benefits associated with the Phantomcell solution. Finally, we provide the system-levelsimulation results for using outdoor small cellsin higher/wider frequency bands in order todemonstrate the potential gains associated withour evolution concept.

The rest of the article is organized as follows.We first present high-level requirements envi-sioning the next decade. We present the pro-posed evolution concept for the next steps afterLTE-A. We identify and discuss the candidatetechnologies related to the proposed evolutionconcept. We present results for system-levelevaluations. Finally, we present our conclusions.

FUTURE REQUIREMENTS

The most important requirement in the future isthe provision of higher system capacity [5]. Overthe past few years, there has been significantgrowth in the volume of mobile data traffic fol-lowing the proliferation of smartphones and newmobile devices that support a wide range ofapplications and services. Last year alone, thevolume of mobile data traffic grew 2.3 times,with a nearly threefold increase in the averagesmartphone usage rate [6]. There is now a gen-eral consensus in the industry that such growthwill continue in the future as well. Many recentforecasts project mobile data traffic to growbeyond 24 times between 2010 and 2015, andthus beyond 500 times in 10 years (2010–2020)assuming the same pace of growth is maintained.Thus, the capacity of future systems needs to beincreased significantly so that they can supportsuch an order of magnitude of growth in trafficvolume. Another important requirement is effi-cient support for a variety of traffic types includ-ing reduction in the impact of the ever increasingvolume of signaling traffic on both the networkand handheld device sides, the support of lowerlatency for the currently proliferating cloudapplications, and the handling of simultaneousconnections from the large number of small datapackets for machine-to-machine (M2M) traffic.

The next requirement is to provide betteruser experience by improving both the achiev-able data rates and fairness in user throughput.The target we set here is to achieve a tenfoldimprovement in the data rate over the nextdecade. We also aim to achieve gigabit-per-sec-ond-order user experience throughput in thewide area.

One more requirement is to design flexibleand cost-efficient network deployments. This

becomes of particular importance as future spec-trum utilization becomes more diverse, rangingfrom lower and narrower frequency bands tohigher and wider frequency bands, as the net-work density is further increased by deployingmore small cells, and as the simultaneous sup-port of diverse environments and deploymentscenarios becomes more important.

There are also other requirements regardingenergy/battery savings and system robustnessagainst emergencies (e.g., tsunamis and earth-quakes). These requirements, however, are animplementation issue for which non-standard-ized solutions would generally be more effective.Thus, we do not set them as a primary target,but rather as an add-on that complements otherrequirements as part of the whole system design.

CONCEPTUAL VIEWS ON SOLUTIONS

Prior to explaining in detail the candidate solu-tions to satisfy future requirements, we describethe proposed concept for future evolution. Inthe future, we believe that wide area enhance-ment will continue to be important, but localarea enhancement will become more importantthan in the past. Here we use the term local areato mainly refer to outdoor dense deploymentsand hotspots where a high volume of trafficneeds to be supported, and better user experi-ence throughput needs to be provided. The keychallenge we identify here resides in how toenhance and integrate the local area with thewide area in such a way that future requirementscan be satisfied.

The proposed concept for future evolutionconsists of two aspects:• Efficient integration of wide and local area

enhancements• Efficient utilization of both lower and higher

frequency bands through frequency-separateddeployments between wide and local areas

INTEGRATION OF WIDE AREA ANDLOCAL AREA ENHANCEMENTS

In Releases 8 to 11, some local area technolo-gies were introduced to LTE and LTE-A. As wemove toward Release 12 onward, the perfor-mance of LTE-B, C, and so on needs to beimproved further, and local area technologieswill play a more important role in the futuresteps of LTE-A. Thus, the performance of the

Figure 1. Future steps for LTE-A (i.e., LTE-B and C) based on the evolutionconcept of eLA as an integrated part of general LTE enhancements.

Common specificationsfor wide and local areas

Local area specificenhancements if sufficient

gain is identified

General enhancements toimprove performance also

in wide areasCA/eICIC/CoMP

for HetNetPico/Femto

eLA

LTE-B, C

LTE-A

LTE

Rel. 8/9

Rel. 10/11

Time

Perf

orm

ance

Evolution

Rel. 12 onward



local area should be further enhanced, and therequired technologies should be specified. Werefer to this as an enhanced local area (eLA), asshown in Fig. 1. However, eLA should be anintegrated part of LTE Release 12 onward alongwith general LTE enhancements. We shouldretain commonality between wide and localareas from the specification point of view, andenhance it as an integrated part of the generalenhancements for the wide area. Specific localarea enhancements can be specified only if a suf-ficient performance gain can be identified. Thus,the proposed way forward is to retain commonspecifications between wide and local areas interms of a fundamental LTE radio interface, andadd eLA specific specifications (e.g., a new carri-er type) only when necessary.

EFFICIENT UTILIZATION OF BOTH LOWER ANDHIGHER FREQUENCY BANDS THROUGH

FREQUENCY-SEPARATED DEPLOYMENTS BETWEENWIDE AND LOCAL AREAS

From the spectrum utilization point of view, thespectrum in the lower frequency bands is becom-ing scarce. Thus, it is crucial to explore and uti-lize higher frequency bands in the future.However, higher frequency bands are difficult toaccommodate in wide areas in macrocellsbecause of either space limitations on the eNodeB (eNB) side, for example, in terms of radio fre-quency (RF) equipment and antenna size, cover-age limitations (e.g., higher path loss), or costissues due to the need to alter the already estab-lished network infrastructure. The proposed wayforward is to use lower frequency bands such asexisting cellular bands in macrocells to providebasic coverage and mobility, and to use separatefrequency bands such as higher frequency bandsin the local area for small cells to provide high-speed data transmission, as shown in Fig. 2. Thefigure shows that in local the area, we can con-sider using a wider spectrum bandwidth in thehigher frequency bands and local-area-specifictechnologies to support smaller and denser celldeployments.

CANDIDATE SOLUTIONSA set of radio access technologies is required inorder to expand the current cellular capacity andperformance, and thus be able to meet futurerequirements and challenges. Hereafter, we pre-sent our identified candidate solutions to addressthe future requirements along with three evolu-tion dimensions: spectrum efficiency, spectrumextension, and network densification (referred toas “The Cube” in [3, 5]).

CANDIDATE TECHNOLOGIES FORENHANCED SPECTRUM EFFICIENCY

General Enhancements — LTE radio access basedon orthogonal frequency-division multiple access(OFDMA) achieves radio link performance thatis close to the Shannon capacity limit for point-to-point communications. However, furtheroverhead reduction or further enhancements forhigher signal-to-interference-plus-noise ratio(SINR) regions could still be considered. InLTE-A, the focus of studies in recent releaseshas shifted to the study of interference solutionsfor multipoint/antenna/multi-user radio links inorder to improve spectrum efficiency. The tech-nical challenges for intracell access (i.e., multi-user multiple-input multiple-output[MU-MIMO) and intercell access (i.e., CoMPtransmission technologies) will continue to bestudied in LTE-B and later. To enhance thespectrum efficiency further, hopefully withoutsignificant increase in the amount of channelstate information (CSI) feedback [7], we identifythe following two key approaches.

Active Beamforming with More Antenna Elements —Three-dimensional (3D) MIMO/beamformingwith active antenna system (AAS) is one of themajor technologies that would be studied inLTE-B [8]. AAS, where not only horizontalantennas but also vertical antenna elements haveindependent RF units, allows flexible beam con-trol in both horizontal and vertical directions, asshown in Fig. 3a1. From a specification perspec-tive, 3D MIMO/beamforming will require the

Figure 2. Frequency-separated deployments between a wide area and a local area.

Local area

Frequency

Very wide(ex. > 3 GHz)

Super wide(ex. > 10 GHz)

Existing cellular bands(high power density for coverage)

Wide area

Higher frequency bands(wider bandwidth for high data rate)

The proposed wayforward is to uselower frequencybands such as exist-ing cellular bands inmacrocells to providebasic coverage andmobility and to useseparate frequencybands such as higherfrequency bands inlocal area for smallcells to provide high-speed datatransmission.



support of more than eight antenna elements.The current physical layer specifications(Release 10) only support up to eight antennaports in the downlink (DL), for example, for theDL reference signals (RSs) and CSI feedbackscheme. Furthermore, extending 3D MIMO/beamforming to MU-MIMO and CoMP opera-tions would need to be investigated.

Massive MIMO, which to some extent isanother variant of 3D MIMO/beamforming withAAS but with very large numbers of antennas(e.g., more than 100 antenna elements), is a keytechnology for small cells in future higher fre-quency bands (e.g., beyond 10 GHz) [9]. Forhigher frequency bands, antenna elements canbe miniaturized, and more elements can beplaced in the same space; thus, very narrowbeamforming can be created, as shown in Fig.3a2. We expect such very narrow beamformingto become essential in higher frequency bands inorder to support practical coverage areas forsmall cells by compensating for the increasedpath loss. Assuming ideal beamforming gain, atwo-dimensional mapping of antenna elementscan compensate for the path loss with a frequen-cy factor of 20 dB/decade. However, there areseveral technical issues toward massive antennatechnologies that need to be resolved, such ashow to achieve accurate beamforming or how tosupport control signaling for mobility and con-nectivity over highly directive links. One possibil-ity to address the control signaling issue is toapply macro-assisted small cells (i.e., the Phan-tom cell), explained later.

Advanced Receiver Cancellation — Receiver cancella-tion technologies do not require much CSI feed-back, and are generally beneficial in both wideand local areas. In LTE Release 11, an advancedreceiver that employs interference rejectioncombining (IRC) was investigated for user equip-ment (UE) with two receiver (Rx) antennas [10].Thus, further study on the IRC receiver withmore Rx antennas (e.g., four or eight) will be anatural extension in LTE-B in order to utilizethe increased degrees of freedom at the UEreceiver for suppressing multiple interferenceresources, as shown in Fig. 3b1. CoMP technolo-gies considering the IRC receiver (e.g., interfer-ence alignment [11]) may be a potential topic forfuture steps beyond LTE-B, although this mightto some extent be achieved through eNB imple-mentations.

In addition to the IRC receiver, we considernonlinear interference cancellation, such as suc-cessive interference cancellation (SIC), as onekey future advanced receiver technology for thefuture radio access. However, one issue regard-ing the SIC receiver for the UE terminal side isthat the SIC receiver requires a condition inwhich interference signals to be cancelled aredecodable at the UE receiver. Thus, it may bechallenging to apply a SIC receiver to cancelintercell interference since those signals are typi-cally difficult to decode. Compared to that, theapplication of SIC to intracell multiple accessschemes can be more feasible. Initial investiga-tions reported that intracell non-orthogonal mul-tiple access using a SIC receiver improvesspectrum efficiency and user fairness [12].

eLA Focused Enhancements — The aforementionedapproaches are generally applicable to both wideand local areas. The following enhancements canbe additionally considered in order to satisfy therequirements specific to local area.

Dynamic Time-Division Duplex and Interference Coordina-tion/Management for Dense Small Cells — In smallcell deployments, it is expected that the numberof UE devices per small cell will not be large,and the traffic pattern in each small cell willchange widely over time depending on userapplications. As a result, spectrum sharingbetween the uplink (UL) and the downlink (DL),so-called dynamic time-division duplex (TDD)[13], would provide gains in terms of spectrumefficiency compared to the semi-static partitionof UL and DL. Study on dynamic TDD isalready underway in Release 11. However, whensmall cells are densely deployed, interferenceissues such as UL-to-DL or DL-to-UL interfer-ence, as shown in Fig. 4a, become more impor-tant. Therefore, interference coordination andmanagement schemes for dense small cells espe-cially for dynamic TDD need to be establishedin LTE-B.

Enhancements for the Higher SINR Region: UL OFDMAand 256-QAM — In small cell deployments, espe-cially in the aforementioned frequency-separateddeployments between wide and local areas, it isexpected that a relatively high SINR can oftenbe observed at the eNB or UE receiver due tolower intercell interference compared to macro-cellular deployments. Therefore, link-level per-formance enhancements for the higher SINRregion (e.g., higher than 20 dB) is beneficial inimproving the user-experience throughput inlocal area. Higher-order modulation such as 256-quadrature amplitude modulation (QAM) canbe effective in improving the throughput perfor-

Figure 3. General enhancements for spectrum efficiency: advanced transmitterand receiver: a1) 3D MIMO/beamforming; a2) massive MIMO in higher fre-quency bands; b1) advanced receiver with more Rx antennas; b2) non-orthogonal multiple access using a SIC receiver .

(a-1)

(a-2)

Very narrowbeamforming

(b-1)

(b-2)

Desiredsignal

Interferencesignals

Power control toachieve user fairness



mance in a high SINR region. For UL enhance-ments in a higher SINR region, the benefits ofOFDMA compared to current single-carrier fre-quency-division multiple access (SC-FDMA)need to be revisited. Also, hybrid access betweenOFDMA and current SC-FDMA [14], as shownin Fig. 4b, can be considered for UL radio access.

SPECTRUM EXTENSION USINGHIGHER FREQUENCY BANDS

Currently, the spectrum below 2.5 GHz is mostlyutilized by existing systems. Thus, spectrumextension to 3.5 GHz and higher frequencybands will be required in order to further expandthe system capacity in the future steps of LTE-A. As described earlier, in our proposed conceptfrequency-separated deployments between wideand local areas are emphasized toward the effi-cient use of higher frequency bands. Thus, spec-trum extension and network densification arehighly correlated. However, when we look intohigher frequency bands, (e.g., beyond 10 GHz),the current LTE radio interface, which was opti-mized for the existing cellular bands (i.e., around2 GHz) would not be optimum. Therefore, at acertain point in the future, it would become nec-essary to redesign the radio interface assumingthe requirements for future newly identifiedspectrum in the World Radiocommunicationconferences (WRC-15) and beyond.

From another point of view, the market sizefor utilizing higher frequency bands needs to besufficiently large. It is therefore desirable fromthe operator perspective that higher frequencybands can be used by many UE devices and formany service areas as much as possible. In thissense, higher frequency bands need to be uti-lized in various deployments, not only for indoorbut also for outdoor deployments. Figure 4shows two typical scenarios for outdoor smallcell deployments, that is, normal high trafficspots such as railway stations in suburban areasin which small cells are sparsely deployed, and asuper high traffic area such as the downtownarea in a large city in which small cells are dense-ly deployed to provide continuous or partial cov-

erage. Furthermore, macrocell deploymentsusing higher frequency bands should be allowed.

NETWORK DENSIFICATION WITHPHANTOM CELL SOLUTION

Macro-Assisted Small Cell: Phantom Cell — In the cur-rent deployments, there are a number of capaci-ty solutions for indoor environments such asWiFi, femtocells, and in-building cells using dis-tributed antenna systems (DASs). However,there is a lack of capacity solutions for the out-door environment that can also support goodmobility and connectivity. Thus, we propose amacro-assisted small cell, called the Phantomcell, as a capacity solution that offers goodmobility support while capitalizing on the exist-ing LTE network. In the Phantom cell solution,the control (C)-plane/user data (U)-plane aresplit as shown in Fig. 5. The C-plane of UE insmall cells is provided by a macrocell in a lowerfrequency band, while for UE in macrocells boththe C-plane and U-plane are provided by theserving macrocell, the same as in a conventionalsystem. On the other hand, the U-plane of UEin small cells is provided by a small cell using ahigher frequency band. Hence, these macro-assisted small cells are called Phantom cells, asthey are intended to transmit UE-specific signalsonly, and the radio resource control (RRC) con-nection procedures between the UE and a Phan-tom cell, such as channel establishment andrelease, are managed by the macrocell. ThePhantom cells are not conventional cells in thesense that they are not configured with cell-spe-cific signals and channels, that is, primary/sec-ondary synchronization signals (PSSs/SSSs),cell-specific reference signals (CRSs), masterinformation block and system information blocks(MIB/SIBs), and so on. Their visibility to the UErelies on macrocell signaling.

The Phantom cell solution comes with a rangeof benefits. One important benefit of macroassistance of small cells is that control signalingdue to frequent handover between small cellsand macrocell and among small cells can be sig-nificantly reduced, and connectivity can be main-

Figure 4. eLA focused enhancements and deployment scenarios for higher frequency bands.

(a) (b)

DFT (SC)Wide area

Local area

Tx data Tx signalIFFT

DL

DL2ULinterference

Super high traffic area(outdoor dense small cells)

ULUL2DL

interference

Normal high traffic spot(outdoor hotspot cell)

S/P (MC)

At a certain point intime in the future, it would becomenecessary toredesign the radiointerface assumingthe requirements forfuture newly identified spectrumin the World radio-communication conferences (WRC-15) andbeyond.



tained even when using small cells and higherfrequency bands. In addition, by applying thenew carrier type (NCT) that contains no orreduced legacy cell-specific signals, the Phantomcell is able to provide further benefits such asefficient energy savings, lower interference trans-missions (and thus higher spectrum efficiency),and reduction in cell planning efforts for densesmall cell deployments since Phantom cells donot rely on legacy cell-specific signals.

Technical Issues Related to the Phantom Cell

Network Architecture — To establish a networkarchitecture that supports the C/U-plane split,interworking between the macrocell and thePhantom cell is required. A straightforwardsolution to achieve this is to support Phantomcells by using remote radio heads (RRHs)belonging to a single macro eNB. This approachcan be referred to as intra-eNB carrier aggrega-tion (CA) using RRHs [15]. However, such atight CA-based architecture has some draw-backs as it requires single-node operation withlow latency connections (e.g., optical fibers)between the macro and Phantom cells. In thefuture, more flexible network architecturesshould be investigated to allow for relaxed back-haul requirements between macro and Phantomcells and to support a distributed node deploy-ment with separated network nodes for each(i.e., inter-eNB CA).

Macro-Assisted Discovery — In frequency-separateddeployments, efficient small cell discovery inhigher frequency bands is an important technicalissue. To achieve efficient discovery (e.g., for UEbattery savings), we propose to use a macro-assisted property in the Phantom cell conceptand to introduce newly defined discovery signals,which are transmitted by small cells in a time-synchronized manner with macro downlink sig-nals. Assisted by the C-plane provided by themacrocell, small cells simultaneously transmitthe discovery signals with a relatively long trans-mission interval (e.g., longer than 100 ms), andthe UE attempts the detection of discovery sig-nals from small cells only in a short time inter-val. Thus, the interfrequency measurement effortand battery consumption of UE to detect thesurrounding small cells in higher frequencybands can be reduced. Furthermore, networkenergy savings when there is no traffic are alsoexpected due to the long transmission interval.The discovery signal should be designed to satis-fy the requirements for small cell deploymentsuch as robustness against intercell interferencefor the scenarios including user-deployed orclosed subscriber group (CSG) cells, support fora large number of sequences to reduce the cellplanning effort, and so on.

Flexible Duplex Support — In frequency-separateddeployments, different duplex schemes, such asfrequency-division duplex (FDD) and TDD, maybe used for lower and higher frequency bands,respectively. In the future, it is desirable to sup-port the Phantom cell (or inter-eNB CA) solu-tion irrespective of the combination of duplexschemes in lower and higher frequency bands.

SIMULATION EVALUATIONSIn this section, we present system-level simulationresults to demonstrate the potential capacity gainsof our evolution concept where outdoor small celldeployments are assumed using higher/wider fre-quency bands. The simulator we used is built in Clanguage and consists of system-level modeling uti-lizing exponential SINR link-to-system-level map-ping [16]. Figure 6 shows the system throughputperformance per square kilometer as a function ofthe transmission bandwidth assuming variable car-rier frequencies ranging from 3.5 to 40 GHz. Themain simulation parameters are also depicted inthe same figure. In this evaluation, we assume 12small cells are randomly located within each sector,where each macrocell site consists of three sectorsand intersite distance is set to 500 m. Thus, the sizeof each sector is about 0.072 km2, and the densityof small cells is 166.7 cells/km2. We also assumethat all U-plane related transmissions are conduct-ed by small cells. In order to evaluate the impact ofpropagation loss increase in higher frequencybands, the distance-dependent path loss model in[17] is assumed. Considering that all small cells aredeployed outdoors to further enhance systemcapacity, we evaluate two cases that differ in termsof the penetration loss. The first one is the casewithout penetration loss assuming all users are out-doors, and the other one is the case with penetra-tion loss (= 20 dB) assuming all users are indoors.We note that the simulation results for the macro-cell only case (without small cells) is 310 Mb/s/km2

assuming a transmission bandwidth of 10 MHz in a2 GHz carrier frequency.

In the case without penetration loss, the sim-ulation results (solid lines) show that spectrumextension using higher frequency bands in smallcells significantly increases the system through-put per square kilometer. For example, a systemthroughput of 31.0 Gb/s/km2, which correspondsto a 100 times gain compared to that for themacrocell only case, is achieved by transmissionbandwidth of 170, 230, 300, and 400 MHz forcarrier frequency of 3.5, 10, 20, and 40 GHz,respectively. Meanwhile, for the case with a 20dB penetration loss included, the simulationresults (dotted lines) show that spectrum exten-sion for the small cells using the higher frequen-cies beyond 10 GHz is not as effective due to theincrease in the compound loss due to path lossand penetration loss. However, spectrum exten-sion in 3.5 GHz carrier frequency remains veryeffective. Based on the results, we would like toexpress the following observations.

Figure 5. Phantom cell: macro-assisted small cell.

Macrocell

Split!!

3.5 GHz(example)

2 GHz(example)

Phantom cell

C-pl

ane

(RRC

)

U-plane



• In 3.5 GHz carrier frequency, outdoor smallcell deployment combined with the spectrumextension is very effective to boost the systemcapacity.

• In higher frequency bands beyond 10 GHz,technology to overcome the propagation loss,such as beamforming (massive MIMO) ishighly required to ensure outdoor small cellsare useful not only for outdoor users but forindoor users as well.

CONCLUSION

In this article we present our views on therequirements, evolution concept, and candidatetechnologies for the future steps of LTE-A. Theestablishment of the evolution concept is of par-ticular importance in setting the evolution frame-work so that 3GPP standardization can focus thespecification work on usage cases and deploy-ment scenarios that really matter not only in theshort term but also in the long term. In the pro-posed evolution concept we emphasize on theimportance of integrating local area into widearea enhancements. Also, we introduce the setof radio access technologies we have identifiedas the most promising for improving spectrumefficiency in both wide and local areas. In partic-ular, we have identified frequency-separateddeployment between wide and local areas as animportant scenario toward efficient utilization ofhigher/wider frequency bands in local area whilestill maintaining basic mobility and connectivityin wide area over existing lower frequency bands.For this scenario, the macro-assisted small cell,called the Phantom cell, with a split of C-planeand U-plane between the macrocell and small(Phantom) cells, is introduced as a new form ofmulticell cooperation. In order to confirm thepotential benefits of our evolution concept, sim-ulation evaluation results assuming outdoorsmall cell deployments using higher/wider fre-quency bands are presented. Outdoor small celldeployment combined with spectrum extensionis shown as an effective approach to boost the

system capacity and achieve the future require-ments.

REFERENCES[1] M. Tanno et al., “Evolved UTRA — Physical Layer

Overview,” Proc. IEEE SPAWC 2007, June 2007. [2] E. Dahlman, S. Parkvall, and J. Sköld, 4G –LTE/LTE-Advanced

for Mobile Broadband, Academic Press, 2011.[3] NTT DOCOMO, “Requirements, Candidate Solutions &

Technology Roadmap for LTE Rel-12 onward,” 3GPPRWS-120010, June 2012.

[4] T. Abe et al., “Radio Interface Technologies for Cooper-ative Transmission in 3GPP LTE-Advanced,” IEICE Trans.Commun., vol. E94-B, no. 12, Dec. 2011, pp. 3202–10.

[5] Y. Kishiyama, “LTE Enhancements and Future Radio Access,”Seminar on Future Wireless Technology, Nov. 2010.

[6] Cisco Whitepaper, “Cisco Visual Networking Index:Global Mobile Data Traffic Forecast Update, 2011-2016,” 15 Feb 2012.

[7] S. A. Ramprashad et al., “Cooperative Cellular NetworksUsing Multi-User MIMO: Tradeoffs, Overheads, andInterference Control Across Architectures,” IEEE Com-mun. Mag., vol. 49, no. 5, May 2011, pp. 70–77.

[8] O. Yilmaz, S. Hamalainen, and J. Hamalainen, “SystemLevel Analysis of Vertical Sectorization for 3GPP LTE,”Proc. ISWCS 2009, Sept. 2009.

[9] T. L. Marzetta, “Non-Cooperative Cellular Wireless WithUnlimited Numbers of Base Station Antennas,” IEEETrans. Wireless Commun., vol. 9, no. 11, Nov. 2010.

[10] Y. Ohwatari et al., “Performance of Interference Rejec-tion Combining Receiver to Suppress Inter-Cell Interfer-ence in LTE-Advanced Downlink,” IEICE Trans.Commun., vol. E94-B, no. 12, Dec. 2011, pp. 3362–69.

[11] R. Cadambe and S. A. Jafar, “Interference Alignmentand Degrees of Freedom of the K-User InterferenceChannel,” IEEE Trans. Info. Theory, vol. 54, no. 8, Aug.2008, pp. 3425–41.

[12] K. Higuchi and Y. Kishiyama, “Non-Orthogonal Accesswith Successive Interference Cancellation for FutureRadio Access,” APWCS2012, Aug. 2012.

[13] J. Li et al., “Dynamic TDD and Fixed Cellular Net-works,” IEEE Commun. Letters, vol. 4, no. 7, July 2000,pp. 218–20.

[14] R. Dinis et al., “A Multiple Access Scheme for theUplink of Broadband Wireless Access,” Proc. IEEEGLOBECOM, vol. 6, Dec. 2004, pp. 3808–12.

[15] M. Iwamura et al., “Carrier Aggregation Framework in3GPP LTE-Advanced,” IEEE Commun. Mag., vol. 48, no.8, Aug. 2010, pp. 60–67.

[16] K. Brueninghaus et al., “Link Performance Models forSystem Level Simulations of Broadband Radio AccessSystems,” Proc. IEEE PIMRC ’05, vol. 4, Sept. 2005.

[17] ITU-R Rep. M.2135, “Guidelines for Evaluation ofRadio Interface Technologies for IMT-Advanced,” Nov.2008.

Figure 6. Throughput gains using higher frequency bands (12 small cells per macro sector).

Transmission bandwidth (MHz)2000

10

0

Syst

em t

hrou

ghpu

t (G

b/s/

km2 )

20

30

40

50

60

400 600 800

Macrocell layout

Number of small cells

Number of users

Moving speed of UE

Antenna configuration

Tx power of small cell

Channel mode

Traffic model

Scheduling algorithm

Distance dependentpath loss

Shadowing standarddeviation

19 cells/3 sectors per cellintersite distance = 500 m

12 per macro sector

30 per macro sector

3 km/h

2 × 2

33 dBm

Pedestrian-A

Full buffer

Proportional fairness

36.7log10(d) + 26log10(f)+ 22.7 (d in m, f in GHz)

8 dB

100x macro only(31.0 Gb/s/km2)

1000

Without PLWith PL (20dB)3.5 GHz10 GHz20 GHz40 GHz

The establishment ofthe evolution conceptis of particular impor-tance in setting theevolution frameworkso that 3GPP stan-dardization can focusthe specificationwork on usage casesand deployment scenarios that reallymatter not only inthe short term but also on the long term.




INTRODUCTIONSince the introduction of mobile technologiesover two decades ago, wireless communicationhas evolved into a utility similar to water andelectricity, needed by almost all people of today’smodern society. To support the ever growingdemand for mobile communications, cellularnetworks have had to evolve from simple localservice providers to massively complex coopera-tive systems. While on one hand the underlyingtransmission methods have been systematicallyadvanced with each generation of wireless tech-nologies, that is:• Time-/frequency-division multiple access

(T/FDMA) in second generation (2G) net-works

• Code-division multiple access (CDMA) in thirdgeneration (3G) networks

• Orthogonal frequency-division multiple access(OFDMA) in fourth generation (4G) networks

to improve the capacity and availability of thesenetworks; on the other hand, the inherentlyenhanced complexity of these transmissionschemes necessitates more intelligent methods ofcooperation and system coordination. Ultimate-ly, the dramatically increasing demand formobile communications [1] requires more andmore intelligent utilization of the radio frequen-cy (RF) spectrum, and enhanced cooperationbetween cells.

Before we begin, the ultimate goals of multi-cell cooperation must be introduced. By coordi-nating resource and power allocation overmultiple cells (or indeed a whole network), thesystem is able to provide simultaneous service tothousands of users. Thus, the main goals ofcooperation can be compiled as: • The optimal utilization of radio resources

within a cellular network• The minimization of interference to neighbor-

ing cells• Hence, the maximization of the number of

simultaneously transmitting usersIn the rest of this introduction, we describe

the most basic forms of network managementwithout cooperation, and highlight the need formore advanced coordination techniques. Follow-ing this, we investigate three areas of multicellcooperation and how they intend to solve theproblem of increased throughput demands forthe limited bandwidth available for wireless sys-tems. Finally, we conclude with a performanceanalysis of an example of such techniques, andthe envisioned further evolution to future mobilenetworks.

FREQUENCY REUSEThe most basic form of interference mitigationutilized in wireless networks is that of frequencyreuse. Consider two mobile stations (MSs) inneighboring cells transmitting on the same bandof frequencies; these will surely interfere witheach other and, consequently, degrade eachother’s signal. Hence, the logical conclusionwould be to allocate orthogonal bandwidths toavoid interference. However, due to the quantity

HARALD BURCHARDT AND HARALD HAAS, UNIVERSITY OF EDINBURGH

ABSTRACTDue to the large growth of mobile communi-

cations over the past two decades, the support-ing cellular systems have continuously needed toexpand and evolve in order to meet the everincreasing demand of wireless connections.While simple frequency reuse and power controlwere enough to manage the demand of thesenetworks at first, multicellular cooperation andcoordination have become paramount to theoperation of larger and more highly utilizedcommunication systems. This article discussesthe development of different techniques forcooperation in large cellular networks, andoffers insights into the need for such an evolu-tion. We investigate various branches of multi-cell cooperation, including user-basedcooperation, system-wide optimization, and theopportunity of multiple-base-station transmis-sion. Furthermore, we offer an example of arecently proposed technique designed for multi-cell cooperation in full frequency reuse orthogo-nal frequency-division multiple access networks.

MULTICELL COOPERATION:EVOLUTION OF COORDINATION AND COOPERATION IN

LARGE-SCALE NETWORKS

BURCHARDT LAYOUT_Layout 1 2/22/13 11:23 AM Page 19


of MSs requesting resources and the limitedbandwidth available, it is clear that not all userscan be allocated different frequencies (i.e., thefrequencies must be reused). Utilizing the samefrequency bands in neighboring cells can bedetrimental to the achievable throughputs ineach cell (as stated in the example); however, asthe wireless signal attenuates directly as a func-tion of the distance from the transmitter, a fre-quency band may be reused in cells that aresufficiently far away from each other. Thus, afrequency reuse scheme results (shown on theleft of Fig. 1) in two neighboring cells neversharing the same set of transmission frequencies,and MSs can transmit virtually interference-free.

Fractional Frequency Reuse —While MSs in neigh-boring cells do not interfere with each other, itis clear that only a portion (1/3 in the examplein Fig. 1) of the already limited system band-width is available in each cell; therefore, thenumber of users servable by the system is dra-matically reduced. Therefore, fractional fre-quency reuse (FFR) is adopted in more modernwireless systems [2], where the center of all cellsutilize the same frequency band, and only thecell edges (which are more vulnerable to inter-ference from other cells as they are closer)employ frequency reuse. An example of this isshown on the right of Fig. 1, and it is clear thatthe bandwidth available in each cell has doubledin comparison to traditional frequency reuse.However, in future systems where full frequencyreuse (i.e., the full bandwidth is used in all cells)is planned, cells will most certainly need tocooperate in order to ensure interference miti-gation for their users.

POWER CONTROLIn general, a wireless base station (BS) willtransmit at maximum power in the downlink inorder to serve the MSs connected to it. Howev-er, in the uplink this can cause substantial diffi-culties such as large interference at aneighboring BS or massive received signal differ-ences (from different users) at the same BS,

referred to as the near-far problem. Therefore,transmit power control (TPC) is utilized in theuplink of most mobile communications systemsto balance the signal-to-noise ratios (SNRs) ofthe MSs in a cell at the BS, and hence easesimultaneous reception and create a fairer sys-tem. In TPC, each MS transmits at a power levelthat is just large enough to overcome the pathloss, or signal attenuation, from itself to its serv-ing BS. Through this, nearby users will transmitwith much lower power than users at the celledge, as a high-power user near the BS wouldessentially drown out farther users. Furthermore,due to the significantly reduced transmit powers,the interference to neighboring cells is limited,additionally easing simultaneous transmission ofusers throughout the network. This will be essen-tial for full frequency reuse networks.

While these basic forms of coordination wereable to satisfy early networks’ service demands,the vast increase in cellular activity cannot becompensated by these single-cell-managementtechniques. Hence, multicell cooperation hasbecome more than necessary.

USER-BASED COOPERATION

Due to the ever increasing complexity and uti-lization of modern mobile communications net-works, it has become more and more difficult toserve the large number of MSs requesting ser-vice. Furthermore, resource allocation on a sin-gle-cell basis is no longer able to sustain thedemands of these networks; hence, users oftenfall into outage. In this section, a class of tech-niques that utilize individual users’ transmissionrequirements to perform multi-cell resource allo-cation is discussed. Here, cells adjust theirresource selection and scheduling based on theMSs in neighboring cells already transmitting. Afew examples of such techniques are discussedbelow.

DYNAMIC FREQUENCY REUSEThe main drawback of static FFR is the inabilityto adapt to an ever changing environment andthe consequent waste of wireless resources.Therefore, in dynamic frequency reuse (DFR) [3,4] the allocation of frequency subbands to differ-ent cells is determined based on the contempo-rary resource allocation and consequentimmediate interference environment in the cellu-lar system. Ideally, we would prefer all frequen-cies to be available in all cells, so at minimum anFFR scheme with a common frequency band inthe cell centers is applied. Subsequently, the fre-quencies allocated to the cell edge may be adapt-ed based on the use of specific frequency bandsin the neighboring cells, and the interferenceincident from these to the cell of interest. Clear-ly, this necessitates some signaling between BSssuch that the possibility of close frequency reusecan be determined in each cell. Therefore, thefrequency and consequent resource allocation ineach cell is adapted dynamically to the behaviorof the cellular system.

DFR may be implemented by a central ordistributed approach. In centralized approaches,resources are assigned to BSs by means of a cen-tral controller, which generally achieves more

Figure 1. On the left is the most basic from of frequency reuse, where the avail-able bandwidth is divided into three equal bands, and distributed such that noneighboring cells utilize the same band. On the right, FFR improves the spa-tial reuse of resources by reusing the same band in the center of the cell, andprotecting the edges through standard frequency reuse.

f f

BURCHARDT LAYOUT_Layout 1 2/21/13 3:16 PM Page 20


efficient resource utilization, at the expense ofadditional signaling and higher complexity of thenetwork infrastructure. In a distributedapproach, each BS autonomously carries out theresource allocation, attempting to individuallyreact to the interference environment. In dis-tributed DFR methods, however, BSs generallymay access only a predefined number of sub-bands, so these approaches provide little flexibil-ity in subband reassignment when interferenceconditions change.

SUCCESSIVE INTERFERENCE CANCELLATIONWhile frequency reuse schemes help protectcells from interference, they are unable to active-ly eliminate co-channel interference (CCI) at theMS. This challenge, however, can be metthrough interference cancellation. To accuratelydefine successive interference cancellation (SIC),we use a definition given in [5], stating that“interference cancellation should be interpretedto mean the class of techniques that demodulateand/or decode desired information, and then usethis information along with the channel esti-mates to cancel received interference from thereceived signal.” In other words, signal process-ing is utilized after reception to improve thequality of the desired information, while remov-ing the interference from the received signal. Anexample of a simple implementation of SIC isshown in Fig. 2.

In SIC, the strongest signal (it has, inevitably,the highest signal-to-noise-plus-interference ratio[SINR]) is detected and decoded first, then thenext strongest, and so on. After the received sig-nals have been reconstructed (through therespective channel estimations), these can besubtracted/cancelled from the compositereceived signal to improve the performance ofthe remaining MSs (i.e., their signal detection).Thus, with each removed signal, the interferencein the composite received signal is reduced, andthe remaining signal SINRs consequentlyenhanced.

It has been shown that SIC can nearly achievethe Shannon capacity for multi-user additivewhite Gaussian noise channels [5]. Many well-known examples of interference cancellation areutilized today, such as the decision feedbackequalizer (DFE) and the original Bell Labs lay-ered space-time (BLAST) system. In the DFE,previously decoded symbols are utilized to can-cel intersymbol interference (ISI) of futurereceived symbols, given that the channel isknown. In BLAST decoding, spatial interferencein multiple-input multiple-output (MIMO) sys-tems can be removed in order to separate spa-tially multiplexed streams of data. Overall, thereare many similarities [5] between the eliminationof: • ISI (equalizers)• Spatial interference (MIMO receivers)• Consequently, multi-user interference cancel-

lationHence, while SIC for multicell cooperation maybe rather novel, the idea of interference cancel-lation has been well proven over the years.

Of course, the great benefits that come fromSIC in multi-user environments come at a price,which is not always satisfiable. In general, the

concept of interference cancellation relies on thepremise that the received signal can be reliablyestimated. This means that not only can thedesired signal be decoded, but also the receivedinterfering signals. For this, a receiver mustknow both what was transmitted by an interferer(i.e., what data was sent with which modulationand coding scheme [MCS]) and the channel overwhich it was sent (i.e., how it has been modi-fied). On one hand, the system may signal amongthe BSs the MCSs of all users simultaneouslytransmitting on the same frequencies such thatthis interference can be cancelled; this would,however, drastically increase the signaling bur-den on the network (e.g., in comparison to fre-quency reuse techniques). On the other hand,even if this information is known, the receivermust perfectly estimate the interfering channelto be able to decode the signal, which is highlyimprobable. Hence, while SIC provides greatpromise for multicell coordination, its require-ments are very difficult to fulfill.

UPLINK INTERFERENCE PROTECTIONIt is clear that the high complexity of SIC and itsheavy reliance on channel estimation make it dif-ficult to implement in cellular communicationnetworks. In addition, multi-cell cooperation andcoordination are more difficult to achieve in theuplink; however, here is where it is usually mostneeded. Because each MS has a much more limit-ed power budget than any BS, reducing interfer-ence in the uplink is paramount to modern mobilecommunications. Therefore, much work (e.g.,TPC, briefly discussed earlier) is dedicated toalleviating the interference, especially that inci-dent on cell edge users in the reverse link. Onesuch technique recently proposed is uplink inter-ference protection (ULIP), summarized here.

Whereas TPC determines MS transmit powerbased on local information, ULIP is a techniquethat actively combats CCI in the uplink of a cel-lular system [6]. The level of uplink interferenceoriginating from neighboring cells (affecting co-channel MSs in the cell of interest) can be effec-tively controlled by reducing the transmit powerof the interfering MSs (i.e., rather than attempt-ing to cancel the signal entirely). This is donebased on the target SINR and tolerable interfer-ence of the vulnerable link. Bands are prioritizedin order to differentiate those (vulnerable/ vic-tim) MSs that are to be protected from interfer-

Figure 2. An example of SIC, where the quality of the desired signal isimproved by successively removing the detected interfering signals from thereceived signal.

Decoderb^

kLinearreceiver

Imperfectchannel

estimationRe-encode

and modulate

Updatecomposite(received)

signal

yk+1(t)=yk(t)-zk(t)

zk(t)

y0(t)



ence and those (aggressor/interfering MSs) thatare required to sacrifice transmission power tofacilitate the protection. Furthermore, MSs arescheduled such that those users with poorertransmission conditions receive the highest inter-ference protection, thus balancing the arealSINR distribution and creating a fairer alloca-tion of the available resources. In addition tointerference protection, the individual powerreductions also serve to decrease the total sys-tem uplink power, resulting in a greener system.

However, while ULIP significantly protectscell edge users from uplink interference, it isclear that a substantial amount of signaling isnecessary to facilitate this protection. Not onlymust the tolerable interference of each user becalculated and sent to its interferers; the inter-fering channel to that user must also be estimat-ed. Fortunately, this can be done at theinterfering MS using the reference signalreceived power (RSRP) of the cell with which itis interfering, and thus does not add any signal-ing complexity to the network.

SYSTEM-WIDE OPTIMIZATION

In the user-target-based cooperation techniquesdiscussed above, it is clear that a significantamount of information must be signaled betweenBSs and MSs in order to facilitate successfulsimultaneous transmission in multicellular sys-tems. Furthermore, because these techniques relyon individual users’ requirements, it is clear thatthey very rarely result in a system-optimal solu-tion. Therefore, another class of techniques isconsidered where, assuming a central controllerwith all necessary information, the multicell coop-eration and resource allocation problem can beexpressed as an optimization problem. Throughthis, such techniques attempt to reach globallyoptimum solutions for a system, at the cost ofnear unlimited knowledge at the controller. Someexamples of these techniques are discussed here.

CAPACITY MAXIMIZATIONIn wireless networks, the resource and powerallocation problem of the system as a whole isoften posed as an optimization problem where

the objective is the maximization of the achiev-able throughput (i.e., sum rate) of the system[7]. Of course, individual user requirements mustalso be met, constraining the set of solutions tothe problem. Therefore, a system optimizationproblem is generally constructed as follows:

where the most common constraints posed arethe user rate requirements and transmit powerlimitations. However, these may be extended tomore detailed QoS requirements such as delay,buffer length, and priority. By solving the aboveproblem, a system-wide optimal resource alloca-tion for the network can be found.

Although such a method seems preferable toany individual-user-driven cooperation tech-niques, there are two significant drawbacks ofsystem optimization that greatly reduce its appli-cability to wireless cellular networks. First of all,the biggest difficulty of such an optimizationproblem is that in almost all cases this problembecomes NP-hard, and is therefore practicallyunsolvable using standard optimization tech-niques. In most cases, the scheduling problem isthen broken down into multiple optimizationproblems, separating the resource allocation andtransmit power determination, resulting in ulti-mately suboptimal solutions. Furthermore,because of the interdependency of, for example,subcarrier allocation and transmit power, solvingthe resource allocation does not guarantee asolution to the transmit power problem. There-fore, the optimality of these techniques is hardlyever achieved.

Second, not only is there a massive computa-tional complexity involved, but also the largeamount of system information that is required bythe central controller to solve these system-wideproblems must be communicated from each BS,implying a tremendous signaling burden on thenetwork. Not only must the individual userrequirements be signaled, but also the desiredand interfering path gains of each MS in the sys-tem must be known at the central controller inorder for an optimal resource and power alloca-tion to be formed. Therefore, while in theorythese problems present optimal network solu-tions, they are in practice almost impossible toimplement.

POWER MINIMIZATIONAnother form of the above optimization prob-lem is, rather than maximize the sum rate, tominimize the total transmit power utilized in thewireless network [8]. Here, clearly, the powerconstraint falls away as it implies the maximiza-tion of network performance with maximumpower.

On the other hand, the individual user

∑

∑

= …

≥ ∀≤ ∀

≤

⎧

⎨

⎪⎪⎪

⎩

⎪⎪⎪

i N

i

i

Objective :max UserRate 1,2, users

subject to:

Constraints :

UserRate RateReq

UserDelay MaxDelay

UserPower TotalPower

ii

i i

i i

i

Figure 3. A visualization of CoMP, where the users within neighboring cellsthat experience identical sets of strongest BSs can be served simultaneouslyand cooperatively by this set of BSs, eliminating interference from what wouldbe the strongest interfering cells.

Area

Cooperation

User group



requirements are the main constraints, whereespecially the rate of each user must be main-tained; otherwise, a system without any transmis-sion would result. This form of optimization isbeneficial in two main ways: • The minimization of interference to other cells

eases power and resource allocation over thenetwork.

• The minimization of transmit power results inmuch greener networks, a topic of fervent dis-cussion for future wireless communications.

However, the same difficulties apply to thisproblem as to the rate maximization variation,and hence the practical implementation of sucha system is highly complex.

LARGE-SCALEMULTIPLE-ANTENNA SYSTEMS

In the previous sections, we have discussed twocontrasting forms of multicell cooperation andcoordination: lower-complexity user-based coop-eration and substantially complex system opti-mization. Furthermore, in a network withthousands of BSs and antennas, the signalingburden required by both of these coordinationdogmas becomes immense. While this cannotnecessarily be avoided, a new class of multicellu-lar coordination that exploits this abundance oftransmission elements is investigated, utilizingmultiple antennas cooperatively to achieve notonly spectral but also energy efficiency gains.This is discussed here.

The recently proposed concept of large-scalemultiple-antenna systems (LSMASs) is a radicalstep toward a new way of conceiving MIMOsystems for energy efficiency optimization [9].LSMASs are extended MIMO systems with tensor hundreds of radiating elements, each oneconsuming a small amount of energy, whichmay be collocated at a BS, spread out on theface of a building, or geographically distributedover an entire network. LSMASs provide aplethora of advantages over conventionalMIMO systems. In particular, these large-scalesystems offer higher data rates, increased linkreliability, and potential energy savings throughthe exploitation of the many degrees of free-dom offered by the many antenna elements.Meanwhile, random impairments such as ther-mal noise and other-cell interference can beaveraged out. While in the past multiple-anten-na systems have been used to provide capacitygains through spatial diversity, energy-efficientnetworks can be achieved by using massivemulti-antenna systems where the many anten-nas are used to form and focus directed beamsto a multiplicity of terminals on the forwardlink, and selectively collect signals from theseterminals on the reverse link. Furthermore, theproposed approach is scalable with the numberof antennas, and it provides energy savings thatincrease with the number of radiating elements.On the other hand, there are several theoretical(capacity limits, power allocation strategies, sig-nal transmission, channel estimation, amount offeedback, etc.) and practical (antenna design,compensation of mutual coupling and spatialcorrelation, etc.) issues that need to be

addressed. Overall, however, LSMASs offer aradical new direction in multicellular coopera-tion to improve not only the spectral efficiencyof wireless communications networks, but alsogenerate large power savings on the existinginfrastructure.

COORDINATED MULTIPOINT TRANSMISSIONOne of the most fervently discussed topics forcooperative multicellular communication infuture network is that of coordinated multipointtransmission (CoMP) [10], where neighboringBSs transmit together to multiple users in theircells. In CoMP, MSs in neighboring cells aregrouped based on the received pilot signalstrength from the surrounding BSs (i.e., usersare grouped if the strongest BSs they see areidentical). Once these groups are formed, eachof the MSs in each group is served simultaneous-ly by the BSs which provide the strongest signalstrengths for that group (Fig. 3). Consequently,these BSs jointly precode their transmissions tothe user group such that the combined receivedsignal from the set of BSs at each MS providesthe user with its desired signal (i.e., which wouldhave been transmitted from its serving BS), with-out receiving any interference from the othercells in its set of strongest BSs.

In [11], the combination mechanisms of suchpreprocessing techniques are described and com-pared: • Joint TDMA (J-TDMA): Only one MS in the

user group is served in each time slot. Thus,transmission is automatically CCI-free, at thecost of reduced spectral efficiency.

• Joint zero-forcing (J-ZF): The joint prepro-cessing matrix F = HH (HHH)–1 is the pseudo-inverse of the aggregate channel matrix Hsuch that each user only receives a noisy ver-sion of its intended data. However, just as ZFreceivers amplify noise, J-ZF precoding gener-ally increases the average transmit power.

• Joint minimum mean square error (J-MMSE):This preprocessing achieves a compromisebetween CCI cancellation and transmitterpower efficiency. Based on the MMSE criteri-on, the precoding matrix is given by F = HH

(HHH + (N0/r)I)–1, where N0/r is a scaledversion of the noise.

Essentially, CoMP eliminates the interferencefrom what would normally be the most stronglyinterfering BSs, consequently enhancing SINRand thus throughput at each MS.

A common problem in the CoMP conceptis that the gain strongly relies on ideal “usercentric” clustering and assignment, meaningall MSs should be served by their individualset of strongest cells. In a large network withirregularly distributed MSs in the cells, the dif-ficulty is to find sufficient users with these setsbeing identical, since due to shadowing thesignal strength of BSs might be widely dis-tributed. Thus, the typical penetration rate ofMSs that might be served by such a single spe-cific set of cells is often very small, causingmany users to be denied service. In [12] partialCoMP, where shifted coverage areas increasethe probabil i ty of MSs sharing a set ofstrongest cells, is suggested to alleviate thisproblem. While the performance gains of par-

LSMASs are extend-ed MIMO systemswith tens or hun-dreds of radiating

elements, each oneconsuming a smallamount of energy,

which may be collo-cated at a BS,

spread out on theface of a building, or geographically

distributed over anentire network.



tial CoMP reach similar levels as those achiev-able via full CoMP, the additional signalingand complexity must be compensated by limit-ing transmission of reference signals by theMSs. However, it is clear that large SINR andthroughput gains are achieved.

While CoMP does not achieve the perfor-mance of system optimization (i.e., albeit atsignificantly reduced complexity), it is able toprovide large gains over user-based coopera-tion schemes. In general, a substantial increasein signaling over traditional intercell coopera-tion techniques is necessary to facilitate CoMP.However, this is true for most of the recentmulti-cell cooperation techniques, as the “sim-ple” solutions are no longer sufficient to pro-vide the large throughput demands of today’snetworks. Furthermore, it is evident that jointprocessing and cooperative transmission tech-niques are necessary for future networks,implying that user clustering techniques willalso need to be optimized to improve systemperformance. In general, however, it is clearthat multicellular cooperation, and specificallyLSMASs, will be paramount to the perfor-mance and development of future wirelesscommunications networks. A performancecomparison of the discussed cooperation tech-niques is shown in Table 1. Pareto optimalscheduling (POS), which combines aspectsfrom user-based cooperation and CoMP, isdescribed next.

PARETO-BASEDOPTIMAL TRANSMISSION SCHEDULING

In our final section, we would like to introducea recently proposed OFDMA resource alloca-tion and power control method [13] based onPareto optimal power control (POPC) [14],which combines some of the aspects of user-based cooperation and CoMP. In POPC, given

a feasible link allocation (i.e., a group of usersin neighboring cells allocated the sameresource block [RB](s)), a transmit power vec-tor P can be found such that all users achievetheir SINR requirements with minimal power.This is, of course, a highly desirable result that,depending on the location and service require-ments of the interfering MSs, is clearly notalways possible. Hence, by scheduling users insuch a manner to maximize the number of fea-sible user groupings (in principle, there can beas many user groups as there are [orthogonal]RBs in the system), the system spectral effi-ciency can be maximized.

However, scheduling users in such a manneris not a trivial task. In POPC, the desired andinterfering path gains of users in neighboringcells allocated the same frequency resource(s)are combined in a matrix F, where

the SINR target of user i is g*i, and Gj,i is thepath gain from MS j in cell j to the BS in cell i.From this, the optimum transmit power vectorP* = (I – F)–1u, where u signifies the SNR ofeach user, can be calculated if the largest eigen-value of F is shown to be less than 1. Effectively,this constraint establishes the stability (i.e., feasi-bility) of the POPC system of equations. Thelargest eigenvalue is found via an eigenvaluedecomposition, which can be highly computa-tionally complex even for small matrices. Thus,in order for a group of users in neighboring cellsto be purposely scheduled onto the same RB, asimple expression for the allocatability of such agroup must be developed.

In [13] the authors, through analytic deriva-

γ=

=

≠

⎧

⎨⎪⎪

⎩⎪⎪

F

i j

G

Gi j

0, if

, if ,ij i j i

i i

*,

,

Table 1. Technique performance comparison: Scores ranging over {– –, –, +, ++} are given in each performance area. ‘– –’ indicatesterrible performance, and ‘++’ indicates superior performance (i.e., ‘++’ in complexity means very low complexity, whereas in spec.eff. it signifies very high spectral efficiency). The general trend is evident that the more optimal and efficient the system is, the morecomplex it becomes.

Method Objective Complexity Signaling Optimality Spec. Eff. Energy Eff.

FFR Interference mitigation ++ ++ – – – – – –

TPC Uplink fairness ++ + – – – – +

DFR Interference management + + – – – –

SIC Interference cancellation – – – – + –

ULIP Interference mitigation + – + + ++

Sys. opt. Objective optimization – – – – ++ ++ ++

LSMAS Efficient transmission – – + + ++

CoMP Cooperative transmission – – + ++ ++

POS Interference management + – – + ++



tion, have found that for three neighboringcells (for any number of cells greater the com-plexity increases exponentially) that if threeMSs i , j , and k , one in each cell, satisfy thecondition

FijFji + FikFki + FjkFkj + FijFjkFki+ FikFjiFkj < 1, (1)

they may be allocated the same resources ineach cell, and POPC can be applied. This isclearly dependent on the individual desiredand interfering path gains, along with theSINR targets of the users. Therefore, a sched-uler might make use of this condition to sched-ule users such that the number of feasiblegroups of MSs is maximized, hence also maxi-mizing the spectral efficiency of these threecells; this is called POS. Of course, in a largenetwork with hundreds of cells, this is not souseful. As indicated in Fig. 4, three neighbor-ing cells may be grouped in the network. Bygrouping three cells with coinciding beam pat-terns (see shaded cells in Fig. 4), POS can beapplied to these three cells, and the clusterwill be relatively shielded from neighboringsectors interference due to the nature of thebeam patterns. This clustering is then tessel-lated over the network, such that POS can beapplied separately in each of these clusterswithout overly excessive CCI from the sur-rounding cells, allowing the MSs to achievetheir transmission requirements. Thus, thisform of POS combines the user-based cooper-ation methods of individual service require-ments with the grouping of users that is soessential in CoMP.

In Fig. 5, a performance comparison to maxi-mum power transmission (utilized in downlinktransmission) and standard TPC (utilized in sys-tem uplink), in terms of spectral and energy effi-ciency is shown, displaying the potential of POSfor future wireless communication networks. Ingeneral, POS has a performance advantage overpower control over all SINR targets (except g*

i =20 dB), whereas also substantial gains over 13%are seen over max. power transmission in themid-SINR (typically the operational) range. ThePOS performance begins to suffer for higherSINRs as not enough feasible MS groupings canbe found. However, on average the POS spectralefficiency is equivalent to that of max. powertransmission.

This becomes even more significant whenconsidered together with the energy efficiency(calculated as data rate per unit of transmittedpower [b/s/W � (bits/J)]) results shown in Fig. 5.As expected, maximum power transmission isthe least energy efficient of the three consideredtechniques. POS, on the other hand, providesmassive energy efficiency benefits for the system,even when compared to power control, withgains of up to 5 dB for higher SINRs. Hence, itis quite clear that POS drastically reduces thetransmit power consumption in a macrocellularnetwork. Considering this together with thespectral efficiency results, it is evident that POSprovides overall superior system performanceover the standard resource and power allocationtechniques.

CONCLUSIONSThroughout this article, we have discussed theevolution of multicell cooperation from moresimple user-requirements-driven cooperationtechniques, over system-wide optimization-based methods, to coordination in LSMASsand cooperative transmission. Clearly, tech-niques optimizing the resource and powerallocation for individual users do not alwaysstrike system optimum decisions, and hencefrom this research moved on to optimizationmethods. While, if solvable, these techniquescan provide system-optimal resource andpower allocations by coordinating over all cellsin a network, it was explained that these meth-ods are highly intractable, and in most situa-tions must be broken down into multiplesubproblems, yielding suboptimal solutions.Finally, the most recent research has consid-ered the opportunity of LSMASs and coopera-tive transmission from multiple BSs in anetwork. CoMP provides large throughputgains for users in the network, but at the costof additional complexity and rather unreliablescheduling. However, it is clear that for futurenetworks to deliver the increasing servicedemands [1], such cooperation over multipleBSs is more than necessary.

We have concluded our article by introduc-ing a recently proposed technique that com-bines some of the aspects of user-basedcooperation and CoMP scheduling. In POS,users in neighboring cells are grouped basedon their desired and interfering path gains, atwhich point POPC can be applied, and largeenergy savings can be achieved without sacri-ficing spectral efficiency. And although a large

Figure 4. Extension of three-cell POS to larger multicellular networks [13].Universal frequency reuse is applied, and the differing coloration of the cellssimply demarks which BS is serving them.



amount of information is needed to determinethe feasibility of a group of MSs, since thisprocedure is performed on a three-cell basis,BSs need only the information from theirimmediate neighbors, limiting the additionalcomplexity and signaling implied. The POSresults, combined with the given comprehen-sive research review, thoroughly show theimmense importance of multi-cell cooperationin today’s, and more significantly for future,wireless networks.

REFERENCES[1] Cisco Visual Networking Index, “Global Mobile Data Traffic

Forecast Update, 2011-2016,” white paper, Feb. 2012.[2] A. Molisch, Multiple Access and the Cellular Principle,

Wiley-IEEE Press, 2011.[3] S. Ali and V. Leung, “Dynamic Frequency Allocation in Frac-

tional Frequency Reused OFDMA Networks,” IEEE Trans.Wireless Commun., vol. 8, no. 8, Aug. 2009, pp. 4286–95.

[4] M. Wang and T. Ji, “Dynamic Resource Allocation forInterference Management in Orthogonal FrequencyDivision Multiple Access Cellular Communications,” IETCommun., vol. 4, no. 6, Apr. 2010, pp. 675–82.

[5] J. G. Andrews, “Interference Cancellation for CellularSystems: A Contemporary Overview,” IEEE WirelessCommun., vol. 12, no. 2, Apr. 2005, pp. 19–29.

[6] H. Burchardt et al., “Uplink Interference Protection andScheduling for Energy Efficient OFDMA Networks,”EURASIP J. Wireless Commun. and Net., May 2012.

[7] Y. Ma and D. Kim, “Rate-Maximization SchedulingSchemes for Uplink OFDMA,” IEEE Trans. Wireless Com-mun., vol. 8, no. 6, Jun. 2009, pp. 3193–205.

[8] D. Kivanc, G. Li, and H. Liu, “Computationally EfficientBandwidth Allocation and Power Control for OFDMA,”IEEE Trans. Wireless Commun., vol. 2, no. 6, Nov. 2003,pp. 1150–58.

[9] G. Wright, “GreenTouch Initiative: Large Scale AntennaSystems,” Spring Meeting, Seoul, South Korea, 2011:http://www.greentouch.org/uploads/documents/pres_gregory wright.pdf.

[10] M. Sawahashi et al., “Coordinated Multipoint Trans-mission/Reception Techniques for LTE-Advanced [Coor-dinated and Distributed MIMO],” IEEE WirelessCommun., vol. 17, no. 3, June 2010, pp. 26–34.

[11] J.-S. Sheu and C.-H. Hsieh, “Joint Preprocessing Techniquesfor Downlink CoMP Transmission in Multipath FadingChannels,” IEEE 75th VTC-Spring, May 2012, pp. 1–5.

[12] W. Mennerich and W. Zirwas, “User Centric Coordinat-ed Multi Point Transmission,” IEEE 72th VTC-Fall, Sept.2010, pp. 1–5.

[13] H. Burchardt et al., “Pareto Optimal Power ControlScheduling for OFDMA Networks,” IEEE 76th VTC-Fall,Sept. 2012.

[14] A. Goldsmith, Wireless Communications, CambridgeUniv. Press, 2005.

BIOGRAPHIESHARALD BURCHARDT ([email protected]) received his B.Sc.in electrical engineering and computer science in 2007from Jacobs University Bremen, and his M.Sc. in communi-cations engineering in 2009 from the Technical Universityof Munich. He is currently working toward his Ph.D. degreeat the Institute for Digital Communications at the Universi-ty of Edinburgh.

HARALD HAAS holds the Chair of Mobile Communications atthe University of Edinburgh. He is also part-time chief tech-nical officer (CTO) of a university spin-out company,pureVLC Ltd. His main research interests are interferencemanagement in wireless networks, multiple antenna con-cepts, and optical wireless communication. He holds 23patents, and has published more than 230 conference andjournal papers. He was an invited speaker at TED Global2011. He currently holds a prestigious Established CareerFellowship of the Engineering and Physical SciencesResearch Council (EPSRC) in the United Kingdom.

Figure 5. System spectral efficiency (SE) and energy efficiency (EE) results forthe various power control techniques in a multi-cellular network. In moderncellular systems, full frequency reuse is applied, utilizing maximum powertransmission in the forward link, and power control in the reverse link.

x 105

20

0.3

0.2

SE (

b/s/

Hz)

0.4

0.5

0.6

0.7

4 6 8 10 12 14 16 18 20

SINR target (dB)20

1

0.5

EE (

b/J)

1.5

2

4 6 8 10 12 14 16 18 20

Max. powerPower controlPOS



http://www.greentouch.org/uploads/documents/pres gregory wright.pdf



INTRODUCTIONThe potential performance benefits of applyingmulticell cooperation for mobile broadband cellu-lar systems depends on numerous factors such asthe network architecture, deployment topology,and the radio standard. As an example, the net-work architecture for wideband code-divisionmultiple access (WCDMA) included a centralizedradio network controller (RNC), hosting majorradio resource management (RRM) algorithmssuch as admission control, mobility control, andpacket scheduling for multiple cells. Nevertheless,it seems that most of today’s WCDMA imple-mentations only utilize multicell RRM in theform of soft handover. With the introduction ofhigh-speed packet access (HSPA) overlay forWCDMA, the architecture moved toward a dis-tributed functionality by moving RRM algorithmssuch as packet scheduling to the base stationnodes [1]. Thus, faster access to radio channelmeasurements was achieved (e.g., for packetscheduling and link adaption), but at the cost ofreduced possibilities for multicell cooperation.

With the introduction of Long Term Evolution(LTE), the architecture evolved further toward afully distributed paradigm by removing the RNC,but at the same time introducing a new inter-base-station interface (called X2) to allow somedegree of multicell coordination and collabora-tion. Since its first introduction, the LTE standardhas evolved significantly toward LTE-Advanced,where numerous spectral efficiency and peak datarate improvements have been introduced. Amongthose, it is worth mentioning improved multiple-input multiple-output (MIMO) techniques, carri-er aggregation (CA), and enhanced intercellinterference management [2].

Despite the many enhancements included inLTE-Advanced, it seems that macro-only net-works may not be able to carry the predictedfuture broadband traffic. Therefore, the next bigleap in cellular system performance improve-ment is estimated to come from the introductionof additional small cells to complement tradi-tional macro site installations (i.e., migratingtoward heterogeneous network [HetNet] topolo-gies). This raises the questions on how smallcells are most spectrally efficiently introduced,and how to integrate them with the macro layerto maximize the performance benefits whilekeeping capital expenditures (CAPEX) andoperational expenditure (OPEX) at tolerablelevels. In this study we further address suchaspects for cases where small cells are in theform of either standalone picos or remote radioheads (RRHs). As it will be demonstrated, bene-fits of LTE-Advanced small cells can beimproved via tight integration and cooperationbetween the macro layer and the small cells in amulticell RRM basis. For this purpose, twostate-of-the-art LTE-Advanced HetNet integra-tion schemes are examined: interference man-agement between macro and small cells, andcooperative inter-site CA between layers. Inboth use cases we highlight the benefit of multi-cell cooperation as well as the importance ofexplicit terminal involvement to push the systemperformance to its maximum. As compared topublished studies (e.g., [3–5]), we present perfor-mance results for highly irregular HetNet sce-narios, where clusters of small cells are onlypresent in some macrocells, calling for additional

BEATRIZ SORET AND HUA WANG, AALBORG UNIVERSITYKLAUS I. PEDERSEN AND CLAUDIO ROSA, NOKIA SIEMENS NETWORKS

ABSTRACTIn this article we present two promising prac-

tical use cases for simple multicell cooperationfor LTE-Advanced heterogeneous network sce-narios with macro and small cells. For co-chan-nel deployment cases, we recommend the use ofenhanced interference coordination, or eICIC,to mitigate cross-tier interference and ensuresufficient offload of users from macro to smallcells. It is shown how the eICIC benefit is maxi-mized by using a distributed inter-base stationcontrol framework for dynamic adjustment ofessential parameters. Second, for scenarioswhere macro and small cells are deployed at dif-ferent carriers an efficient use of the fragmentedspectrum can be achieved by using collaborativeinter-site carrier aggregation. In addition to dis-tributed coordination/collaboration between basestation nodes, the importance of explicit termi-nal assistance is highlighted. Comprehensive sys-tem-level simulation results illustrate theperformance benefits of the presented tech-niques.

MULTICELL COOPERATION FOR LTE-ADVANCEDHETEROGENEOUS NETWORK SCENARIOS

SORET LAYOUT_Layout 1 2/21/13 4:05 PM Page 27


macro to macro coordination. A recent surveyon Third Generation Partnership Project (3GPP)LTE-Advanced HetNet features can be found in[3], while more information on the role of termi-nals in HetNet scenarios is available in [6]. Inaddition, there is ongoing research towardadvanced coordinated multipoint (CoMP) trans-mission and reception techniques within 3GPP,as reported in [2]. However, advanced CoMPschemes are outside the scope of this article, aswe focus here on simpler downlink LTE-Advanced multicell cooperation schemes forHetNet scenarios.

The rest of the article is organized as follows:We first outline the system model and addressedproblem. Second, we analyze enhanced intercellinterference coordination (eICIC) for co-chan-nel cases, emphasizing the latest research ideas.Afterward, cases with dedicated carriers formacro and small cells are studied, where cooper-ative inter-site CA aims at exploiting the frag-mented bandwidth. In both cases, system-levelperformance results are presented to demon-strate the benefits of multicell cooperation interms of end-user-experienced throughput. Thearticle is closed with some concluding remarks.

SYSTEM MODEL AND PROBLEM STATEMENT

Figure 1 illustrates the considered LTE-Advanced HetNet scenario with macro base sta-tions (called eNBs) and small cells. The macroeNBs are assumed to transmit with 46 dBmpower per sector and have 14 dBi antenna gain(including feeder loss), which results in an equiv-alent isotropic radiated power (EIRP) of 60dBm. Small cells only have an EIRP of 35 dBmin the example in Fig. 1. In traditional homoge-neous networks, the eNB offering the highest

signal strength indicator is selected as the serv-ing eNB for the user equipment (UE). In LTE,either the reference signal received power(RSRP) or the reference signal received quality(RSRQ) is used as the comparative metric in thecell-selection procedure. While RSRP onlyreflects the received power from each cell, RSRQis defined as RSRP divided by the total receivedpower, capturing the channel quality of therespective resources. Applying the RSRP orRSRQ cell selection criterion in HetNets leads toa downlink imbalance problem: The coverage ofthe macrocell is much larger than that of thesmall cell, resulting in a small number of UEdevices being served by the small eNB. The cov-erage area of the small cell is not only limited byits transmition power, but to a large extent by thepotential interference experienced from themacro eNB. This imbalance can be corrected byexpanding the range of the small cell. Thus, apositive bias denoted as range extension (RE)offset is added to the RSRP or RSRQ measuredfrom small eNBs, pushing more UE devices tothis layer. The latter is illustrated in Fig. 1, wherethe basic coverage area of small cells is plottedwith darker blue, while the range extended smallcell coverage is depicted with lighter blue.

The performance and the need for multicellcooperation techniques for such a HetNet sce-nario depend heavily on the carrier deploymentstrategy of macro eNBs and small cells. Here weconsider two distinct cases:• Co-channel deployment, where macro and

small cells occupy the same bandwidth• Dedicated carrier deployment where macro

and small cells use different carriersThe co-channel case has the advantage of beingable to exploit the full system bandwidth at bothmacro and small layers, but the disadvantage of

Figure 1. Sketch of the considered LTE-Advanced HetNet scenario.

(i) Co-channeldeployment

(ii) Dedicatedcarrier

Small cell: Tx power:Antenna gain:EIRP:

30 dBm5 dBi

35 dBm

Macro eNB: Tx power:Antenna gain:EIRP:

46 dBm14 dBi

60 dBm

Carrier 1

Carrier 1

Carrier 2

Signal

Interference

UE in inter-site CAmode, i.e. receivingdata from both cells.

Small cell connectedUE. Subject to

interference frommacro eNB.

Extended coverage areaof small cell with RE

Coverage areaof macro

Basic coveragearea of small cell

without RE

The next big leap incellular system per-formance improve-ment is estimated tocome from the intro-duction of additionalsmall cells to com-plement traditionalmacro site installa-tions (i.e., migratingtoward heteroge-neous network topologies).



having interference between the two base stationlayers. The dedicated carrier case has the benefitof no inter-layer interference, but the disadvan-tage of not using the full system bandwidth atboth macro and small cells, resulting in frag-mented spectrum between the two layers. Ingeneral, co-channel deployment is preferred foroperators with moderate LTE spectrum (e.g., 10MHz), whereas dedicated carrier deploymentsare attractive in situations with larger availabilityof bandwidth (20 MHz or more) and high pene-tration of small cells. As the characteristics forthese two deployment cases are clearly different,they also call for different multicell cooperationtechniques.

For the co-channel case we investigate multi-cell cooperation in the form of inter-layer inter-ference management. More specifically, we studythe applicability of time-domain eICIC betweenmacro and small cells [3]. As compared to previ-ous eICIC studies we focus on the inter-sitecoordination signaling required to have suchsolutions working in practice, as well as recentlyconsidered eICIC enhancements on top of thebasic functionality defined for LTE-AdvancedRelease 10 (Rel-10). New state-of-the-art eICICperformance results are presented for realisticscenarios. For the dedicated carrier case there isno inter-layer interference, and therefore noneed to apply eICIC. However, the performancecan be further improved by using collaborativeinter-site CA between macro and small cells,such that the UE devices will essentially beserved on a larger bandwidth simultaneouslyfrom multiple cells. For the inter-site CA casewe assume that the small cells are RRHs, beingconnected to macro eNBs via high-bandwidthlow-latency fibers. Thus, all baseband processingfor the small cells (RRHs) could be placed inthe macro eNB, offering further opportunitiesfor various optimizations such as joint multicellpacket scheduling. Referring to the 3GPP termi-nology, the dedicated carrier scenario withmacro and RRHs is denoted CA scenario num-ber four [7]. As discussed in detail in the follow-ing sections, successful application of eICIC andinter-site CA collaboration techniques requiresexplicit UE assistance and involvement [6].

CO-CHANNEL INTERFERENCE COORDINATIONIn the co-channel deployment example in Fig. 1we assume small cells in the form of outdoorpicos, but all the techniques presented here arealso valid for other types of small cells. Here,only users in the very close vicinity of the pico-eNB will be served by it, offering limited off-load gains. This is due to the high interferencereceived from the macro layer, which essentiallyprevents picos from using larger RE to increasethe offload of users. eICIC techniques (intro-duced in 3GPP Rel-10 specifications) alleviatethis interference problem and enable applicationof moderate to high values of RE. The basicprinciple is to establish coordinated resourcepartitioning between the macro and pico layerwith the aim of providing some resources free ofmacro interference to pico UE devices. This isenabled by preventing the macro eNB fromtransmitting on certain subframes, the so-calledalmost blank subframes (ABS), as illustrated inFig. 2. In this way, victim pico UE devices in theextended area may be scheduled within the sub-frames overlapping the muted frames of themacrocell, mitigating cross-tier interference. Themuting ratio determines the periodicity of theABS on a time-domain basis (4 over 8 subframesin the example in Fig. 2). The concept relies onaccurate time and phase synchronizationbetween all base station nodes. The name“almost” blank subframes refers to the fact thatthe aggressor cell still transmits the most essen-tial information in order to provide support tolegacy UE devices. Thus, ABS does includetransmission of common reference signals (CRS)and, depending on the subframe number, alsochannels such as the physical broadcast channel(PBCH), and the primary and secondary syn-chronization signals (PSS/SSS). The interferencefrom PBCH and PSS/SSS can be handled by notconfiguring subframes where those channelsappear as ABS, as well as by using time shiftsbetween the macro and pico layers to avoid, orminimize, the probability of experiencing colli-sions of those channels between layers. Theinterference from CRS does, however, appear inevery ABS. Ultimately, the macro transmission

Figure 2. Basic principle of eICIC.

4 8 12 16 Subframe

Tx power [dBm]

46MacroeNB1

MacroeNB2

PicoeNB3

0

MacroeNB1

MacroeNB2

X dB

ABSPicoeNB3

0

0

CRS interference

46

30

0

Protectedsubframes

Normalsubframes

Pico eNB3 can scheduleUEs in the extended area

eNBs can exchangeand request informa-

tion on ABS pat-terns, and

dynamically changethe configuration,

taking into considera-tion not only the

overall system per-formance, but alsothe individual QoS

requirements of indi-vidual users.



power in ABS is not zero, but on averagereduced by X dB compared to normal transmis-sion.

Numerous studies in the literature (e.g., [4])have confirmed that eICIC performance dependsheavily on numerous factors such as:• ABS muting patterns at the different macro-

cells• RE settings for the picocells• UE traffic distributions

3GPP specifications support several mecha-nisms for distributed configuration of eICICrelated parameters by means of coordinationamong eNBs, as summarized in Fig. 3. Thus,eNBs can exchange and request information onABS patterns, and dynamically change the con-figuration, taking into consideration not only theoverall system performance, but also the individ-ual QoS requirements of individual users. Otherself-organizing network (SON) features likemobility load balancing (MLB) and mobilityrobustness optimization (MRO) facilitate coor-dinated adjustment of mobility parameters suchas RE [8]. Specifically, MLB allows tuning thehandover and cell selection bias for cell pairssuch as macro and picocells to balance the loadbetween layers, and MRO monitors failed hand -overs to fine-tune mobility parameters and mini-mize the probability of radio link failures(RLFs). Thus, MRO essentially limits settings ofRE that would otherwise result in risk of RLF.More details on mobility aspects in HetNet sce-narios can be found in [9]. The X2 mechanismsfor ABS configuration, and the MLB and MROprocedures form a simple but effective inter-eNB coordination protocol that facilitatesautonomous network self-adjustment to maxi-mize the benefits of eICIC. Notice that theinter-eNB coordination mechanisms summarizedin Fig. 3 are applicable for macro to macro,macro to pico, and pico to pico. Besides thecoordination among eNBs, the explicit involve-ment of the UE in the process is also important.When eICIC is enabled, highly time-variantinterference fluctuations are likely to happen,

which basically means that UE measurementreports should be time-restricted to have valuefor the network. For example, UE channel stateinformation (CSI) measurements for pico UEdevices are recommended to be time restrictedso that the eNB receives CSI feedback corre-sponding to normal and protected subframes ofthe macro. Finally, measurement restrictions forradio link monitoring (RLM) and mobility mea-surements shall also be configured for terminalsto maximize the gain from eICIC [3, 6].

The described coordination mechanisms areavailable in 3GPP Rel-10 specifications. FurtherRel-11 eICIC (FeICIC) work item aims atstrengthening the eICIC concept and boostingthe overall performance. Particularly, two mainnovelties are envisioned in the new release: CRSinterference cancellation (IC) and low-powerABS (LP-ABS).

UE CRS INTERFERENCE CANCELLATIONOne of the main UE receiver enhancements in3GPP Rel-11 is the application of CRS IC toremedy the residual CRS interference duringABS. When the macro is muting a subframe,some resource elements have zero transmissionpower while resource elements with CRS havenormal transmission. Assuming 2 ¥ 2 MIMO,the CRS overhead during ABS is approximately9 percent, corresponding to an average trans-mission power of ABS, which is roughly 10 dBbelow that of normal subframe transmissions.To fully benefit from macro muting, pico UEmay apply CRS IC during subframes where themacro uses ABS, by estimating the CRS inter-ference from neighboring cells and subtractingit from the received signal. In the ideal case, allpico UE devices would suppress the interfer-ence from all macro cells with no residual inter-ference. However, a more realistic non-idealCRS IC assumption entails terminals cancellingonly the most relevant interferers with an inter-ference cancellation error [5]. Deciding the suit-able number of cancelled interferers is the keyaspect here, and it depends on the UE capabili-

Figure 3. Multicell and eNB-UE coordination for co-channel deployments.

UE feedback:

1) CSI reporting2) Mobility events

UE configuration:

1) Time-domain RRM, CSI, and RLM measurement restrictions.2) Potential CRS IC assistance.3) Mobility event configuration.

ABS coordinationABS status, ABS information, invoke, etc.

Mobility load balancing (MLB)Exchange of load information, negotiation and change of

mobility parameters such as RE

Mobility robustness optimization (MRO)Root cause identification to minimize radio link failures,

dropped calls, unnecessary handover, etc..

eNB #1(macro or pico)

eNB #2(macro or pico)

Peer-to-peer basestation coordination

To fully benefit frommacro muting, pico UE may applyCRS IC during subframes where themacro uses ABS, by estimating theCRS interferencefrom neighboringcells and subtractingit from the received signal.



ties but also on the user location. Thus, pico UEdevices in the extended area are in the best con-dition for estimating the CRS from the domi-nant interfering macros and, at the same time,those that mainly require and benefit from CRSIC. The network can provide additional assis-tance for the UE to further ease the CRS IC, asalso indicated in Fig. 3.

LOW-POWER ABSDuring protected subframes the macro layermay, instead of stopping data transmission,reduce the transmission power, scheduling strict-ly users in the vicinity of the macro eNB. Thisoption is referred to as low-power ABS, in con-trast to (zero-) ABS. LP-ABS brings additionalflexibility for optimizing the system under differ-ent conditions, enabling second order optimiza-tions. Nonetheless, reducing the macrotransmission power also has a cost in terms ofadditional complexity. As happens with ABS,essential information (CRS and other controlchannels) required by the system has to be trans-mitted unchanged during the reduced powersubframes for backward compatibility. Thus,CRS IC is still relevant for LP-ABS operation.The introduction of LP-ABS mode can be fullyexploited by means of extra coordinationbetween layers. Therefore, additional informa-tion exchange could include the exchange of sug-gested (configured) LP-ABS power reductionbetween pico and macro layers (and vice versa).In order to respect the currently defined eNBtransmitter radio frequency (RF) requirements,power reductions of 3 dB and 6 dB are typicallyconsidered for LP-ABS.

EICIC PERFORMANCEA network layout as defined in [10] is simulated.The network topology consists of hexagonalthree-sector macrocells with an inter-site dis-tance of 500 m and four randomly placedhotspots per macrocell area. According to 3GPPsimulation guidelines, one picocell is deployed ineach hotspot. The system-level simulator followsthe LTE specifications, including detailed mod-eling of major RRM functionalities such aspacket scheduling, hybrid automatic repeatrequest (HARQ), link adaptation, 2 ¥ 2 closedloop MIMO with precoding and rank adapta-tion. The available bandwidth is 10 MHz, andProportional Fair (PF) scheduling is applied. InFig. 4 we show the performance gain of eICICunder different assumptions. The performancegain in 5th percentile and 50th percentile end-user throughput (macro and pico users) with fullbuffer traffic is shown, compared to the scenariowith no eICIC and no RE. For the case withouteICIC and no RE, only 40 percent of the usersare offloaded to the picocells, meaning that themajority of users (60 percent) are served by themacro layer. However, by enabling eICIC with4/8 of subframes as ABS and using 12 dB of RE,the offload increases to 80 percent. Thus, only20 percent of the users are served by the macrolayer, resulting in more resources per macro usereven if it is muting 50 percent of the subframes.As a result, a significant eICIC gain for 12 dB isobserved in Fig. 4, assuming RSRP-based cellselection. Comparing the three sets of perfor-

mance results for 12 dB RE clearly shows theimportance of having explicit UE support in theform of CRS IC. The cases with ideal and non-ideal CRS IC are fairly close, while there isnoticeable performance degradation when CRSIC is not applied (non-ideal UE CRS IC refer-ring to cases where only a few interferers arepartially suppressed [5, 6]). Decreasing the REoffset from 12 dB to 9 dB means that a smallernumber of users are offloaded to picocells.Therefore, a smaller ABS muting ratio should beapplied in order to keep the trade-off betweenmacro loss and pico gain. In this case, the maxi-mum eICIC gain is achieved when reducing themuting ratio to 2/8. Applying LP-ABS with 6 dBpower reduction in 4/8 subframes is anothervariant configuration of eICIC. Here we useideal CRS IC and only 6 dB RE in coherencewith the power reduction at the macro layer.Using 6 dB RE means, of course, that fewerusers are offloaded, but this is partly compensat-ed by the fact that for LP-ABS, the macro is stillable to schedule users in every subframe. Fur-thermore, the possibility of scheduling macrousers during LP-ABS leads to a higher optimalmuting ratio (4/8) even for such small RE offset.The last set of results reported in Fig. 4 is forthe case where RSRQ is used for cell selection(i.e., handover). For such cases, the UE devicesare configured to only measure on picocells dur-ing subframes where the macros apply ABS.This essentially means that only low values ofthe RE are needed, as the RSRQ metric includesimplicit knowledge of the eICIC interferencereduction. Comparing the cases with RSRP andRSRQ (both with ideal CRS IC) seems to indi-cate that the picocell footprint is slightly bettercontrolled by using RSRQ.

UNIFORM VS. NON-UNIFORM TOPOLOGIESAll results in Fig. 4 assume uniform HetNetcases, that is, with the same number of picos inall macrocells. We next evaluate the effective-ness of eICIC techniques in more realistic non-

Figure 4. FeICIC performance gain under different conditions.

10

0

Gai

n co

mpa

red

to n

o eI

CIC

and

0dB

RE

[%]

Gain 5 percentileGain 50 percentile

20

30

40

50

60

70

80

90

Ideal CRS ICRE=12dB

Non-idealCRS IC

RE=12dB

No CRS ICRE=12dB

Ideal CRS ICRE=9dB

LP-ABSRE=6dB

ideal CRS IC

RSRQRE=2dB

ideal CRS IC



uniform HetNet scenarios. In particular, we con-sider the scenario pictured in Fig. 5a, wherepico cells are only present in the three centermacrocells, while small cells are not present inall other macrocells. Such deployment cases areenvisaged to be typical for initial LTE-A HetNetrollouts. These topologies can also benefit fromefficient cooperation among macro eNBs, takinginto account that only some cells (center cells inFig. 5) can gain from eICIC. Particularly, thejoint use of LP-ABS and ABS can significantlyimprove the performance of pico UE in theextended area. Let us assume an RE of 12 dBfor the picos and center macrocells using ABS in4/8 subframes. Due to the deployment of pico-cells in the center macrocells, neighboringmacrocells may be configured with ABS or LP-ABS in order to offer sufficient cumulativemacrocell interference reduction to the picos. Byapplying LP-ABS in the first tier of interferingcells, the interference suffered by victim pico UEis mitigated, and at the same time, the perfor-mance of macro UE in the neighboring cells willnot degrade much (the experienced signal-to-interference-plus-ratio [SINR] will not be signifi-cantly affected since the center macros areassumed to use ABS). Figure 5b shows theeICIC performance gain in this scenario. Noticethat the performance gain is computed account-ing for the users in the three center cells andthose in the surrounding cells. Thus, this includesboth effects of gains from center macrocellsbeing able to offload more users to the picos andthe potential degradation in surrounding macrosfrom using LP-ABS. Two different values of theLP-ABS power reduction in the neighbor cellsare simulated (3 dB and 6 dB), as well as thecases with no muting and zero-power ABS. Herethe eICIC performance gain is on the order ofapproximately 30–40 percent, illustrating apromising use case for applying LP-ABS inmacrocells without picos.

The inter-macro coordination described herecan be dynamically configured through the X2interface. As mentioned before, the addition ofpower reduction to the existing exchanged mes-sages between eNBs could facilitate the LP-ABS

operation. Thus, a macro with picos would getthe ABS information from all surroundingmacrocells, and on detection of one macro notapplying ABS it would report its own ABS con-figuration and suggest reducing the power inorder to minimize the interference caused topico UE.

COLLABORATIVE INTER-SITE CAThe second considered HetNet scenario is thecase where macro and small cells are deployedat different carriers (Fig. 1, case (ii)). In thiscase there is obviously no need for eICIC asthere is no interference between macro andsmall cells. Here, we therefore recommendexploiting collaborative inter-site CA in order tofully utilize the fragmented radio resources. Par-ticularly, we propose applying CA between themacro and the most dominant small cell whenev-er a UE device is in a position where it can hearthat cell. To be able to apply such inter-site CAtechniques, we assume that the small cells areimplemented as RRHs. Thus, the macro andRRHs are connected via a high-bandwidth low-latency fiber with the majority of baseband pro-cessing hosted in the macro sites. Referring tothe established LTE-Advanced CA terminology[11, 12], we propose that UE devices alwayshave the macrocell configured as their primarycell (PCell), whereas RRHs are configured assecondary cells (SCells) whenever possible. UEdevices configured to operate with inter-site CAnaturally have access to higher transmissionbandwidth, and therefore the opportunity to beserved at higher data rates. The fact that UEdevices always have a macro as PCell results inrobust mobility performance as RLF monitoringis only performed for PCell, which typically isfound to be good for macro-only carriers. RSRQmeasurement reports from terminals are used todecide if UE is configured on both macro andsmall cells, or whether it is only connected toone layer. Compared to RSRP-based measures,the RSRQ here has the advantage of implicitload information as the interference on the twoconsidered carriers is likely to be significantly

Figure 5. Intercell coordination for non-uniform topologies: a) scenario; b) performance gain.

(b)(a)

5 percentile0

5

0

Gai

n co

mpa

red

to n

o eI

CIC

and

0dB

RE

[%]

Zero ABSLP-ABS 6dBLP-ABS 3dBNo muting

Zero ABSLP-ABS, zero-ABS or no mutingNo muting

Macro sitePico site

10

15

20

25

30

35

40

50 percentile

The inter-macro coordinationdescribed here canbe dynamically configured throughthe X2 interface. As mentionedbefore, the additionof the power reduction to theexisting exchangedmessages betweeneNBs could facilitatethe LP-ABS operation.



different. Aggressive settings for configuring asmall cell as an SCell can be used, as the net-work can choose to deactivate the SCell if it isnot being used for data transmission to reduceUE power consumption [11]. As the packetscheduler for the small cells (RRH) is physicallylocated in the macro eNB, joint multicell packetscheduling for those UE devices configured withinter-site CA is feasible. In order to illustrate theapplication of such techniques, let us consider asimple extension of standard PF scheduling tosuch cases. Joint PF scheduling across multiplecells basically refers to how the PF metric iscomputed per cell. In a joint PF scheduler, thedenominator of the PF metric is updated as thesum of the past average scheduled throughputover all cells where the UE has been scheduledin the past [13]. It simply requires informationexchange on the average scheduled throughputbetween the scheduler for macro and small cells.In that way, the joint scheduler essentially offersfast and efficient inter-site load balancingbetween macro and small cells, thereby allowingfor more equitable distribution of radioresources among UE.

PERFORMANCE OF COLLABORATIVE INTER-SITE CAIn order to further illustrate the benefits of col-laborative inter-site CA for the downlink, wepresent performance results from extensive sys-tem-level simulations. These simulations arebased on the same assumptions as in the previ-ous section for eICIC evaluation (in line with3GPP HetNet simulation assumptions in [10]).However, here we assume that macro and smallcells are deployed at two different 10 MHz carri-ers. The intra-site CA case with the two carriersdeployed at the macrocell can be found in [11].In the simulations, we assume that all UEdevices are configured with inter-site CAbetween the macro and the strongest RRH,while load balancing is managed by the joint PFscheduler. In order to illustrate how the inter-site CA performance varies vs. the offered sys-tem load, we apply a simple birth-death trafficmodel with fixed payload size per call. The callarrival follows a homogeneous Poisson process,while the payload size per call is fixed at 10 Mb.The average offered load per macrocell area iscalculated as the product of the equivalent userarrival rate (per macrocell area) and the payloadsize. Figure 6 shows the 5th percentile and 50thpercentile end-user throughput vs. the offeredload. These results are for the case with fourRRHs per macrocell area and with/withoutinter-site CA. For the cases without inter-siteCA, the UE devices are only served by one cellat a time, determined by the RSRQ measure-ment report from the terminal. It is observedthat both the 5th and 50th percentile userthroughput with inter-site CA are significantlyhigher than without inter-site CA. The end-user-experienced gain of inter-site CA is up to 60percent under low load conditions. However, thegain brought by inter-site CA decreases as theload increases. At high load, the performancewith inter-site CA is almost the same as withoutinter-site CA. This behavior can be explained asfollows: At low offered load (small number ofUE devices), UE using inter-site CA benefits

from larger transmission bandwidth andincreased multi-user diversity. When the offeredload is high (large number of UE devices in eachcell), it does not really matter whether a UEdevice is assigned on one cell only or assignedon both cells using inter-site CA since the sys-tem is saturated, and the scheduler tries to allo-cate the available resources among all UE in afair manner. However, at high load, the use ofinter-site CA still offers advantages in terms ofenhanced HetNet mobility robustness and fasterinter-layer load balancing via the use of jointpacket scheduling for both macro and RRHs. Itis worth mentioning that the highest gain withinter-site CA (at low load) is lower than the the-oretical 100 percent gain achievable by doublingthe transmission bandwidth since the SINRexperienced at the macro and small cell layers istypically quite different.

CONCLUSIONS

In this article we have demonstrated the bene-fits of using tight collaboration between LTE-Advanced macro and small cells, including theimportance of having explicit terminal involve-ment and assistance. Two practical feasibleschemes have been studied. For co-channeldeployment cases, the advantages of usingeICIC were demonstrated. eICIC allows higheroff-load of users to the small cells, resulting inoverall improvements in end-user-experiencedthroughput. One of the enablers for successfulapplication of eICIC is the use of distributedinter-site muting pattern adjustment, mobilityload balancing, and robustness optimization.For dedicated carrier deployment cases, thebenefit of inter-site CA between macro andsmall cells has been shown. Using such tech-niques obviously offers higher peak data ratedue to larger bandwidth accessibility for the ter-minals, but also related benefits in terms offaster load balancing and easier mobility man-

Figure 6. 5 percentile and 50 percentile UE throughput with/without inter-siteCA, bursty traffic model.

Offered load [Mbps]200

10

0

5 pe

rcen

tile

and

50

perc

enti

le U

E th

roug

hput

[M

bps]

20

30

40

50

60

70

40 60 80 100

5percentile with inter-site CA5percentile without inter-site CA50 percentile with inter-site CA50 percentile without inter-site CA



agement as there is no interference between thenetwork layers. For both of the considered mul-ticell RRM collaboration techniques, explicitterminal support is needed to harvest the fullbenefit. UE CRS IC and mechanisms for config-uring measurement restrictions are required foreICIC cases, as well as CA capability for themulticarrier scenario. In conclusion, we haveessentially shown that the introduction of smallcells benefits from being tightly integrated withthe macro layer in order to unleash the full per-formance potential.

REFERENCES[1] H. Holma and A. Toskala, Eds., LTE for UMTS, OFDMA and

SC-FDMA Based Radio Access, 2nd ed., Wiley, 2011.[2] P. Bhat et al., “LTE-Advanced: An Operator Perspective,”

IEEE Commun. Mag., vol. 50, no. 2, Feb. 2012, pp. 104–14.[3] A. Damnjanonic et al., “A Survey on 3GPP Heteroge-

neous Networks,” IEEE Wireless Commun., vol. 18, no.3, June 2011, pp. 10–21.

[4] Y. Wang and K. I. Pedersen, “Performance Analysis ofEnhanced Inter-cell Interference Coordination in LTE-Advanced Heterogeneous Networks,” Proc. IEEE VTC-Spring 2012, May 2012.

[5] B. Soret, Y. Wang, and K. I. Pedersen, “CRS InterferenceCancellation in Heterogeneous Networks for LTE-Advanced Downlink,” Proc. IEEE ICC 2012, Int’l. Wksp.Small Cell Wireless Networks, June 2012.

[6] A. Damnjanovic et al., “UE’s Role in LTE Advanced Het-erogeneous Networks,” IEEE Commun. Mag., vol. 50,no. 2, Feb. 2012, pp. 164–76.

[7] M. Iwamura et al., “Carrier Aggregation Framework in3GPP LTE-Advanced,” IEEE Commun. Mag., vol. 48, no.8, Aug. 2010, pp. 60–67.

[8] S. Hämäläinen, H. Sanneck, and C. Sartori, Eds., LTESelf-Organising Networks (SON), 1st ed., Wiley, 2012.

[9] S. Barbera et al., “Mobility Performance of LTE Co-Channel Deployment of Macro and Pico Cells,” Proc.IEEE WCNC 2012, Apr. 2012, , pp. 2863–68

[10] 3GPP Technical Report 36.814, “Further Advancementsfor E-UTRA Physical Layer Aspects,” v. 9.0.0, Mar. 2010,www.3gpp.org.

[11] K. I. Pedersen et al., “Carrier Aggregation for LTE-Advanced: Functionality and Performance Aspects,” IEEECommun. Mag., vol. 49, no. 6, June 2011, pp. 89–95.

[12] S. Zukang et al., “Overview of 3GPP LTE-Advanced CarrierAggregation for 4G Wireless Communications,” IEEE Com-mun. Mag., vol. 50, no. 2, Feb. 2012, pp. 122–30.

[13] Y. Wang et al., “Carrier Load Balancing and PacketScheduling for Multicarrier Systems,” IEEE Trans. Wire-less Commun., vol. 9, no. 5, May 2010, pp. 1780–89.

BIOGRAPHIESBEATRIZ SORET received her M.Sc. and Ph.D. degrees intelecommunication engineering from the University ofMálaga, Spain, in 2002 and 2010. From 2001 to 2003 shewas a research assistant in the same university. From 2003to 2005 she worked in Cetecom Spain developing protocoltest systems for Bluetooth and GSM/UMTS. From 2005 to2011 she was an assistant professor at the University ofMálaga. She is currently a postdoctoral researcher at Aal-borg University in collaboration with Nokia Siemens Net-works, Denmark. Her current research work is focused onradio resource management for heterogeneous networks.Other main research interests include quality of service andcross-layer design for wireless communications and queue-ing systems.

HUA WANG received his Bachelor’s degree from ZhejiangUniversity, China, in 2003, and his M.Sc. and Ph.D. degreesfrom Technical University of Denmark in 2005 and 2009,respectively. From 2005 to 2006 he worked as a radioengineer at Danske Telecom, Denmark. From 2008 to 2009he was a visiting scholar at the University of Texas atAustin. He is currently a postdoctoral researcher at AalborgUniversity in collaboration with Nokia Siemens Networks,Denmark. His main research interests are in the area ofradio resource management and performance analysis forsmall cell enhancements in heterogeneous networks.

KLAUS INGEMANN PEDERSEN received his M.Sc. in electricalenginerring and Ph.D. degrees in 1996 and 2000 from Aal-borg University, Denmark. He is currently with NokiaSiemens Networks in Aalborg, Denmark, where he is work-ing as a senior wireless network specialist. His current workis related to radio resource management and 3GPP stan-dardization of LTE and LTE-Advanced. The latter includeswork on carrier aggregation, heterogeneous networks,mobility, interference management, and system perfor-mance assessment. He is the author/co-author of approxi-mately 100 technical publications on a wide range oftopics, as well as an inventor on several patents. He hasalso contributed to books covering topics such as WCDMA,HSPA, LTE, and LTE-Advanced.

CLAUDIO ROSA received his M.Sc. and Ph.D. degrees fromAalborg University in 2000 and 2005, respectively. He alsoreceived an M.Sc. degree in telecommunication engineer-ing from Politecnico di Milano, Italy, in 2003. Curently heworks as a wireless network specialist at Nokia SiemensNetworks in Aalborg. His work is principally related to thestandardization of LTE-Advanced, with particular focus oncarrier aggregation evolution and small cell enhancementsin future releases of the 3GPP specifications. Since 2003 hehas been authoring/co-authoring several publications pri-marily in the field of uplink radio resource management,and has edited books on HSPA, LTE, and LTE-advanced.

At high load, the useof inter-site CA stilloffers advantages interms of enhancedHetNet mobilityrobustness and fasterinter-layer load balancing via use of joint packetscheduling for bothmacro and RRHs.


http://www.3gpp.org


The authors at The Uni-versity of Texas at Austinacknowledge the supportof the Office of NavalResearch (ONR) (GrantN000141010337), theDARPA IT-MANET pro-gram (Grant W911NF-07-1-0028), and the ArmyResearch Labs (GrantW911NF-10-1-0420).


INTRODUCTIONInterference is a major impairment to success-ful communication in commercial and militarywireless systems. In cellular systems, interfer-ence is created when different base stationsshare the same carrier frequency due to fre-quency reuse. Intercell interference reducesdata rates throughout the cells and causes out-ages at the cell edges. In local area networks,interference is created when different accesspoints share the same channel. The mediumaccess control protocol attempts to deal withinterference by avoiding packet collisions (over-lapping transmissions). This conservativeapproach leads to underutilization of systembandwidth. Similarly, neighboring nodes in adense mobile ad hoc network interfere if theyshare the same time and frequency resources.The medium access control protocol again lim-

its the number of simultaneous conversationsand consequently the system performance.Interference is thus a critical impairment inmost wireless systems.

Communication in the presence of interfer-ence is often analyzed using an abstractionknown as the interference channel. In theexample interference channel of Fig. 1, threedifferent transmitters wish to communicatewith three receivers. Each transmitter has amessage only for its paired receiver. Assumingthe transmitters share the same time and fre-quency resources, each transmission createsinterference at the unintended receivers. Theremay be other sources of interference not illus-trated, such as jamming in military networksor self-interference created from nonlinearitiesin the radio frequency components; thosesources are not captured in the basic interfer-ence channel.

Traditional methods for dealing with interfer-ence often revolve around giving each user exclu-sive access to a fraction of the communicationresources. In frequency-division multiple access(FDMA), the system bandwidth is dividedamong the transmitters; for example, in Fig. 1each transmitter would be given a third of thetotal bandwidth. In time-division multiple access(TDMA), users take turns transmitting on aperiodic set of transmission intervals. In randomaccess systems (e.g., carrier sense multipleaccess), transmitters listen to see if the channelis available and then transmit if it is. Randomaccess protocols are typically much less efficientthan preassigned orthogonal access like FDMAor TDMA since the spectrum may not be fullyutilized and collisions may occur. Regardless ofthe access protocol, the unifying concept remainsavoiding interference by limiting the number ofoverlapping transmissions. If simultaneous trans-missions are allowed, the resulting interferenceis often treated as noise, and a loss in data rateensues.

Recently, a new concept for communicationin interference channels, called interferencealignment (IA), was proposed in [1, 2]. IA is acooperative interference management strategy

OMAR EL AYACH, THE UNIVERSITY OF TEXAS AT AUSTINSTEVEN W. PETERS, KUMA SIGNALS, LLC

ROBERT W. HEATH JR., UNIVERSITY OF TEXAS AT AUSTIN

ABSTRACTInterference alignment is a revolutionary

wireless transmission strategy that reduces theimpact of interference. The idea of interfer-ence alignment is to coordinate multiple trans-mitters so that their mutual interference alignsat the receivers, facilitating simple interferencecancellation techniques. Since interferencealignment’s inception, researchers have investi-gated its performance and proposed severalimprovements. Research efforts have been pri-marily focused on verifying interference align-ment’s ability to achieve the maximum degreesof freedom (an approximation of sum capaci-ty), developing algorithms for determiningalignment solutions, and designing transmis-sion strategies that relax the need for perfectalignment but yield better performance. Thisarticle provides an overview of the concept ofinterference alignment as well as an assessmentof practical issues including performance inrealistic propagation environments, the role ofchannel state information at the transmitter,and the practicality of interference alignmentin large networks.

THE PRACTICAL CHALLENGES OFINTERFERENCE ALIGNMENT

HEATH LAYOUT_Layout 1 2/21/13 4:07 PM Page 35


that exploits the availability of multiple signalingdimensions provided by multiple time slots, fre-quency blocks, or antennas. The transmittersjointly design their transmitted signals in themultidimensional space such that the interfer-ence observed at the receivers occupies only aportion of the full signaling space. An amazingresult from [2] is that alignment may allow thenetwork’s sum data rate to grow linearly, andwithout bound, with the network’s size. This is insharp contrast with orthogonal access strategieslike FDMA or TDMA, where sum rate is moreor less constant since, regardless of network size,only one pair of users can communicate in agiven time/frequency block. Since the develop-ment of IA, there has been work on a variety oftopics to understand its theory and potentialapplications. A good summary of key results isgiven in [3].

The objective of this article is to review IAwith a focus on making it practical. First, wedescribe IA in more detail, summarizing rele-vant practical issues that we believe are criti-cal for its successful deployment. Then wereview different techniques for computingalignment solutions and more general inter-ference management solutions. Because com-puting these solut ions rel ies heavi ly onchannel state information (CSI), we describethe two competing CSI acquisition techniques,reciprocity and feedback, and focus on theirrespective benefits and limitations. We high-light the fact that the dimensions needed foralignment and the overhead of CSI acquisi-tion both rapidly increase with network size.This places a limit on the gains achievable viaIA in large networks. For this reason we sum-marize work on partial connectivity and userclustering, which leverages network topologyinformation to reduce the requirements ofalignment. We conclude with a discussion ofthe research challenges that remain in realiz-ing IA in practice.

We note that the objective of this article is toprovide a high-level introduction to the linearprecoding type of IA, which exploits CSI for allusers. Deeper discussions of other forms of IA,including blind alignment, are found in [3].

LINEAR INTERFERENCE ALIGNMENT:CONCEPT

Interference alignment in its simplest form is aprecoding technique for the interference chan-nel. It is a transmission strategy that linearlyencodes signals over multiple dimensions such astime slots, frequency blocks, and antennas. Bycoding over multiple dimensions, transmissionsare designed to consolidate (i.e., align) the inter-fering signals observed at each receiver into alow-dimensional subspace. By doing so, IA maxi-mizes the number of non-interfering symbolsthat can be simultaneously communicated overthe interference channel, otherwise known as themultiplexing gain. Interestingly, achieving thechannel’s maximum multiplexing gain, alsoknown as degrees of freedom, implies that thesum rates provided by IA can approach sumcapacity at high signal-to-noise ratio (SNR).

To illustrate the IA concept, consider thefour-user system of Fig. 2, where real valued sig-nals are coded over three dimensions and com-municated over real valued channels. In thisfour-user system, a receiver observes a total ofthree interference signals, each of which is repre-sented as a vector in the real 3D space. Withoutcareful structure, the three interference signalswill occupy all three signal dimensions at thereceiver; the signals do not fortuitously align. IAallows users to cooperatively precode their trans-missions such that each three interference signalsare fully contained in a two-dimensional space.Alignment thus leaves one dimension in whichusers can decode their messages free from inter-ference by projecting the received signal onto thesubspace orthogonal to the interference.

The IA visualization in Fig. 2 translates intoan equally intuitive mathematical representation.Consider a K-user interference channel in whicheach user i transmits a set of Si information sym-bols encoded in the Si ¥ 1 vector si. Symbols siare precoded using a matrix Fi and observed ateach receiver l after propagating over the matrixchannel Hl,i; the dimensions of Fi and Hl,i will bemade clear shortly. For such a system, thereceived signal at user i is

where the vector vi is the noise observed at receiv-er i. When only the time or frequency dimensionis used for precoding (e.g., when coding over abandwidth of B subcarriers), Hl,i is a B ¥ B diago-nal matrix; diagonal since subcarriers are orthogo-nal. In a multiple-input multiple-output (MIMO)system with NT transmit antennas and NR receiveantennas, Hl,i is NR ¥ NT with no special struc-ture.

Regardless of the dimension used to aligninterference, IA can be summarized as calculat-ing a set of precoders F l such that any givenuser, even using a simple linear receiver Wi, cancancel the interference it observes from all otherusers (i.e., Wi*Hi,lFl = 0 "l ≠ i) without nullingor destroying its desired signal Wi*Hi,iFi. Earlywork on IA showed that a system’s ability to find

y H F s H F s vi i i i i ii

i= + +≠∑, , ,ll

l l

Figure 1. Illustration of a three transmit/receive pair interference channel. Eachtransmitter creates a signal that is interpreted as interference by its unintendedreceiver.

Received signal

Interference

Direct link

TX1 RX1

TX2 RX2

TX3 RX3



such IA precoders is directly related to the num-ber of signal dimensions over which it can code.Intuitively, the more time slots, frequency blocks,or antennas available for precoding, the moreflexibility a system has in aligning interference.The problem of characterizing the number ofcoding dimensions needed to ensure the feasibil-ity of IA has been studied extensively [4].

LINEAR INTERFERENCE ALIGNMENT:CHALLENGES

Interference alignment relies on some assump-tions that must be relaxed before it is adopted inpractical wireless systems. In this section, webriefly discuss some of the most pressing chal-lenges facing the transition of IA from theory topractice.

DIMENSIONALITY AND SCATTERINGIA is achieved by coding interference over multi-ple dimensions. Intuitively, the more interferingsignals that need to be aligned, the larger thenumber of dimensions needed to align them.

When using the frequency domain for align-ment, prior work has shown that the number ofsignaling dimensions needed to achieve good IAperformance grows faster than exponentially withthe number of IA users. Properly aligning evenas few as four users requires a potentially unrea-sonable number of subcarriers and a correspond-ingly large bandwidth [3]. This dimensionalityrequirement poses a major challenge for IA inpractical systems. The dimensionality require-ment is relatively milder in the case of MIMOIA where more users can be supported as longas the number of antennas grows linearly withnetwork size [4]. For this reason, IA seems mostlikely to be implemented in MIMO systems. Wethus place a special focus on MIMO IA.

SIGNAL-TO-NOISE RATIOIA is often degrees of freedom optimal, meaningthat the sum rates it achieves approach the chan-nel’s sum capacity at very high SNR. At moder-ate SNR levels, however, the sum rates resultingfrom IA may fall short of the theoretical maxi-mum. As a result, IA may be of limited use tosystems with moderate SNR unless IA algo-rithms are further improved. Examples of suchalgorithms are discussed later.

CHANNEL ESTIMATION AND FEEDBACKChannel state information, be it at the transmit-ter or receiver, is central to calculating IA pre-coders. As a result, sufficient resources must beallocated to pilot transmission, and in somecases to CSI feedback, to ensure the availabilityof accurate CSI. Since IA precoders must berecalculated when the channel changes apprecia-bly, the overhead of CSI acquisition in high-mobility fast-fading systems can limit the gainsof IA. For this reason, low-overhead signalingstrategies must be devised to properly trade offCSI quality with CSI acquisition overhead.

SYNCHRONIZATIONIA via linear precoding is a transmission strategyfor the coherent interference channel. Thus, IA

requires tight synchronization to remove anytiming and carrier frequency offsets betweencooperating nodes. In the absence of sufficientsynchronization, additional interference termsare introduced to the signal model, renderingthe IA solution ineffective. Synchronizationstrategies that leverage GPS satellite signalscould help fulfill this requirement.

NETWORK ORGANIZATIONNodes cooperating via IA must not only syn-chronize, but also negotiate physical layerparameters, share CSI, and potentially self-orga-nize into small alignment clusters if full networkalignment proves to be too costly. In the absenceof centralized control, distributed network proto-cols must be redesigned around this more com-plex and cooperative physical layer.

FURTHER DESCRIPTION OF CHALLENGESThe following three sections are devoted to sur-veying the research efforts invested in overcom-ing some of the challenges discussed in thissection. We present algorithmic improvementsthat primarily target IA’s low SNR performance,we explore methods of obtaining the CSIrequired by IA, and we address the challenge ofapplying IA to larger networks.

COMPUTING INTERFERENCEALIGNMENT SOLUTIONS

There are some simple formulas for calculatingIA precoders in some special cases [2]. To enableIA in general network settings, however,researchers have relied on developing iterativeIA algorithms. Since then, the algorithmic focushas largely been on MIMO IA.

The earliest method for finding MIMO IAprecoders was the distributed solution presentedin [5]. The idea for the algorithm is that at eachiteration, users try to minimize the extent towhich their signal leaks into the other users’desired signal subspaces. After the algorithmconverges, the IA condition Wi*Hi,lFl = 0 should

Figure 2. A diagram illustrating interference alignment. At each receiver, threeinterferers collapse to appear as two. This enables interference-free decodingin a desired signal subspace.



ideally be satisfied and the desired signal spacesbe free of interference. While the algorithm per-forms well at high SNR, it can be far from opti-mal in badly conditioned channels or at lowSNR. The reason is that while this algorithmproperly aligns interference, it is oblivious towhat happens to the desired signal power duringthe process.

The first attempt to improve low SNR perfor-mance is the Max-SINR algorithm of [5]. Insteadof minimizing interference power at each itera-tion, the improved algorithm accounts for thedesired channel by iteratively maximizing theper-stream signal-to-interference-plus-noise ratio(SINR). By relaxing the need for perfect align-ment, Max-SINR outperforms IA at low SNRand matches its performance at high SNR. Otheralgorithms have also relaxed the perfect align-ment requirement and adopted more directobjectives such as maximizing network sum rate.For example, [6] highlights the equivalencebetween maximizing sum rate and minimizingthe signal’s sum mean square error and uses theproperties of the equivalent minimum meansquare error (MMSE) problem to give a simpleiterative algorithm in which precoders can beupdated in closed form or via traditional convexoptimization.

Suboptimal low SNR performance is admit-tedly not the only limitation of IA, and manyalgorithms have been developed to address vari-ous shortcomings. Most early work on IA, forexample, focused on the case where all interfer-ing users are able to cooperate. The ability tocooperate, however, is fundamentally limited byconstraints such as the number of antennas andcooperation overhead. As a result, large net-works will inevitably have uncoordinated inter-ference, or colored noise, which motivates theimproved IA algorithms in [7]. Another limita-tion of precoding for the MIMO interferencechannel is that it often requires sharing entirechannel matrices, potentially incurring a largeoverhead penalty, as we discuss later. The workin [8] proposes a distributed transmission strate-gy inspired by game theory in which both pre-coders and transmit powers are iterativelyadjusted to maximize sum rate by sharing scalarquantities known as “interference prices.” Thestrategy in [8] thus replaces channel matrix feed-back with several iterations of scalar feedback.Such a strategy could potentially reduce IA’soverhead.

While editorial constraints have limited thediscussion to the solutions of [2, 5–8], a largenumber of noteworthy solutions have been devel-oped to improve on IA performance. The inter-ested reader is encouraged to see the extensivelist of IA-inspired algorithms and results inhttp://www.profheath.org/research/interfer-encealignment/.

OBTAINING CSI IN THEINTERFERENCE CHANNEL

Calculating IA precoders requires accurateknowledge of the interference generated by eachtransmitter. The premise is that if a transmitterknows the “geometry” of the interference it cre-

ates, it can conceivably shape it to mitigate itseffect. Two methods of obtaining this knowledgeare reciprocity and feedback.

INTERFERENCE ALIGNMENT VIA RECIPROCITYIn time-division duplexed systems where trans-missions on the forward and reverse links over-lap in frequency and are minimally separated intime, propagation in both directions will be iden-tical. In such systems, channels are said to bereciprocal. Reciprocity enables IA by allowingtransmitters to infer the structure of the interfer-ence they cause by observing the interferencethey receive.

Consider, for example, the IA strategy pro-posed in [5] in which the precoders Fi and thecombiners Wi are updated iteratively to reducethe power of uncanceled interference Wi*Hi,lFlto zero. To do so, transmitters begin by sendingpilot data using an initial set of precoders.Receivers then estimate their interferencecovariance matrix and construct combiners thatselect the receive subspaces with least interfer-ence. When the roles of transmitter and receiverswitch, thanks to reciprocity, that same receivesubspace that carried the least interferencebecomes the transmit direction which causes theleast interference on the reverse link. Iteratingthis subspace selection on the forward andreverse links results in a set of precoders satisfy-ing the IA conditions.

While [5] at each iteration chooses the sub-space with least interference, a variety of moresophisticated objectives may be considered. Forexample, [9] considers an update rule that mini-mizes the signal’s mean square error, resulting inimproved sum rate performance. Regardless ofthe subspace selection rule, the general frame-work for precoding with reciprocity is shown inFig. 3a and proceeds as follows:• Forward link training: Transmitters send pre-

coded pilot symbols using a set of initial pre-coders. Receivers estimate forward channelparameters and compute combiners that opti-mize a predefined objective.

• Reverse link training: Receivers send precodedpilot symbols using the combiners from theabove step as transmit precoders. Transmittersin turn optimize their combiners/precodersand initiate a second training phase with theupdated precoders.

• Iterations: Communicating pairs iterate theprevious steps until convergence.

• Data transmission: Payload data is then com-municated.Relying on reciprocity for precoding over

the interference channel has a number ofpotential drawbacks. First, iterating over the airincurs a non-negligible overhead due to therecurring pilot transmissions. While the resultsin [9] consider pilot overhead, more work isneeded to settle the viability of reciprocity. Sec-ond, reciprocity may not suffice for all IA-based algorithms.

For example, one of the algorithms in [7]attempts to improve IA by considering sourcesof uncoordinated interference. Since the uncoor-dinated interference observed by the transmit-ters and receivers is not “reciprocal,” reciprocitycannot be used. Third, reciprocity does not hold

Calculating IA pre-coders requires accu-rate knowledge ofthe interference gen-erated by each trans-mitter. The premiseis that if a transmit-ter knows the“geometry” of theinterference it cre-ates, it can conceiv-ably shape it tomitigate its effect.Two methods ofobtaining this knowl-edge are reciprocityand feedback.


http://www.profheath.org/research/interferencealignment/


in frequency duplexed systems, and ensuringreciprocity with time duplexing requires tightlycalibrated radio frequency (RF) devices.

INTERFERENCE ALIGNMENT WITH FEEDBACKSeveral IA results have also considered systemswith CSI feedback. A general feedback frame-work is shown in Fig. 3b. In such a system, thetransmitters first send training sequences withwhich receivers estimate the forward channels.Receivers then feed back information about theestimated forward channels, potentially aftertraining the reverse link. After feedback, thetransmitters have the information needed to cal-culate IA precoders. Feedback, however, invari-ably introduces distortion to the CSI at thetransmitters and incurs a non-negligible over-head penalty. Therefore, the difficulty lies indesigning low-overhead low-distortion feedbackmechanisms for IA.

An established method to provide high-quali-ty feedback with low overhead is limited feed-back (i.e., channel state quantization). Limitedfeedback was first considered in [10] for single-antenna systems where alignment is achieved bycoding over orthogonal frequency-division mult-plexed (OFDM) subcarriers. The feedback strat-egy in [10] leverages the fact that IA solutionsremain unchanged if channels are scaled orrotated, which allows efficient quantization via

what is known as Grassmannian codebooks.Unfortunately, maintaining proper alignmentrequires that the accuracy of quantized CSIimprove with SNR, which in turn implies thatthe quantization codebook size must scale expo-nentially with SNR. The complexity of quantizedfeedback, however, increases with codebook sizeand large Grassmannian codebooks are difficultto design and encode. Furthermore, this strategyrelies on the CSI’s Grassmannian structure forefficient quantization. As a result, it cannot beapplied to systems where the CSI exhibits nospecial structure. This alienates a main case ofinterest, namely IA in multiple antenna systemswhere the CSI to be fed back is the set of chan-nel matrices.

To support MIMO IA and overcome theproblem of scaling complexity, IA with analogfeedback was considered in [11]. Instead ofquantizing the CSI, analog feedback directlytransmits the channel coefficients as uncodedquadrature and amplitude modulated symbols.Since no quantization is performed, the onlysource of distortion is the thermal noise intro-duced during training and feedback. As a result,the CSI quality naturally increases with the SNRon the reverse link. This implies that IA’s multi-plexing gain is preserved as long as the forwardand reverse link SNRs scale together.

Feedback, however, poses a number of chal-

Figure 3. The two CSI acquisition paradigms considered in IA literature: a) reciprocity-based strategies that infer the forward channelfrom reverse link pilots; b) feedback strategies that rely on explicitly communicating forward channel information to the transmitters.

Iterate untilconvergence

Estimate forwardchannel

Calculate IAprecoders

Estimate reversechannel

Transmitters train forward channels

Transmitters send precoded data

Receivers send feedback to all

transmitters

Receivers train reverse channels

(b)

Time

RX1

RXk-1

RXk

RXk+1

RXK

Time

TX1

TXk-1

TXk

TXk+1

TXK

Optimize receivefilters




Transmitters send precoded pilots

Transmitters send precoded data

Receivers send

precoded pilots

Receivers send

precoded pilots

(a)

Time

RX1

RXk-1

RXk

RXk+1

RXK

Transmitters sendprecoded pilots

Time

TX1

TXk-1

TXk

TXk+1

TXK



lenges. Not only does the required quality of CSIscale with SNR, but the required quantity alsoscales super-linearly with network size. If feed-back is inefficient, IA’s overhead could dwarf itspromised theoretical gains. Second, IA relies onsharing CSI with interfering transmitters: multi-cell systems that share transmitter CSI over abackhaul may suffer from queuing delays thatrender CSI obsolete on arrival. While work onaddressing these issues is underway, mostresearch still considers systems with a limitednumber of users. In large-scale networks, howev-er, practical challenges are amplified manifold.

INTERFERENCE ALIGNMENT INLARGE SCALE NETWORKS

Consider applying IA, or any other interferencechannel precoding strategy, to a large-scale net-work, such as a regional cellular network. In thissetting, the need for signaling dimensions grows,the overhead of CSI acquisition explodes, andthe task of synchronization becomes daunting.This grim outlook, however, is a byproduct of anaive overgeneralization of the basic interfer-ence channel. In large-scale networks not allinterfering links are significant, and thus not allof them should be treated equally.

To examine the feasibility of IA in large net-works, [12] considered a large network whereineach user receives interference from a finite sub-set of users, such as the first tier of interferers ina grid-like cellular system. Using this model, [12]showed that the number of antennas needed fornetwork-wide alignment is a linear function of thesize of the interfering subset rather than the sizeof the entire network. This implies that perfectalignment in an infinitely large network is theoret-ically possible with a finite number of antennas.This partially connected model, however, is ulti-

mately a simplification of network interference.While some interferers may indeed be too weakto merit alignment, that threshold remainsunclear. Moreover, neglecting interferers that fallbelow that threshold, as is implicitly done in thepartially connected model, may not be optimal.Exploring the optimal transmission strategy in thisgray area between full connectivity and true par-tial connectivity is the focus of [13].

The work in [13] considers a wireless networkand assigns a finite channel coherence time anddifferent path loss constants to each link. Thissetup allows [13] to gauge the gain from aligninga set of users vs. the CSI acquistion overheadinvolved. On one hand, it is noticed that thelarge coherence time in static channels dilutesthe cost of CSI and enables alignment over thewhole network. On the other hand, fast fadingchannels do not allow enough time to acquirethe CSI needed for IA. In this latter case, sys-tems perform better with lower overhead strate-gies such as TDMA. In between the twoextremes, it is noticed that systems benefit mostfrom a hybrid IA/TDMA strategy wherein small-er subsets of users cooperate via alignment, andTDMA is applied across groups. The objectivethen becomes finding an optimal user grouping,or network partitioning, for which [13] providesvarious algorithms based on geographic parti-tioning, approximate sum rate maximization, andsum rate maximization with fairness constraints.In all grouping solutions only long-term pathloss information is used, which further reducesthe overhead of IA by avoiding frequentregrouping.

The IA research community has made suc-cessful initial attempts at demonstrating that thereach of IA can extend well beyond the informa-tion theoretic interference channel through worksuch as [7, 12] and many others. Perhaps themost crucial next step toward realizing IA gainsis to rethink the upper layers above IA in thecommunication stack (e.g., the medium accesslayer). These protocols have traditionally beenbuilt around point-to-point physical layers andmust be redesigned to enable more cooperativemulti-user physical layers instead. Preliminarywork on this front is already in progress throughthe prototyping work in [14], which extends theearlier prototyping efforts by the same authors.

A PRACTICAL PERFORMANCE EVALUATION

In this section we provide numerical results onthe performance of IA with a special focus onthe concepts discussed throughout the article.Figure 4 compares the performance of the algo-rithms presented earlier in a three-user systemwith two antennas per node. For this assessmentboth simulated channels and measured MIMO-OFDM channels from the testbed in [15] areconsidered. Simulated channels are assumed tobe Rayleigh fading with equal average receivedpower on all links. Simulated channels are fur-ther assumed independent across all users andantennas. This constitutes a baseline of highlyscattered channels, considered ideal for IA. Todemonstrate the effect of limited scattering, wereport measurements from [15] corresponding toa scenario in which antennas are spaced half a

Figure 4. A performance comparison for three precoding solutions for the inter-ference channel over both measured and simulated channels.

SNR (dB)50

5

0

Sum

rat

e (b

/s/H

z)

10

15

20

25

30

35

10 15 20 25 30 35

IA over measured channelsWMMSE over measured channelsMAX SINR over measured channelsIA over simulated channelsWMMSE over simulated channelsMAX SINR over simulated channelsWMMSE over simulated ch. (10 initializations)



wavelength apart. The results in Fig. 4 verify that IA, in this

case, achieves a multiplexing gain of three aspredicted in theory. Multiplexing gain can becalculated by examining the slope of the sumrate curves. The same results, however, alsodemonstrate IA’s suboptimality at low SNR,where precoding algorithms such as theWMMSE algorithm in [6] or the MAX-SINRalgorithm in [7] consistently outperform. Finally,Fig. 4 shows that the correlation present in mea-sured channels can have significant effects onthe performance of IA. Interestingly, in suchmeasured channels, the algorithms in [6, 7] pro-vide large gains over IA even at a significantSNR of 35dB.

In Fig. 5 we relax the assumption of perfectchannel knowledge and consider a five user sys-tem that acquires CSI through training and ana-log feedback, thus suffering from both CSIdistortion and CSI acquisition overhead. Theresults indicate that, at pedestrian speeds, theslow channel variation enables acquiring accurateCSI at a low overhead cost, which translates intosubstantial gains over non-cooperative transmis-sion strategies such as TDMA. The same resultsindicate that in very high mobility settings, whereCSI acquisition is costly, one may elect to eithercoordinate a smaller user group or altogetheradopt a lower overhead strategy such as TDMA.In a large wireless network, grouping users basedon overhead constitutes selecting from a continu-um of network structures. For example, Fig. 6demonstrates how even a simple six-user systemshould be partitioned differently as the overheadcost of full IA varies. The throughput maximizingstrategy transitions from full six-user IA in staticchannels to basic TDMA for fast fading channels,passing through a range of “hybrid” networkstructures in which users cluster into small coop-erating groups.

FUTURE RESEARCH DIRECTIONS

In this section we present some active areas ofinterest and general topics for further researchon interference alignment.

Algorithms: Algorithms remain a hot topic forIA research as a number of “extra features” canbe incorporated or further improved such as com-plexity, low SNR performance, distributed com-putation, and robustness to CSI imperfections.

Feedback: Temporal correlation can beleveraged to both improve the accuracy ofCSI as well as reduce the overhead of feed-back. Limited feedback strategies such as [10]can be generalized to better support MIMOIA. The development of l imited feedbackstrategies can benefit from the lessons learnedin the MIMO single-user and broadcast chan-nels where feedback can be significantly com-pressed by exploit ing the mathematicalproperties of the quantized CSI. Finally, addi-tional work is needed to better understand theeffects of feedback overhead and feedbackdelay.

IA in multihop networks: Preliminary workon multihop IA suggests that relays can greatlyreduce the coding dimensions needed to achievea network’s degrees of freedom, and otherwise

simplify the optimal transmission strategies [3].The importance of practical precoding algo-rithms for the relay-aided interference channel isfurther amplified by the standards community’sgrowing interest in deploying relays in futurewireless systems.

IA in modern cellular networks: To make IAa viable multicell cooperation technique, IAresearch must characterize the effect of cellularsystem complexities such as scheduling, resourceallocation, and backhaul signaling delay. IAmust also be re-evaluated using models thatmore accurately resemble modern cellular sys-tems, which could include heterogeneous infra-structure such as macrocells, picocells, smallcells, relays, and distributed antennas.

CONCLUSION

Interference alignment is a transmission strategythat promises to improve throughput in wirelessnetworks. This article reviews the key concept oflinear interference alignment, surveys recentresults on the topic, and focuses on bringing theconcept closer to implementation by discussingsome of its main limitations. After the key hur-dles in areas such as low SNR performance, CSIacquisition overhead, synchronization, and dis-tributed network organization can be overcome,interference alignment will be ready for practicalimplementation.

REFERENCES[1] M. Maddah-Ali, A. Motahari, and A. Khandani, “Com-

munication over MIMO X Channels: Interference Align-ment, Decomposition, and Performance Analysis,” IEEE

Figure 5. Effective throughput vs. SNR for a five user system at varying normal-ized Doppler spreads. Each transmitter is equipped with 3 antennas, andevery user communicates via one spatial stream. The figure demonstrates howas Doppler increases, so does overhead and CSI error, resulting in a drop ineffective sum rate. Channels are assumed to be Rayleigh and user speeds arecalculated for a system with a coherence bandwidth of 300 kHz at a carrierfrequency of 2 GHz.

SNR (dB)50

5

0

Effe

ctiv

e su

m t

hrou

ghpu

t (b

its/

s/H

z)

10

15

20

25

30

35

40

45

50

55

10 15 20

Pedest

rian sp

eeds

Vehicu

lar sp

eeds

TDMA baseline (no overhead)

25 30 35

5 users at v=10 km/hr5 users at v=160 km/hr5 users at v=320 km/hr5 users at v=480 km/hr



Trans. Info. Theory, vol. 54, no. 8, 2008, pp. 3457–70. [2] V. Cadambe and S. Jafar, “Interference Alignment and

Degrees of Freedom of the K-User Interference Chan-nel,” IEEE Trans. Info. Theory, vol. 54, no. 8, Aug.2008, pp. 3425–41.

[3] S. Jafar, Interference Alignment — A New Look at Sig-nal Dimensions in a Communication Network, NowPublications, 2011.

[4] M. Razaviyayn, G. Lyubeznik, and Z.-Q. Luo, “On theDegrees of Freedom Achievable through Interference Align-ment in A MIMO Interference Channel,” IEEE Trans. SignalProc., vol. 60, no. 2, Feb. 2012, pp. 812–21.

[5] K. Gomadam, V. Cadambe, and S. Jafar, “A DistributedNumerical Approach to Interference Alignment and Appli-cations to Wireless Interference Networks,” IEEE Trans.Info. Theory, vol. 57, no. 6, June 2011, pp. 3309–22.

[6] Q. Shi et al., “An Iteratively Weighted MMSE Approachto Distributed Sum — Utility Maximization for a MIMOInterfering Broadcast Channel,” IEEE Trans. Signal Proc.,vol. 59, no. 9, Sept. 2011, pp. 4331–40.

[7] S. W. Peters and R. W. Heath, Jr., “Cooperative Algo-rithms for MIMO Interference Channels,” IEEE Trans.

Vehic. Tech., vol. 60, no. 1, Jan. 2011, pp. 206–218. [8] C. Shi et al., “Local Interference Pricing for Distributed

Beamforming in MIMO Networks,” Proc. IEEE MILCOM,Oct. 2009.

[9] C. Shi, R. A. Berry, and M. L. Honig, “Adaptive Beam-forming in Interference Networks via Bi-DirectionalTraining,” ArXiv preprint arXiv:1003.4764, Mar. 2010,http://arxiv.org/abs/1003.4764

[10] H. Bölcskei and I. Thukral, “Interference Alignmentwith Limited Feedback,” Proc. IEEE Int’l. Symp. Info.Theory, July 2009, pp. 1759–63.

[11] O. El Ayach and R. W. Heath, Jr., “Interference Alignmentwith Analog Channel State Feedback,” IEEE Trans. WirelessCommun., vol. 11, no. 2, Feb. 2012, pp. 626–36.

[12] M. Guillaud and D. Gesbert, “Interference Alignmentin the Partially Connected K-User MIMO InterferenceChannel,” Proc. Euro. Signal Proc. Conf., Barcelona,Spain, Sept. 2011, pp. 1–5.

[13] S. W. Peters and R. W. Heath, Jr., “User Partitioning forLess Overhead in MIMO Interference Channels,” IEEE Trans.Wireless Commun., vol. 11, no. 2, 2012, pp. 592–603.

[14] K. Lin, S. Gollakota, and D. Katabi, “Random AccessHeterogeneous MIMO Networks,” Proc. ACM SIGCOMMConf. 2011, pp. 146–57.

[15] O. El Ayach, S. Peters, and R. W. Heath, Jr., “The Feasi-bility of Interference Alignment over Measured MIMO-OFDM Channels,” IEEE Trans. Vehic. Tech., vol. 59, no.9, Nov. 2010, pp. 4309–21.

BIOGRAPHIESOMAR EL AYACH [S’08] ([email protected]) received hisB.E. degree in computer and communications engineeringfrom the American University of Beirut, Lebanon, in 2008. Hecompleted his M.S. in electrical engineering degree in 2010at the University of Texas at Austin (UT Austin) where he isnow a Ph.D. student. He was an intern at the University ofCalifornia, Berkeley, in the summer of 2007 and an intern atSamsung Electronics in the summers of 2011 and 2012.

STEVEN W. PETERS [S’03] ([email protected]) isco-founder and chief executive officer of Kuma Signals, LLCin Austin, Texas. He received B.S. degrees in electrical engi-neering and computer engineering from the Illinois Instituteof Technology in Chicago in 2005, and an M.S.E. degree inelectrical engineering from UT Austin in 2007, where he isalso working toward a Ph.D. in electrical engineering.

ROBERT W. HEATH JR. [S’96, M’01, SM’06, F’11] ([email protected]) is a professor in the Electrical and Communica-tions Engineering Department at UT Austin and the directorof the Wireless Networking and Communications Group. Hereceived his B.S. and M.S. in electrical engineering from theUniversity of Virginia and his Ph.D. in electrical engineeringfrom Stanford University. He is the president and CEO ofMIMO Wireless Inc. and chief innovation officer at KumaSignals LLC. He is the recipient of the David and DorisLybarger Endowed Faculty Fellowship in Engineering and isa registered Professional Engineer in Texas.

Figure 6. The effective sum rate achieved by full network IA, TDMA, or a parti-tioned network in which smaller groups coordinate via IA. As overheadincreases, the plot indicates that it is often optimal to allow fewer users tocommunicate simultaneously.

Number of orthogonal groups21

5

0

Sum

rat

e b/

s/H

z

10

15

20

25

30

35

3 4 5 6

Basic TDMA6-user IA

10% overhead

30% overhead

50% overhead

70% overhead

90% overhead


http://arxiv.org/abs/1003.4764





INTRODUCTIONWireless communication has become essential toour lives in many ways, through a variety of ser-vices as well as devices ranging from pocketphones to laptops, tablets, sensors, and con-trollers. The advent of multimedia dominatedtraffic poses extraordinary constraints on datarates, latency, and, above all, spectral efficiency.In order to deal with the expected saturation ofavailable resources in currently used bands, newwireless systems are designed based on:• Greater densification of infrastructure equip-

ment (small cells)• Very aggressive spatial frequency reuse, which

in turn results in severe interference conditionsfor cell-edge terminalsThe role played by multiple antenna combin-

ing in mitigating interference by means of zeroforcing (ZF) (or related criteria) beamforming iswell established. Over the last few years, thecombination of multiple-antenna approachestogether with the concept of cooperation amonginterfering wireless devices was explored, show-ing strong promise (see [1] and references there-in). In particular, transmitter (TX)-based

cooperation allows for avoidance of interferencebefore it even takes place (e.g., multicell multi-ple-input multiple-output [MIMO] or “joint pro-cessing coordinated multipoint [CoMP]”), orhelps to shape it in a way which makes it easierfor the receivers (RX) to suppress it (e.g., align-ment). TX cooperation methods can be catego-rized depending on whether the data messagesintended at the users must be known at severalTXs simultaneously or not. For systems notallowing such an exchange (e.g., due to privacyregulations or low backhaul capabilities), inter-ference alignment (IA) has been shown to beinstrumental [2]. In contrast, when user datamessage exchange is made possible by a specificbackhaul routing architecture, multicell, a.k.a.“network” MIMO, methods offer the best theo-retical benefits [3].

A distinct advantage of TX cooperation overconventional approaches relying on egoisticinterference rejection lies in the reduced numberof antennas needed at each RX to ZF residualinterference. This gain is further amplified whenuser data messages exchange among TXs ismade possible. For instance, in the case of threeinterfering two-antenna TXs, relying on RX-based interference rejection alone requires threeantennas to ZF the interference at each RX,while just two are needed when coordination isenabled via IA [4]. Furthermore, if the threeuser messages are exchanged among the TXs,thus enabling network MIMO precoding, justone antenna per TX and RX is sufficient to pre-serve interference-free transmission.

However, the benefits of multiple antennatransmit cooperation are gained at the expenseof requiring channel state information (CSI) atthe TXs. Indeed, whether one considers cooper-ation with or without user data sharing, the TXsshould in principle acquire the complete CSIpertaining to every TX and RX pair in the net-work. This is also the case for distributedschemes (e.g., [5]) where the computation ofprecoders typically relies on iterative techniqueswhere each iteration involves the acquisition oflocal feedback. However, as local feedback isupdated over the iterations, this approach implic-itly allows each TX to collect information aboutthe precoders and channels of other TXs, henceamounting to an iterative global CSI acquisitionat all TXs.

PAUL DE KERRET AND DAVID GESBERT, EURECOM

ABSTRACTMultiple-antenna “based” transmitter cooper-

ation has been established as a promising tooltoward avoiding, aligning, or shaping the inter-ference resulting from aggressive spectral reuse.The price paid in the form of feedback andexchanging channel state information betweencooperating devices in most existing methods isoften underestimated, though. In reality, feed-back and information overhead threatens thepracticality and scalability of TX cooperationapproaches in dense networks. Hereby weaddresses a “Who needs to know what?” prob-lem when it comes to CSI at cooperating trans-mitters. A comprehensive answer to this questionremains beyond our reach and the scope of thisarticle. Nevertheless, recent results in this areasuggest that CSI overhead can be contained foreven large networks provided the allocation offeedback to TXs is made non-uniform and toproperly depend on the network’s topology. Thisarticle provides a few hints toward solving theproblem.

CSI SHARING STRATEGIES FOR TRANSMITTERCOOPERATION IN WIRELESS NETWORKS

DE KERRET LAYOUT_Layout 1 2/21/13 4:08 PM Page 43


At first glance, CSI feedback and sharingrequirements grow unbounded with networksize. Since over the air feedback and backhaulexchange links are always rate and latency limit-ed, the practical application of TX cooperationin dense large networks is difficult.

In this article, we challenge the common viewthat interfering TXs engaging in a cooperativescheme can or should share global (network-wide) CSI. Instead, we formulate the problem ofa suitable CSI dissemination (or allocation) poli-cy across transmitting devices while maintainingperformance close to the full CSIT sharing sce-nario. We report a couple of findings revealinghow the need for CSIT sharing can be alleviatedby exploiting specific antenna configurations ordecay property of signal strength vs. distance,hence making TX cooperation distributed andscalable. We use interference alignment and net-work MIMO as our driving scenarios.

More specifically, for the cooperation sce-nario without user data message sharing wherealignment of interference is sought, we showhow perfect alignment is possible in certainantenna topologies without knowledge of allthe channel elements at some TXs. For the net-work MIMO scenario, this is not the case, andwe illustrate instead how power decay vs. dis-tance can be exploited to substantially reducethe CSI sharing requirements while fulfillingoptimal asymptotic rate performance condi-tions. A common trait behind the findings isthat different cooperating TXs can (and oftenmust) live with their own individual partial ver-sion of the global CSIT. Hence, CSIT represen-tation quality is bound to be non-uniform acrossTXs. Consequently, we discuss briefly the prob-lem of multiple-antenna precoding with TX-dependent CSI.

BRIEF NOTIONS IN MULTIPLE-ANTENNATRANSMITTER COOPERATION

We consider fast fading multiple-antenna wire-less networks where the transmission can bemathematically represented by writing yi, thereceived signal at RX i, as

(1)

where xi is the signal emitted by TX i and Hij isa matrix containing the channel elementsbetween TX j and RX i. The transmitted sym-bols x = [x1, …, xK]T are then obtained fromthe user’s data symbols s = [s1, …, sK]T bymultiplication with a precoder T (i.e., x = Ts).If the user’s data symbols are not sharedbetween the TXs, the precoder T is restrictedto a particular block-diagonal structure, whileit can otherwise take any form. The receivedfilter gi

H is then applied to the received signal yito obtain an estimate of the transmitted datasymbol.

Here, we briefly discuss the leading tech-niques for MIMO-based cooperation with orwithout user data message exchange. We pointout commonly made assumptions in terms ofCSIT sharing and feedback design.

INTERFERENCE ALIGNMENT FORINTERFERENCE CHANNELS

When the user’s data symbols are not sharedbetween the TXs, the setting is referred to as aninterference channel (IC) in the communicationtheoretic literature. In MIMO ICs, a methodcalled interference alignment (IA) has recentlybeen developed and shown to achieve the maxi-mal number of degrees of freedom (DoF), orpre-log factor, in many cases [2, 4]. As a conse-quence, IA has attracted a lot of interest in thecommunity. In this work, we take the DoF asour key performance metric, such that we focuson IA schemes.

IA is said to be feasible if the antenna config-uration (i.e., the distribution of antenna ele-ments at the TXs and RXs) yields enoughoptimization variables to allow for interference-free transmission of all user data symbols, whichmeans fulfilling [4]

"i, "j ≠ i, g iHHij tj = 0. (2)

Intuitively, IA consists of letting the TXs coordi-nate among themselves to beamform their sig-nals such that the interferences received at eachof the RXs are confined in a subspace of reduceddimensions, which can then be suppressed by lin-ear filtering at the RXs with a smaller number ofantennas.

PRECODING IN THE NETWORK MIMO When the user’s data symbols are sharedbetween the TXs, the TXs form a distributedantenna array, and a joint precoder can beapplied at the transmit side [3]. Consequently,this setting becomes similar to the single TXmulti-user MIMO downlink channel, and theinterference between the TXs can be completelycanceled, for example, by applying a global ZFprecoder T µ H–1.

LIMITED FEEDBACK VS. LIMITED SHARINGThe limited feedback capabilities have been rec-ognized as a major obstacle for the practical useof the precoding schemes described above. Con-sequently, a large body of literature has focusedon this problem, and both efficient feedbackschemes and robust transmission schemes havebeen derived for network MIMO [6, 7] and IA[8, 9].

However, all these works assume that theimperfect channel estimates obtained via limitedfeedback are perfectly shared between all the TXantennas. This is a meaningful assumption whenthe TXs are collocated but less realistic other-wise, as we shall now see.

CSIT Sharing Issues — One obstacle to the sharingof global CSIT follows from the fact that theamount of CSI that has to be exchanged increas-es very quickly with the number of TXs. In fact,each TX needs to obtain the CSI relative to thefull multi-user channel, which consists of (NK)2

scalars in a K-user setting with N antennas ateach node.

In addition, acquiring the CSI at a particu-lar TX can be realized by either direct broad-cast of the CSI to al l l istening TXs or an

∑= +≠

y H x H xi ii i ij jj i

A distinct advantageof TX cooperationover conventionalapproaches relyingon egoistic interfer-ence rejection lies inthe reduced numberof antennas neededat each RX to ZFresidual interference.This gain is furtheramplified when userdata messagesexchange amongTXs is made possible.



over-the-air feedback to the home base stationalone, followed by an exchange of the localCSIs over the backhaul, as it is currently advo-cated by Third Generation Partnership Project(3GPP) Long Term Evolution — Advanced(LTE-A) standards [10]. See the illustrationsfor such scenarios in Fig. 2. Note thatexchange over the backhaul can involve fur-ther quantization loss and may lead to a dif-ferent CSI aging at each TX due to protocollatency. In either case, the channel estimatesavailable at the various TXs will not be exactlythe same. This leads to a form of CSI discrep-ancy that is inherent in the cooperation amongnon-collocated TXs.

In order to capture multiple-antenna precod-ing scenarios whereby different TXs obtain animperfect and imperfectly shared estimate of theoverall multi-user channel, we denote by H(j) thenetwork-wide channel matrix estimate availableat TX j. Consequently, the precoding schemes inFig. 1 have to be modified to take into accountthat each TX will compute its precoder based onits own channel estimate. Thus, TX j transmits xj= ej

TT(j)s based on the knowledge of H(j) only. The fundamental questions that arise are:

• How complete and accurate should the esti-mate H(j) be for each j while operating underreasonable CSI overhead constraints?

• How should precoders be designed given thelikely discrepancies between various channelestimates?Although these questions prove to be difficult

and to a large extent remain open, we shed somelight on the problem for two key scenarios in thefollowing sections.

ALIGNING INTERFERENCE WITHINCOMPLETE CSIT

Let us first consider an IC without a user’sdata sharing. Feasibility studies for IA are typi-cally carried out under the assumption of fullCSIT. However, one can show that IA feasibili-ty and the CSIT model are in fact tightly cou-

pled notions. Assume, for instance, that all theRXs were given a generous number of anten-nas equaling or exceeding the number of TXs;it is well known that the interference could besuppressed at the RXs alone, and no precod-ing, and hence no CSIT, is necessary. Thisexample suggests the existence of a trade-offbetween the number of antennas and the CSIsharing requirements. Thus, it is possible todesign IA algorithms using less CSIT than con-ventionally thought, without performancedegradations by exploiting the availability ofextra antennas at a subset of devices. Morespecifically, the problem of finding the minimalCSIT allocation that preserves IA feasibilitycan be formulated. The minimality refers tothe size of a CSIT allocation, defined as thetotal number of scalars sent through the multi-user feedback channel.

We differentiate between antenna configura-tions where IA is feasible, and the number ofantennas at the TXs and the RXs provide justenough optimization variables to satisfy align-ment conditions, denoted as tightly feasible, andthe ones where extra antennas are available,denoted as super feasible. Furthermore, we call aCSIT allocation strictly incomplete if at least oneTX does not have the complete multi-user CSI.With such concepts in place, the following lessoncan be drawn.

TIGHTLY FEASIBLE ICSA strictly incomplete CSIT allocation impliesthat some TXs compute their precoders in orderto fulfill IA inside a smaller IC formed by a sub-set of RXs and a subset of TXs. Most of thetime, this creates additional constraints for theoptimization of the other precoders, whichmakes IA infeasible. However, it can be shownthat IA feasibility can be preserved under thefollowing condition [11].

Lesson 1 — In a tightly feasible IC, there exists astrictly incomplete CSIT allocation preserving IAfeasibility if there exists a tightly feasible sub-ICstrictly included in the full IC.

Figure 1. The description of a distributed IA algorithm is done in (a) while (b) represents the distributed precoding in a Network MIMOwith user's data sharing. The matrix Ei is a matrix which selects the rows of the multi-user precoder T corresponding to the antennas atTX i.

s 3

s^2

s 1

CSI

CSI

T=[t1, t2, t3]=fJP(H)

T=[t1, t2, t3]=fIA(H)

T=[t1, t2, t3]=fIA(H)

T=[t1, t2, t3]=fIA(H)x3=E3

TTs

x1=E1TTs

x2=E2TTs

x2=t2s2

x1=t1s1

x3=t3s3

s^2

s^3s^1

s=[s1, s2, s3]Ts2

s1 s=[s1, s2, s3]T

s=[s1, s2, s3]Ts3

(b)(a)

CSI

CSI

CSI

T=[t1, t2, t3]=fJP(H)

T=[t1, t2, t3]=fJP(H)

CSI



Exploiting this result, a CSIT allocation algo-rithm is derived in [11] along with an algorithmthat achieves IA based on this incomplete CSITallocation. In a few words, it consists of givingto each TX the CSI relative to the smallesttightly feasible sub-IC to which it belongs. Theprecoders are then designed to align interfer-ence inside this smaller sub-IC, thus requiringonly a part of the total CSIT, while IA feasibili-ty is preserved. We will see in the simulationsresults presented in the following that the reduc-tion in CSIT size is significant. In fact, thereduction of the CSIT allocation feeds on theheterogeneity of the antenna configuration suchthat the more heterogeneous the antenna con-figuration is, the larger is the saving brought byusing the minimal CSIT allocation. This is par-ticularly appealing in regard to future networks,where mobile units and base stations from dif-ferent generations with different numbers ofantennas are likely to coexist.

SUPER-FEASIBLE ICSIn super-feasible settings, the additional anten-nas can be used to reduce the size of the mini-mal CSIT allocation. But how to optimallyexploit these additional antennas to reduce thefeedback size is a very intricate problem. Still, alow-complexity heuristic CSIT allocation can bederived [11]. The main idea behind the algo-rithm is to let some TXs or RXs ZF less inter-ference dimensions such that smalltightly-feasible settings are formed inside theoriginal setting.

The effective CSIT reduction is illustratedin Fig. 3 for a three-user IC. The results areaveraged over 1000 random distributions ofthe antennas across the TXs and RXs. If 12antennas are distributed between the TXs and

the RXs, the setting is tightly feasible, and theprevious CSIT allocation policy for tightly fea-sible sett ings is used. With more than 12antennas, the algorithm exploits every addi-tional antenna to reduce the size of the CSITallocation.

When the setting is tightly feasible, the reduc-tion in feedback size requires neither a DoFreduction nor any additional antenna and comesin fact “for free”: It simply results from exploit-ing the heterogeneity in the antenna numbers atthe TXs and RXs.

A CSIT ALLOCATION POLICY FORNETWORK MIMO

When the user’s data symbols are jointly pre-coded at the TXs, complete CSIT allocation, inthe sense defined above, is needed in all practi-cally relevant scenarios. So in this case, a differ-ent notion of reduced CSIT sharing must beadvocated. The essential ingredient of thisapproach is the classical intuition that a TXshould have a more accurate estimate for chan-nels creating the strongest interference (i.e.,originating from devices in the close neighbor-hood). This means the fact that interferencedecays with path loss can be exploited in princi-ple to reduce the CSIT sharing requirements.This concept was recently introduced in [12]. Amathematical tool known as generalized DoFcomes in handy to capture the effect of pathloss on the multiplexing gain of cooperatingnetworks with partially shared CSIT. Addition-ally, a simplified model referred to as a Wynermodel is used in this context to aid analyticaltractability and provide first insights into thisproblem.

Figure 2. Possible scenarios for the feedback of the CSI.

Quant./delay

H(1)H(1)

H(3)

H(3)

CSI broadcastscenario

CSI sharingscenario

H(2)

H(2)

h1h1

h1(1)

h1(3)

h1(3)

h1(1)

h2(1)

h2(2)

h2(2)

h2(3)

h3(1)

h3(3)

h3(2)

h3(3)

h3

h3

h2h2

hiT

H=[h1,h2h3]T

Quant./delay

Quant./delay



GENERALIZED DEGREES OF FREEDOM

TX cooperation methods are often evaluatedthrough the prism of DoF performance. Unfortu-nately, the DoF is essentially path-loss-indepen-dent, such that a DoF analysis fails to properlycapture the behavior of a large (extended) net-work MIMO. An extension of the notion of DoF,introduced in [13] as the generalized DoF, offers amuch better grip on the problem as it can bettertake path loss models into account. Upon definingthe interference level g as g = log(INR)/log(SNR)with SNR denoting the signal-to-noise ratio andINR the interference-to-noise ratio, it is possibleto define the generalized DoF as the DoFobtained when the SNR and INR both tend toinfinity for a given interference level g.

For ease of exposition, we consider scenarioswhere all the TXs and RXs have single anten-nas. The CSI is distributed, meaning that eachTX has its own channel estimate based on whichit computes its transmit coefficient without fur-ther exchange of information with the otherTXs. We denote the estimate at TX j by H(j) andits ith row, which corresponds to the channelfrom all TXs to RX i, by hi

(j). We consider a digi-tal quantization with a number of bits quantiza-tion hi

(j) equal to Bi(j). Therefore, TX j computes

its own version of the precoding matrix T(j)

based on its own estimate H(j). It then transmitsxj = ej

TT(j)s.

THE ONE-DIMENSIONAL WYNER MODELIn the simple 1D Wyner model [14], the TXs areregularly placed along a line, and only thedirect neighboring TXs emit non-zero interfer-ence. The channel is thus represented by atridiagonal matrix. Furthermore, we assumethat the interference from the direct neighborsis attenuated with a coefficient m = Pg –1,according to the generalized DoF model. Thetransmission in the Wyner model is schemati-cally presented in Fig. 4.

Our objective is to evaluate how small Bi(j)

can be while guaranteeing the same DoF as asystem with perfect CSIT. Obviously, sharing themost accurate CSIT to all the TXs is a possiblesolution, but the size of the CSI required at eachTX then grows unbounded with the number ofusers K, making this solution both inefficient andunpractical. In contrast, a much more efficientCSI sharing policy achieving the maximal gener-alized DoF, denoted as distance-based, is sum-marized below [12].

DISTANCE-BASED CSIT ALLOCATIONWe are interested in a CSIT allocation strategy,referred to below as distance-based, wherebyeach TX receives a number of CSI scalars thatremains bounded as the number of users Kincreases.

The distance-based CSIT allocation isobtained by setting for all i, j,

Bi(j)Dist = �([1 + (g – 1)Ái – jÁ]+

+ 2[g+(g – 1)Ái – jÁ]+)log2(P)˘ (3)

where [•]+ is equal to zero if the argument isnegative and to the identity function otherwise,and �•˘ is the ceiling operator. It can be shown

that the CSIT allocation {Bi(j)Dist}i,j allows the

maximal generalized DoF to be achieved (i.e.,those achieved in a system with perfect feed-back) [12]. The proof is based on the off-diago-nal exponential decay of the inverse of thetridiagonal channel matrix.

Setting g = 1 (no significant path loss attenu-ation) in the previous equation, a conventionalCSIT allocation is obtained where all channelsare described with the same number of bits at allTXs (uniform allocation). For g < 1, however,the number of bits allocated decreases with thedistance Ái – jÁ between the considered TX andthe index of the quantized channel until no bit atall is used for quantizing the channel if the dis-tance Ái – jÁ between the RX and the TX is larg-er than �1/(1 – g)˘ . Crucially, this solutionallocates to each TX a total number of bits thatno longer grows with K as only the CSI relativeto a neighborhood is shared at each TX.

The different CSIT allocations are comparedin Fig. 5 for K = 15 users and g = 0.5. The con-ventional CSIT allocation consists in providingthe best quality to all the TXs, while the otherstrategies have the same size as the distance-based CSIT allocation, but the feedback bits areshared uniformly and according to a convention-al clustering of size 3. It can be seen from Eq. 3that the ratio between the sizes of the distance-based CSIT allocation and the conventionalCSIT allocation is independent of the SNR.Here, the distance-based CSIT allocation repre-sents only 10 percent of the size of the conven-tional CSIT allocation.

Hence, the distance-based CSIT allocationachieves the maximal number of generalizedDoF with only a small share of the total CSIT,and outperforms the other schemes of compari-son. Additionally, the user’s data sharing canalso be reduced to a neighborhood without lossof performance. Consequently, this scheme canbe seen as an alternative to clustering in which

Figure 3. Average CSIT allocation size in terms of the number of antennasrandomly distributed across the TXs and RXs for K = 3 users.

Total number of antennas randomly distributed across the TXs and the RXs1312

20

0

Ave

rage

siz

e of

the

CSI

T al

loca

tion

40

60

80

100

120

140

160

14 15 16 17

Conventional (complete) CSIT allocationHeuristic proposed algorithmReduced CSIT allocation through exhaustive search



the hard boundaries of the clusters are replacedby a smooth decrease of the level of coopera-tion.

PRECODING IN THE NETWORK MIMOCHANNEL WITH DISTRIBUTED CSIT

As becomes clear from the previous section, anefficient CSI dissemination policy naturally leadsto a significant reduction of the CSI sharingrequirements. As a consequence, the CSIT isrepresented non-uniformly across the TXs. Sincenon-uniform sharing is the best strategy in orderto maximize performance under a given feedback

overhead constraint, it is a natural consequencethat some users’ channels will be coarselydescribed at certain TXs and more accurately atothers. Interestingly, the problem of designingprecoders that can accommodate such a peculiarCSIT scenario is by and large open. In particular,new robust precoding schemes should be devel-oped, as conventional precoders are designedunder the assumption that the same imperfectCSIT is shared perfectly among the TXs.

Let us consider the model of distributed CSIdescribed earlier where TX j receives its ownchannel estimate H(j) with the ith row, denotedby h i

(j), obtained using Bi(j) quantizing bits.

Assume a network where each TX has roughlythe same average path loss to each RX. TheDoF, which can be achieved with limited feed-back, is studied in [7] for the single TX MIMOdownlink channel. We can extend this to the set-ting of non-uniform CSI to gain insight into thedesign of efficient precoders. In this case, CSIscaling coefficients a i

(j) are introduced anddefined by the limit of Bi

(j)/((K – 1)log2(P)) whenP goes to infinity.

ZF is widely used and well known to achievethe maximal DoF in the MIMO downlink chan-nel with perfect CSIT [7]. One may wonder howconventional ZF performs in the presence ofCSI discrepancies brought by imperfect sharing.The answer is strikingly pessimistic: The sumDoF achieved can be shown to be equal to justK mini,j ai

(j) [15]. Intuitively, this can be restatedas follows.

Lesson 2 — In the network MIMO with distributedCSI, the worst channel estimate across the TXsand the users limits the DoF achieved by each userusing ZF precoding.

This is in strong contrast to the single TX casestudied in [7] where the quality of the feedbackof user i relative to hi only impacts the DoF ofuser i. It shows clearly the disastrous impact ofCSI non-uniformity, since one inaccurate esti-mate at one TX degrades the performance of allthe users. One may also wonder whether conven-tional-type robust precoders [6] can offer a betterresponse. The answer is negative, unfortunately.Instead, a novel precoder design is needed that istailored to the non-uniform CSIT sharing model.

Preliminary results to this end [15] suggestthat it is possible to dramatically improve theDoF in certain scenarios. For instance, in thetwo-TX network, a scheme referred to as active-passive (AP) ZF, consisting of letting the TXwith degraded CSIT arbitrarily fix its transmitcoefficient while the other TX compensates tozero-out the interference, can be shown to recov-er the optimal DoF.

The average rate achieved with conventionalZF, AP ZF, and ZF with perfect CSIT are com-pared in Fig. 6. In that case, the sum rate of con-ventional ZF saturates at high SNR, while APZF is more robust and achieves a better DoF.

OPEN PROBLEMS

New concepts for CSIT sharing in wireless net-works have been derived, and their potentialto reduce signaling overhead has been shown.

Figure 4. Schematic representation for the Wyner model considered.

......

TX j

RX j

H(j-1) H(j) H(j+1)

djdj-1 dj+1

µ aj-2 µ bj µ ajµ bj-1 µ aj-1 µ bj+1 µ bj+2 µ bj+2

Figure 5. Average rate per user in terms of the SNR. The distance-based CSITallocation, the uniform allocation, and the clustering one have all a size equalto 10 percent of the size of the conventional CSIT allocation.

SNR (dB)100

2

0

Ave

rage

rat

e pe

r us

er [

(b/s

/Hz/

user

)

4

6

8

10

12

14

16

18

20

20 30 40 50 60

Perfect CSIDistance-based CSI allocationUniform CSI allocationClustering of size 3Without TX cooperation



We have presented some insights into a newproblem that presents serious challenges, butalso opportunities for the future. This leads tonew intriguing open questions. First, IA algo-rithms with incomplete CSIT are based on aDoF-preserving criterion only (i.e., on the per-formance at asymptotically high SNR). Theimpact of the incomplete CSIT on the perfor-mance at finite SNR should then be investigat-ed to obtain practical solutions. Similarly,robust precoding schemes for the MIMO net-work with distributed CSI have so far consid-ered DoF only as a metric. By and large,precoding over network MIMO with distribut-ed CSI remains a challenging problem.Regarding the optimization of the CSIT allo-cation in a network MIMO channel, we believethat the distance-based CSIT allocation hasthe potential to impact the design of cellularsystems in practical settings. However, it hasbeen examined for simplified channel modelsonly and should be adapted to more realisticscenarios.

CONCLUSION

The uniform allocation of CSI resourcestoward all TXs does not lead to efficient use ofthe feedback and backhaul resources: TheCSIT dissemination should be designed to allo-cate to each TX the right CSI, in terms ofwhich elements to share and accuracy. Forboth network MIMO and IA, the CSIT sharingrequirements have been significantly loweredby designing appropriate CSIT disseminationpolicies, thus making TX cooperation morepractical and thereby paving the way for largeperformance improvement. Additionally, theproblem of robust precoding schemes takinginto account the inconsistency between theCSIs at the TXs has been tackled and shown tobe a key element.

ACKNOWLEDGMENTThis work has been performed in the frameworkof the European research project HARP, whichis partly funded by the European Union underits FP7 ICT Objective 1.1 — The Network of theFuture.

REFERENCES[1] D. Gesbert et al., “Multi-Cell MIMO Cooperative Net-

works: A New Look at Interference,” IEEE JSAC, vol. 28,no. 9, pp. 1380–1408, Dec. 2010.

[2] V. R. Cadambe and S. A. Jafar, “Interference Alignmentand Degrees of Freedom of the K-User InterferenceChannel,” IEEE Trans. Info. Theory, vol. 54, no. 8, Aug.2008, pp. 3425–41.

[3] M. K. Karakayali, 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.

[4] C. Yetis et al., “On feasibility of interference Alignmentin MIMO Interference Networks,” IEEE Trans. SignalProcess., vol. 58, no. 9, Sept. 2010, pp. 4771–82.

[5] K. Gomadam, V. R. Cadambe, and S. A. Jafar, “A Distribut-ed Numerical Approach to Interference Alignment andApplications to Wireless Interference Networks,” IEEE Trans.Info. Theory, vol. 57, no. 6, June 2011, pp. 3309–22.

[6] M. B. Shenouda and T. N. Davidson, “Robust Linear Pre-coding for Uncertain MISO Broadcast Channels,” Proc.IEEE Int’l Conf. Acoustics, Speech and Signal Process-ing, 2006.

[7] N. Jindal, “MIMO Broadcast Channels with Finite-RateFeedback,” IEEE Trans. Info. Theory, vol. 52, no. 11,

Nov. 2006, pp. 5045–60.[8] H. Bolcskei and I. J. Thukral, “Interference Alignment

with Limited Feedback,” Proc. IEEE Int’l. Symp. Info.Theory, 2009.

[9] O. E. Ayach and R. W. Heath, “Interference Alignmentwith Analog Channel State Feedback,” IEEE Trans. Wire-less Commun., vol. 11, no. 2, Feb. 2012, pp. 626–36.

[10] S. Sesia, I. Toufik, and M. Baker, LTE — The UMTSLong-Term Evolution: From Theory to Practice, 2nd ed.,Wiley, 2011.

[11] P. de Kerret and D. Gesbert, “Interference Alignmentwith Incomplete CSIT Sharing,” submitted to IEEETrans. Wireless Commun., Nov. 2012.

[12] —, “CSI Feedback Allocation in Multicell MIMO Chan-nels,” Proc. IEEE ICC, 2012.

[13] R. Etkin, D. Tse, and H. Wang, “Gaussian InterferenceChannel Capacity to Within One Bit,” IEEE Trans. Info.Theory, vol. 54, no. 12, Dec. 2008, pp. 5534–62.

[14] A. D. Wyner, “Shannon-Theoretic Approach to a Gaus-sian Cellular Multiple-Access Channel,” IEEE Trans. Info.Theory, vol. 40, no. 6, Nov. 1994, pp. 1713–27.

[15] P. de Kerret and D. Gesbert, “Degrees of Freedom ofthe Network MIMO Channel with Distributed CSI,” IEEETrans. Info. Theory, vol. 58, no. 11, Nov. 2012, pp.6806–24.

BIOGRAPHIESPAUL DE KERRET [S] ([email protected]) graduated fromEcole Nationale Superieure des Telecommunications de Bre-tagne, France, in 2009 and obtained a diploma degree inelectrical engineering from Munich University of Technology(TUM), Germany. He also earned a four-year degree in math-ematics at the Universite de Bretagne Occidentale, France, in2008. From January to September 2010 he was a researchassistant at the Institute for Theoretical Information Technol-ogy, RWTH Aachen University, Germany. Since October 2010he has been working toward a Ph.D. degree in the MobileCommunications Department at EURECOM, France.

DAVID GESBERT [F] ([email protected]) is a professor andhead of the Mobile Communications Department, EURE-COM, France. He received his Ph.D. in 1997 from TelecomParis Tech. From 1997 to 1999 he was with the Informa-tion Systems Laboratory, Stanford University. In 1999 hewas a founding engineer of Iospan Wireless Inc, San Jose,California, a startup company pioneering MIMO-OFDM(now Intel). He has guest edited six special issues, andpublished about 200 papers and several patents in thearea of wireless networks, among which three won inter-national awards. He co-authored the book Space TimeWireless Communications: From Parameter Estimation toMIMO Systems (Cambridge Press, 2006).

Figure 6. Average rate per user in terms of the SNR for a1(1) = 1, a2

(1) = 0, a2(1)

= 0.5, a2(2) = 0.7.

SNR (dB)0

0

Ave

rage

rat

e pe

r us

er (

b/s/

Hz)

10

12

8

6

4

2

35 4030252015105

ZF with perfect CSIActive-passive ZFConventional ZF

DE KERRET LAYOUT_Layout 1 2/22/13 11:24 AM Page 49




The work of C.-B. Chae(corresponding author)was in part supported bythe KCC (Korea Commu-nications Commission),Korea, under the R&Dprogram supervised by theKCA (Korea Communi-cations Agency) (KCA-2012-911-01-106) and theMinistry of KnowledgeEconomy under the ITConsilience Creative Pro-gram (NIPA-2012-H0201-12-1001).


INTRODUCTIONIn wireless cellular systems, a geographicalregion is partitioned into small cells. Withineach cell, a transmitter is dedicated to servingreceivers within that cell. Within this cell struc-ture, all voice and data traffic that the receiversrequest and generate will be transmitted fromand to the transmitter dedicated to serving them.One of the key performance indicators thatdecides the maximum rate at which thetransceivers communicate with each other is thereceived signal-to-interference-plus-noise ratio(SINR) — the higher the SINR, the higher themaximum rate. Transmissions from neighboringco-channel transmitters, however, result in out-of-cell interference, degrading the SINR, espe-cially to the receivers located at the cell edge. Tomitigate the effects of out-of-cell interference,multicell cooperative processing between thetransmitters has been actively investigated [1–5].

Multicell cooperative processing may also bean important technology for small cell networks.To deal with the increasing data traffic due to

smart phones, wireless cellular systems such asLong Term Evolution (LTE) networks will off-load portions of their traffic onto smaller cellssuch as picocells, femtocells, and relays, whichtogether with the currently existing macrocellsform a heterogeneous network. As these cellsget smaller, and more cells are packed into thesame amount of space, the receiver may seemore diverse and stronger out-of-cell interfer-ence; thus, a novel way of handling out-of-cellinterference is required. Therefore, in emergingwireless cellular standards such as Third Genera-tion Partnership (3GPP) LTE-Advanced, coordi-nated multipoint (CoMP) transmission andreception techniques, which use the core conceptof multicell cooperative processing, are beingactively discussed [6].

In multicell cooperative processing, the trans-mitters cooperate to jointly combat interferencefrom adjacent cells and fading to receivers. Thiscooperation requires exchanging some informa-tion such as downlink channel information andmay require multiple antennas. The barriers pre-venting such cooperation have become less cost-ly; the wired communication betweentransmitters (e.g., X2 interface in the 3GPPLTE-Advanced networks) enable the transmit-ters to exchange information quite reliably withonly marginal delay. The decrease in the price ofradio frequency (RF) electronic componentsenable the transmitters and receivers to use mul-tiple-input multiple-output (MIMO) technolo-gies. By combining beamforming/precodingtechniques enabled by MIMO with multicellcooperative processing, the post-processingSINR per receiver can be increased, thus furtherimproving the system throughput.

If the number of data streams designated to areceiver is smaller than the number of receiveantennas, the receiver can utilize the receiveantennas to combat the other-cell interference.Additional degrees of freedom at each receiverenable such operation, even with the use of lin-ear receive filters [2, 4, 5]. When the number ofreceive antennas is not greater than the numberof streams to decode, or transceiver coordina-tion is not feasible due to inaccurate feedback,

INSOO HWANG, UNIVERSITY OF TEXAS AT AUSTINCHAN-BYOUNG CHAE, YONSEI UNIVERSITY

JUNGWON LEE, SAMSUNG ELECTRONICS U.S. R&D CENTERROBERT W. HEATH, JR., UNIVERSITY OF TEXAS AT AUSTIN

ABSTRACTMulticell cooperation may play an important

role in achieving high performance in cellularsystems. Multicell cooperation with a singlereceive antenna at the mobile station has beenwidely investigated. Applying cooperation withmultiple receive antennas, allowed in severalemerging wireless standards, has met with somechallenges. This is primarily because the transmitprecoding/beamforming vector and receive pro-cessing have to be jointly optimized in multiplecells to combat out-of-cell interference. In thisarticle, we first discuss the role of the receiveantennas in a multicell environment, and thenreview recently proposed multicell cooperativealgorithms and receive antenna techniques fordifferent interference statistics. Finally, we high-light recent work on the fundamental limits ofcooperation and the possibility of overcomingsuch limits by using multiple receive antennas inmulticell cooperative networks.

MULTICELL COOPERATIVE SYSTEMS WITHMULTIPLE RECEIVE ANTENNAS

CHAE LAYOUT_Layout 1 2/21/13 4:09 PM Page 50


more advanced receiver algorithms to suppressthe other-cell interference can also be used.Advanced receiver algorithms such as the jointmaximum likelihood (ML) detector may givebetter error rate performance through decodingthe desired signal and interference signal jointlyrather than treating the interference as noise [7].

Another potential benefit of using multiplereceive antennas in multicell cooperative net-works is the measurement of network utility per-formance. The capacity region of the MIMOinterference channel with multiple receive anten-nas at each receiver—the system model of inter-est in this article—is still unknown for mostcases. The degrees of freedom of the MIMOinterference channel has recently been well stud-ied [8], and a better measure of the achievabilitythat takes interference signal strength intoaccount has been proposed in [9]. Despite itsimportance, the role of the receive antenna inmulticell cooperative coordinated processing hasyet to be fully understood.

To achieve this diverse set of objectives, inthis article, we first introduce basic CBF systems,each of which is mainly categorized by the typeof information sharing. Next, we investigate therole of multiple receive antennas in CBF (CBF)systems and introduce advanced receiver pro-cessing techniques to efficiently mitigate theother-cell interference. Then we introduce recenttheoretical findings on the fundamental limits ofcooperation and discuss their expansion to moregeneral multicell scenarios. Finally, we reviewmulticell CBF in emerging standards, mainlyfocusing on 3GPP LTE-Advanced systems, andthen follow with some conclusions.

COORDINATED BEAMFORMING SYSTEMSPrior work on multicell cooperative processingcan be categorized as either:• Coordinated beamforming (CBF)• Joint processing (JP)• Spatial sensing (SS)In this section, we present the key concepts ofthe three different multicell cooperative process-ing techniques.

CELLULAR NETWORKS: COORDINATEDBEAMFORMING WITH NO DATA SHARING

In CBF, the transmitters exchange only thedownlink channel information, and each receiverreceives a stream of data from only one trans-mitter. The receiver in CBF is illustrated as MSkin Fig. 1. In this case, the multicell MIMO chan-nel is the MIMO interference channel, and thegoal of CBF is to jointly design optimized trans-mit and receive beamformers to minimize theeffect of out-of-cell interference. A number ofCBF algorithms have been proposed in the liter-ature [4, 8], with interference alignment beingone of the technical enablers.

In CBF, data for a receiver is only availableat and transmitted from one cooperating basestation for a time-frequency resource. Userscheduling and beamforming decisions, however,are made with coordination among transmitterscorresponding to the cooperating set. The trans-mit beamforming and receive combining vectors

are generated to reduce the unnecessary inter-ference to other receivers scheduled within thecoordinated cell. The scheduler is targeted tomaximize the SINR or minimize the leakagepower to other receivers while maximizing thesignal power to the desired receiver. The latter isoften called Max-SLNR (signal-to-leakage-plus-noise ratio) scheduler and often leads to bettersum rates than the Max-SINR scheduler.Through scheduling, the desired set of receiversis selected so that the transmit and/or receivebeamforming vectors are chosen to reduce theinterference to/from other neighboring users,while increasing the serving cell’s signal strength.The cell edge receiver’s SINR can be improvedby CBF, resulting in a cell-edge throughputimprovement.

When the receivers are equipped with multi-ple receive antennas, transceiver algorithms atthe cooperating transmitters and receivers needto be jointly optimized to improve the sum rate.In a two-cell system with multiple receive anten-nas, non-iterative CBF algorithms called inter-ference-aware CBF (IA-CBF) were proposed in[4]. It was proven that the proposed algorithmsare optimal in terms of the degrees of freedomof the two-cell MIMO channel, where the mobilestations have two receive antennas. An asymp-totic expression of the achievable sum rate ofthe proposed system with respect to the numberof receive antennas was also derived. Interesting-ly, as the number of receive antennas increases,the proposed algorithm achieves not only fulldegrees of freedom but also the sum rate of thepoint-to-point upper bound.

Figure 1. Coordinated multicell processing system model (a three-cell sce-nario). Receiver MSk is in a coordinated beamforming scenario, receiver MSlis under a joint processing scenario, and receiver MSm is in a spatial sensingscenario.

Desired signal

Interference

Desired signal

Desired signal

BSk

BSlMSk

BSm

MSm

MSl

Cell k

Cell m

Inter-BS backhaul

Cell l



Generalization of IA-CBF algorithms to morethan two-cell scenarios is also presented in thesame paper [4]. The authors proposed a novel(physical) beam-switching mechanism that inten-tionally creates a beam conflict; a conflict that inthe conventional system was, as much as possi-ble, avoided. A beam sequence optimizationproblem is NP-hard; simple algorithms are dis-cussed by prioritizing the transmitter to decidethe beam index first based on its own channelinformation. Next the transmitter determines itsown beam index that creates a beam conflictwith the top priority transmitter. The philosophyused here is that instead of mitigating other-cellinterference and treating them as backgroundnoise, the transmitters can create a strong inter-ference link and use it to further minimize thebackground interference. This can be done byadditional receiver processing, so the importanceof multiple receive antennas in CoMP hasbecome more obvious.

CELLULAR NETWORKS: JOINT PROCESSING WITH

PERFECT DATA SHARING

In joint processing, each user receives a datastream from multiple transmitters. To enable it,downlink data streams as well as downlink chan-nel information must be exchanged via inter-transmitter cooperation. The joint processingtransforms the MIMO interference channel (inthe coordinated case) into the MIMO broadcastchannel in a case of ideal cooperation becausemultiple coordinated transmitters effectivelyform a distributed super base station. The capac-ity region of the MIMO broadcast channel hasbeen studied and analyzed [10]. The receiver injoint processing is illustrated as MSl in Fig. 1.

Similar to spatial division multiple access(SDMA) technologies in the MIMO broadcastchannel, in joint processing, transmit beamform-ing is performed to eliminate the other-cellinterference at the transmitter while thereceivers with single receive antenna performssimple decoding. Typically the transmitter com-putes a transmit precoding vector using theaggregated channel knowledge. Linear precodingis one of the simplest solutions, although zero-forcing channel inversion linear precoder suffersfrom excessive power penalty. Regularized chan-nel inversion precoder is hard to find the exactregularization factor. Nonlinear pre-processingsuch as vector perturbation (VP) or Tomlinson-Harashima precoding (THP) can be applied ontop of the linear precoder, showing a better sumrate performance. The role of multiple receiveantennas at the receiver side, however, has notbeen well addressed.

The network CBF (N-CBF) algorithm pro-posed in [11] approaches the joint processingsystem in different ways. Instead of multipletransmit antennas with a single receive antenna,the paper focuses on the multicell downlinkchannel where each transmitter is equipped withone transmit antenna and each receiver isequipped with multiple receive antennas. Linearand nonlinear CBF algorithms are proposed forboth clustered broadcast channels and full broad-cast channels that achieve near multicell capaci-

ty. The proposed algorithms minimize the largesteigenvalue of the inverse of the effectivematched channel matrix to minimize the trans-mit power consumption and simultaneously max-imize the effective channel gain. This can beachieved through the fact that the effectivechannel, R = H*H where H is the channelmatrix, goes to the identity matrix as the numberof receive antennas grows large; the more receiveantennas the system uses, the better throughputcan be obtained even with a single transmitantenna. The expansion of this problem withmultiple transmit antennas is nontrivial and maygive better sum rate performance.

It is known that joint processing provideshigher system throughput than CBF, but at thecost of downlink data stream exchange betweentransmitters. We note that the down side of jointprocessing is the requirement that the transmit-ters exchange downlink data streams. The priceto be paid for such additional cooperation, how-ever, may be decreasing. For instance, in a 3GPPLTE-Advanced network [6], the transmitters arealready connected over a wired backhaul line, sodata exchange between transmitters is becomingeasier. Recent demand for higher data rate sup-port and the use of multiple receive antennas atthe receiver changes the aspect of joint process-ing from classical soft handoff or macro diversityto joint transmit and receive beamformer/pre-coder design in CoMP.

COGNITIVE NETWORKSSo far, we have focused on homogeneous cellu-lar networks and addressed the role of thereceive antennas. In this section, we considercognitive networks and also investigate howmuch gain can be achieved due to multiplereceive antennas. As the standardization of het-erogeneous networks become more concrete,cognitive multicell network is also drawing lotsof attention. The use of femtocells in a heteroge-neous network may use the same licensed bandas the macro transmitter, but the femtocell maynot be allowed to create any noticeable interfer-ence to the legacy receivers. When the primarytransmitters are active, the secondary receiverwhich is attached to the secondary cell (e.g.,femtocells) suffers interference from the primarytransmitters. This type of interference has moti-vated new research into joint transceiver designsin cognitive multicell cooperative networks. Thereceiver MSm in Fig. 1 fits into this scenario.

In cognitive multicell cooperative networks,most prior work has focused on the design ofprecoding matrices to suppress interference tothe primary receivers. In conventional cognitivemulticell systems, the secondary receiver treatsinterference from the primary transmitter as anadditive noise. It is thus mainly an interferenceavoidance approach that minimizes the signalleakage to the primary receivers. The interfer-ence to the secondary receivers is considered asecondary problem, although the secondaryreceivers are also desired to communicate reli-ably. Again, multiple receive antennas at the sec-ondary receivers may help to achieve bothobjectives simultaneously:• To prevent interference to the primary

receivers

Recent demand forhigher data rate sup-port and the use ofmultiple receiveantennas at thereceiver changes theaspect of joint pro-cessing from classicalsoft-handoff ormacro diversity tojoint transmit and receive beam-former/precoderdesign in multicellcooperative processing.



• To remove the interference, due to primarytransmissions, at the secondary receiver

This can be done with the help of multiplereceive antennas at the secondary receiver.

The spatial sensing CBF (SS-CBF) algo-rithms proposed in [5] address this problem.With single-antenna primary terminals andtwo-antenna cognitive terminals, a l ineartransceiver design has been introduced under aglobal channel state information (CSI) assump-tion. Depending on the required informationof the secondary transceiver, the approach andthe role of multiple receive antennas may bedifferent. If the secondary transceiver has onlythe local CSI, which consists of the channelknowledge of primary transmitter-secondaryreceiver and secondary transceivers, the sec-ondary transmitter constructs the projected-channel singular value decomposition (P-SVD)to enable error-free communication to the sec-ondary receiver. If the channel between prima-ry transceivers is known to the secondarytransceiver in addition to the local CSI, thesecondary transceivers can adopt a joint trans-mitter-receiver optimization, which has provedto be optimal under the zero-interference con-straint at both the secondary transmitter andreceiver. If the local CSI and side information(precoding and postcoding information of theprimary transceivers) is known to the sec-ondary transceivers, iterative precoding anddecoding is proposed, outperforming the P-SVD approach of the local CSI.

In all three of the proposed SS-CBF algo-rithms, multiple antennas at the secondarytransceiver are used in two ways: to suppressinterference to the primary receivers and fromthe primary transmitters, as well as to maxi-mize the achievable rate of the secondary link.The conventional approach in cognitive multi-cell cooperative networks is focused solely onconstructing a transmitter by treating the inter-ference from the primary transmissions to thesecondary receiver as additive noise. The pro-posed SS-CBF jointly designs the secondarytransceivers, resulting in an achievable ratesimilar to that of equivalent point-to-pointMIMO channels. As the number of antennas atthe second transceivers increase, the perfor-mance of SS-CBF with global CSI converges tothat of the point-to-point MIMO channel,while the other two SS-CBF algorithms alsoshow the achievable rates very close to thatbound.

ADVANCED RECEIVER ALGORITHMSIf the number of receive antennas is insufficientor full coordination between transmitters andreceivers is not feasible, the receiver may utilizeits receive antennas in different ways. Anadvanced receiver technique may also berequired when a small group of adjacent inter-ferers are not fully synchronized. Thus, theother-cell interference becomes non-Gaussian[12]. Assuming the receiver can estimate boththe desired channel and interference channel,advanced interference mitigation algorithms canbe used. Depending on the statistics of interfer-ence, the appropriate receiver algorithms candiffer. Table 1 presents the three possible receiv-er algorithms under three types of out-of-cellinterference, which are minimum mean squareerror (MMSE), interference whitening withMMSE (IW-MMSE) ,or MMSE with interfer-ence rejection combining (MMSE-IRC), andjoint detection.

In this section, we introduce the advancedreceiver algorithms that can be used in multicellcooperative CBF systems. Note that the use ofadvanced receivers is also possible on top of theCBF algorithms introduced earlier. Typically,when the channel is slowly varying so that chan-nel knowledge at the transmitter can be accu-rate, CBF can be effectively utilized. In an agilechannel condition, the receiver may applyadvanced receiver techniques rather than relyingon full coordination. The advanced receive algo-rithm can thus be a supplementary to CBF algo-rithms, not a competitive one.

WHITE GAUSSIAN INTERFERENCE: MMSE FILTERIf the out-of-cell interference comes from manynon-dominant interferers, and the interferencefrom each of the interferers is uncorrelated, theaggregated interference seen by the receiver canbe assumed to be white Gaussian. If the varianceof the interference is known or precisely mea-sured by the receiver, the receiver can treat thewhite Gaussian interference as noise. Since theinterference-plus-noise covariance matrixbecomes an identity matrix with the same diago-nal entry sI2, it can be treated as additive whiteGaussian noise, and the MMSE receiver pro-vides the best link quality to the receiver in mul-ticell cooperative systems.

In a multicell coordinated system with whiteGaussian interference, the MMSE filter per-forms differently in different SINR levels. If the

Table 1. Comparisons of receiver processing. Interference channel, interference plus noise variance, andinterference covariance matrix are respectively denoted by HI, sI

2, and RI.

Interference Scenario Possible Receiver Algorithm Required Parameters

White Gaussian Many interferers withoutdominant one MMSE sI

2

ColoredGaussian

A few interferers withdominant one IW-MMSE or MMSE-IRC RI

Modulatedsymbol

Single or a few stronginterferers Joint detection HI, modulation order

In all three of theproposed SS-CBF

algorithms, multipleantennas at the sec-

ondary transceiverare used in two

ways: to suppressinterference to the

primary receiversand from the prima-

ry transmitters, aswell as to maximizethe achievable rate

of the secondarylink.



signal power is noticeably higher than the inter-ference-plus-noise power, the receiver mayignore the interference and perform as a zero-forcing decorrelator. The detector simply dividesthe received signal by the serving cell channelgain and then slices it to the nearest signal con-stellation point. If the interference-plus-noisepower is relatively higher than the signal power,the MMSE receiver matches with the servingcell channel. In the CBF system, the MMSE fil-ter maximizes the post-processing SINR whenthe interference is white Gaussian, regardless ofthe SINR levels.

COLORED GAUSSIAN INTERFERENCE:IW-MMSE OR MMSE-IRC

The noise-plus-interference term at the receivermay contain either intercell interference orintracell interference, or a mix of both plus addi-tive Gaussian noise. In general, such interfer-ence is not white, especially when the number ofinterferers is small and there are a few dominantinterferers. The noise-plus-interference term isthen spatially colored with its covariance matrixhaving non-zero off-diagonal terms. The coloredinterference-plus-noise term can cause a seriousproblem to the detector and decoder, which isgenerally designed assuming interference andspatially temporally white noise. Assuming theinterference-plus-noise covariance matrix is non-singular, a spatial whitening filter matrix needsto be applied on the receive signal such that thetransformed interference becomes white, with aslight sacrifice to the desired signal strength.

The interference whitening filter W trans-forms the interference-plus-noise term n into aspatially white Gaussian vector. The transformednoise after whitening, Wn, has the covariance ofan identity matrix, E{Wn(Wn)*} = WE{nn*}W*= INr. This yields WRW* = INr, and after takingthe eigen-decomposition on the effective chan-nel matrix R, R = USU*, the whitening matrixcan be obtained as W = S–1/2U*. After thewhitening filter is applied, an MMSE detectorcan be used to eliminate the white Gaussiannoise.

An alternative way to detect the symbol withcolored interference is to use interference rejec-tion combining (IRC). In the MMSE-IRC filter,the non-diagonal interference covariance matrixRI is added to the MMSE whitening matrix,enabling the receiver to reject the interference.Note that the MMSE-IRC receiver can be biasedor unbiased, depending on the use of a normal-ization factor. The biased MMSE-IRC receivermaximizes the received SINR, while the unbi-ased MMSE-IRC filter minimizes the MSE.With a proper scaling, however, the two MMSEfilters show identical coded block error rate per-formance [13]. Thus, we may observe that thetwo MMSE-IRC filters may result in similarthroughput performance.

The IW-MMSE/MMSE-IRC detector is oneof the simplest solutions to combat (slightly)non-Gaussian interference. By sacrificing thedesired channel gain a bit, the interference plusnoise can be approximated as Gaussian, so aconventional detector and decoder can be uti-lized. The IW-MMSE/MMSE-IRC receiver,

however, can perform well only with high modu-lation order, high code rate in a weak to medi-um interference region. As thesignal-to-interference level increases, the packeterror rate performance becomes floored. If theinterference is highly non-Gaussian, which typi-cally happens for out-of-cell interference withlow modulation order, IW-MMSE and MMSE-IRC are not desirable approaches.

INTERFERENCE WITH KNOWN MODULATIONORDER: JOINT DETECTION

In practical multicell cooperative systems, thetransmit symbols are modulated, so the receiverneeds to detect the transmitted symbol using aproper demodulator. The modulation order can bechanged over time to adapt the transmission rateclose to the instantaneous channel capacity. If themodulation order of both the desired signal andinterference signal are known to the receiver, thereceiver can attempt to decode the desired symbolsby jointly decoding the desired symbol and inter-ference symbol. In [7], a significant performancegain is shown to be obtained if the detectors explic-itly take into account the modulation formats ofthe desired and interference signals.

Unlike the conventional interference cancelerand successive interference canceler, which workwell only in noise-limited and interference-limit-ed regimes, respectively, the joint ML detectorcan cover variety of interference range. Specifi-cally, the joint ML detector incorporates knowl-edge of the interfering channel and the finitemodulation of the interference, turning interfer-ence-limited transmission system into a noise-limited one. The joint ML detector detects thedesired and interfering signals jointly, then dis-cards the detector output for the interference,which aids the detection of the desired signal. Asthe computational complexity can easily be dou-ble that of the interference-ignorant ML detec-tor, a less complex joint minimum distance (MD)detector can also be used. In a multicell cooper-ative network, however, the joint MD detector isslightly worse than the joint ML detector in alow SNR region.

The joint detection approach is beneficialwhen the channels are fast fading. In such a con-dition, channel knowledge at the transmitter iseasily outdated; hence, rigorous coordinatedprocessing is somewhat infeasible. The jointdetector, however, performs well in such an agilechannel condition as it turns an interference-lim-ited channel into a noise-limited channel. It isalso shown that the full diversity gains can beachieved with the joint detector even in the pres-ence of interference, so the transmit diversityscheme is advantageous. The joint detector isthus quite complementary to multicell coopera-tive CBF systems with multiple receive antennas.

MULTICELL COOPERATIVE PROCESSING INEMERGING STANDARDS

The next generation wireless cellular systems willuse coordinated multicell joint processing. In the3GPP LTE-Advanced systems, CoMP is consid-ered one of the strongest and best proven tools

As the signal-to-inter-ference level increas-es, the packet errorrate performancebecomes floored. Ifthe interference ishighly non-Gaussian,which typically hap-pens for the out-of-cell interference withlow modulationorder, the IW-MMSEand MMSE-IRC is nota desirable approach.



to improve cell edge throughput. It is also proventhat CoMP gives a higher gain than the tradi-tional approaches of cellular network planningsuch as frequency reuse, sectoring, and spreadspectrum. Despite the fact that the standardiza-tion of CoMP is still in progress and scheduledto be finalized in the first half of 2013, it ismeaningful to review the practical scenarios andlimitations. It is noteworthy that, based on theproposals submitted to the Release 12 (Rel-12)LTE-Advanced workshop, CoMP will be furtherevolved and enhanced to meet the requirementsfor so-called fifth generation (5G) systems. Inthis section, we briefly introduce the scenarios ofinterest and key challenges when implementedin real-world multicell cellular systems.

COMP ALGORITHMS AND SCENARIOSThe scenarios and supported CoMP algorithmsin Rel-11 3GPP LTE-Advanced are introducedin [6]. In the standard, both CBF and JP areconsidered as the main CoMP algorithms, andJP is further decomposed into joint transmissionand dynamic point selection.

In coordinated scheduling/CBF (CS/CB),which contains both coordinated scheduling andCBF, the desired data stream is transmitted onlyfrom a transmitter in the CoMP cooperating set.The resource block is assigned to the receiverwith CBF by scheduling of the serving cell aftercoordination among multiple coordinated cells.The transmit beamforming and receive combin-ing vectors are generated to reduce the unneces-sary interference to or from other receiver(s)scheduled within the coordinated cell. The bestserving set of users will be selected so that thetransceiver beamforming vectors are chosen toreduce the interference to/from other neighbor-ing receivers, while increasing the serving cellsignal strength. The other-cell interference canbe mitigated, and the cell edge receiver through-put can be improved.

In joint transmission (JT), a data stream issimultaneously transmitted from multiple trans-mitters to a receiver in a given time-frequencyresource. Unlike the CBF receiver, the receiverin joint processing tries to coherently combinethe symbols from multiple transmitters; if thedata streams from multiple transmitters are syn-chronized, the receiver can have a (coherent)combining gain, which additionally gives bothSNR and diversity gains. To enable coherentcombining at the receiver, a strict time-frequen-cy synchronization should be enforced, and thephase difference from different transmittersneeds to be fed back to the transmitter. In thissense, joint transmission may be more demand-ing and fragile than CS/CBF.

In dynamic point selection (DPS), data istransmitted from one transmitter within thecooperating transmitters in a time-frequencyresource, and the designated transmitter isdynamically selected. The other transmitter(s)may mute so the receiver(s) do not see out-of-cell interference. The transmitting point maychange in time depending on the feedback indi-cating the most desirable transmitting point bythe receiver. Dynamic point selection alsoincludes dynamic cell selection, which is a CoMPtransmission scheme that can be seen as a natu-

ral extension of existing scheduler implementa-tions. Unlike joint transmission, dynamic pointselection does not need the interpoint informa-tion. Dynamic point selection supports CoMPscenarios 3 and 4 (explained later), where low-power remote radio heads (RRHs) are deployedas a heterogeneous network. In situations wherethe mobile station penetrates across the RRHtransmission boundaries, dynamic point selectionshows similar gain to joint transmission.

In Table 2, the differences of the three CoMPalgorithms are presented. It is expected thatjoint transmission may give the highest through-put gain at the cost of additional feedback andhigher use of inter-base-station communication.Dynamic point selection is simpler to implementbut preferable only for the heterogeneous net-work scenarios in which the mobile station seesmore cell edges so that received SINRs are simi-lar to each other. Coordinated scheduling/CBFis probably the simplest and one of the mostrobust schemes with decent performance gain.

The four CoMP scenarios considered in theLTE-Advanced specifications and supportedCoMP algorithms are presented in Table 3. Sce-narios 1 and 2 are designed for classical homo-geneous deployments, while scenarios 3 and 4are for the heterogeneous deployment. Scenario3 targets typical picocell heterogeneous deploy-ments, and scenario 4 represents a distributedantenna system (DAS) where the points dis-tributed within the macrocell can be thought ofas RRHs. Scenario 4 differs from scenario 3 inthe fact that all points belong to the same cell.The DAS approach using RRHs (scenario makesit easier to handle the out-of-cell interferenceand to apply joint processing than the picocell(scenario 3), although the orthogonal cell-specif-ic pilot sequence design issue ought to beresolved. Note that in all scenarios, two receiveantennas are the default receive antenna config-uration; hence, aforementioned CoMP algo-rithms can be exploited.

PRACTICAL ISSUES IN RECEIVER PROCESSINGThe main practical issues of CoMP include feed-back channel design and downlink reference sig-nal design as they determine the accuracy of thechannel information at the transmitter andreceiver, respectively. This receiver processing isalso practically important because the receiverneeds to learn about the channel with limitedreference signals. Although the discussions arestill in the consensus building stage, it is mean-ingful to learn about the working agreementsmade so far in the 3GPP LTE-Advanced stan-dard.

Feedback Channel Design — To enable CoMP opera-tion, each receiver feeds back the necessaryinformation to the transmitter via reliable feed-back channel. The channel quality indicator(CQI) is the main feedback information. WithCQI, along with other feedback informationsuch as precoding matrix index and rank indica-tor, the transmitter may be able to learn aboutthe necessary downlink and interference chan-nel, and to schedule a transmission with anappropriately selected modulation and codingindex. As the transmitter solely relies on the

The main practicalissues of CoMP

include feedbackchannel design anddownlink reference

signal design as theydetermine the

accuracy of the channel information

at the transmitterand receiver, respectively.



information fed back from the receiver, thedesign of feedback channel is important inCoMP operation.

One of the main issues of the CoMP feed-back design is CQI generation because the defi-nitions of CQI vary in different CoMPalgorithms. A CQI without considering interfer-ence is the maximum eigenvalue of the effectivedownlink channel, while a CQI with interferenceis the maximum eigenvalue of the effective chan-nel multiplied by the MMSE-IRC whiteningmatrix. With the use of an ML detector, CQIcomputation can become more complicated. TheCQI thus depends on the receiver algorithm aswell as CoMP algorithms.

Increased hypotheses for CQI generation perCoMP scenarios and ambiguity on the CQI met-ric with a certain receiver algorithm may cause aserious problem to both the transmitter and thereceiver. One typical issue is what interferenceshould be taken into account in CQI computa-tion; the interference from cooperating transmit-ters may be eliminated after proper coordination,so it may not be considered while computing theCQI. Interference from non-interfering transmit-ters should be considered as interference in theCQI computation metric. However, as the coop-erating sets may change in time, the reportedCQI may not represent the true link quality, andthe transmitter may not detect this problem sincethe CQI computation metric is not transparent tothe transmitter. This makes the feedback design atthe receiver harder.

Downlink Reference Signal Design — The downlinkreference signal is used for the receiver to esti-mate the multicell channel. The reference signal

needs to be carefully designed to meet the per-formance requirement. If the density of the ref-erence signal in the time-frequency grid is toohigh, significant spectral efficiency loss occurs. Ifthe reference signals are too sparse, the multicellchannel measurement can become very coarse.In 3GPP LTE-Advanced systems, a cell-specificreference signal with adequate reference signaldensity is considered for CoMP operation. Insuch a cell-specific reference signal, the scram-bling pattern and cyclic shifting depends on thecell identification (ID) number.

In some CoMP scenarios, especially CoMPscenario 4 where the RRHs use the same cellIDs, the reference signal from multiple transmis-sion points can easily collide. Even for the CoMPscenarios with different cell IDs, as the numberof transmitting nodes gets higher, the referencesignal collision probability gets higher. To pro-vide additional orthogonality between referencesignals from different transmitters, the 3GPPLTE-Advanced systems introduce virtual cellIDs. The reference signal sequence is initiatedand scrambled, not in the cell-specific mannerusing the cell ID, but by the node-specific virtualID. By doing this, additional orthogonalityamong reference signals can be obtained.

The second issue is the possibility of differentgranularity of channel estimation intervals in theserving and interfering cells. In general, CoMPoperation can be done when the receiver doesnot move or moves very slowly, so the channelcoherence time is strictly longer than the round-trip delay. The issue is that the interferencechannel does not change significantly in time,and its estimation need not be as accurate asthat of the serving cell channel. In most cases,

Table 2. A summary of the characteristics for each CoMP algorithm.

CS/CB JT DPS

Performance gain Medium High Medium to high

Feedback overhead Medium High High

Requirements on frequency/time sync Low High High

Backhaul requirement Medium High High

Relative phase information at the transmitter Not required Required Not required

Relative amplitude information Required Required Not required

Table 3. CoMP scenarios and supported algorithms in 3GPP LTE-Advanced.

Scenarios and descriptions Equivalent system CoMP algorithms

1. Homogeneous network with intrasite coordination Single-cell coordination CS/CB, JT

2. Homogeneous network with high-power RRHs Multicell coordination CS/CB, JT

3. Heterogeneous with low-power RRHs (different cell IDs) Macro-pico voordination CS/CB, JT, DPS

4. Heterogeneous with low-power RRHs (same cell ID) Distributed sntennadystem (DAS) CS/CB, JT, DPS

Increased hypothesesfor CQI generationper CoMP scenariosand ambiguity onthe CQI metric withcertain receiver algorithm may causea serious problemboth to the transmitter and the receiver.



only the covariance of the interference can beutilized. The interference channel estimationperiod can be longer, so the receiver can save itspower and devote more resources to serving cellchannel estimation.

Receiver Architecture — In 3GPP LTE-AdvancedCoMP, the default receive antenna number is setto two or four. This enables the receiver to applyan interference mitigation process. Specifically,the receiver algorithms considered in [6] areMMSE (mandatory) or MMSE-IRC (recom-mended). For the MMSE-IRC receivers, bothbiased and unbiased receivers are consideredeven though the throughput performance differ-ence may be marginal regardless of the biased-ness. The performance of the receiver dependson the accuracy of the covariance matrix estima-tion and frequency selectivity of the covariancematrix. Joint detection has yet to be consideredin CoMP literature, but it can also be utilized ifthe modulation order of the interference signalis precisely known to the receiver.

SYSTEM PERFORMANCE EVALUATION RESULTSTo consider the practical issues discussed in

the previous sections, we present the system per-formance evaluation results. We mostly use thesimulation parameters presented in [6]. We usea 2 ¥ 2 antenna configuration with dual polar-ized antennas and 6 ms feedback delay underfull buffer traffic. For intercell interference mod-eling, we explicitly consider seven configurableintercell interference links to the mobile station,and the remaining links are modeled withRayleigh distribution. Among the scenarios inTable 3, we simulate a homogeneous network(scenario 1) and a heterogeneous network withlow power RRHs with difference cell IDs (sce-nario 3).

We first present a simulation result for rankstatistics with different transmit modulationorders. The rank statistic is an indicator that amobile station is capable of receiving more thanone stream using multiple receive antennas. Thebenefit of using additional antennas at thereceiver can be interpreted as the probability tochoose rank 2 under a good channel condition.The simulation results with the use of the IWreceiver is presented in Table 4. As shown in thetable, the mobile station tends to choose morethan one stream in a homogeneous scenario withhigh probability. The receiver processing withmultiple receive antennas can benefit in such agood channel. For a heterogeneous network,however, the UE sees more diverse sources ofinterference. It tends to reject the interferenceusing part of the receive antennas; hence, thechances that the mobile station is scheduled withmore than one stream are limited. The simula-tion results show the benefit of using multiplereceive antennas in both cases.

Simulation results with different transmitterand receiver techniques discussed in this articleare also presented in Table 5. We compare thecell average throughput and cell edge through-put (5 percent user throughput) using the base-line IW receiver with JD and CoMP. By doingso, we show how to handle the way interferenceaffects the system performance. In Table 5, we

observe that the use of an advanced receiveantenna and coordination technique improvesthe cell edge throughput the most. In a homoge-neous scenario, the cell edge throughput isimproved by 52 percent when CoMP with jointdetection is applied. In a heterogeneous sce-nario, CoMP is the best technique to improvecell edge user throughput, while CoMP with JDis best for average cell throughput. This demon-strates that both multicell cooperative processingand the advanced receive technique are goodmethods to overcome other-cell interference inpractical systems.

FUNDAMENTAL LIMITS OF COOPERATION

Multicell cooperation is promising as it promisesto convert the multicell network from interfer-ence-limited to noise-limited. Coordinationbetween transmitters and receivers or the use ofan advanced receive algorithm enables the trans-formation. The use of multiple receive antennasis helpful to alleviate other-cell interference. Thenatural question is whether such benefits are stillfruitful in a large multicell network. When thecooperating transmitters form a cluster, a recentarticle [14] derives fundamental limits of cooper-ation in a general uplink setting. Based on theiranalysis, the spectral efficiency gets saturated asthe transmit power increases in clustered large-scale multicell networks.

Surprisingly, in [14], it is shown that satura-tion of the spectral efficiency at sufficiently hightransmit powers is unavoidable in large net-works. A key insight used in that paper is thatnot all channels can be estimated due to timevariation; thus, there is residual interference thatlimits the performance. Since the interferencepower is proportional to the transmit power,increasing the transmit power will not solve theproblem. This results in the collapse of the mul-tiplexing gain in the higher SNR region. Thepaper also claims that the fundamental limita-tions cannot be overcome through any othertechnological advances. Based on the insightsfrom [14], the discrepancy in terms of through-put gains between theory [4, 5, 11] and practice[6] in multicell cooperative processing comesfrom this fundamental limit, mainly from thelatency of the backhaul link or limited feedbackof the practical cellular systems.

Reference [14] provides intuition on howlarge multicell networks are likely to perform. Itwould be interesting to study how multiplereceive antennas impact the theory derived in[14], for example, to show if they can extend theregion of multicell cooperation where capacityscales with SNR. The results in small-scale mul-

Table 4. Rank statistics: homogeneous (left column) and heterogeneous (righttwo columns) scenarios.

Macro UE(homogeneous)

Macro UE(heterogeneous)

Pico UE(heterogeneous)

Rank 1 54.90% 92.4% 84.2%

Rank 2 45.10% 7.6% 15.8%



ticell cooperative systems [4, 5, 11] show thatadditional receive antennas help achieve a betterspectral efficiency. An investigation of the trade-offs between numbers of antennas at a singletransmitter and coordination of many transmit-ters is an interesting topic for the purpose of anyfundamental preference for centralized (e.g.,massive MIMO) or distributed (e.g., distributedantenna systems) architectures. This may givesome insights on the evolution of the multicellcooperative network for the next generation cel-lular standards and the role of multiple receiveantennas therein.

CONCLUSIONS

This article introduces multicell cooperative sys-tems with multiple receive antennas andadvanced receiver techniques. In particular,coordinated beamforming, joint processing, andspatial sensing techniques are introduced,explaining their potential use of multiple receiveantennas. Advanced receiver algorithms, whichinclude MMSE, interference rejection combining(IRC), interference whitening, and joint detec-tion in different interference statistics were alsointroduced. CoMP, as envisioned by emergingwireless standards, is reviewed. Finally, the fun-damental limits of cooperation are discussed,and the potential role multiple receive antennasmay play is reviewed. Multiple receive antennasand receiver processing have become moreimportant in multicell cooperative networks,both theoretically and practically.

REFERENCES[1] D. Gesbert et al., “Multicell MIMO Cooperative Net-

works: A New Look at Interference,” IEEE JSAC, vol. 28,no. 9, Dec. 20, pp. 1380–4810.

[2] P. Marsch and G. P. Fettweis, Coordinated Multi-Pointin Mobile Communications: From Theory to Practice,Cambridge Univ. Press, 2011.

[3] H. Dahrouj and W. Yu, “Coordinated Beamforming forthe Multicell Multi-Antenna Wireless Systems,” IEEETrans. Wireless Commun., vol. 9, no. 5, May 2010, pp.1748–95.

[4] C.-B. Chae et al., “Interference-Aware Coordinated Beam-forming in a Multi-Cell System,” IEEE Trans. Wireless Com-mun., vol. 11, no. 10, Oct. 2012, pp. 3692–3703.

[5] K. Lee et al., “MIMO Transceiver Designs for SpatialSensing in Cognitive Radio Networks,” IEEE Trans. Wire-less Commun., vol. 10, no. 11, Nov. 2011, pp. 3570–76.

[6] 3GPP Long Term Evolution, “Coordinated Multi-PointOperation for LTE Physical Layer Aspects,” tech. rep.TR36.819, 2011.

[7] J. Lee, D. Toumpakaris, and W. Yu, “Interference Miti-gation via Joint Detection,” IEEE JSAC, vol. 29, no. 6,June 2011, pp. 1172–84.

[8] T. Gou and S. A. Jafar, “Degrees of Freedom of the KUser M ¥ N MIMO Interference Channel,” IEEE Trans.Info. Theory, vol. 56, no. 12, Dec. 2010, pp. 6040–57.

[9] R. H. Etkin, D. N. C. Tse, and H. Wang, “Gaussian Interfer-ence Channel Capacity to Within One Bit,” IEEE Trans. Info.Theory, vol. 54, no. 12, Dec. 2008, pp. 5534–62.

[10] H. Weingarten, Y. Steinberg, and S. Shamai, “TheCapacity Region of the Gaussian Multiple-Input Multi-ple-Output Broadcast Channel,” IEEE Trans. Info. Theo-ry, vol. 52, no. 9, Sept. 2006, pp. 3936–64.

[11] C.-B. Chae, S. Kim, and R. W. Heath, Jr., “NetworkCoordinated Beamforming for Cell-Boundary Users: Lin-ear and Non-Linear Approaches,” IEEE J. Select. Topicsin Sig. Proc., vol. 3, no. 6, Dec. 2009, pp. 1094–105.

[12] C. Seol and K. Cheun, “A Statistical Inter-Cell Interfer-ence Model for Downlink Cellular OFDMA NetworksUnder Log-Normal Shadowing and Multipath RayleighFading,” IEEE Trans. Commun., vol. 57, no. 10, Oct.2009, pp. 3069–77.

[13] Z. Bai et al., “On the Equivalence of MMSE and IRCReceiver in MU-MIMO Systems,” IEEE Commun. Lett.,vol. 15, no. 12, Dec. 2011, pp. 1288–90.

[14] A. Lozano, R. W. Heath, Jr., and J. G. Andrews, “Fun-damental Limits of Cooperation,” submitted for publi-cation, Mar. 2012; http://arxiv.org/abs/1204.0011.

BIOGRAPHIESINSOO HWANG ([email protected]) is a Ph. D candidatein the Electrical and Communications Engineering Depart-ment at the University of Texas, Austin (UTA)under thesupervision of Prof. Robert W. Heath Jr. He received hisM.S. in electrical engineering from the University ofSouthern California and his B.S. from POSTECH, Korea.He has more than five years of experience in wireless sys-tems research at Qualcomm Research (San Diego), Sam-sung Electronics US R&D Center (San Diego),Telecommunications R&D Center (Korea), and SamsungAdvanced Institute of Technology (Korea). He is currentlyworking on next generation cellular systems and stan-dards, focusing on investigation of the limits and chal-lenges in large-scale heterogeneous and smal l cel lnetworks.

CHAN-BYOUNG CHAE [SM’12] ([email protected]) is anassistant professor in the School of Integrated Technolo-gy, Yonsei University, Korea. Before joining Yonsei Uni-versity, he was with Bell Labs, Alcatel-Lucent, MurrayHill, New Jersey, as a member of technical staff, andHarvard University, Cambridge, Massachusetts, as apost-doctoral research fellow. He received his Ph.D.degree in electrical and communications engineeringfrom UTA. He was the recipient of the IEEE ComSoc APOutstanding Young Researcher Award in 2012 and IEEEVTS Dan. E. Noble Fellowship Award in 2008. He cur-rently serves as an Editor for IEEE Transactions on Wire-less Communications, and his research interest includesresource management in energy-efficient networks andnano networks.

JUNGWON LEE ([email protected]) is a seniordirector at Samsung Mobile Solutions Lab in San Diego,California. He received his Ph.D. degree in electrical engi-neering from Stanford University. He has co-authored morethan 50 papers as well as numerous standards contribu-tions for various standards bodies, and he holds over 60U.S. patents. He currently serves as an Editor for IEEE Com-munications Letters, and his main research interests lie inwireless and wireline communication theory.

ROBERT W. HEATH, JR. [F’11] ([email protected]) is aprofessor in the Electrical and Communications EngineeringDepartment at UTA and director of the Wireless Network-ing and Communications Group. He received his B.S. andM.S. in electrical engineering from the University of Vir-ginia, and his Ph.D. in electrical engineering from StanfordUniversity. He is the president and CEO of MIMO WirelessInc. and chief innovation officer at Kuma Signals LLC. He isthe recipient of the David and Doris Lybarger EndowedFaculty Fellowship in Engineering and is a registered Pro-fessional Engineer in Texas.

Table 5. Cell throughput: homogeneous (left two columns) and heterogeneous(right two columns) scenarios.

Cell avg.throughput

Cell edgethroughput

Cell avg.throughput

Cell edgethroughput

IW 1.13 (Mb/s) 0.016 9.62 0.054

JD 1.27 0.023 9.89 0.062

CoMP 1.14 0.017 9.58 0.067

CoMP with JD 1.34 0.024 10.09 0.066

CHAE LAYOUT_Layout 1 2/22/13 11:39 AM Page 58

http://arxiv.org/abs/1204.0011







INTRODUCTIONDuring the standardization process of ThirdGeneration Partnership Project (3GPP) LongTerm Evolution (LTE)-Advanced [1], a numberof techniques have been proposed to boost thepeak data rates, such as carrier aggregation anddownlink multiple-input multiple-output(MIMO) up to eight layers. By taking advan-tages of these techniques, peak rates up to 100Mb/s for high mobility and 1 Gb/s for low mobil-ity are envisaged. However, intercell interferenceis preventing ubiquitous user experience of suchhigh data rate (i.e., the data rate achieved by celledge users is a small fraction of the peak datarate). Nonuniform signal-to-interference-plus-noise ratio (SINR) distribution of a typicalhomogeneous network with 500 m inter-site dis-tance is depicted in Fig. 1a. Over 20 dB SINR isachieved in the cell center; however, it can onlyreach 5 dB or below in most areas due to inter-cell interference.

The heterogeneous network is an emergingnetwork deployment. In a heterogeneous net-

work the layer of a planned homogeneousmacro network is overlaid by layers of low-power nodes (LPNs), such as picocells andremote radio heads (RRHs). The macrocellprovides wide-area coverage, and LPNs pro-vide coverage for hotspots and offload trafficfrom the macrocell. The gain of offloadingtraffic is limited due to imbalanced transmis-sion power (i.e., LPN coverage is shrunk by thehigh-power macrocell). To extend the LPNcoverage, an offset to the received power mea-surement is introduced in cell selection, allow-ing more user equipment (UE) to be attachedto the LPN. The problem of this approach isthat it dramatically increases the interferencethe macrocell imposes on UE attached to theLPN by coverage expansion. Figures 1b and 1cshow SINR distribution of a heterogeneousnetwork deployment without and with coverageexpansion, respectively. Two LPNs are placedrandomly in the coverage of each macrocell. Itcan be seen that the expanded region suffersvery low SINR, which is depicted by deep colorin the figure.

From information theory i t i s a lreadyknown that intercell interference can be seenas an opportunity if cells process signals coop-eratively. Coordination techniques can beclassif ied as semi-static coordination anddynamic coordination according to therequirement of information sharing acrosscells. Semi-static coordination techniques,including fractional frequency reuse (FFR)[2] , soft frequency reuse (SFR) [3] , andenhanced intercell interference coordination(eICIC) [4], require little time-insensitiveinformation exchange across cells. The basicidea of FFR is to create partitions betweencell edge and cell center UE based on SINR.Resources of cell center UE in every cell arereused, , thus removing adjacent cell interfer-ence to cel l edge UE. In the case of SFR,edge UE devices are restricted to those allo-cated orthogonal resources by coordination ofneighboring cells as in FFR. Higher per-cell

SHAOHUI SUN, PEKING UNIVERSITY AND STATE KEY LABORATORY OF WIRELESS MOBILECOMMUNICATIONS (CATT)

QIUBIN GAO, YING PENG, YINGMIN WANG, STATE KEY LABORATORY OF WIRELESS MOBILECOMMUNICATIONS (CATT)

LINGYANG SONG, PEKING UNIVERSITY

ABSTRACTIntercell interference management has

become a critical issue for future cellular mobilesystems. Coordinated multipoint transmission/reception, or CoMP, is an effective way of man-aging intercell interference, and has been regard-ed as a key technology of LTE-Advanced. Thisarticle first provides an overview of downlinkCoMP techniques specified in 3GPP LTE Rel-11, which mainly focuses on transmissionschemes, channel state information reporting,interference measurement, and reference signaldesign. Then uplink CoMP is discussed in briefas most of the coordination gain can be achievedby implementation with little standardizationsupport. Evaluation results are provided to showthe efficiency of CoMP. The challenges as wellas possible solutions for future CoMP standard-ization are also discussed.

INTERFERENCE MANAGEMENT THROUGH COMP IN3GPP LTE-ADVANCED NETWORKS

GAO LAYOUT_Layout 1 2/21/13 4:10 PM Page 59


resource utilization is achieved by reusing allresources in every cell but with a lower trans-mit power given to center UE to mitigate theincreased interference. Both FFR and SFRare supported since LTE Release 8 (Rel-8).

In LTE Rel-10, the time-division multiplexing(TDM) eICIC principle was exploited to miti-gate interference from macrocell to picocells inheterogeneous network. The macrocell transmitsan almost blank subframe (ABS) or multimediabroadcast over single-frequency network(MBSFN) subframe with a certain periodicity,but the picocells transmit normal subframes allthe time. The macrocell is not allowed to sched-ule any UE in an ABS or MBSFN subframe.The UE connected to picocells in the expandedregion can be scheduled during transmission ofan ABS or MBSFN from the macrocell, whileUE close to picocells may be scheduled in anysubframe.

Semi-static coordination techniques devel-oped in LTE Rel-8/9/10 can provide benefits inmanaging intercell interference with relativelylittle backhaul overhead and low implementationcomplexity. However, the achievable benefit islimited as it is not able to follow the dynamicinterference condition caused by dynamicscheduling and beamforming. Moreover, semi-statically preserved resources result in low effi-ciency of resource utilization.

Dynamically coordinating the transmissionand reception of signals at multiple cells couldlead to further performance improvement.Exist ing dynamic multi-cel l coordinatingschemes include network MIMO [5], distribut-ed MIMO [6], multicell MIMO [7], and so on.These schemes are conceptually the same, pro-viding promising improvements of spectralefficiency and more uniform throughput distri-bution. They are unified into coordinated mul-tipoint (CoMP) transmission/reception in3GPP discussion as a key technology of LTE-Advanced. CoMP was initially brought forwardin the study item of LTE-Advanced [1] tomeet the requirements of IMT-Advanced [8]in 2008. However, there was no consensus onthe gain of CoMP under realistic conditions,and it was not included in LTE Rel-10.Instead, a study item on CoMP was proposedto continue to study the performance benefitsand specification impacts. After further study,performance gain was identified, and a workitem was approved to specify CoMP in LTERel-11.

In the following sections, standardized CoMPin 3GPP LTE-Advanced is described. Downlinkand uplink CoMP are introduced. Evaluationresults are presented to show performance gainsof CoMP. Conclusions are drawn, and futurechallenges are also discussed.

COMP IN 3GPP LTE-ADVANCED

Both downlink and uplink CoMP are standard-ized to support four identified CoMP scenar-ios. Downlink CoMP involves channel stateinformation (CSI) reporting from UE, interfer-ence measurement, and reference signal design,which require much standardization work.Uplink CoMP, on the other hand, can beimplemented with little standardization sup-port. To optimize performance of uplinkCoMP, uplink reference signal and uplinkpower control are enhanced in a CoMP-favor-able manner.

Figure 1. SINR distribution: a) homogeneous macro network; b) heteroge-neous network without coverage expansion; c) heterogeneous network withcoverage expansion.

(a)

-200-250

200

250

150

100

50

20dB

0

-50

-100

-150

-200

-250

15dB

10dB

5dB

0dB

-5dB

-150 -100 -50 0 50 100 150 200 250

(b)

-200-250

200

250

150

100

50

20dB

0

-50

-100

-150

-200

-250

15dB

10dB

5dB

0dB

-5dB

-150 -100 -50 0 50 100 150 200 250

(c)

-200-250

200

250

150

100

50

20dB

0

-50

-100

-150

-200

-250

15dB

10dB

5dB

0dB

-5dB

-150 -100 -50 0 50 100 150 200 250



SCENARIOS

Four scenarios were defined in the 3GPP LTE-Advanced CoMP study [9]. The defined scenar-ios are deployed in numerical evaluations, andstandardization is aimed at providing support forall four scenarios. It is worth noting that 3GPPstandardization focuses on scenarios that arewith backhaul of low latency and sufficientcapacity in LTE Rel-11. The four defined sce-narios are illustrated in Fig. 2.

Scenario 1: Homogeneous network with intra -site CoMP. The coordination takes placebetween three sectors of the same site in ahomogeneous macro network. This is the mostprevalent deployment scenario at the currentstage, and CoMP can be applied in this scenariowithout any additional investment in networkinfrastructure.

Scenario 2: Homogeneous network with high-transmission-power RRHs. RRHs with equaltransmission power to that of the eNodeB areemployed to construct a homogeneous network.The coordination area consists of RRHs con-trolled by the same eNodeB. More CoMP gaincould be achieved than with scenario 1 due to apossibly larger coordination area.

Scenario 3: Heterogeneous network with low-power RRHs within macrocell coverage. A num-ber of low-power RRHs connected to aneNodeB are deployed in macrocell coverage.Each RRH forms a small cell with physical cellidentity (PCI) independent from the macrocell.As explained in the introduction, interferenceimposed by a macrocell on UE associated withRRH severely degrades system performance.The interference problem could be resolved byapplying CoMP in this scenario.

Scenario 4: Heterogeneous network with low-power RRHs within macrocell coverage wherethe transmission/reception points created by theRRHs have the same PCI as the macrocell. Themain difference from scenario 3 is that no newcell is formed by RRH in this scenario, whichresults in distinct control channel design andmobility management. However, from aspect ofdata transmission, there is no difference betweenthe two scenarios since the network topology isthe same.

In scenario 1, the coordinated cells are collo-cated, and it does not suffer from the restrictionsof underlying backhaul. In scenarios 2–4 theRRHs are connected to the eNodeB via opticalfiber, and it is assumed that information isexchanged without latency and capacity con-straints. For the sake of simplicity, a macrocellor an RRH is called a transmission point (TP) indownlink and receiving point (RP) in uplink.

DOWNLINK COMPFor downlink CoMP techniques, specificationefforts mainly focus on transmission schemes,CSI reporting, interference measurement, refer-ence signal design, and control signaling [9].Transmission schemes are not mentioned explic-itly in the specification; however, all the specifi-cation items are to support CoMP transmissionschemes. CSI reporting facilitates link adapta-tion in the network, while interference measure-ment and reference signal are essential inderiving a CSI report. Besides common stan-dardization support for frequency-division duplex(FDD) and time-division duplex (TDD) systems,special attention was paid to TDD to exploitchannel reciprocity.

General Specification Support for FDD and TDDTransmission Schemes — In the framework of 3GPP,downlink CoMP transmission is categorized asdynamic point selection (DPS), dynamic pointblanking (DPB), joint transmission (JT), andcoordinated scheduling/beamforming (CS/CB).The four categories of transmission schemes areshown in Fig. 3.

With DPS, a TP is dynamically selectedaccording to the instantaneous channel conditionon per-subframe basis. The TP with the best linkquality is selected to exploit channel variationsopportunistically (i.e., selection diversity gain isachieved). In general, the gain of selection diver-sity in addition to the already achieved frequen-cy and spatial diversity is marginal. To getfurther gain, DPB can be used in conjunctionwith DPS. By DPB, the dominant interferer(s) toUE in the coordination area is identified andmuted dynamically (i.e., no signal is transmittedfrom the identified TP). By muting the dominantinterferer, SINR of the UE may be enhanced

Figure 2. CoMP scenarios in 3GPP LTE-Advanced.

Scenario 1

Scenario 3

Scenario 2

RRH eNodeBRRH

eNodeB

Different cells

Scenario 4

RRH eNodeBRRH

Same cell

eNodeBRRH

RRH

In the framework of3GPP, downlink

CoMP transmission iscategorized asdynamic point

selection (DPS),dynamic point

blanking (DPB), jointtransmission (JT),

and coordinatedscheduling/beam-forming (CS/CB).



significantly as a few dominant interferers maycontribute an absolute majority of the totalinterference. A TP is muted only when there isgain on the utility of the whole coordinationarea. Muting the dominant interferer causes per-formance loss of the particular TP. If perfor-mance improvements of those beneficiariesoutweigh the loss, the TP can be muted; other-wise, it should remain active to transmit data. Itseems unfair to UE connected to the muted TP;however, since scheduling is dynamically per-formed, those UE devices may also benefit frommuting other TPs in subsequent subframes. Onthe whole, it is beneficial to all UE if thescheduling algorithm is designed carefully.DPS/DPB is especially useful in CoMP scenarios3 and 4 as a muted macrocell benefits when mul-tiple RRHs reside in the macrocell. It is verylikely that the boosted RRH performance couldoutweigh the loss of the macrocell.

JT represents the case where two or moreTPs are transmitting signal to a single UEdevice on the same time-frequency resource.JT is categorized into coherent and non-coher-ent JT, depending on whether coherent com-bining of signals from multiple TPs is targeted.Coherent JT offers the best performanceamong all considered CoMP schemes, since itallows multiple TPs to completely nullify inter-ference toward co-scheduled UE. As evidencedin [10], coherent JT is especially sensitive toimperfections of CSI, so it would be prematureto include it in the standard considering thecurrent state of the art. Therefore, there is noexplicit support for coherent JT in LTE Rel-11; nevertheless, it can be implemented inTDD by channel reciprocity. Non-coherent JTis not able to nullify interference across multi-ple TPs due to the lack of relative phase infor-mation across TPs. The SINR at UE could beboosted by non-coherent JT, but with area-splitt ing-gain loss, as multiple TPs couldinstead schedule multiple separate UE devices.Only UE devices located on the coverage edgeof two TPs are able to get benefit from non-coherent JT, for the area-splitting-gain losscould be compensated by the boosted through-put in low SINR regions. The gain of non-coherent JT is more prominent in a networkwith low traffic load, as vacant resources ofneighboring TPs could be borrowed to performnon-coherent JT.

With CS/CB, the scheduling and beamform-ing across different TPs are aligned to reduceinterference. Coordinating scheduling decisions(i.e., which specific UE is selected for transmis-sion) can help alleviate strong interference con-ditions. In addition, a wise selection ofbeamforming weights may further contribute toreduce interference. Ideally, scheduling andbeam selection should be carried out jointly tooptimize performance. However, it is impracticalto make joint decisions on UE and beams acrossmultiple TPs. A simple and prevalent way toimplement CS/CB is an iterative scheduling pro-cedure. Each cell revisits the choice of UE andunderlying transmit beam(s) based on schedulingdecisions and beams decided by other cells inthe previous iteration. An updated schedulingdecision not only accounts for the utility of thescheduled UE but also for the utility of the vic-tim UE that has been tentatively scheduled byother cells in the previous iteration. The goal ofeach updating is to maximize a global utility(e.g., the total weighted throughput with thecoordination area). Only a few numbers of itera-tions are needed to reach a near-optimal solu-tion.

In practice, there is no clear boundarybetween transmission schemes; that is, combina-tion of transmission schemes is possible. Forexample, an RRH is selected for data transmis-sion for UE, and the scheduling and beams deci-sions are coordinated across multiple RRHswhile the macrocell is muted. In LTE Rel-11, allof the above transmission schemes can be sup-ported in a transparent manner from the specifi-cation perspective.

CSI Reporting — Channel state information at thetransmitter is necessary to take advantage ofmultiple antenna techniques. LTE Rel-8 employsimplicit feedback; that is, the reported CSI is aset of transmission parameters recommended byUE corresponding to a transmission hypothesis.The parameters are derived under the transmis-sion hypothesis. CSI includes the precodingmatrix indicator (PMI), rank indicator (RI), andchannel quality indicator (CQI). The RI is thetransmission rank or number of independentdata streams that can be supported by the chan-nel in spatial multiplexing transmission. ThePMI is an index to a codeword in a predefinedcodebook comprising precoding matrices. The

Figure 3. Illustration of CoMP transmission schemes: a) dynamic point selection; b) dynamic point blank-ing; c) joint transmission; d) CS/CB.

TP1 TP2

Dynamic selection

(a)

TP1 TP2

Coherent/non-coherentcombining

(c)

TP1 TP2

TP1 TP2

Dynamic blanking

(b)

(d)

With CS/CB, the scheduling andbeamforming acrossdifferent TPs arealigned to reduceinterference. Coordinating scheduling decisions(i.e., which specificUE is selected for transmission) can help alleviatestrong interferenceconditions.



indexed codeword can be used as transmit beam-forming weight for UE directly or used in deriv-ing transmit beamforming weight. The CQIreflects the quality of the target link, which is areference for modulation and coding scheme(MCS) selection.

A transmission hypothesis in CoMP compris-es two parts: signal hypothesis and interferencehypothesis. The signal hypothesis specifies TP(s)from which the packet data is assumed to betransmitted, and the interference hypothesisstands for interference suffered during theassumed data transmission. For DPS, CSI oftransmission hypotheses with signal hypothesisfrom each candidate TP is needed at the net-work to facilitate dynamic selection of TPs. Inaddition, for DPB, CSI of transmission hypothe-ses with the same signal hypothesis but differentinterference hypotheses (e.g., whether an inter-fering TP is muted or not) are necessary for thenetwork to make a decision on whether or not tomute the interfering TP.

To maximize flexibility in scheduling in thenetwork, CSI of all possible transmissionhypotheses should be reported by UE. However,the feedback overhead and UE implementationcomplexity is proportional to the number ofhypotheses. To limit feedback overhead whileretaining flexibility, Rel-11 UE can be config-ured to report CSI corresponding to one ormore transmission hypotheses. CSI correspond-ing to one transmission hypothesis is defined asa CSI process. A CSI process is determined bythe association of a signal hypothesis and aninterference hypothesis, where the signal hypoth-esis and interference hypothesis are measuredthrough CSI-RS and interference measurementresource (IMR), respectively. Note that there isno transmission hypothesis corresponding to JTand CS/CB in Rel-11 feedback, since that CSIcan be calculated from the CSI of the DPS/DPBtransmission hypothesis in the network withoutsignificant performance loss.

Assuming there are two TPs involved in CSIreporting, the possible CSI processes are illus-trated in Table 1. In each transmission hypoth-esis, the signal can be transmitted from eitherTP1 (TP1 on, TP2 off) or TP2 (TP2 on, TP1off), and the interference from the interferingTP can be either muted (off) or not (on). Thereare four possible CSI processes in the consid-ered example. For CSI1 in Table 1, the signalfrom TP1 is “on” and the signal from TP2 is“off,” which means that the signal is transmit-ted from TP1 only. Regarding the interferencehypothesis, the interference from TP1 is “off”and that from TP2 is “on,” which means thatthe assumed transmission experiences interfer-ence from TP2. The corresponding transmissionhypothesis is DPS from TP1. Unlike CSI1, inter-ference from TP2 is “off” for CSI2, and thusCSI2 can be used directly for DPB transmissionfrom TP1 while muting TP2. CSI with JT andCS/CB transmission hypotheses are not sup-ported in LTE Rel-11, and the required CSIcould be inferred from CSI2 and CSI4. It isworth noting that by the association of signalhypothesis and interference hypothesis, CSIprocesses besides the listed ones could also beconfigured.

Interference Measurement — To derive the CSI ofeach CSI process, appropriate interference cor-responding to the interference hypothesis shouldbe measured by UE. In addition, accuracy isanother requirement since CoMP performanceis sensitive to accuracy of interference measure-ment. To facilitate CSI reporting, LTE Rel-11introduces a UE-specific interference measure-ment resource (IMR) for UE to measure inter-ference. Each CSI process is linked with aconfigured IMR, and the interference hypothesisof the CSI process is derived on the IMR.

Each IMR occupies a subset of resource ele-ments (REs) that are muted intentionally on cer-tain TPs. No signal is transmitted from thoseTPs on IMR occupied REs. UE simply receivesthe signal on those REs to get interference esti-mation. The network should coordinate the sig-nal transmission on configured IMR for TPs inthe coordination area to ensure that the signalon the IMR matches the interference hypothesis.To elaborate more clearly, Fig. 4 depicts severalIMR configurations. On IMR1, both TP1 andTP2 are muted; the received signal on those REsrepresents interference from TPs other than TP1and TP2. Similarly, UE could estimate interfer-ence out of {TP1} and {TP2} on IMR2 andIMR3, respectively.

Reference Signal Design — The design philosophyof LTE-Advanced reference signal (RS) is toseparate RS for CSI reporting (CSI-RS) andRS for demodulation (DM-RS). For CSIreporting, RS sparse in time and frequencydomain is sufficient. The sparsity significantlyreduces system overhead. For demodulation,higher RS density is necessary to meet demodu-lation requirements. DM-RS only needs to betransmitted in the resources scheduled for datatransmission. It is transmitted in the same wayas data, and thus the transmitted data can bedemodulated by referring to DM-RS withoutknowing the actual CoMP transmission scheme.This gives the network sufficient flexibility inchoosing a CoMP transmission scheme to opti-mize performance.

A CSI-RS-resource refers to some predefinedtime/frequency/code resources used for transmis-sion of CSI-RS from a certain TP. To supportCoMP transmission, multiple CSI-RS-resourcesneed to be configured for UE. The configuredset of CSI-RS-resources is defined as a CoMP

Table 1. Example CSI process configuration for two TPs.

CSI process

Transmission hypothesis

Signal hypothesis Interference hypothesis

TP1 TP2 TP1 TP2

CSI1 on off off on

CSI2 on off off off

CSI3 off on on off

CSI4 off on off off



measurement set. Each CSI-RS-resource in aCoMP measurement set is used to track thechannel of one TP. A CSI process is associatedwith a CSI-RS-resource in the CoMP measure-ment set to derive the signal hypothesis of theCSI process.

Special Considerations for TDD — In a TDD system,channel reciprocity can be exploited to reducefeedback overhead and improve CSI reportingaccuracy. CSI obtained by channel reciprocity isfree of quantization and feedback compressionerrors, and thus more accurate than that report-ed by UE. CQI reporting from UE is still need-ed to deliver information about downlinkinterference, which is different from interferencemeasured on an uplink channel. In principle,reported CQI along with CSI by channelreciprocity enables all kinds of transmissionschemes. Even with CQI reporting, the feedbackoverhead is still greatly reduced by the absenceof PMI/RI reporting.

Although coherent JT is not supported explic-itly in Rel-11, it can be implemented in a TDDsystem to achieve significant performance gain.A prerequisite of coherent JT is that channelreciprocity holds over multiple TPs. Antennacalibration across multiple TPs should be carriedout to ensure overall channel reciprocity.

Channel-reciprocity-based schemes rely heav-ily on the accurate reception of an uplink sound-ing reference signal (SRS). Coordination of SRStransmission among multiple TPs is beneficial inimproving channel estimation accuracy. By suchcoordination, the SRS transmission in one cell

(TP) is orthogonal to other cells in the coordina-tion area, reducing interference imposed on SRStransmission.

The targets of SRS transmission may be locat-ed at separate positions, and the distances fromUE to those targets are different. The transmis-sion power should be set to guarantee that thefurthest TP can receive the SRS correctly.Enhanced transmission power is also beneficialto uplink CoMP in deriving uplink CSI at multi-ple RPs.

UPLINK COMPFor uplink CoMP, uplink CSI is available in thenetwork without resource-consuming transmis-sion on uplink for CSI reporting. Terminals needalmost no modification to support uplink CoMPtransmission. It is easier to implement CoMP inuplink than in downlink from the UE perspec-tive. Cooperation schemes are implementation-specific in the network, except that possiblebackhaul transmission needs to be standardizedin future. To optimize uplink CoMP, uplink ref-erence signal and uplink power control areenhanced in a CoMP-favorable manner.

Uplink CoMP includes joint reception (JR)of the received signal at multiple RPs and/orcoordinated scheduling (CS) decisions amongRPs to control interference and improve cover-age.

Joint reception (JR): The basic conceptbehind JR is to utilize antennas at different sites.The signals received by the RPs are jointly pro-cessed to produce the final output. Receiveddata needs to be transferred between the RPs

Figure 4. Interference measurement resource (IMR) configuration examples.

IMR1 IMR2 IMR3

PDSCH transmission

l = 6l = 0l = 0 l = 6l= 6l= 0l = 0 l= 6

Even-numbered slots Odd-numbered slots

A. TP1

Even-numbered slots Odd-numbered slots

B. TP2

Muted for interference measurement

Subc

arri

ers



for them to operate. The amount of exchangedinformation between cooperating RPs dependson to what extent the received signals are pre-processed before information exchange. Quan-tized baseband samples convey the mostcomplete information for joint processing. Itachieves the best performance, while the load onbackhaul is also the highest. Further processingat RPs can lower the load of backhaul, but withless CoMP gain. There is clearly a trade-offbetween CoMP gain and backhaul overhead.

Coordinated scheduling (CS): UE schedulingand precoding selection decisions are alignedamong RPs in a coordination area to minimizeinterference. A much reduced load is achievedin the backhaul network because only CSI andresource allocation information needs to beshared among cooperating RPs.

In a heterogeneous network, the transmis-sion power of various nodes may differ dramati-cally. For example, transmission power of amacro node may be 10~20 dB higher than thatof a lower-power RRH. As a result, the cellboundary between a macro node and an RRHnode (similar received signal level betweennodes) is actually very close to the RRH node.It is very likely that cell edge UE connected to amacro node is closer to an RRH. In this case,the RRH received signal strength of the UEuplink transmission is stronger than that of themacro node. That is, the uplink transmission ofsuch UE can target the RRH instead of themacro node. The transmission power can thusbe lowered to reduce interference and extendbattery life. The transmission of uplink anddownlink are decoupled in this scenario. Notethat this can be viewed as a special type of jointreception.

It is worth noting that uplink CoMP coopera-tion schemes are not limited to those mentionedin this article. Network is free to choose anyschemes that benefit uplink transmission.

EVALUATIONS

Extensive evaluation work has been carried outduring the study phase of CoMP in 3GPP [9]. Bythe collected results in [9], performance benefitsof both downlink and uplink CoMP are identi-fied. In this section, evaluation results are givento verify the performance of downlink CoMPaccording to the latest progress in 3GPP. Thesimulation is implemented via Microsoft VisualC++ 2008.

EVALUATION RESULTS FOR SCENARIOS 1 AND 2Coherent JT is evaluated in scenarios 1 and 2.CSI is obtained by channel reciprocity in thenetwork. CQI for each cell is reported by UE.Each cell first performs single cell scheduling toselect at most two UE devices without consider-ing intercell coordination. The transmit beam-forming of each selected UE is then jointlycalculated to suppress interference to co-sched-uled UE in the coordination area. Figure 5ashows that coherent JT provides gain for cellcenter UE as well as cell edge UE because spa-tial reuse is allowed by multiple-user transmis-sion. If scheduling decisions are made jointlyover the coordination area, further gain could be

expected. Other detailed evaluation assumptionscan be found in [9].

EVALUATION RESULTS FOR SCENARIOS 3 AND 4The combined scheme of DPS/DPB and non-coherent JT is evaluated in scenarios 3 and 4.TPs within a CoMP coordination area are joint-ly scheduled, and each CoMP coordinationarea is scheduled independently. Within eachcoordination area, exhaustive search is utilizedto schedule the UE group and the correspond-ing transmission schemes (DPS, DPB, and non-coherent JT) that achieve the highest totalweighted estimated throughput. UE reports allof the four CSI processes (CSI1–CSI4) to thenetwork as shown in Table 1. For DPS andDPB, the reported CSI is used directly, whilefor non-coherent JT, the network figures outrequired CSI from reported CSI. Both intra-sector and intra-eNodeB cooperation are eval-uated. The coordination area includes onemacrocell and four RRHs within the macrocell(sector) in intra-sector cooperation. For intra-eNodeB cooperation, coordination is carriedout between three collocated macrocells (sec-tors) and 12 RRHs connected to those macro-cells (sectors). From a data transmissionperspective, scenarios 3 and 4 are identical;thus, only one set of results is given. As seen inFig. 5b, DPS/DPB and non-coherent JT pro-vide throughput gain mainly for UE with medi-an to low rates. Furthermore, the largercoordination area provides additional gain.Other detailed evaluation assumptions arealigned with [9].

Figure 5. Evaluation results of CoMP schemes: a) scenarios 1 and 2; b) sce-narios 3 and 4.

(a)

Average throughput gain

15.74%

18.11%

Cell edge throughput gain

18.10%

5.00%

0.00%

Scenario 1 Scenario 2

10.00%

15.00%

20.00%

(b)

Average throughput gain Cell edge throughput gain

28.47%

5.00%

0.00%

Intra-cell coordination Intra-eNodeB coordination

15.00%

10.00%

20.00%

25.00%

30.00%

14.42%

8.25% 8.81%

18.64%



CONCLUSIONS AND FUTURE WORKThis article has presented an overview of thestandardization work on CoMP in 3GPP LTE-Advanced. Due to the limited timeframe of LTERel-11 standardization, only basic functions ofCoMP are supported. The CoMP technique isfar from mature, and many problems remain tobe solved.

1) Backhaul-efficient CoMP. StandardizedCoMP in 3GPP assumes that an ideal backhaulconnection exists. However, the available back-haul types are diverse in practice, such as T1/E1and microwave. On these practical backhaultypes, assumption of infinite capacity and zerolatency are no longer valid. Smart and backhaul-efficient cooperation techniques are needed.

2) Efficient CSI feedback. CoMP perfor-mance gain, especially coherent JT, relies heavilyon the accuracy of CSI in the network. In practi-cal systems, the accuracy of CSI in the networkis limited by the capacity of uplink transmission.How to design efficient feedback over the limit-ed uplink to support aggressive CoMP schemessuch as coherent JT poses big challenges onresearch and standardization.

3) Multiple-site antenna calibration for TDD.As seen from evaluation results, coherent JTexploiting channel reciprocity provides signifi-cant gain for TDD. However, channel reciprocitydoes not hold strictly. To ensure channelreciprocity and hence maximize the capability ofcoherent JT in TDD, antennas across multipleTPs need to be jointly calibrated. Self-calibrationis difficult to implement, especially for scenarios2–4. Over-the-air (OTA) calibration seems nec-essary in these scenarios. With the constraint offeedback overhead, the content of informationreported by UE to network and the calibrationalgorithm need further study.

ACKNOWLEDGMENTThe authors acknowledge the support from theNational Science and Technology Major Projectof China under Grant 2011ZX03003-001-01 andthe National Basic Research Program of Chinaunder Grant 2012CB724103.

REFERENCES[1] 3GPP TR 36.814 v9.0.0, “Evolved Universal Terrestrial

Radio Access (E-UTRA) and Evolved Universal TerrestrialRadio Access Network (E-UTRAN); Further Advance-ments for E-UTRA Physical Layer Aspects (Release 9),”TSG RAN.

[2] G. Boudreau et al., “Interference Coordination and Can-cellation for 4G Networks,” IEEE Commun. Mag., vol.47, no. 4, Apr. 2009, pp. 74–81.

[3] J. Li, N. Shroff, and E. Chong, “A Reduced-Power Chan-nel Reuse Scheme for Wireless Packet Cellular Net-works,” IEEE/ACM Trans. Net., vol. 7, no. 6, Dec. 1999,pp. 818–32.

[4] 3GPP R2-106897, “Introduction of Enhanced ICIC,”3GPP TSG RAN WG2 meeting #72.

[5] H. Huang et al., “Increasing Downlink Cellular Throughputwith Limited Network MIMO Coordination,” IEEE Trans.Wireless Commun., vol. 8, no. 6, June 2009, pp. 2983–89.

[6] A. Saleh, A. Rustako, and R. Roman, “Distributed Antennasfor Indoor Radio Communication,” IEEE Trans. Commun.,vol. 35, no. 12, Dec. 1987, pp. 1245–51.

[7] D. Gesbert et al., “Multi-Cell MIMO Cooperative Net-works: A New Look at Interference,” IEEE JSAC, vol. 28,no. 9, Dec. 2010, pp. 1380–408.

[8] ITU-R Rep. M.2134, “Requirements Related to Technical Per-formance for IMT-Advanced Radio Interface(s),” 2008.

[9] 3GPP TR 36.819 v11.1.0, “Evolved Universal TerrestrialRadio Access (E-UTRA) and Evolved Universal TerrestrialRadio Access Network (E-UTRAN); Coordinated Multi-Point Operation for LTE Physical Layer Aspects (Release11),” TSG RAN.

[10] S. Annapureddy et al., “Coordinated Joint Transmis-sion in WWAN,” IEEE Commun. Theory Wksp., May2010.

BIOGRAPHIESSHAOHUI SUN ([email protected]) received a B.S. in autocontrol engineering and an M.S. in computer engineeringfrom Xidian University in 1994 and 1999, respectively, anda Ph.D. in communication and information systems fromXidian University in 2003. He has been deeply involved inthe development and standardization of LTE/LTE-Advancedsince 2005. His research area of interest includes multiple-antenna technology, heterogeneous wireless networks, andrelay.

QIUBIN GAO ([email protected]) received his B.S. andPh.D. in control science and engineering from TsinghuaUniversity, China. He is currently a senior research engi-neer at the Datang Wireless Mobile Innovation Centerof the China Academy of Telecommunication Technolo-gy (CATT). His current research interests include physi-ca l layer des ign for mobi le communicat ion,multiple-antenna technology, CoMP, and system perfor-mance evaluation. He is inventor/co-inventor of morethan 100 patents in wireless communications, andauthor/coauthor of a number of journal and conferencepapers.

YING PENG ([email protected]) received her M.Sc. and Ph.D.in electrical and electronic engineering from the Universityof Bristol, United Kingdom. She is presently a senior stan-dardization researcher in the Wireless Mobile InnovationCenter of CATT. Her research interests include 3GPP/ITUstandardization, LTE/LTE-A physical layer design, CoMP,heterogeneous networks, and cognitive radio. She receivedFirst Prize in “4G TD-LTE-Advanced” of the Science andTechnology Prize from CCSA in 2011.

YINGMIN WANG [M] ([email protected]) received hisPh.D. degree in signal and information systems from XidianUniversity, Xi’an, China. He joined R&D of TD-SCDMA sys-tems at the Datang Mobile Communications EquipmentCo., Ltd., Beijing, China, in 2000. He has served as chiefscientist since 2010 in the Datang Telecom Technology andIndustry Group. He is currently leading the Datang WirelessMobile Innovation Center. His current research interestsinclude digital communication signal processing and wire-less networks. He has applied for more than 150 patents inwireless communications and published a number ofpapers.

LINGYA N G SONG [SM] ( l [email protected])received his Ph.D. from the University of York, UnitedKingdom, in 2007, where he received the K. M. StottPrize for excellent research. He worked as a postdoctor-al research fellow at the University of Oslo, Norway,and Harvard University, Cambridge, Massachusetts,until rejoining Philips Research UK in March 2008. InMay 2009, he joined the School of Electronics Engineer-ing and Computer Science, Peking University, China, asa full professor. He is a co-inventor of a number ofpatents (standards contributions), and author or co-author of over 100 journal and conference papers. Hehas received best paper awards at three internationalconferences.

DPS/DPB and non-coherent JT provide throughputgain mainly for UEwith median to lowrates. Furthermore,the larger coordination areaprovides additionalgain.









INTRODUCTIONMultiple-input multiple-output (MIMO) trans-mission can greatly increase capacity when thesignal-to-interference-plus-noise ratio (SINR) ishigh. In many practical systems (e.g., full fre-quency reuse cellular systems), however, SINR islow, especially near the cell edge. In these sce-narios, increasing the number of antenna ele-ments might not yield significant performanceimprovement. For this reason, spectral efficiencyof cellular systems with MIMO is far below thepromised value.

Consider as a starting point a single-cellMIMO system. For ease of exposition, from hereon we assume that each mobile station (MS) hasa single antenna, while the base station (BS) hasmultiple antenna elements. In the downlink, theBS can now form a beam, such that the desiredsignal is concentrated at the location of an MS,while minimizing (interfering) energy to the MSsthat do not want this signal. We note that beam-forming is usually done based on the instanta-neous channel realizations, not just on theaverage directional characteristics of the chan-nel. The BS can form multiple beams simultane-ously, and thus provide multiple data streams to

the MSs at the same time. This is the principleof multi-user MIMO (MU-MIMO). However,MU-MIMO does not, by itself, alleviate theproblem of intercell interference (ICI). Withoutcoordination among different cells, the beamsformed by one BS might “point” toward an MSin a neighboring cell, and thus provide concen-trated interference energy to an MS of interest(Fig. 1a).

To avoid the ICI, multiple BSs can be con-nected via backhaul links so that they can collab-orate and thus act as a single huge MIMOsystem. This effectively eliminates ICI and fur-thermore enhances the desired signal (similar tomacro diversity in soft handover), thus ensuringhigh SINR even at the cell edge. This conceptwas (to our knowledge) first elaborated in [1]and has since then attracted broad interest in thecellular industry, because it can improve both cellaverage and cell edge throughput. Currently, it ismostly known as coordinated multipoint (CoMP)transmission, although BS cooperation and net-work MIMO [2] are often used synonymously. In2008, a study item was started for CoMP in ThirdGeneration Partnership Project (3GPP) LongTerm Evolution (LTE)-Advanced; it is also apart of Advanced WiMAX.We only considerinter-BS CoMP, where multiple geographicallyseparated BSs cooperate. While intra-BS CoMP(which exploits coordination between sectors cov-ered by the same BS) has many aspects in com-mon, its particular features are beyond the scopeof this article.

CoMP is generally divided into two cate-gories: CoMP joint processing (CoMP-JP) andCoMP coordinated beamforming (CoMP-CB),depending on what kinds of information areshared among BSs. An illustration of CoMP-JPand CoMP-CB is shown in Figs. 1b and 1c.

For CoMP-JP, both data and channel stateinformation (CSI) need to be shared via back-haul links between each BS and a central unit(CU). The CU could be a separate entity, possi-bly collocated with one of the BSs. Alternative-ly, each BS can serve as a CU — a conceptcalled decentralized cooperation in the litera-ture. This approach allows the cooperating BSsto behave like a single large multi-antenna BSwith distributed antenna elements. The dis-tributed array forms beams toward all the users

CHENYANG YANG, SHENGQIAN HAN, XUEYING HOU, BEIHANG UNIVERSITYANDREAS F. MOLISCH, UNIVERSITY OF SOUTHERN CALIFORNIA

ABSTRACTCoordinated multipoint, or CoMP, transmis-

sion has been recognized as a spectrally efficienttechnique for full frequency reuse cellular sys-tems, in which base stations cooperate to reduceor eliminate intercell interference. However,there are still many obstacles before it can beput into practical use. In this article, we first dis-cuss the features of CoMP systems and channelsthat are distinct from single-cell multi-antennasystems. We then give an overview of state-of-the-art approaches for coping with the factorsthat limit the potential of CoMP. A major issueis the acquisition of channel state information,which creates different challenges for TDD andFDD systems. Another set of challenges arisesfrom the limited capacity available on the back-haul connections between the cooperating basestations. Both the fundamentals of possible solu-tions and their relations to cellular standards arediscussed.

HOW DO WE DESIGN COMP TO ACHIEVE ITSPROMISED POTENTIAL?

MOLISCH LAYOUT_Layout 1 2/21/13 4:11 PM Page 67


in its coverage area (i.e., the cells covered by allthe cooperating BSs) simultaneously, employingall available BS antennas for each beam. Conse-quently, ICI is turned into desired signals. Onthe downside, this approach places highdemands on the backhaul links and requires sig-nal-level synchronization as well as data-levelsynchronization among BSs.

In contrast, CoMP-CB only needs the coordi-nated BSs to share CSI and the scheduling infor-mation. CoMP-CB, also called spatial ICIcoordination (ICIC), retains the concept of cells:

in each cell, the BS forms beams toward theusers in such a way that it not only increases thedesired signal strength towards the desired userin its own cell, but also reduces interferencetowards the users in the adjacent cells. In thisapproach, each BS needs only the data for theusers in its own cell, as well as the CSI andscheduling information of the adjacent cells.This significantly reduces (compared to CoMP-JP) the demands for backhaul links becausesharing CSI among BSs needs much lower capac-ity than sharing data [3]. However, CoMP-CBonly passively avoids ICI rather than proactivelyexploiting it.

While CoMP is useful — and challenging —both for the uplink and downlink of cellular sys-tems, we focus in this article on the downlink,due to the fact that limitations of downlink trans-mission speed are currently considered the moreimportant bottleneck of cellular communications.Since the number of antennas available at the(coordinated) large BS is much larger than thaton the (individual) MSs, CoMP-JP should beused in conjunction with MU-MIMO transmis-sion to fully exploit the abundant spatialresources provided by the BS cooperative system.

Since the above description pointed out greatsimilarities of downlink CoMP-JP to single-cellMU-MIMO, one might wonder why the mani-fold and well explored techniques for imple-menting MU-MIMO cannot be applied in astraightforward manner. As a matter of fact,there are three important obstacles to such asimplistic approach:• Single-cell MIMO differs in some subtle but

important aspects from a true CoMP-JP setup,thus requiring changes in the transmissionstrategies.

• The acquisition of CSI is more difficult inCoMP systems.

• There are restrictions on the sharing of infor-mation between the cooperating BSs due tothe limitations of the backhaul network.

In the following three sections, we deal withthose issues one by one.

SINGLE-CELL MIMO VS. COMP-JPCoMP-JP differs from single-cell MIMO even ifthe BSs are synchronized and connected withperfect backhaul, which comes from the dis-tributed BSs and practical limitations.

PBPCIt has been widely recognized that CoMP issubject to a per-BS power constraint (PBPC),while most of the optimization of beamform-ing, scheduling, and power allocation for sin-gle-cell MIMO is subject to a sum powerconstraint (SPC). Since power cannot beshared among BSs, directly applying the trans-mit strategies optimized under SPC will leadto optimistic results, especially for heteroge-neous networks including macro, micro, andpico BSs with very different transmit powers.In homogeneous networks where multiple BSshave the same transmit power, the transmis-sion schemes designed for maximizing the sumrate under PBPC perform close to those underthe SPC.

Figure 1. Illustration of non-CoMP, CoMP-JP and CoMP-CB. For the CoMPsystems, BS1 is the master BS of MS1, from which the average channel gain isstronger: a) non-CoMP: no coordination between BSs, and strong interfer-ence might be caused to adjacent cell edge users; b) CoMP-JP: BSs cooperateby jointly serving multiple users in their covered area; c) CoMP-CB: BSscooperate by avoiding interference to adjacent cell edge users.

• Local channel of MS1: α11h11• Cross channel of MS1: α12h12• Global channel of MS1: g1 = [α11h11 α12h12]

BS2

BS2

BS1

BS1

MS1

MS1

MS2

MS2

α22h22

α12h12α11h11

α21h21

Data of MS1M

S2

(a)

(c)

(b)

CU

Signal Signal

Strong inter-cellinterference

BS2BS1

Data of

Data of MS1

MS1

MS2Data of M

S2

CSICSI

CSI

CSI

BackhaulBackhaul

Backhaul

Backhaul



ASYNCHRONOUS INTERFERENCE

Since BSs are not collocated, interference from dif-ferent BSs received at each MS are asynchronous [4].This cannot be compensated by the timing-advancetechnique conventionally applied in cellular systems,because the degree of freedom of the timing advanceis used up by ensuring that the signals from the BSsarrive synchronously at a desired MS. For orthogonalfrequency-division multiple access (OFDMA) sys-tems, this problem can simply be solved by prolong-ing the cyclic prefix. Considering that cell size iscontinually reduced and CoMP is more desirable forhigh-density networks, even such a prolonged cyclicprefix does not need to be very long.

DYNAMIC CLUSTERINGDue to the prohibitive complexity and overhead,it is not possible to allow all BSs (which mightspan a whole city) to cooperate. Moreover, a

CoMP system with many BSs in a networkexhibits negligible performance gain comparedto one where only a few BSs are cooperating [5,Chapt. 7]. As a pragmatic trade-off, cooperativeclusters can be formed, within which severaladjacent BSs jointly transmit. Dynamic clusteringoutperforms fixed clustering, but it leads to adynamic overall number of transmit antennas.Considering the flexibility and scalability, chan-nel training and feedback mechanisms need tobe redesigned for such a cooperative network.

NON-I.I.D. GLOBAL CHANNELSThe global channel for each MS in a CoMP-JPsystem is a stacking of multiple single-cell chan-nel vectors, as shown in Fig. 1b. The distributedantennas yield a special non-independent andidentically distributed (i.i.d.) global channel,where the average channel energies of the linksbetween multiple BSs and each MS differ. As a

Figure 2. CSI acquisition procedures in a) TDD and b) FDD CoMP systems, where the specific steps from S1 to S5 are respectivelyshown in each procedure.

g1||g1||

Y1jBS

Y1jBS S1

MS

Y11BSS1j

BS S11BS

• Imperfect reciprocity • Per-BS self-calibration

• Global codebook for quantizing global CDI

• Per-cell codebook for quantizing per-cell CDI

• Unreciprocal global CSI

Baseband

Baseband

Baseband

... ...Y1j

BS Y11BSS1j

BS S11BS... ...

CSI acquisition procedure: CSI acquisition procedure:

(a)

S1

S1 S1

S1

uj

S1α11h11

α12h12

S1

: Calibrating system antennaS2

S2

S2

DLReconstructglobal CDI

by feedback

CDI feedback (indicesof codewords)

CSI

(in

dice

s of

code

wor

ds)

Quantizing g1 using:• Global codebook• Per-cell codebook

CU

CU

UL

BS1

BS1

BS1

° h11j =

∆= a11jh11j,U

∆= a11

h11j,U

° a111 = ... = a11j = ... = a11Nt

a11 INta12 INt

° g1 = g1,U A1

° Global CDI: g_

1 =

° Global codebook: cg = {cg(i)},i = 1,...,2Bg

cg(i) is 1 x 2Nt codeword Bg is number of bits for quantizing g

_1

[α11h11 α12h12]=

°A1 =

Y1MS

Y1MS S1j

BS

S1MS

c11j c11j

BS1

BS2

BS2

MS1

MS1

MS1

S2

: Sending training signals from MSsS3

S3

S3

S3

: Estimating channels at BSs: a^11 h^11,U, α^12 h^12,US4

S4

S4

S4

: Forwarding channel estimates to CUS5

S5

S5

: Reconstructing global channel at CU: g^1 = [α^11 h^11,U, α^12 h^12,U]

(b)

S1 : Sending training signals from multiple BSsS2 : Estimating global channel by each userS3 : Quantizing CSI (selecting codeword)S4 : Feeding back codebook indicesS5 : Reconstructing global channel at CU

(for per-cell codebooks)

u1

h11||h11||° Global CDI: h

_11 =

° Per-cell codebook:

1||g1||

C1p = {c1

p(i)},i = 1,...,2B1p,

c1p(i) is 1 x Nt codeword

B1p is number of bits for quantizing h

_11



result, the statistics of the global channel dependon the MS location (i.e., path loss and shadow-ing). An ongoing research goal is to investigatehow such non-i.i.d. channel statistics and thestacked channel structure can be exploited toreduce the overhead to gather CSI for downlinkmultiuser precoding and scheduling.

CHALLENGES RELATED TOCHANNEL INFORMATION ACQUISITION

CSI at the transmitter (CSIT) is essential for allkinds of CoMP transmission to obtain their fullbenefits. It is well known that the performancegain of MU-MIMO is largely dependent on theCSIT quality, and the same holds true for CoMP.To facilitate downlink spatial precoding andscheduling, the CU needs to gather CSIT fromall coordinated BSs to all MSs in their servingcells. In time-division duplexing (TDD) systems,

the CSI is estimated at each BS by uplink train-ing via exploiting channel reciprocity. In fre-quency-division duplexing (FDD) systems, sinceuplink and downlink operate in different fre-quency bands, the CSI is estimated at each MSby downlink training, and then is fed back to theMS’s master BS via uplink channels. The chan-nel acquisition procedures in TDD and FDDsystems are illustrated in Fig. 2.

In this section, we assume perfect backhaul,that is, the backhaul links connecting the CUand multiple BSs within each cluster have unlim-ited capacity and zero latency. With such anassumption, CoMP-JP could realize the fullpotential of BS cooperation transmission if theglobal channels of multiple MSs were perfectlyavailable at the CU. We focus on various issuesassociated with obtaining the global CSI in TDDand FDD systems.

TDD SYSTEMSImperfect Channel Reciprocity — TDD was expectedto be more desirable for CoMP because it canobtain CSIT by exploiting the reciprocitybetween uplink and downlink channels. Unfortu-nately, only uplink and downlink propagationchannels are reciprocal. The baseband equivalentchannel is no longer reciprocal due to the differ-ent characteristics of radio frequency (RF)chains used in reception and transmission inpractical systems.

Self-calibration is a popular antenna calibra-tion method in single cell systems. It adjusts allantennas at one BS to achieve the same RF ana-log gain as that of a reference antenna, andhence ensures a constant scalar ambiguitybetween the uplink and downlink channels forall antennas. As shown in Fig. 2a, the equivalentdownlink and uplink channels between BS1 andMS1 are related by h11 = a11h11,U, where a11 is acomplex ambiguity factor. Such a scalar ambigui-ty does not affect the performance of single-cellsingle-user systems. However, when self-calibra-tion is employed at each BS in CoMP systems, itwill lead to Nb ambiguity factors between theuplink and downlink channels at different coor-dinated BSs (i.e., a11 and a12 for the two BSsCoMP illustrated in Fig. 2a). Then the globalequivalent uplink channel g1,U and downlinkchannel g1 of MS1 is related by g1 = g1,UA1,where A1 is the diagonal ambiguity matrix asshown in Fig. 2a.

The multiple ambiguity factors in the globalequivalent channels lead to imperfect downlinkglobal CSIT even if the uplink channel estima-tion is perfect. Such an ambiguity is more detri-mental than channel estimation errors. Becauseit is a kind of multiplicative noise rather thanadditive noise, it will hinder co-phasing of coher-ent CoMP transmission. Under some circum-stances, sharing all data among BSs might not beadvantageous anymore, given such undesirablemultiplicative noise [6].

One possible way to avoid this dilemma isover-the-air calibration. It has been applied insingle-cell systems to achieve the same goal asself-calibration for one BS. In CoMP it can beemployed to calibrate the antennas among theBSs, where the channel is estimated simultane-ously through two different approaches:

Figure 3. The probability density function (PDF) of cos2 q; q is the anglebetween the channels of two MSs located in different places d1 and d2. Theglobal channels between two MSs in different cells (e.g., d1 = –100 m and d2= 100 m) are orthogonal in high probability since cos2 q Æ 0. By contrast,the channels between two MSs in the same cells (e.g., d1 = 100 m and d2 =100 m) have the same direction in high probability since cos2 q Æ 1. Whenboth MSs are located at the cell edge (d1 = d2 = 0), their channel directionsare distributed uniformly within (0, 2p).

cos2θ0.20

1

0

PDF

2

3

4

5

6

7

8

9

0.4

• cos2 θ =

• g1 = [α11h11 α12h12] g2 = [α21h21 α22h22]

|g1g2H|2

||g1||2||g2||2

d1 = -100, d2 = 100 d1 = -100, d2 = 100

d1 = -50, d2 = 100

d1 = 0, d2 = 0

d1 = -50, d2 = 100

(b)

(a)(d=250m)

(c)

0.6 0.8

-d d1

BS1 BS2MS2MS1

1

d0

α21h21α11h11 α 22h22

α12h12

d2



reciprocity of propagation channels, and mea-surement on the downlink and feedback; knowl-edge of those two channel estimates allowsmultiple ambiguity factors to be estimated. Theperformance of this method depends on theaccuracy of the channel estimation. To improvethe performance, the calibration needs to beobtained from the measurements of multipleuplink and downlink frames or multiple users.

In practice, the interference experienced atthe BSs and MSs are also not reciprocal. As aresult, the actual SINR to assist downlink modu-lation and coding selection (MCS) should beestimated at the MS, then fed back to the BSs.

Training Overhead — To estimate the global chan-nel for downlink cooperative transmission, theuplink training overhead is proportional to thenumber of MSs and transmit antennas at eachMS, which is roughly Nb times the overhead ofsingle-cell systems, where Nb is the number ofthe coordinated BSs. To ensure that the down-link spectral efficiency gain over non-CoMP isnot “eaten up” by the uplink training overhead,it is paramount to reduce the training overhead,especially for relatively fast fading channels.

The trade-off between the performance gainof CoMP-JP and the required overhead forchannel estimation is nontrivial. For a givencoherence time, the training length as well as thenumber of cooperative BSs can be optimizedsuch that they maximize the net throughput(excluding the uplink overhead). This trade-offdepends on the MS location and SNR as ana-lyzed in [7]. Under which circumstances the ulti-mate performance of CoMP-JP — withoptimized training overhead — will outperformnon-CoMP systems is still an open problem.

A simplified form of this parameter optimiza-tion is a switching between CoMP and non-CoMP transmission modes. For users thatrequire asymptotically zero training overhead(e.g., static users), CoMP-JP always outperformsnon-CoMP transmission. For users requiringhigh training overhead, the net data rate ofsome cell-center MSs under CoMP transmissionmay be even lower than non-CoMP. This sug-gests developing transmission mode selectioneither from a system perspective or independent-ly from each MS’s perspective.

We can also reduce the required CSI by dif-ferentiating what we use it for. For instance,channel direction information (CDI) is essentialfor MU-MIMO precoding, and the channelnorms and channel angles among MSs are essen-tial for multi-user scheduling. If using full CSIT(i.e., perfect instantaneous CSIT), the spatialscheduling needs enormous training overheadeven in single-cell systems. Fortunately, in CoMPsystems, the channel orthogonality between MSsis largely dependent on their locations, as shownin a numerical result in Fig. 3. Intuitively, theglobal channels of two cell-center users are moreorthogonal, since they are separated geographi-cally. This indicates that for MSs in differentcells their average channel gains can be exploit-ed for scheduling [8] and possibly even for pre-coding. Since average channel gains vary slowly,they can be obtained at longer intervals, so train-ing overhead can be reduced accordingly.

Non-Orthogonal Training — In CoMP systems, train-ing signals for all MSs in multiple cells should bemutually orthogonal to obtain optimal channelestimation. However, in systems complying withthe LTE standard, the training sequences ofMSs in the same cell are orthogonal, but thosefor the MSs in different cells are not orthogonal(and are not identical). Orthogonal training forthe MSs in different cells demands intercell sig-naling and protocols to coordinate the trainingsequences among cells. From the viewpoint ofsystem compatibility and complexity, the trainingsequences of MSs in different cells are preferredto be non-orthogonal.

In fact, due to the non-i.i.d. feature of CoMPchannels, orthogonal training for intercell globalCSI may not be necessary. When the globalchannel is estimated under a minimum meansquare error (MMSE) criterion, non-orthogonaltraining for the MSs in different cells leads toacceptable performance degradation for down-link precoding [9].

FDD SYSTEMSIn order to save uplink resources for channelfeedback, the feedback is done using a quantiza-tion codebook known at both the BS and MS.Such limited feedback techniques have beeninvestigated extensively in the context of single-cell MIMO.

In the following we consider the feedback ofCDI, which is essential for precoding. We do notaddress the feedback of channel quality indica-

Figure 4. Average quantization accuracy of the global CDI, E {cos2(–_h, h)},

v.s the overall number of bits for each user, Bsum, which is reproduced fromFig. 6 in [10]. The number of coordinated cells Nb = 2 or 3, each BS hasfour antennas. An optimal bit allocation among the per-cell CDIs and thephase ambiguity (PA) with independent codeword selection (ICS) proposedin [10] is compared with other two feedback schemes: (i) optimal bit alloca-tion with a low complexity joint codeword selection (JCS) method, where onlyper-cell CDIs are quantized and no bits are allocated for quantizing the PA;(ii) the overall bits are equally allocated for the per-cell CDI quantization andno bits are allocated for PA quantization, where ICS is used. “Without PA” inthe legend means without PA feedback.

Bsum (bit)108

0.1

0

1- A

vera

ge q

uant

izat

ion

erro

r

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

12 14 16 18 20 22 24

Bit alloc. with ICS proposed in [10]Bit alloc. without PA and with low complexity JCSEqual bit alloc. without PA and with ICS

Nb = 3

Nb = 2



tion (CQI) facilitating user scheduling and MCS,which largely depends on the employed beam-forming.

The CDI quantization is chosen from a code-book with unit norm vectors of size 2B, where Bis the number of feedback bits. Each MS quan-tizes its CDI,

_h, for example, by choosing the

closest codeword to the CDI as measured by theinner product (which reflects the quantizationaccuracy), Ô

_hH hÔ = cos(–

_h, h), where h is the

quantized CDI. Then each MS feeds back B bitsto indicate the index of this codeword in thecodebook. It has been shown that for MU-MIMO, quantization errors lead to severethroughput limits at high SNR levels. The feed-back overhead increases with SNR to ensure aconstant rate loss compared to the ideal case.

In CoMP systems, the dimension of the glob-al CDI may vary dynamically, and the statisticsof the CDI depends on each MS’s location. Thisimplies that every MS needs a unique codebook.This is unrealistic due to the prohibitive com-plexity of generating codebooks as well as select-ing the codewords. Considering networkscalability and compatibility, it is highly desirableto design a per-cell codebook-based feedbackstrategy, where the existing codebooks can bereused to quantize each single-cell channel asshown in Fig. 2b. Although such a structured

codebook is suboptimal, its performance can beenhanced by representing the global CDI in anoptimal manner.

Feedback Overhead — To increase the networkspectral efficiency, a fundamental question is:how much uplink overhead is required to achievethe downlink performance gain of CoMP-JPover single-cell MU-MIMO?

To reduce the feedback overhead, the non-i.i.d. feature of CoMP channels should beexploited. This can be realized by optimizing thesizes of the per-cell codebooks. In [10], a scalinglaw of the feedback overhead for each user inCoMP-JP systems was provided. The analysisshowed that the feedback overhead of CoMP-JPis about Nb times of that of single-cell MIMO,which depends on user location. When a userhas equal average per-cell channel gains, itsoverhead is largest. When a user is in the cellcenter, its overhead is less because only thechannels to one of the BSs is strong enough toneed a fine quantization (large size codebook).

Alternatively, we can switch the transmissionmodes between CoMP-JP and non-CoMP, ordesign spatial scheduling exploiting the channelstatistics, as in the TDD case. We can alsoemploy selective feedback [11], where the CSI ofweak links of each user are not fed back.

How to Represent the Global CDICodeword selection: Considering that a global

channel is a stacking of multiple per-cell chan-nels, each MS can select codewords either jointlyto minimize the quantization error of global CDIor independently to minimize the quantizationerror of each per-cell CDI.

Independent codeword selection leads tosevere performance degradation for global CDIquantization due to phase ambiguities amongper-cell CDIs, especially for cell edge MSs [12].Such a phase ambiguity does not affect the per-formance of a single-cell limited feedbackMIMO system. However, it introduces multi-plicative noise to the downlink global CDI anal-ogous to the imperfect channel reciprocity in aTDD CoMP system. The phase ambiguities canbe either compensated by feedback or mitigatedby joint codeword selection.

On the other hand, simply selecting multiplecodewords jointly does not necessarily lead tominimal quantization error of the global CDI. Inorder to achieve this, the joint codeword selec-tion should consider the structure of the globalCDI reconstruction, which affects the overallquantization accuracy [12]. In addition, to enjoythe optimality of the judiciously developed jointcodeword selection, its prohibitive complexityneeds to be reduced.

Codebook bit allocation: When we quantizethe global CDI with per-cell codebooks, the fea-tures of the CoMP channel give rise to newparameters that can be optimized. Since differ-ent per-cell CDIs have different contributions tothe global CDI, we can allocate bits to differentper-cell codebooks. The optimal bit allocationthat minimizes the average global CDI quantiza-tion error (i.e., 1 – E {cos2(–

_h, h)}) under an

overall constraint on the number of feedbackbits turns out to be similar to water filling: more

Figure 5. Cumulative distributed function (CDF) of the net throughput of usersaveraged over small-scale fading channels. We consider the urban macrocellscenario of 3GPP. The path loss factor is 3.76, and the cell radius R = 250m. The cell edge SNR (i.e., the received SNR of the user located at distance Rfrom the BS) is set as 5 dB, where the inter-cluster interference is regarded aswhite noise. Nb = 3, each BS has four antennas, and 10 users are randomlyplaced in each cell. The downlink training overhead includes both the com-mon and dedicated pilots. The overhead of CoMP-JP and non-CoMP sys-tems occupy 28.1% and 12.6% of the overall downlink resources,respectively, which are typical values in 3GPP. The systems where all users areserved by CoMP-JP or non-CoMP are also simulated, either with or withoutconsidering the overhead. The performance of a mode switching method witha distance threshold is shown. As addressed earlier, when the training over-head is considered, it is shown that with CoMP-JP the net throughputs of thecell center users become inferior to those with non-CoMP.

Average throughput of users (b/s/Hz)20

0.1

0

CD

F

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

4 6 8 10 12

Mode switching

CoMP-JPNon-CoMPCoMP-JP without overheadNon-CoMP without overhead



bits will be allocated to stronger channels [10].Moreover, since different users have very differ-ent SINRs, we can also allocate the total numberof bits among the users to maximize the weight-ed sum rate under an overall uplink feedbackconstraints in the cluster.

In Fig. 4, we provide simulation results forthe average global CDI quantization accuracy ofseveral feedback schemes. It shows the impact ofthe bit allocation among the per-cell CDIs andthe phase ambiguity as well as the impact of thecodeword selection.

Common Pilot Optimization — To facilitate channelfeedback and data detection at the MS, trainingsignals are transmitted for channel estimation,which consume downlink resources for datatransmission. We can further distinguishbetween:• “Common pilots” for channel feedback, which

are transmitted from a particular transmitantenna and allow the MS to estimate thechannel to this particular antenna

• “Dedicated pilots” for data detection, whichare transmitted for each data stream with thesame precoding settings as the actual userdata

For common pilots, the induced overhead growsin proportion to the overall number of cooperat-ing BSs, while for dedicated pilots, the overheadis in proportion to the overall number of datastreams intended for all MSs, and thus does notnecessarily increase when CoMP is used.

Analogous to the uplink pilot optimization ina TDD system, downlink pilot optimization thatmaximizes net throughput is also a nontrivialtrade-off between the performance gain achievedby CoMP and the reduced downlink spectralefficiency resulting from spending additionaltime-frequency resources on non-payload trans-missions.

IMPERFECT BACKHAUL

In existing cellular systems, the backhaul linksamong BSs are not perfect as assumed. On onehand, a limited-capacity backhaul does not allowBSs to share a large amount of data. On theother hand, the CSI shared among BSs may havequantization error and severe latency if the exist-ing X2 interface (interface for communicationbetween BSs) in LTE systems is used. The back-haul imperfection is even more severe in parts ofheterogeneous networks, such as femto-cells,which hinders the application of CoMP. Whilebackhaul links can be upgraded by high-speedoptical fiber without technical challenges, thiscreates very high costs to the operators andtherefore may not be realized in the near future.

The impact of limited-rate backhaul is largestfor uplink CoMP, since in that case (quantized)analog signals need to be conveyed on the back-haul. Although not as stringent as in the uplinkcase, downlink CoMP-JP might also have toreduce its throughput to accommodate thecapacity limitation of the backhaul links [5, Ch.12.2]. In that case, it cannot achieve its full per-formance potential.

For both TDD and FDD systems, outdatedchannel information can considerably decrease

the effectiveness of CoMP. The latency of theX2 interface (which arises from the IP-basedprotocols in the backhaul networks), which cre-ates delays in the distribution of the CSI to dif-ferent cooperating BSs, may reach 10 ms andmore. This is more significant than theturnaround time of a TDD system (or the feed-back latency of CSI in an FDD system), whichcreates delays between the measurement of CSIand its availability at one particular BS. Recentinvestigations have shown that CoMP-JP canbenefit more from channel prediction than non-CoMP due to the non-i.i.d. channel feature [13].If predicted channels instead of estimated chan-nels are used for precoding, the performancedegradation of CoMP-JP will be largely alleviat-ed.

In the following, we discuss several possibleapproaches to mitigate the effect.

SWITCHING BETWEENDIFFERENT TRANSMISSION MODES

CoMP-CB needs much less backhaul capacitythan CoMP-JP, since it only shares the CSIT,not the user data [3]. Considering that both ofthese schemes are able to eliminate interference,CoMP-CB may outperform CoMP-JP understringent backhaul capacity constraints [14].Consequently, mode switching between thesetwo CoMP transmission modes — adaptive tolocation and number of MSs — will provide bet-ter overall throughput.

Another natural way is to switch betweenCoMP-JP and non-CoMP. Since cell center usersexperience lower ICI, they will not benefit asmuch from CoMP as cell edge users. Intuitively,we can simply divide the users in each cell with athreshold based on their average channel gains.Since only the users to be served by CoMP needto share their data among the BSs, the backhaulload can be controlled by judiciously selecting thethreshold. In Fig. 5, we provide simulation resultsof mode switching, where the results for pureCoMP-JP and pure non-CoMP schemes areshown as a baseline, either with or without con-sidering the training overhead. In the consideredsimulation setting (as shown in the caption of thefigure), the distance threshold to divide the usersinto the CoMP-JP and non-CoMP users is 90 m,and 62 percent of the users prefer to be servedby CoMP. After mode switching, the backhaulload is reduced by 27 percent.

Except for these “hard” mode switchingmethods where all data of some cell edge usersare shared among the backhaul, partial coopera-tion among BSs is possible where partial data ofall users are shared. For example, “soft” trans-mit mode switching can be operated with ratesplitting, where the downlink data of each user issplit into common and private parts, and onlythe common data is shared among the cooperat-ing BSs [15].

INTERFERENCE COORDINATIONIf the backhaul capacity is too limited, such thatno data are able to be shared among BSs, we arefaced with an interference channel problem ininformation theoretic terminology. When onlyCSIT is shared, each BS serves multiple MSs in

For the commonpilots, the inducedoverhead grows in

proportion to theoverall number ofcooperating BSs,

while for dedicatedpilots, the overhead

is in proportion tothe overall number

of data streamsintended for all MSs,

and thus does notnecessarily increase

when CoMP is used.



its own cell and coordinates with other BSs. Insuch a scenario, PBPC is no longer a perfor-mance limiting factor, and inter-BS calibrationin TDD systems is unnecessary.

CoMP-CB is one of the popular ways to coor-dinate ICI. When full CSIT for MSs, both withinthe cell and in adjacent cells, is available at eachBS, various criteria can be used to design theprecoding, such as zero forcing (ZF), MMSE,and maximal signal-to-leakage-plus-noise ratio(SLNR). To implement centralized beamform-ing, scheduling, or power allocation, these chan-nels need to be forwarded to a CU via backhaullinks. The disadvantage of CoMP-CB is its limit-ed performance gain. When MU-MIMO precod-ing is applied, the maximal multiplexing gain ofCoMP-CB equals the number of transmit anten-nas at each BS, which is the same as single-cellMU-MIMO systems.

To reduce the impact of outdated CSITcaused by the X2 interface latency, statisticalCSIT (long-term averaged CSIT), such as theangular power spectrum of the MSs in othercells, or scheduling results can be shared amongBSs. It can also be combined with instantaneouslocal CSIT for ICI avoidance. However, this onlyperforms well when each per-cell channel ishighly spatially correlated. Other CoMP-CBschemes that take latency on the X2 interfaceinto account are described in [5, Sec. 5.3.2].

CONCLUSION

We have addressed fundamental limiting factorsthat prevent CoMP from achieving its full poten-tial, in particular the training or feedback over-head to gather channel information, and thebackhaul constraints. We have briefly summa-rized critical issues in channel acquisition andtypical ways to tackle the technical challenges inTDD as well as FDD systems. Considering thescenarios with imperfect backhaul, we have dis-cussed possible ways to coordinate interference.Although various innovative approaches havebeen devised in the literature, the downlinkspectral efficiency gain of CoMP is still obtainedat the price of high uplink and downlink over-head. To provide high spectral efficiency fromthe network viewpoint, many challenging issuesremain to be solved, especially for large-scalesystems, including cooperative clustering selec-tion, various ways of overhead reduction, trans-mit strategy optimization with imperfectbackhaul links, and decentralized and distributedrealization.

REFERENCES[1] A. F. Molisch, Internal Memo, AT&T Labs-Research,

2001; see also A. F. Molisch et al., “Method and Appa-ratus for Reducing Interference in Multiple-Input-Multi-ple-Output (MIMO) Systems,” U.S. Patent 7,912,014(filed 2002, granted 2011).

[2] M. K. Karakayali, 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] D. Samardzija and H. Huang, “Determining BackhaulBandwidth Requirements for Network MIMO,” EUSIP-CO, 2009.

[4] H. Zhang et al., “Asynchronous Interference Mitigationin Cooperative Base Station Systems,” IEEE Trans. Wire-less Commun., vol. 7, no. 1, Jan. 2008, pp. 155–65.

[5] P. Marsch and G. P. Fettweis, Coordinated Multi-Pointin Wireless Communications: From Theory to Practice,Cambridge Univ. Press.

[6] S. Han et al., “Coordinated Multi-Point Transmission withNon-Ideal Channel Reciprocity,” Proc. IEEE WCNC, 2011.

[7] J. Hoydis, M. Kobayashi, and M. Debbah, “Optimal ChannelTraining in Uplink Network MIMO Systems,” IEEE Trans.Signal Proc., vol. 59, no. 6, June 2011, pp. 2824–33.

[8] S. Han et al., “Channel Norm-based User Scheduling inMulticell Cooperative Transmission Systems,” Proc. IEEEGLOBECOM, 2009.

[9] X. Hou, C. Yang, and B. K. Lau, “Impact of Non-Orthog-onal Training on Terformance of Downlink Base StationCooperative Transmission,” IEEE Trans. Vehic. Tech.,vol. 60, no. 9, Nov. 2011, pp. 4633–39.

[10] F. Yuan and C. Yang, “Bit Allocation Between Per-CellCodebook and Phase Ambiguity Quantization for Limit-ed Feedback Coordinated Multipoint Transmission Sys-tems,” IEEE Trans. Commun., vol. 60, no. 9, Sep. 2012,pp. 2546–59.

[11] A. Papadogiannis et al., “Efficient Selective FeedbackDesign for Multicell Cooperative Networks,” IEEE Trans.Vehic. Tech., vol. 60, no. 1, Jan. 2011, pp. 196–205.

[12] D. Su, X. Hou and C. Yang, “Quantization based onPer-Cell Codebook in Cooperative Multi-Cell Systems,”Proc. IEEE WCNC, 2011.

[13] L. Su, C. Yang, and S. Han, “The Value of Channel Pre-diction in CoMP Systems with Large Backhaul Latency,”Proc. IEEE WCNC, 2012.

[14] N. Seifi et al., “Coordinated Single-Cell vs. Multi-CellTransmission with Limited-Capacity Backhaul,” Proc.IEEE ACSSC, Nov. 2010.

[15] R. Zakhour and D. Gesbert, “Optimized Data Sharing inMulticell MIMO with Finite Backhaul Capacity,” IEEE Trans.Signal Proc., vol. 59, no. 12, Dec. 2011, pp. 6102–11.

BIOGRAPHIESCHENYANG YANG [SM’08] is a full professor at Beihang Univer-sity, China. She was supported by the 1st Teaching andResearch Award Program for Outstanding Young Teachers ofHigher Education Institutions of the Chinese Ministry of Edu-cation during 1999–2004. She is an Associate Editor for IEEETransactions on Wireless Communications and was Chair ofthe Beijing chapter of IEEE Communications Society. Herrecent research interests include network MIMO, energy-effi-cient transmission, and interference management.

SHENGQIAN HAN [S’05, M’12] received his B.S. degree incommunication engineering and Ph.D. degree in signaland information processing from Beihang University in2004 and 2010, respectively. From 2010 to 2012 heheld a postdoctoral research position with the School ofElectronics and Information Engineering, Beihang Uni-versity. He is currently a lecturer ay the same school. Hisresearch interests are multiple antenna techniques,cooperative communication, and energy-efficient trans-mission in the areas of wireless communications andsignal processing.

XUEYING HOU received her B.S. degree from Beihang Univer-sity in 2007. She is currently working toward her Ph.D.degree in signal and information processing in the Schoolof Electronics and Information Engineering, Beihang Uni-versity. From September 2009 to June 2010 she was a vis-iting student with the School of Electrical Engineering,Royal Institute of Technology (KTH), Stockholm, Sweden.Her research interests include cooperative communication,channel estimation, and limited feedback techniques inwireless systems.

ANDREAS F. MOLISCH [S’89, M’95, SM’00, F’05] ([email protected]) is a professor of electrical engineering at the Universi-ty of Southern California. His research interests include themeasurement and modeling of mobile radio channels,ultra-wideband communications and localization, coopera-tive communications, multiple-input multiple-output sys-tems, and wireless systems for health care. He is a Fellowof the AAAS and IET, an IEEE Distinguished Lecturer, amember of the Austrian Academy of Sciences, as well asthe recipient of numerous awards.

Although variousinnovative approach-es have beendevised in the litera-ture, the downlinkspectral efficiencygain of CoMP is stillobtained at the priceof a high uplink anddownlink overhead.To provide high spec-tral efficiency fromthe network view-point, many chal-lenging issuesremain.





THE STRUCTURE OF A MULTICELLCOOPERATIVE NETWORK

The requirement for increased high-data-rate cov-erage has led to intensive research activities inacademia and industry aiming to extend limits ofcurrent communication networks with the con-straint of scarce bandwidth resource. Departingfrom the conventional cellular network operationparadigm, where interaction between cells was lim-ited mainly to handover procedures, multicell coop-eration (MCC) resembles a more proactiveapproach by utilizing and rethinking the opportuni-ties for networks with multiple user terminals(UTs) and base stations (BSs). The key idea inMCC is to exploit the resources of multiple dis-tributed BSs and UTs in order to more efficientlycombat fading and interference, which are the mainlimiting factors for current communication systems.

BS COOPERATIONBase station cooperation has been one of theactive research topics in the area of MCC forthe past decade and is currently being consid-

ered for inclusion in future releases of LongTerm Evolution — Advanced (LTE-A) in theform of cooperative multipoint (CoMP) technol-ogy [1]. In BS cooperation, the resources of dis-tributed BSs are utilized to mitigate or evenexploit intercell interference (ICI) through coor-dinated beamforming/coordinated scheduling(CB/CS) or joint processing and transmission(JPT), respectively. By tuning the transmittingBS’s radiation pattern and signaling, CB/CS aimsto strike a balance between maintaining optimalperformance of UTs and minimizing the level ofunintended interference caused to other mobiles.While CB/CS is primarily a downlink techniquewith a single BS performing the coordinatedtransmission, JPT involves multiple BSs sharing,cooperatively processing, transmitting, andreceiving uplink and downlink streams.

Promising performance gains from state-of-the-art BS cooperation techniques do not comefor free: there are certain requirements to enableany cooperative communication scheme, whichresult in increased coordination and processingcomplexity. In general, to enable cooperation itmay be necessary to:• Collect system information (e.g., channel state

information [CSI] or even raw signal samples)[2] from other nodes

• Process this information to make a decision onhow to act cooperatively

• Deliver this decision with individual instruc-tions to all involved partiesClearly, it is desirable to completely eliminate

some of the above steps, or at least minimizeassociated overhead. Relaxation of these require-ments in BS cooperation typically involves theselection of a subset of collaborating BSs [1],which in turn can be based on intelligent selec-tion of cooperating nodes (e.g., incentive-basedrecruitment) [3].

Given such a nontrivial task — to efficientlycoordinate multiple distributed transceivers — itbecomes crucially important to adequately modelthe environment where this communicationtakes place to ensure that the way we accountfor the key influencing parameters (e.g., interfer-

ANVAR TUKMANOV, ZHIGUO DING, AND SAID BOUSSAKTA, NEWCASTLE UNIVERSITYABBAS JAMALIPOUR, UNIVERSITY OF SYDNEY

ABSTRACTRecognized as an efficient technology allowing

enhanced high-data-rate coverage, multicell coop-eration builds on collaborative operation of dis-tributed network elements aiming to reduce theimpact of intercell interference, or even to exploitit as a source of additional information. However,performance gains arising from this collaborationcome at a price of increased complexity and coor-dination among involved network elements. Animmediate question is how to adequately model amulticell cooperative network, while striking thebalance between analytic complexity and preci-sion of design decisions. In this article we high-light two categories of multicell network models,Wyner models and random network models, andtheir impact on the design and analysis of MCCsystems. Recent examples of realistic yet tractablemulticell network models are discussed.

ON THE IMPACT OFNETWORK GEOMETRIC MODELS ON

MULTICELL COOPERATIVE COMMUNICATION SYSTEMS

TUKMANOV LAYOUT_Layout 1 2/21/13 4:12 PM Page 75


ence and fading) allows us to predict the actualsystem performance.

MULTICELL SYSTEM MODELSIn developing MCC techniques, researchers takedifferent approaches to modeling multicell net-works. Current multicell cooperative networkmodels generally are variations of the Wynermodel, where network elements have determin-istic locations, while the cellular network itselfhas a simple regular macroscopic structure [2].While it is well understood that this is a simplis-tic model, the insights obtained through thisapproach still provide valuable guidance for thepractical design of current MCC communicationsystems. However, the main drawback of theWyner model is that obtained performance met-rics, conclusions, and design guidelines are usu-ally applicable only to the specific studied systemsetup. In other words, the extent to which suchresults are valid for different setups is notknown. The impact of this issue is that large-scale simulations are required for hypothesistesting in systems designed using the Wynermodel.

This was the key motivating factor for intro-ducing an alternative approach to system model-ing, which would capture the random nature ofcommunication networks and the impact of net-work topologies on various performance metrics.The first efforts to capture such macroscopicproperties for both wired and wireless communi-cation networks appeared in the late 1990s [4].Specifically, in [4] the point process theory and

stochastic geometry were used to obtain networkperformance metrics that to a greater extentaccounted for the natural randomness of wire-less networks. Recently these methods foundapplications in the area of MCC [5–7], wherenetwork topology has been shown to have partic-ular importance (e.g., due to introduced propa-gation delay and propagation path loss).

The main contribution of this article is incomparing existing and emerging approaches tomodeling of MCC systems, and highlightingassociated advantages and limitations. Our aimis to attract attention to more recent advancesthat, in conjunction with expertise in traditionalmethods, may ultimately support the design offuture communication systems. We first reviewthe Wyner model, which is used in the majorityof published works on MCC, is described. Fol-lowing that, heoretical tools that allow a depar-ture from such a deterministic approach arebriefly introduced to facilitate the discussion onrandom network topologies for MCC. Remarksthen conclude the article.

WYNER-TYPE NETWORK TOPOLOGIES

Introduced in the early 1990s, a general Wynermodel (WM) consists of M identical, regularlyplaced cells with K fixed users per cell. All net-work nodes are asumed to be equipped with asingle antenna. In a linear or one-dimensional(1D) WM, the BSs are aligned along a line suchthat a cell i has two adjacent cells with BSsindexed i – 1 and i + 1, respectively. In a moreadvanced 2D setup, a planar WM is representedby a regular hexagonal structure, where the cen-tral cell is surrounded by six adjacent cells.While the latter seems to be the more realisticof these two setups, the majority of researchresults on achievable rates and outage probabili-ties of MCC techniques exist primarily for thelinear WM, mainly due to its analytic tractability[2]. Figure 1a depicts three BSs for the 1D setup,where each cell spans over a segment of length2R, and each BS is located at the center of itscorresponding cell. The position of the referenceuser terminal (UT) k is not explicitly used inperformance evaluation.

Recall that the reason for introduction ofmulticell models is to study coexistence of multi-ple BSs in terms of, for example, ICI. In theoriginal WM, the contribution of each cell to ICIat the UT k in downlink is assumed to be limitedto L = 2 adjacent cells, where parameter L iscalled interference span. In particular, the aver-age channel gain between cells i and j is charac-terized by a single, globally available, anddeterministic parameter ai,j Œ [0,1], which is thesame for all UTs in cell i. For the serving BSsuch homogeneous channel gain ai,i = 1. If thesechannel gains a are also symmetric, such a setupwith no randomness is referred to as GaussianWM, and allows system performance to be stud-ied through a single design parameter a. In con-trast, a fading WM in addition to the staticparameter a includes a random coefficient hi,k toaccount for the fading channel between BS i andUT k. In this way, while the network topologyremains static, channel uncertainty is introducedthrough the fading coefficient hi,k, which is inde-

Figure 1. Multicell network topologies in 1D: a) linear Wyner network modelwith fading; b) linear network model with fixed BSs and random UT locationand account for fading and path loss; c) linear network model with randomBSs and UT locations with fading and path loss.

R

R

... ...

0

BS i-1 BS i

(a)

UT k

α⋅hi-1,k hi,k α⋅hi+1,k

BS i+1

2R 4R 6R

... ...

0

BS i-1 BS i

(b)

UT k

l(ri-1,k)⋅hi-1,k l(ri+1,k)⋅hi+1,kl(ri,k)⋅hi,k

ri,k

BS i+1

2R 4R 6R

... ...

0

BS i-1 BS i

(c)

UT k

l(ri-1,k)⋅hi-1,k

l(ri+1,k)⋅hi+1,k

l(ri,k)⋅hi,k

ri,kri+1,kri-1,k

BS i+1

R 2R 4R 6R



pendent for each communicating/interfering pairof network elements. Introduction of fading intothe model allows the multipath effects on top ofthe static interference picture to be accountedfor. Other extensions and variations of the linearWM include non-symmetric channel gains,extended interference span L > 2, for perfor-mance results; the interested reader is referredto [2].

In the following, outage probability will beused as a performance metric, which corre-sponds to the probability of the event that thesignal-to-interference-plus-noise ratio (SINR) atthe receiver falls below some predefined thresh-old. Note that for the Gaussian WM with staticparameter a, SINR is a constant parameter andadmits no notion of outage probability; whereasfor versions of WM with randomness, and ofcourse for random models to be discussed fur-ther, outage probability is a tool to capture theeffect of randomness of SINR.

The key advantage of WM is its simplicity,which allows tractable models for various signalprocessing methods and communication proto-cols for MCC. Currently, with the aid of thismodel, capacity results are available for CB/CS,JPT, including the account for limited inter-BSbackhaul, limited feedback, relaying, and variousoptimization methods for MCC. However, thereremains a question on the extent to which WMcaptures performance characteristics for realisticnetwork geometries. This issue was first system-atically addressed in [8], where it was found thata WM can adequately represent systems withlarge amounts of averaging (e.g., when the num-ber of users is large). However, such results areonly valid for uplink, whereas if downlink com-munication is considered, the 1D WM wasreported to fail in capturing the impact of ran-dom user locations in terms of outage probabili-ty and average throughput. In order to betterdescribe the random nature of wireless networks,researchers have developed alternativeapproaches for modeling networks, which arediscussed in the following section.

A STOCHASTIC GEOMETRY APPROACH TORANDOM WIRELESS NETWORKS

To elaborate on the impact of random networkmodeling on the design of MCC systems, in thissection, we first provide an overview of instru-ments used to describe random multicell net-work topologies and to account for randomnessin node locations in wireless networks in general.

MOTIVATIONReal-world patterns of deployed BSs may havelittle in common with regular structures used inWMs due to, for example, inhomogeneous signalpropagation conditions, availability of BS deploy-ment sites, or regulatory requirements. To illus-trate the impact of different assumptions onnetwork performance, let us consider a simpleillustrative example where assumptions based onthe linear WM will be gradually relaxed towardrandom placement of both the BSs and UTs.

We are interested in the outage probability Pof a user in cell i due to interference from adja-

cent BSs i – 1 and i + 1. Consider the linearWM with fading in Fig. 1a as a baseline sce-nario. Specifically, fixed BSs are located in cen-ters of cells covering segments of 2R each, whileinterference from adjacent cells is characterizedby a symmetric design parameter a for bothcells. Recall from earlier that the design parame-ters a are homogeneous for all users in a givencell; therefore, the specific location of the stud-ied user in this scenario plays no role in perfor-mance analysis. Channels between UT k and BSsundergo independent Rayleigh-distributed fad-ing, and additive white Gaussian noise (AWGN)power is assumed to be equal and normalized at1.

In the second scenario in Fig. 1b, a gridmodel with randomly located reference UTs andstatic BSs is considered: the reference user loca-tion is now a uniformly distributed random vari-able, characterized by the distance ri ~ U(0,R) tothe serving BS. In addition to fading, channelgains between BSs and UTs now include bound-ed path loss function l(ri) = (1+ri

b)–1, where b isthe path loss exponent for the propagation envi-ronment.1 Distances and values of associatedpath loss functions between UT k and adjacentBSs can be explicitly found once ri is known,since BSs are located in cell centers.

In the third scenario random BSs and UTlocations are assumed: BSs are now uniformlydistributed on the same total coverage segmentof length 6R, such that cell dimensions become(dependent) random variables, and BSs are notnecessarily located in cell centers (Fig. 1c).Specifically, coverage segments for BSs i – 1, i,and i + 1 are now segments of random lengthsXi–1, Xi, and Xi+1, respectively, subject to thecondition Xi – 1 + Xi + Xi+1 = 6R. The distancefrom the reference UT k to its home BS i is uni-formly distributed conditioned on the size of thehome cell, that is, ri ~ U(0, XiÔbi), where bi is thelocation of BS i.

Outage performance for the above three sce-narios is illustrated in Fig. 2. Specifically, 106

network realizations were simulated to capturethe impact of network topologies on outage. Thepower of interfering BSs was set to 1, and thedesign parameter a for the WM has been chosenas an average of the channel gains from either ofthe adjacent BSs to a uniformly placed UT as in[5]. Simulation results show that although theslopes for outage probability curves match for allthree scenarios, there is a significant differencein performance of all three models caused by theassumptions and factors involved in networkmodeling. Specifically, the WM provides a 10 dBoptimistic picture compared to the case whenthe UT’s location becomes random and path lossis accounted for explicitly. In addition, outageperformance for the scenario with random BSs isapproximately 1.5 dB worse compared to therandom UTs and fixed BSs scenario.

While a more realistic system model is desir-able, in simulation or analysis of large-scale sys-tems it can be impractical to explicitly accountfor individual effects of multiple factors; but onecan attempt to capture significant properties andthe impact of network topology via certain prob-abilistic law. Clearly, such a probabilistic modelwould also address only a limited number of

1 Different path loss mod-els can be used dependingon the application. Forexample, the mentionedbounded model is onemethod to avoid ambigui-ties of a less complicatedunbounded model l(ri) =ri

–b, which are associatedwith the increase in thereceived power for ri < 1and with a singularity at ri= 0. Another model l(ri)= C(ri/rref)–b is used inapplications, where anestimate C of the signalpower is available for cer-tain reference distance rref< ri in the antenna far-field.

While a more realis-tic system model isdesirable, in simula-

tion or analysis oflarge-scale systems itcan be impractical toexplicitly account forindividual effects ofmultiple factors; butone can attempt tocapture significantproperties and theimpact of network

topology via certainprobabilistic law.



aspects of the real-world process; however, net-work topology and space are factors that arebelieved to be key for the design of future gen-eration networks. Fortunately, these types ofinteractions can be described using point processtheory and stochastic geometry, where networkelements can be modeled as point patterns [4,9].

BASICS OF POINT PROCESSESThe core element in moving from the static WMto a more general setup is the point process. Apoint process (PP) F in d-dimensional space Rd

is a pattern of points distributed in R d withrespect to some probabilistic law, where point xŒ F can represent a location of a BS or a user.Here d is the number of space dimensions, so d= 1 corresponds to a PP on line and d = 2 cor-responds to a PP on a Euclidean plane. Suchprobabilistic law for any PP can be mathemati-cally described using finite-dimensional (fi-di)distributions of the form

Pr(F(B1) = n1, …, F(Bk) = nk), (1)

where the notation F(Bi) = ni states that thereare exactly n points of some PP F in a subset Bof space Rd. These fi-di distributions, in turn,allow distributions of random distances betweenpoints of the PP to be obtained, which are ofparticular importance for MCC processing, sincethey are associated with variable signal arrivaltimes and statistical internode channel character-ization.

Different types of PPs exist, capturing variouslevels of interactions between member points.Examples of such interactions could be minimalinternode distance, node clustering, and repellingor attraction conditions, which in practice couldtranslate into minimal inter-BS distance orgrouping of users around business and shoppingcenters. The ultimate desired result of applying

PP theory to modeling MCC topologies is thesame as for Wyner-type models: obtaining anaccurate prediction of the performance of somecommunication technique for a typical user.2

And, as with a WM, PP-based models requirecertain assumptions to be made for analyticaltractability.

POINT PROCESS MODELS FOR MCC AND THEIRLIMITATIONS

The simplest and most commonly used PPs formodeling nodes in wireless networks are thePoisson point process (PPP) and the binomialPPs (BPP). The reason for such common appli-cation of PPPs is that they allow for analyticallytractable description of statistical properties ofnetworks, where nodes are independently dis-tributed (e.g., ad hoc networks) [4]. Specifically,the counts of points of the PPP in any numberof disjoint subsets Bi are independent Poissondistributed random variables (see [10] for distri-bution expressions). In general, the level of cap-tured correlation between points is the mainlimiting factor for any PP-based model becauseof associated analytical complexity. In the caseof relatively simple PPPs, points are assumed tobe absolutely uncorrelated (i.e., the number n1of points of a PPP in a subset B1 conveys noinformation about the number n2 of points inanother disjoint subset B2. While this is a clearlyunrealistic assumption for MCC due to depen-dence in BS placements, it works well for ad hocnetworks [9], and even such simple PPs can stillbe used to obtain performance bounds (ratherthan precise results) for such metrics as mutualinformation, outage probability, coverage andlevel of interference in MCC systems.

Point process theory is gradually movingtoward capturing general types of interactionsbetween points of a PP [10]. One useful exampleis the BPP, where the numbers of points in dis-joint subsets Bi become correlated through thetotal number of points

so that any ni is now binomial-distributed.Accounting for this type of correlation adds pre-cision to the network model; however, this alsoincreases the complexity of performance analysis[11]. High-precision results for stronger pointcorrelations (e.g., minimal internode distanceand clustering) are possible through moresophisticated PPs, which still use PPPs and BPPsas building blocks.

For instance, hard-core PP (HCPP) modelspoint patterns where elements lie no closer thana certain distance h [10, p. 162]. HCPP has beenrecently applied to modeling of spatial correla-tion among active transmitters in networks withcarrier-sense multiple access [12], and can be ofparticular importance for modeling MCC net-works. The presence of a BS at a certain loca-tion in practice implies that neighboring BSsmust be geographically separated by some dis-tance. This real-world inter-BS distance dependson such factors as the type of terrain and popu-lation density; hence, the HCPP provides a sim-

N nii

= ∑

Figure 2. Outage probability as a function of transmit SNR in the presence ofinterference from adjacent cells for WM and random UT, and random BSmodels.

Transmit SNR50

10-3

10-4

Out

age

prob

abili

ty

10-2

10-1

100

10 15 20 25 30 35 40

Static BSsRandom BSsWyner

2 In the literature onstochastic geometry andrandom networks, theterm “typical user” refersto a “randomly chosenuser,” which itself is for-mally defined using Palmtheory [4, 10]. To see theneed for a formal defini-tion, consider, for exam-ple, a possible ambiguitywhen a realization of a PPcontains no points, sothere is nothing fromwhich to choose.



plified version of BS placement correlation thatallows avoiding highly unlikely BS locations.Recall that pure PPP-based models lack anymechanism to capture such dependence, or to pre-vent impractical placement of BS masts in the net-work model. Consider models of cellular networksin Fig. 3 and Fig. 4 for example. In the PPP-basedmodel in Fig. 3 some BSs are placed very closely,such as three BSs at approximately (4, –6). In con-trast, Fig. 4 illustrates the case when BSs arerestricted to lie closer than h = 1, which leads to amore realistic network realization. It is importantto stress that the fixed minimal distance h is still asimplification of the real-world correlation, and inactual deployed networks h may depend on vari-ous parameters. To extend the flexibility in model-ing of BS placement correlations further, soft-corePP (SCPP) [10, p. 164] can be used to account forinhomogeneous terrain properties through vari-able distance h.

In overall, PP-based models naturally pro-duce results for a typical user’s performance,which is one of the benchmarks for practicalcommunication system designs [4]. The key tasksinvolved with PP-based network modeling are to:• Build a PP model to capture interactions or

correlations between network elements• Calculate desired metrics based on this model

Therefore, the main limitation for PP-basedapproaches is in analytical complexity stemmingfrom accounting for correlations between net-work elements, such as inter-BS distance. This isin contrast to Wyner-type models, where aninherent drawback is in large-scale simulations,necessary to generalize and verify resultsobtained for a static setup. Table 1 compareskey features of PP-based random models andWyner-type models.

RANDOM NETWORK TOPOLOGIES FORMULTICELL COOPERATION

Different to the case with a WM for fixed setups,there is no generally accepted network modelaccounting for random network topology forMCC. Rather, there is a common set of toolsthat can be applied to modeling certain networkproperties. In the following, we discuss recentexamples of application of such instruments tomulticell scenarios.

FIXED BSS AND RANDOM USER LOCATIONSJoint processing of uplink transmission from uni-formly distributed users by two fixed BSs wasconsidered in [7] for 1D topology. Specifically,two BSs were fixed at certain positions on a lineto form two identical cells with the same radiusR. Each BS was equipped with N antennas, andcould share both CSI and signal data throughunlimited backhaul links in order to jointlydecode user signals with another BS. There wereK uniformly distributed users in each cell.Although this setup resembles the WM (Fig. 1a),the key difference is that the performance resultsare averaged over random user locations usingthe density function of user distribution.

The main advantage of such an approachover WMs with static locations is that it hasbeen demonstrated to provide insights into the

performance of a typical user, avoiding computa-tionally intensive simulations. Specifically, withthe results for such typical user performance,one can obtain optimal placement of the BSs interms of maximizing mutual information throughvarying system parameters in asymptotic expres-sions for the mutual information.

DISTRIBUTED ANTENNA SYSTEMWITH RANDOMNESS

Although the account for random user distribu-tion has allowed for efficient design and opti-mization in the linear setup with two cooperatingcells in [7], more advanced models are requiredto describe planar scenarios. One of the firstattempts to address the performance of a ran-dom multicell planar network was presented in[5], where the impact of placement of multipledistributed antenna systems (DAS) on the inter-ference experienced by a user was addressed.

A cluster with multiple single antenna ele-ments distributed randomly around a BS wasconsidered in [5] as a building block, so the mul-ticell network consisted of multiple such DASclusters. Antenna elements (AEs) in each clusterwere modeled as a BPP with N AEs per cell,while users were represented as points of ahomogeneous PPP with density l. For such asetup, outage probability performance of a typi-cal user in a single cell was reported to be closeto the case with fixed placement of the AEs,which indicates that even random placement ofAEs can deliver the same performance gains.The multicell scenario, however, was studied in[5] with the assumption that AEs and user posi-

Figure 3. Completely randomly distributed BSs and UTs on a plane. Each UTis associated with the nearest BS, forming Voronoi tessellations. Blue linesrepresent cell borders, red points are randomly placed BSs, and small blackpoints are the UTs. Illustration is zoomed from 20 ¥ 20 scale.

x

BSs = 402 λbs = 1; UTs = 40536 λut = 100

-8-10

-8

-10

z

-6

-4

-2

0

2

4

6

8

10

-6 -4 -2 0 2 4 6 8 10



tions are fixed, for which case it has been demon-strated that selecting a single AE for transmis-sion is preferable for a multicell environmentover blanket transmission, due to reduced ICI ofthe former approach. It is important to stressthat in [5] network topology exhibits randomnessonly for derivation of outage performance for asingle cell.

RANDOM BS AND USER LOCATIONSGreater flexibility in multicell topology was con-sidered recently in [6]. Specifically, interferencecoordination between multiple randomly dis-tributed BSs serving multiple randomly distribut-

ed users on a plane was studied. Locations ofBSs and UTs were modeled as independenthomogeneous PPPs, and BSs were clustered intocooperating groups with respect to an overlaidregular hexagonal lattice. Each user was associ-ated with the nearest BS, such that each cell inthe network has irregular shape and size,depending on proximity of other BSs. Figure 3illustrates the setup.

The presented cellular network topology is dif-ferent from regular hexagonal structures not onlyvisually, but also through distribution of distancesto interfering BSs. For example, the distance tothe nth nearest BSs in the setup in Fig. 1c is arandom variable, the probability distribution func-tion (PDF) of which can be found for PPPs inclosed form [10]. Although placement of BSs canbe only approximately described using PPPs, byusing such a PDF we can get a rough estimate ofthe distance to the nearest (and probably thestrongest) interfering BS. Note that this interfer-ing BS will have index n = 2, because the BS withn = 1 would be the serving BS. Therefore, suchstatistics can play an important role in MCC mod-els by providing estimates for the distances.

In the network in Fig. 3, BSs share CSI todirect their radiation patterns toward the intend-ed users, avoiding transmission toward unintend-ed stations. Clearly, the decisions made by a BSwill depend on the realization of node locations,which is visibly different from regular placement.For such a setup, shapes of outage probabilitiesfor a typical mobile station and for the one locat-ed close to the center of cooperating clusterwere compared analytically in [11]. Intuitively,the chance of outage for the nodes located clos-er to the cell center should be lower than thatfor a typical user; however, the analysis in [6]based on PP theory allowed quantification of theapproximate rate of outage probability decay.

CONCLUSION AND FUTURE DIRECTIONS

The aim of this article was to highlight currentapproaches to network modeling with applica-tions to multicell cooperation, motivated by the

Figure 4. BSs modeled as a realization of a hard-core point process with h = 1,so the minimal inter-BS distance is h = 1.

x

BSs = 276 λbs = 1; UTs = 39783 λut = 100

-8-10

-8

-10

z

-6

-4

-2

0

2

4

6

8

10

-6 -4 -2 0 2 4 6 8 10

Table 1. Comparison of key features of Wyner models and random models based on point processes.

Feature Wyner models Random models

BS locations Deterministic Points of a point process

UT locations Not used Points of a point process

Cell shape Line segment (1D); Generally hexagon or circle (2D) Irregular, typically Voronoi tesselations

Path loss Implicitly included in design parameter a Explicit path loss model l(ri)

Fading Explicit (except Gaussian WM) Explicit

Advantages Simplicity; analytical tractability; account for essentialchannel characteristics

Model real processes (e.g. interference); account for ran-domness, direct results for typical user

Disadvantages

Replace real processes (e.g. interference); static topol-ogy; hypothesis testing and generalization by simula-tions required, hence computational complexity toimplement/verify/generalize results

Analytical complexity to create accurate point processmodel for systems with correlated nodes; analytical com-plexity to obtain performance metrics; modelrevision/new model required for different communicationstrategies



fact that associated assumptions drive such per-formance-determining parameters as signalpropagation delay, attenuation, and interferencepicture. These approaches can generally be clas-sified into:• Regular models, represented by the Wyner

model family, where network topology is staticand channel gains are typically described usinga single parameter to capture channel attenu-ation and interference

• Random network models, which use point pro-cesses to model placements and interactionsbetween network elementsTo understand future research directions in

MCC network modeling, it is important to onceagain recall the ultimate aim of such modeling.A network model is required to provide a ver-sion of the environment or a setup where devel-oped communication techniques will operate.Ideally, one would want such a version to repli-cate all important features of the real environ-ment; however, usually detailed modeling isassociated with high complexity. Given the grow-ing number of mobile terminals and evolution ofcommunication techniques, it is a questionwhether the growth in computational power willbe sufficient to continue delivering acceptableresults for techniques developed based onWyner-type constructions. More important, nottaking network topology into account may limitour understanding of the operation of moderncommunication systems. On the other hand, thelimitations of PP-based models originate mainlyfrom analytical complexity, although surprisinglysimple closed form expressions are already avail-able for systems with little dependence betweennetwork elements.

Therefore, further research into analyticallyfeasible yet accurate PP-based models is neces-sary to capture or approximate the dependenciesin BS placement and mobile user clustering. Ide-ally, such models could not only reduce or elimi-nate the need for large-scale simulations, butalso deliver more accurate results and advanceour understanding of MCC systems. If closedform solutions are intractable, asymptotic resultscould still provide valuable insights into opera-tion and design of modern MCC networks.

REFERENCES[1] E. Hossain, D. I. Kim, and V. K. Bhargava, Cooperative Cel-

lular Wireless Networks, Cambridge Univ. Press, 2011.[2] D. Gesbert et al., “Multi-Cell MIMO Cooperative Net-

works: A New Look at Interference,” IEEE JSAC, vol. 28,no. 9, Dec. 2010, pp. 1380–408.

[3] Z. Hasan, A. Jamalipour, and V. Bhargava, “CooperativeCommunication and Relay Selection Under AsymmetricInformation,” Proc. IEEE WCNC ’12, Apr. 2012, pp.2373–78.

[4] F. Baccelli et al., “Stochastic Geometry and Architectureof Communication Networks,” Telecommun. Sys., vol.7, no. 1–3, 1997, pp. 209–27.

[5] J. Zhang and J. Andrews, “Distributed Antenna Systemswith Randomness,” IEEE Trans. Wireless Commun., vol.7, no. 9, Sept. 2008, pp. 3636–46.

[6] K. Huang and J. Andrews, “A Stochastic-GeometryApproach to Coverage in Cellular Networks With Multi-CellCooperation,” Proc. IEEE GLOBECOM ’11, Dec. 2011.

[7] J. Hoydis et al., “Analysis of Multicell Cooperation with Ran-dom User Locations via Deterministic Equivalents,” Proc.10th Int’l. Symp. Modeling and Optimization in Mobile, AdHoc and Wireless Networks, May 2012, pp. 374–79.

[8] J. Xu, J. Zhang, and J. Andrews, “On the Accuracy ofthe Wyner Model in Cellular Networks,” IEEE Trans.Wireless Commun., vol. 10, no. 9, Sept. 2011, pp.3098–3109.

[9] M. Haenggi et al., “Stochastic Geometry and RandomGraphs for the Analysis and Design of Wireless Net-works,” IEEE JSAC, vol. 27, no. 7, 2009, pp. 1029–46.

[10] D. Stoyan, W. S. Kendall, and J. Mecke, StochasticGeometry and Its Applications, 2nd ed. Wiley, 2008.

[11] A. Tukmanov et al., “On the Broadcast Latency in FiniteCooperative Wireless Networks,” IEEE Trans. Wireless Com-mun., vol. 11, no. 4, Apr. 2012, pp. 1307–13.

[12] H. ElSawy and E. Hossain, “Modeling Random CSMAWireless Networks in General Fading Environments,”Proc. IEEE ICC ’12, 10–15 June 2012, pp. 6522–6989.

BIOGRAPHIESANVAR TUKMANOV [S’09] ([email protected])received his B.Eng. degree in mobile communications fromKazan State Technical University in 2007 and his M.Sc.degree in communications and signal processing fromNewcastle University in 2009, all with distinction. Since2010 he has been working toward his Ph.D. degree atNewcastle University. From 2008 to 2010 he was anIP/MPLS network engineer with Tattelecom, Kazan. Hisresearch interests are in cooperative diversity, stochasticgeometry, and information theory.

ZHIGUO DING [S’03, M’05] received his B.Eng. in electricalengineering from the Beijing University of Posts andTelecommunications in 2000, and his Ph.D. degree in elec-trical engineering from Imperial College London in 2005.From July 2005 to June 2010, he worked with Queen’sUniversity Belfast, Imperial College, and Lancaster Universi-ty. Since October 2010 he has been with Newcastle Univer-sity, currently as a reader. His research interests arecross-layer optimization, cooperative diversity, statisticalsignal processing, and information theory.

SAID BOUSSAKTA [S’89, M’90, SM’04] received his Ingenieurd’Etat degree in electronic engineering from the NationalPolytechnic Institute of Algiers (ENPA), Algeria, in 1985and his Ph.D. degree in electrical engineering (signal andimage processing) from the University of Newcastle uponTyne, United Kingdom, in 1990. From 1990 to 1996 hewas with the University of Newcastle upon Tyne as asenior research associate in digital signal and image pro-cessing. From 1996 to 2000 he was with the Universityof Teesside, United Kingdom, as a senior lecturer in com-munication engineering. From 2000 to 2006 he was withthe University of Leeds as a reader in digital communica-tions and signal processing. He is currently a professor ofcommunications and signal processing at the School ofElectrical, Electronic and Computer Engineering, Universi-ty of Newcastle upon Tyne, where he is lecturing onadvanced communication networks and signal processingsubjects. His research interests are in the areas of fastDSP algorithms, digital communications, communicationsnetworks systems, cryptography and security, and digitals ignal / image processing. He has authored and co-authored more than 250 publications. He is a Fellow ofthe IET.

ABBAS JAMALIPOUR [S’86, M’91, SM’00, F’07] received hisPh.D. degree from Nagoya University, Japan. He is theProfessor of Ubiquitous Mobile Networking with theSchool of Electrical and Information Engineering, Univer-sity of Sydney, Sydney, NSW, Australia. He is a Fellow ofthe Institute of Electrical, Information, and Communica-tion Engineers (IEICE) and the Institution of EngineersAustralia, an IEEE Distinguished Lecturer, and a Techni-cal Editor of several scholarly journals. He has been aChair of several international conferences, including IEEEICC and IEEE GLOBECOM, General Chair of IEEE WCNC’10, as well as technical program chair of IEEE PIMRC’12 and IEEE ICC ’14. He is Vice President — Confer-ences and a member of the Board of Governors of theIEEE Communications Society (ComSoc). He is the recipi-ent of several prestigious awards, including the 2010IEEE ComSoc Harold Sobol Award for Exemplary Serviceto Meetings and Conferences, the 2006 IEEE ComSocDistinguished Contribution to Satellite CommunicationsAward, and the 2006 IEEE ComSoc Best Tutorial PaperAward.

To understand futureresearch directions inMCC network model-

ing, it is importantto once again recallthe ultimate aim of

such modeling. Anetwork model is

required to provide aversion of the envi-ronment or a setup

where developedcommunication tech-niques will operate.




1 The static power con-sumption of a BS refers tothe power consumption ofthe BS when there are noactive users in the cover-age of the BS.


INTRODUCTIONGreening is not merely a trendy concept, but isbecoming a necessity to bolster social, environ-mental, and economic sustainability. Naturally,green communications has received much atten-tion recently. As cellular network infrastructuresand mobile devices proliferate, an increasingnumber of users rely on cellular networks intheir daily lives. As a result, the energy con-sumption of cellular networks keeps increasing.Therefore, greening cellular networks is attract-ing tremendous research efforts in bothacademia and industry [1]. Meanwhile, it hasbeen shown that, with the aid of multicell coop-eration, the performance of a cellular network interms of throughput and coverage can beenhanced significantly. However, the potential ofmulticell cooperation on improving the energyefficiency of cellular networks remains to beunlocked. This article overviews the multicellcooperation solutions for improving the energyefficiency of cellular networks.

The energy consumption of a cellular net-work is mainly drawn from base stations (BSs),which account for more than 50 percent of theenergy consumption of the network. Thus,improving the energy efficiency of BSs is crucialto green cellular networks. Taking advantage ofmulticell cooperation, the energy efficiency ofcellular networks can be improved from threeaspects. The first one is to reduce the number ofactive BSs required to serve users in an area [2].The solutions are to adapt the network layoutaccording to traffic demands. The idea is toswitch off BSs when their traffic loads are belowa certain threshold for a certain period of time.When some BSs are switched off, radio coverageand service provisioning are taken care of bytheir neighboring cells. The second aspect is toassociate users with green BSs powered byrenewable energy. Through multicell coopera-tion, off-grid BSs enlarge their service area whileon-grid BSs shrink their service area. Zhou et al.[3] proposed the handover parameter tuningalgorithm and power control algorithm to guidemobile users to associate with BSs powered byrenewable energy, thus reducing on-grid powerexpenses. Han and Ansari [4] proposed an ener-gy-aware cell size adaptation algorithm namedICE. This algorithm balances the energy con-sumption among BSs, enables more users to beserved with green energy, and therefore reducesthe on-grid energy consumption. Envisioningfuture BSs to be powered by multiple types ofenergy sources (e.g., the grid, solar energy, andwind energy), Han and Ansari [5] proposed tooptimize the utilization of green energy for cel-lular networks by cell size optimization. The pro-posed algorithm achieves significant main gridenergy savings by scheduling the green energyconsumption along the time domain for individ-ual BSs, and balancing the green energy con-sumption among BSs for the cellular network.

The third aspect is to exploit coordinatedmultipoint (CoMP) transmissions to improve theenergy efficiency of cellular networks [6]. Onone hand, with the aid of multicell cooperation,the energy efficiency of BSs on serving cell edgeusers is increased. On the other hand, the cover-age area of BSs can be expanded by adoptingmulticell cooperation, thus further reducing thenumber of active BSs required to cover a certainarea. In addition to discussing the multicell

TAO HAN AND NIRWAN ANSARI, NEW JERSEY INSTITUTE OF TECHNOLOGY

ABSTRACTRecently, green communications has received

much attention. Cellular networks are amongthe major energy hoggers of communication net-works, and their contributions to the globalenergy consumption increase fast. Therefore,greening cellular networks is crucial to reducingthe carbon footprint of information and commu-nications technology. In this article, we overviewthe multicell cooperation solutions for improvingthe energy efficiency of cellular networks. First,we introduce traffic-intensity-aware multicellcooperation, which adapts the network layout ofcellular networks according to user trafficdemands in order to reduce the number of activebase stations. Then we discuss energy-awaremulticell cooperation, which offloads traffic fromon-grid base stations to off-grid base stationspowered by renewable energy, thereby reducingthe on-grid power consumption. In addition, weinvestigate improving the energy efficiency ofcellular networks by exploiting coordinated mul-tipoint transmissions. Finally, we discuss thecharacteristics of future cellular networks, andthe challenges in achieving energy-efficient mul-ticell cooperation in future cellular networks.

ON GREENING CELLULAR NETWORKS VIAMULTICELL COOPERATION

HAN LAYOUT_Layout 1 2/21/13 4:13 PM Page 82


cooperation solutions, we investigate the chal-lenges for multicell cooperation in future cellu-lar networks.

TRAFFIC INTENSITY AWAREMULTICELL COOPERATION

There are two reasons traffic demands of cellu-lar networks experience fluctuations. The firstone is the typical day-night behavior of users.Mobile users are usually more active in terms ofcell phone usage during the day than during thenight, and therefore, traffic demands during theday are higher than those at night. The secondreason is the mobility of users. Users tend tomove to their office districts during workinghours and come back to their residential areasafter work. This results in the need for largecapacity in both areas at peak usage hours, butreduced requirements during off-peak hours.However, cellular networks are usually dimen-sioned for peak hour traffic, and thus, most ofthe BSs work at low work load during off-peakhours. Due to the high static power consump-tion,1 BSs usually experience poor energy effi-ciency when they are operating under a lowwork load. In addition, cellular networks are typ-ically optimized for the purpose of providingcoverage rather than operating at full load.Therefore, even during peak hours, the utiliza-tion of BSs may be inefficient regarding energyusage. Adapting the network layout of cellularnetworks according to traffic demands has beenproposed to improve the energy efficiency of cel-lular networks. The network layout adaptation isachieved by dynamically switching BSs on/off.Figure 1 shows several scenarios of network lay-out adaptations. Figure 1a shows the originalnetwork layout, in which each BS has three sec-tors. For the green cell, if most of the trafficdemands on it are coming from sector three, andthe traffic demands in sectors one and two arelower than a predefined threshold, the green cellcould switch off sectors one and two to saveenergy, and the users in the sectors that areswitched off will be served by the neighboringcells. In this case, the network layout after theadaptation is shown in Fig. 1b. If traffic demandsfrom sector three of the green cell also decreasebelow the threshold, the entire green cell isswitched off, and the network layout is adapted,as shown in Fig. 1c. Under this scenario, theradio coverage and service provisioning in thegreen cell are taken care of by its active neigh-boring cells. When a BS is switched off, theenergy consumed by its radio transceivers, pro-cessing circuits, and air conditioners can besaved. Therefore, adapting the network layout ofcellular networks according to traffic demandscan save a significant amount of energy con-sumed by cellular networks.

While network layout adaptation can poten-tially reduce the energy consumption of cellularnetworks, it must meet two service requirements: • The minimum coverage requirement• The minimum quality of service (QoS) require-

ments of all mobile usersTherefore, in carrying out network layout adap-tation, multicell cooperation is needed to guar-

Figure 1. Illustrations of network layout adaptations for macro cellular sites: a)original network layout; b) BS partially switched off; c) BS entirely switchedoff.

(a)

(b)

(c)

33

2

1

33

2

1

33

2

1

33

2

1

33

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1

33

2

1



antee service requirements. Otherwise, it willresult in a high call blocking probability andsevere QoS degradation. For example, two adja-cent BSs both experience low traffic demands.However, only one BS can be switched off tosave energy, and the other BS should be activeto sustain the service provisioning in both cover-age areas. In this case, if both BSs are switchedoff according to their own traffic demands, theirsubscribers will lose connections. Therefore,cooperation among BSs is essential to enabletraffic-intensity-aware network layout adapta-tion.

COOPERATION TO ESTIMATE TRAFFIC DEMANDSNetwork layout adaptation is based on the esti-mated traffic demands at individual cells. Thetraffic demand estimation at individual cellsrequires the cooperation of their neighboringcells. To avoid frequently switching BSs on/off,they are switched off only when their trafficdemands are less than a predefined threshold fora minimum period, T. Therefore, the estimatedtraffic demands should represent the trafficdemands at individual cells for at least a timeperiod of length T. Hence, traffic demands at aBS consist of three parts: traffic demands fromusers who are currently attached to the BS, traf-fic demands from users who will hand over toother BSs, and traffic demands from users whowill hand over to the BS from its neighboringBSs. While the first two components can bemeasured and estimated by individual BSs, theestimation of the third component requirescooperation from neighboring cells. Two factorscontribute to the handover traffic from theneighboring cells. The first one is the user mobil-ity. A user’s motion, including the direction andvelocity, can be measured by various signal pro-cessing methods. Therefore, individual BSs canpredict:

• How many users will hand over to other cellsin the near future

• To which cells these users are highly likely tohand over

Such information is important for their neigh-boring BSs to estimate their traffic demands.The second reason for handover traffic is theswitching off of neighboring BSs. If one of theneighboring cells is switched off, the users underits coverage will be handed over to its neighbor-ing cells, thus increasing traffic demands of theneighboring cells. Therefore, cooperation amongradio cells is important for the traffic demandestimation at individual cells.

COOPERATION TO OPTIMIZE THESWITCHING OFF STRATEGY

With the traffic demand estimation, cellularnetworks optimize the switching on/off strate-gies to maximize the energy saving while guar-anteeing users’ minimal service requirements.Currently, most existing switching on/off strate-gies are centralized algorithms, which assumethat there is a central controller that collectsthe operation information of all BSs and opti-mizes network layout adaptations. Three meth-ods have been proposed to determine whichBSs are to be switched off. The first one is ran-domly switching off BSs with low traffic loads.This method mainly applies to BSs in the nightzone, where few users are active. The methodrandomly switches off some BSs to save energy,and the remaining BSs provide coverage forthe area. The second method is a greedy algo-rithm, which enforces BSs with higher trafficloads to serve more users and switches off BSswith no traffic load. The third method is basedon the user–BS distance. The required trans-mission power of BSs for serving users dependson the distance between users and BSs. The

Figure 2. Energy utilization of the cellular network: a) on-grid energy consumption; b) renewable energy consumption.

The renewable energy generation rate (kWh/h)

(a)

0

1

0

The

on-g

rid

ener

gy c

onsu

mpt

ion

(kW

h)

2

3

4

0.5

x102

x102

1.0 1.5 2.0

The renewable energy generation rate (kWh/h)

(b)

0

3

2

The

rene

wab

le e

nerg

y co

nsum

ptio

n (k

Wh)

4

5

6

7

0.5 1 1.5 2

SSFCSO without sleepCSO

SSFCSO without sleepCSO



longer the distance, the greater the transmis-sion power required in order to meet users’minimal service requirements. Therefore, theuser–BS distance can be an indicator of theenergy efficiency of cellular networks: theshorter the average user–BS distance, the high-er the energy efficiency. Hence, the algorithmtends to switch off BSs with the longestuser–BS distance to improve the energy effi-ciency of cellular networks.

Centralized algorithms, however, require thechannel state information and traffic load infor-mation of every cell. Collecting this informationcentrally may impose tremendous communica-tion overheads, and thus reduce the effectivenessof centralized algorithms in improving energyefficiency. Therefore, distributed algorithms arefavored, especially for heterogeneous networksthat consist of various types of cells such asmacrocells, microcells, picocells, and femtocells.To enable distributed algorithms, individual cellsmay cooperatively form coalitions, and share thechannel state information and traffic load infor-mation. Based on the shared information, indi-vidual BSs optimize their operation strategies tominimize the total energy consumption of theBS coalition.

ENERGY AWARE MULTICELL COOPERATION

With the worldwide penetration of distributedelectricity generation at medium and low volt-ages, it is proved that renewable energy such assustainable biofuels, and solar and wind energycan be effectively uti l ized to substantiallyreduce carbon emission. Therefore, researchershave proposed to power BSs with renewableenergy to save the on-grid energy consumption

and reduce the carbon footprint. NokiaSiemens Networks [7] has developed off-gridBSs that rely on a combination of solar andwind power supported by fuel cell and deepcycle battery technologies. Compared to on-grid BSs, off-grid BSs achieve zero carbonemission and save a significant amount of on-grid energy. However, due to the dynamicnature of renewable resources, renewable ener-gy generation highly depends on environmentalfactors such as temperature, light intensity, andwind velocity. In addition, because of the limit-ed battery capacity, generated energy shouldbe util ized well to avoid energy overflow.Therefore, how to optimize the utilization ofrenewable energy in cellular networks is not atrivial problem. One intuitive solution for thisproblem is the energy-aware cell breathingmethod. Here, “energy-aware” pertains to twoperspectives. The first one is the awareness ofenergy sources such that users are guided todraw energy from off-grid BSs rather than on-grid BSs. This can be achieved by enlarging thecell sizes of off-grid BSs and shrinking those ofon-grid BSs [3]. The second perspective is theawareness of the amount of energy storagesuch that off-grid BSs with higher amounts ofenergy storage are forced to serve a largerarea. In this way, energy overflow can be avoid-ed, thus enabling the renewable energy genera-tion system to harvest more energy fromrenewable sources [4]. The energy-aware cellbreathing technique is a step forward to fur-ther traffic-intensity-aware network layoutadaptations. In addition to the traffic loadinformation and channel state information,energy-aware cell breathing requires multiplecells to cooperatively share the energy informa-

Figure 3. Energy utilization of the cellular network: a) joint transmission; b) cooperative beamforming; c) cooperative relaying; d) dis-tributed space-time coding.

Null steeringBeamforming

(a)

Base station 1 Base station 2

User 1

User 2

(b)

(c) (d)

Base station 1 Base station 2

Base station

User 1 User 2

Phase 1Phase 2

Time slot 1Time slot 2

Primary base station

Secondary user Relay node

Relay node

Relay node

Secondary base station

Primary userMobile users



tion, including the estimated amount of energyarrival and the amount of energy storage, andto coordinate to minimize the energy consump-tion of cellular networks.

A CASE STUDYEnvisioning future BSs powered by multipletypes of energy sources (e.g., on-grid energy andrenewable energy), we investigate the optimalenergy utilization strategies in cellular networkswith both on-grid and off-grid BSs. In such cel-lular networks, BSs are powered by renewableenergy if they have enough renewable energystorage; otherwise, the BSs switch to on-gridenergy to serve mobile users. To optimize ener-gy utilization, BSs cooperatively determine theircell size through energy-aware cell breathingbased on the proposed cell size optimization(CSO) algorithm [5]. By cooperative cell breath-ing, the on-grid energy consumption of the cel-lular network over a period of time consisting ofseveral cell size adaptations is minimized. Theoptimal energy utilization strategy includes twoparts: the multi-stage energy allocation policyand the single-stage energy consumption mini-mization. The multi-stage energy allocation poli-cy determines the amount of renewable energyallocated at individual off-grid BSs during eachcell size update. The single-stage energy con-sumption minimization involves two steps. Thefirst step is to minimize the number of activeon-grid BSs by offloading traffic cell breathing.On-grid BSs shrink their cell sizes to enforcethe associated users to hand over to off-gridBSs, and the off-grid BSs with higher amountsof energy storage enlarge their cell sizes toabsorb more users. The second step is to mini-mize the number of active BSs (both on-gridand off-grid) by applying the sleep mode check-ing algorithm. When traffic demands on individ-ual BSs are lower than a threshold, the BSs seekto put themselves into sleep mode by cooperat-ing with their neighboring cells. If the BS is insleep mode, its associated users will be handedover to its neighboring cells. In this case, if theneighboring BSs have enough energy storage toserve the handed over users, and the total ener-gy consumption of the cellular network isreduced, the BS is switched into sleep mode tosave energy.

Figure 2 shows the comparisons between theCSO algorithm and the strongest-signal firstmethod, which always associates a user with theBS with the strongest received signal strength.Figure 2a shows the on-grid energy consumptionat different renewable energy generation rates.

In this simulation, there are 400 mobile users inthe mobile network. As the renewable energygeneration rate increases, the on-grid energyconsumption of the mobile network is reduced.When the renewable energy generation rate islarger, more electricity is generated from renew-able energy. Therefore, more BSs can servemobile users using electricity generated byrenewable energy instead of consuming on-gridenergy. When the renewable energy generationrate is larger than 0.6 kWh/h, the CSO algorithmachieves zero on-grid energy consumption byoptimizing the cell size of each BS, while theSSF algorithm still consumes a significantamount of on-grid energy. When the renewableenergy generation rate is low, the CSO algo-rithm saves more than 200 kWh electricity thanthe SSF algorithm. As the generation rateincreases, the gap between the energy consump-tion of CSO and that of SSF narrows becausethe CSO algorithm has already achieved zeroenergy consumption, and the energy consump-tion of SSF reduces due to increased availabilityof renewable energy. Figure 2b shows the renew-able energy consumption at different energygeneration rates. When the energy generationrate is less than 0.7 kWh/h, the renewable energyconsumptions of all the simulated algorithmsincrease as the renewable energy generation rateincreases. This is because the larger the renew-able energy generation rate, the more renewableenergy can be used to serve mobile users. Inaddition, the CSO algorithm consumes morerenewable energy than the SSF algorithm doesbecause the CSO algorithm optimizes the cellsizes of BSs and maximizes the utilization ofrenewable energy in order to reduce the on-gridenergy consumption. When the renewable ener-gy generation rate is larger than 0.7 kWh/h, therenewable energy consumption of the CSO algo-rithm does not change much, while that of SSFkeeps increasing. This is because when therenewable energy generation rate is larger than0.7 kWh/h, the on-grid energy consumption ofthe CSO algorithm is zero, as shown in Fig. 2a.This indicates that all the users are served byrenewable energy. Therefore, the renewableenergy consumption of the CSO algorithmreflects the total energy consumption of the net-work, which is similar under different energygeneration rates in the simulation.

ENERGY EFFICIENT COMP TRANSMISSION

Coordinated multipoint (CoMP) transmission isa key technology for future mobile communica-tion systems. In CoMP transmission, multipleBSs cooperatively transmit data to mobile usersto improve their receiving signal quality. BSsshare the required information for cooperationvia high-speed wired links. Based on the infor-mation, BSs determine joint processing andcoordinated scheduling strategies. While CoMPtransmission has been proved to be effective inimproving the data rate and spectral efficiencyof cell edge users [8], its potential for improvingthe energy efficiency of cellular networks has notbeen fully exploited. In this section, we discussthe potentials of CoMP transmission in greeningcellular networks.

Figure 4. Cooperation to increase coverage area.

Base station 1

Cooperative coverage area

Base station 3Base station 2



INCREASE ENERGY EFFICIENCY FORCELL EDGE COMMUNICATIONS

BSs usually have lower energy efficiency in serv-ing the users at the cell edge than in serving theusers within the cell. This is not only becausecell edge users are far away from BSs andrequire more transmission power, but alsobecause cell edge users experience more inter-ference from neighboring cells. Multicell cooper-ation can improve energy efficiency for cell edgecommunications [6].

Joint Transmission — Joint transmission exploitsthe cooperation among neighboring BSs totransmit the same information to individualusers located at the cell edge [9]. Figure 3ai l lustrates the joint transmission scheme.User 2 associates with BS 2 while user 1 asso-c iates with BS 1 . Instead of serving theirassociated users independently, BSs 1 and 2cooperatively transmit information to theirusers. For example, if both BSs apply time-division multiple access (TDMA) in the firsttime slot, both BS 1 and 2 transmit the sameinformation to user 2. The user combines thesignals from both BSs to decode the informa-tion. In the second time slot, the BSs cooper-atively serve user 1. Due to joint transmission,cell edge users experience higher receivingsignal strength and lower interference, hencerequiring less transmission power from bothBSs. Thus, joint transmission improves theenergy efficiency of the cell edge communica-tions.

Cooperative Beamforming — In cooperative beam-forming, a BS forms radio beams to enhance thesignal strength of its serving users while formingnull steering toward users in its neighboringcells. As shown in Fig. 3b, BS 1 forms the radiobeam toward its associated user, user 1, toincrease the user’s receiving signal strength. Onthe other hand, in order to reduce the interfer-ence to user 2 who is served by a neighboringBS, BS 1 forms the null steering toward user 2.BS 2 executes the same operation as BS 1.Therefore, through cooperative beamforming,the signal-to-noise ratio (SNR) of both user 1and user 2 is enhanced. Thus, less transmissionpower is required for BSs to serve the cell edgeusers, and the energy efficiency of cell edgecommunications is increased.

Cooperative Relaying — Taking advantage of cogni-tive radio techniques, cooperation can be exploit-ed between a primary cell and a secondary cellas shown in Fig. 3c. The secondary BS coopera-tively relays the signal from the primary BS tothe primary user. The primary user combines thesignals from both the primary and secondary BSsto decode the information. To stimulate cooper-ative relaying, the primary BS grants the sec-ondary BS some spectrum that can be utilizedfor communications between the secondary BSand secondary users. By cooperating with sec-ondary BSs, the transmission power requiredfrom primary BSs to serve the cell edge users isreduced. Therefore, the cooperative relayscheme reduces the power consumption of pri-mary BSs [10]. However, due to the time-varying

Figure 5. Future cellular networks with holistic cooperation.

Coverage area of high-power green BS

High-power BS

Low-power BS

Secondary BS

Users

Coverage area of low-power BS

Coverage area of secondary BSCoverage area of high-power on-grid BSTransmission beam of high-power BSTransmission beam of low-power BSTransmission beam of secondary BSCommunication link between BSs

Owing to joint trans-mission, cell edgeusers experiencehigher receiving

signal strength andlower interference,therefore requiring

less transmissionpower from both

BSs. Thus, the jointtransmission

improves energy efficiency of the

cell edge communications.



wireless channels, the scheduling delay involvedin the relaying process may increase the outageprobability. Fan et al. [11] showed that the out-age probability caused by the scheduling delaycan be alleviated by optimizing the transmissionpower of the network nodes.

Distributed Space-Time Coding — Distributed space-time coding [12] explores the spatial diversityof both the relay nodes and the mobile usersto combat mult ipath fading, and can beapplied to improve the spectral and energyefficiency of cell edge users. As illustrated inFig. 3d, the distributed space-time codingscheme consists of two phases. In the firstphase, the BS broadcasts a data packet to itsdestination and the potential relays. In thesecond phase, the potential relays that candecode the data packet cooperatively transmitthe decoded data packet to the destinationusing a suitable space-time code. By exploringspatial diversi ty , distr ibuted spaced-t imecoded cooperative transmission improves thediversity gain of cellular networks. However,the multiplexing gain of the wireless systemmay be sacrificed [13]. Zou et al. [13] pro-posed an opportunistic distributed space-timecoding (O-DSTC) scheme to increase themultiplexing gain. Taking advantage of DSTCcooperative transmission, the spectral efficien-cy is increased. As a result, fewer transmis-sions are required in order to transmit a givennumber of data packets. Therefore, the ener-gy eff ic iency of the cel lular networks isenhanced.

ENABLE MORE BSS TO ENTER SLEEP MODEAs discussed in the previous subsection, CoMPtransmission (e.g., joint transmission) enhancesthe receiving SNR of cell edge users, implyingthat BSs are able to cover a larger area with thesame transmission power. Therefore, under lowtraffic demands, adopting the CoMP transmis-sion can reduce the number of BSs required tocover an area, thus improving the energy effi-ciency of cellular networks [14].

As presented earlier, traffic-intensity-awaremulticell cooperation enables BSs with low traf-fic demands to go into sleep mode to save ener-gy. The users in the off cells will be served bytheir active neighboring cells. However, due tothe limitation of BSs’ maximal transmissionpower, a few BSs should stay awake to providecoverage in the area. By applying CoMP trans-mission, the number of required active BSs canbe further reduced. As illustrated in Fig. 4, theinner circle of BS 1 is the original coverage area,while the outer circle indicates the coverage areaafter cooperation with its neighboring BS, BS 3.The green area is the additional coverage areaachieved by multicell cooperation. Under thecondition of lower traffic demands, if there is nocooperation among BSs, three BSs have to stayawake to cover the whole area. However, if mul-ticell cooperation is enabled, BS 1 and BS 3 canprovide services to the users in the area by apply-ing cooperative transmission; thus, BS 2 can beswitched into sleep mode. Therefore, the totalenergy consumption of cellular networks isreduced.

FUTURE CELLULAR NETWORKS WITHMULTICELL COOPERATION

With advances of cellular networks and commu-nication techniques, future cellular networks willbe featured as heterogeneous networks in termsof both network deployments and communica-tion techniques, as shown in Fig. 5.

Regarding network deployments, heteroge-neous networks refer to deploying a mix of high-power and low-power BSs in order to satisfy thetraffic demands of service areas. High-power BSs(e.g., macro and micro BSs) are deployed to pro-vide coverage to a large area, while low-powerBSs are deployed to provide high capacity withina small coverage area. In addition, based onenergy supplies, BSs can be further divided intotwo types: on-grid BSs, and off-grid BSs poweredby green energy such as solar power and windpower. Therefore, future heterogeneous net-works consist of four types of BSs: high-poweron-grid BSs, high-power off-grid BSs, low-poweron-grid BSs, and low-power off-grid BSs. Interms of technology diversity, heterogeneousnetworks consist of a variety of technologiessuch as multiple input multiple output (MIMO),cooperative networking, and cognitive network-ing. Taking advantage of the increasing networkdiversity, BSs have more cooperation opportuni-ties, and thus can achieve additional energy sav-ings. However, realizing optimal multicellcooperation is not trivial.

Coalition Formulation — The problem of coalitionformation is to determine the coalition size andmembers of the coalition. Although the problemof coalition formation is well studied in the fieldof economics, and some game theory methodshave been applied to solve the coalition forma-tion problem in wireless networks, few modelscan be directly applied to form the coalitions ofBSs for the purpose of reducing the energy con-sumption of cellular networks. This is becausethe original coalition formation problem is toform a coalition to maximize the benefit of allthe members in the coalition, while the problemof forming coalitions of BSs is to maximize totalbenefits of the coalitions. In addition, due to thediversity of BSs and radio access technologies, asuccessful coalition may consist of BSs and tech-niques that are complementary to each other tomaximize the energy savings of the coalition.Hence, novel models or methods of coalitionforming must be developed to facilitate energy-efficient multicell cooperation.

Green Energy Utilization — For the sake of environ-mental friendliness, off-grid BSs powered byrenewable energy are favored over on-grid BSs.In order to optimize the utilization of off-gridBSs, the fundamental design issue is to effective-ly utilize the harvested energy to sustain trafficdemands of users in the network. The optimalutilization of green energy over a period of timedepends on the characteristics of the energyarrival and energy consumption at the currentstage as well as in future stages, and on thecooperation strategies of the neighboring cells orcooperating cells.

Taking advantage ofthe increasing network diversity,BSs have more cooperation opportunities, andthus can achieveadditional energysavings. However,realizing optimalmulticell cooperationis not trivial.



Incentive Mechanism — Note that in future hetero-geneous networks, multicell cooperation mayinvolve cells not owned by the same operator.For example, the coalition may consist of a BSfrom a different operator, a secondary BS oper-ated via cognitive radio, and a relaying cellformed by mobile users. To incentivize coopera-tion among these cells, a simple but effectiveincentive mechanism should be exploited.

CONCLUSION

This article discusses how to reduce the energyconsumption of cellular networks via multicellcooperation. We focus on three multicell coop-eration scenarios that enhance energy efficiencyof cellular networks. The first one is traffic-intensity-aware multicell cooperation, in whichmultiple cells cooperatively estimate trafficdemands and adapt the network layout based onthe estimated traffic demands. Through networklayout adaptation, the number of active BSs canbe reduced, thus reducing the energy consump-tion of cellular networks. The second scenario isenergy-aware multicell cooperation, in which off-grid BSs powered by green energy are enforcedto serve a large area to reduce the on-grid powerconsumption. The third one is energy-efficientCoMP transmission, in which the overall energyconsumption is reduced by improving the energyefficiency of BSs in serving cell edge users. Inaddition, we discuss the diversity of future cellu-lar networks in terms of radio access technolo-gies and BS characteristics, and the challenges tooptimally exploit multicell cooperation by capi-talizing on diversities.

REFERENCES[1] L. Correia et al., “Challenges and Enabling Technologies

for Energy Aware Mobile Radio Networks,” IEEE Com-mun. Mag., vol. 48, no. 11, Nov. 2010, pp. 66–72.

[2] Z. Niu, “Tango: Traffic-Aware Network Planning andGreen Operation,” IEEE Wireless Commun., vol. 18, no.5, Oct. 2011, pp. 25–29.

[3] J. Zhou et al., “Energy Source Aware Target Cell Selectionand Coverage Optimization for Power Saving in CellularNetworks,” Proc. 2010 IEEE/ACM Int’l Conf. Green Com-puting and Commun. & Int’l Conf. Cyber, Physical andSocial Computing, Hangzhou, China, Dec. 2010.

[4] T. Han and N. Ansari, “ICE: Intelligent Cell Breathing toOptimize the Utilization of Green Energy,” IEEE Com-mun. Letters, vol. 16, no. 6, June 2012, pp. 866–69.

[5] T. Han and N. Ansari, “Optimizing Cell Size for EnergySaving in Cellular Networks with Hybrid Energy Sup-plies,” IEEE GLOBECOM ’12, Dec. 2012, pp. 3–7.

[6] F. Heliot, M. Imran, and R. Tafazolli, “Energy EfficiencyAnalysis of Idealized Coordinated Multi-Point Commu-nication System,” IEEE VTC-Spring ’11, Budapest, Hun-gary, May 2011.

[7] “E-Plus, Nokia Siemens Networks Build Germany’s FirstOff-Grid Base Station,” http://www.nokiasiemensnet-works.com/news-events/press-room/press-releases/e-plus-nokia-siemens-networks-build-germany-s-first-off-grid-base-station

[8] R. Irmer et al., “Coordinated Multipoint: Concepts, Per-formance, and Field Trial Results,” IEEE Commun. Mag.,vol. 49, no. 2, Feb. 2011, pp. 102–11.

[9] M. Sawahashi et al., “Coordinated Multipoint Transmis-sion/Reception Techniques for LTE-Advanced (Coordi-nated and Distributed MIMO),” IEEE Wireless Commun.,vol. 17, no. 3, June 2010, pp. 26–34.

[10] S. Jayaweera, M. Bkassiny, and K. Avery, “AsymmetricCooperative Communications based Spectrum Leasing ViaAuctions in Cognitive Radio Networks,” IEEE Trans. Wire-less Commun., vol. 10, no. 8, Aug. 2011, pp. 2716–24.

[11] L. Fan et al., “Outage Probability Analysis and Power Allo-cation for Two-Way Relay Networks with User Selectionand Outdated Channel State Information,” IEEE Commun.Letters, vol. 16, no. 5, May 2012, pp. 638–41.

[12] J. N. Laneman and G. W. Wornell, “Distributed Space-Time-Coded Protocols for Exploiting Cooperative Diver-sity in Wireless Networks,” IEEE Trans. Info. Theory, vol.49, no. 10, Oct. 2003, pp. 2415–25.

[13] Y. Zou, Y.-D. Yao, and B. Zheng, “Opportunistic Dis-tributed Space-Time Coding for Decode-and-ForwardCooperation Systems,” IEEE Trans. Signal Proc., vol. 60,no. 4, Apr. 2012, pp. 1766–81.

[14] M. Herlich and H. Karl, “Reducing Power Consumptionof Mobile Access Networks with Cooperation,” 2ndInt’l. Conf. Energy-Efficient Computing and Network-ing, New York City, NY, May 2011.

BIOGRAPHIESTAO HAN [S’08] received his B.E. in electrical engineeringand M.E. in computer engineering from Dalian Universityof Technology and Beijing University of Posts and Telecom-munications, respectively. He is currently a Ph.D. candidatein the Department of Electrical and Computer Engineering,New Jersey Institute of Technology (NJIT), Newark. Hisresearch interests include mobile and cellular networks,network optimization, and energy-efficient networking.

NIRWAN ANSARI [S’78, M’83, SM’94, F’09] received hisB.S.E.E. (summa cum laude with a perfect GPA) from NJIT,his M.S.E.E. from the University of Michigan, Ann Arbor,and his Ph.D. from Purdue University, West Lafayette, Indi-ana. He joined NJIT in 1988, where he is a professor of elec-trical and computer engineering. He has also assumedvarious administrative positions at NJIT. He has been a visit-ing (chair/honorary) professor at several universities. His cur-rent research focuses on various aspects of broadbandnetworks and multimedia communications. He has servedon the Editorial/Advisory Boards of eight journals. He waselected to serve on the IEEE Communications Society (Com-Soc) Board of Governors as a member-at-large (2013–2014)as well as IEEE Region 1 Board of Governors as the IEEENorth Jersey Section Chair. He has chaired ComSoc techni-cal committees, and has actively organized numerous IEEEinternational conferences/symposia/workshops, assumingleadership roles as Chair or TPC Chair for various confer-ences, symposia, and workshops. He has authored Compu-tational Intelligence for Optimization (Springer 1997) withE.S.H. Hou and Media Access Control and Resource Alloca-tion for Next Generation Passive Optical Networks (Springer,2013) with J. Zhang, and edited Neural Networks inTelecommunications (Springer 1994) with B. Yuhas. He hasalso contributed over 400 publications, over one third ofwhich were published in widely cited refereed journals/mag-azines. He has been granted over 15 U.S. patents. He hasalso guest edited a number of special issues, covering vari-ous emerging topics in communications and networking.He has frequently been selected to deliver keynote address-es, distinguished lectures, and tutorials. Some of his recentrecognitions include a couple of best paper awards, severalExcellence in Teaching Awards, a Thomas Alva EdisonPatent Award (2010), an NJ Inventors Hall of Fame Inventorof the Year Award (2012), and designation as a ComSocDistinguished Lecturer (2006–2009).

The coalition mayconsist of a BS froma different operator,

a secondary BS oper-ated via cognitive

radio, and a relayingcell formed by

mobile users. Toincentivize coopera-

tion among thesecells, a simple buteffective incentivemechanism should

be exploited.

HAN LAYOUT_Layout 1 2/22/13 11:43 AM Page 89

http://www.nokiasiemensnetworks.com/news-events/press-room/press-releases/e-plus-nokia-siemens-networks-build-germany-s-first-off-grid-base-station


This work was supportedin part by the U.S.National Science Founda-tion under Grant CNS-205726 and the U.S.National Science Founda-tion CAREER Awardunder Grant ECCS-0348694.


INTRODUCTIONRecent years have witnessed the rapid emer-gence of a variety of multimedia wireless servicesover cellular networks, such as video confer-ences, mobile TV, music downloading, onlinegaming, and so on. Motivated by the concept ofubiquitous computing, these promising wirelessservices are designed to be available for peopleanytime and anywhere. Quality of service (QoS)provisioning plays a crucial role in supporting

multimedia wireless services in cellular networks.Consequently, the research community has madesignificant efforts in the development of variousadvanced and efficient wireless cellular network-ing architectures to support the diverse QoSrequirements. Among them, multicell coopera-tion networking is receiving significant researchattention from both academia and the industry.This is because multicell cooperation enablessignificant performance improvement of cellularnetworks in terms of the coverage of high-data-rate services and system throughput. As a result,multicell cooperation is being adopted in variouscommunication standards, including Third Gen-eration Partnership Project (3GPP) [1] LongTerm Evolution (LTE)-Advanced , or LTE-A[2], and IEEE 802.16m [3] for the next genera-tion wireless networks.

However, multicell cooperation also imposesthe new challenges associated with the two mutu-ally conflicting constraints between thepower/energy efficiency/saving and the QoS pro-visioning required by mobile users, which differfrom conventional cellular network systems. Onone hand, to ensure the demanded wirelesstransmission capacity for guaranteeing the strin-gent QoS provisioning over time-varying chan-nels covering the sizable wireless cells, a largeamount of power supply and consumption isdesired/needed at the base stations to sustain therequired wireless channel qualities [4, 5]. On theother hand, one of the main design goals/criteri-ons of multicell cooperation networks is to mini-mize the power/energy consumption [1, 6] amongall base stations through multicell cooperationand power scheduling, which, however, may sig-nificantly affect the QoS provisioning requiredby mobile users. Some standardization andresearch efforts have been made to addressthese problems. For instance, as a promisingtechnique under the multicell cooperation frame-work in the LTE-A standard [2], the cooperativemultipoint or coordinated multipoint (CoMP)[7] approach was proposed to improve the cover-age of high data rates and increase the overallmulicell cooperation network throughput whileefficiently mitigating intercell interference.While CoMP has the potential to improve QoS

XI ZHANG AND PING WANG, TEXAS A&M UNIVERSITY

ABSTRACTMulticell cooperation is emerging as a

promising wireless communication technique tosignificantly improve the performance of cellularnetworks in terms of the coverage of high-data-rate services and the system throughput for thenext generation wireless networks. On the otherhand, multicell cooperation also imposes thenew challenges associated with the two mutuallyconflicting constraints between the power/energyefficiency/saving and the QoS provisioningrequired by mobile users, which differ from theconventional cellular networks. To overcome theabove problem, we propose the semi-Markovdecision process model-based stochastic opti-mization scheme for the optimal trade-offbetween power saving and QoS provisioningover multicell cooperation networks. Our objec-tive is to efficiently minimize the power con-sumption at base stations while guaranteeingdiverse QoS provisioning for mobile usersthrough multicell cooperation power scheduling.To achieve this goal, we apply finite-stateMarkov chains to model the mobile users’ densi-ty, mobility, multicell cooperation power-profilescheduling, and QoS provisioning. Using thesemodels, we formulate the SMDP-based stochas-tic optimization problem and develop an effi-cient iteration algorithm to implement ourSMDP-based scheme. We evaluate our proposedschemes through numerical analyses, which showthat our proposed schemes significantly outper-form the other schemes in terms of minimizingpower consumption while guaranteeing therequired QoS provisioning.

OPTIMAL TRADE-OFF BETWEEN POWER SAVINGAND QOS PROVISIONING FOR

MULTICELL COOPERATION NETWORKS

The authors proposethe semi-Markov decision processmodel-based stochastic optimizationscheme for the optimal trade-offbetween power saving and QoS provisioning over multicell cooperationnetworks.

ZHANG LAYOUT_Layout 1 2/21/13 4:25 PM Page 90


provisioning, it does not explicitly address anyspecific/concrete power scheduling schemeswhich take into account the trade-off betweenpower saving and QoS provisioning over multi-cell cooperation networks. In [8] the authorsstudied a joint power and subcarrier allocationalgorithm that aims at minimizing network powerconsumption with a lower-bounded data rateconstraint. The authors of [9] developed a sleep-ing mode control scheme under the given per-formance requirements. In [10] the authorsderived an optimal sleep mode policy to maxi-mize a multiple-objective function of QoS andthe power consumption. However, the above-mentioned research works only consider the sim-ple power scheduling action of either totallyswitching off the base stations or keeping themin a fully operating state in a deterministic man-ner [11]. As a result, this type of deterministicswitching-on/off mode control scheme is lesseffective for the important application scenarioswhere the mobile users’ traffic loads are relative-ly heavy and their variations are highly random,especially when taking the users’ mobility/loca-tion and dynamic traffic load density/distributionvariations into account.

To overcome the aforementioned problems,in this article we propose the semi-Markov deci-sion process (SMDP) model-based stochasticoptimization scheme to achieve the optimaltrade-off between power saving and QoS provi-sioning over multicell cooperation networkswith the dramatic variations in mobile users’traffic load density and mobility. The objectiveof our proposed stochastic optimization frame-work is to efficiently minimize the power con-sumption at the base stations while guaranteeingthe diverse QoS provisioning for mobile usersthrough multicell cooperation power scheduling.To achieve this goal, we apply finite-stateMarkov chain techniques to develop a set ofanalytic models to characterize mobile users’density, mobility, and multicell cooperationpower-profile scheduling. Based on these mod-els, we formulate the SMDP-based stochasticoptimization problem to optimize the trade-offbetween power saving at base stations (BSs) andQoS provisioning to mobile users over multicellcooperation networks. We also develop an effi-cient iteration algorithm to implement our pro-posed SMDP-based optimization scheme. Thenwe conduct numerical analyses to evaluate ourproposed schemes, which show that our pro-posed SMDP-based stochastic control processconverges to the unique optimal solution andsignificantly outperforms the other schemes interms of minimizing the power consumption atBSs while guaranteeing the required QoS provi-sioning to mobile users.

The rest of this article is organized as follows.First, we develop the system models for optimiz-ing the trade-off between power saving and QoSprovisioning over our multicell cooperation net-works. Then, we formulate the SMDP-basedstochastic optimization problem to derive theoptimal power scheduling policy to optimize thetrade-off between power saving and QoS provi-sioning through multicell cooperation. Finally,we conduct numerical analyses to evaluate ourproposed schemes. The article then concludes.

THE SYSTEM MODELS

MULTICELL COOPERATION NETWORK MODELWe consider cellular networks where multicellcooperation is employed to minimize the powerconsumption at BSs while guaranteeing QoS pro-visioning to mobile users (MUs). Our cellularnetwork model is illustrated by an example shownin Fig. 1. It is a cellular network with five cells.As shown in Fig. 1a, one central cell is surround-ed by four neighboring cells. BSs are located atthe respective center of the cells, illustrated byhollow triangles. As shown in Fig. 1b for eachcell’s internal architecture, a number of MUs,represented by solid dots, are randomly distribut-ed in the cell with each MU located at the vari-able radius distance, denoted by d, between theBS and the MU. Clearly, the MUs’ traffic loaddistributions among wireless cells are indepen-dent of each other. Within each cell, the BS canperceive or sense the users’ information onmobility, dynamic behaviors, and traffic load spa-tial distributions, which all vary with time. TheMUs’ traffic load and distributions in each cellare characterized by the number of MUs, whichvaries with time. Thus, the power control of theBS in each cell needs to be dynamically sched-uled and cooperated according to the traffic loadvariations in order to minimize the total powerconsumption while guaranteeing the QoS formobile users.

TRAFFIC LOAD MODEL FOR MOBILE USERSFor each wireless cell, our MUs’ traffic loadmodel consists of two components:• Mobile users’ density model• Mobile users’ mobility modelwhich are detailed as follows, respectively.

Mobile Users’ Density Model — We use the finite-state Markov chain to model the MU density ineach wireless cell. Assuming the maximal num-ber of MUs that a cell can support is (D – 1),our MU density Markov chain in each cell canhave at most D state levels, with each level rep-resenting the number of MUs existing in the cell.Each cell’s BS acquires/senses the current stateof MU density through the sensing mechanismat the BS. Under the finite-state Markov chainmodel, the next state of user density in the cell is

Figure 1. Wireless cellular network model: a) muticell cooperation networkmodel; b) single cell model with a base station at the center of the cell, whered denotes the radius distance from the base station of the cell to the mobileuser.

(a) (b)

d

Mobile users (MUs)Base station (BSs)



determined by the current state of user densityand the Markov transition probability. Let D ={0, 1, …, (D – 1)} represent the Markov-chainstate space of user density in cell i, Ni(t) denotethe number of MUs in cell i with Ni(t) Œ D, andfgihi(t) be the transition probability that the stateof user density Ni(t) changes from state gi tostate hi, where gi, hi Œ D. Then the state transi-tion probability matrix of user density is definedfor cell i: Fi(t) = [fgihi(t)], where fwizi(t) = Pr{Ni(t + 1) = hiÔNi(t) = gi}, " gi, hi Œ D.

Mobile Users’ Mobility Model — We use the finite-state Markov chain to model the MU mobility interms of MUs’ spatial distributions in each wire-less cell. According to users’ different locationdistributions (e.g., an MU moves from the centerlocation of the BS to the edge of the cell), wecan shut down the BS or reduce its power forpower/energy saving. To accurately model theimpact of user mobility for cell power saving, weapply the k-means clustering algorithm [12, 13]to categorize the MUs into b classes accordingto the radius distances from the BS to the MUs.As shown in Fig. 2, for example, the MUs ineach cell are classified into three classes (b = 3)illustrated by three different-sized boxes, respec-tively. The numbers of users in each class variesas MUs move around. Assuming that there areK (£ (D – 1)) users in cell i, the spatial distribu-tion state of k MUs’ locations in cell i is denotedby Li(t) = (li,1, …, li,k), where k = {1, 2, …, K}and li,k represents the current location (mea-sured by the radius distance from the corre-sponding BS) of user k in cell i. Since there areb possible locations for each MU, there aretotally bk possible spatial location distributionsfor any k (£ K) mobile users in cell i. The BS canacquire/sense the current state of MUs’ spatiallocation distribution through a sensing mecha-nism, and the next state is determined by thecurrent state and the user mobility Markov-chaintransition probability. Let L = {L1, …, LbK}stand for the user mobility Markov chain statespace for K users’ spatial distributions in a cell,and let jUiVi(t) denote the probability that thestate of users’ location distribution in cell imoves from location state Ui to location state Viat time t, where {Ui, Vi Œ L. Then we have thebK ¥ bK user mobility Markov chain state transi-tion probability matrix of cell i, which is definedas Qi(t) = [jUiVi(t)], where jUiVi(t) = Pr{Li(t +

1) = ViÔLi(t) = Ui}, " Ui = (ui,1, …, ui,k, …,ui,K), Vi = (vi,1, …, vi,k, …, vi,K) Œ L, where ui,k,vi,k are the values that li,k can take, respectively.

POWER CONSUMPTION MODELFOR BASE STATIONS

We use the finite-state Markov chain to modelthe transmit power scheduling in each cell.Assuming our transmit power the Markov chainin each cell has E power-state levels, with eachlevel representing a transmit power level at theBS in a cell. Then we let E = {0, 1, …, (E – 1)}represent the power scheduling Markov chainstate space in cell i, Ji(t) denote the implementedtransmit power of cell i where Ji(t) Œ E at time t,and ywi,zi(t) be the Markov-chain state transitionprobability that the transmit power state of cell imoves from state wi to state zi at time t, wherewi, zi Œ E. The BSs are assumed to be able totransmit at one of E discrete power levels, andeach level corresponds to a state in the powerscheduling Markov chain. At each decision time,all cells have the associated power profilesdenoted J(t) = [e1, …, ei, …, eN], where Ndenotes the total number of cells in multicellnetworks, and ei denotes the transmit power ofcell i. Then the transmit power Markov-chainstate transition probability matrix of cell i isdefined as Yi(t) = [ywi,zi(t)], where Ywi,zi(t) =Pr{Ei(t + 1) = ziÔEi(t) = wi " wi}, zi Œ E. Forexample, we set the BS transmit power to haveE = 4 different power levels: 0, 5, 10, and 15 W,which are the values we will use in our numeri-cal performance analyses, described later. Also,for the smoothness and convenience of powercontrol, in this article our transmit-power con-trol can only increase or decrease the transmitpower across the neighboring transmit powerlevels. These imply that in this particular case,our power control Markov chain state space incell i is represented by {0, 5, 10, 15}, and theMarkov chain state transition can only takeplace across adjacent states.

The total operating power, denoted by P, of aBS to cover a single cell usually includes twoparts: a constant part Pconst, the power that isindependent of the number of MUs and theirdistances, and the dynamic part, which variesproportionately with the traffic intensity denotedby r (which is equal to (D – 1)), supplying thepower needed to transmit the signal from the BSto all the mobile users in a cell, and depends onr of the BSs. Therefore, the power consumptionin cell i can be calculated by Pi = Pconst + riPtran(i)[14], where ri is the traffic intensity of cell i, whichvaries over time, and Ptran(i) is the transmit powerdynamically allocated to cell i.

OBJECTIVE FUNCTIONS ANDQOS CONSTRAINTS FOR MOBILE USERS

The proposed optimal power scheduling for mul-ticell cooperation scheme is based on thestochastic control theory. Due to the time-vary-ing traffic distribution of wireless cells, we needto derive the optimal wireless-cell power schedul-ing policy to efficiently implement multicellcooperation, which aims at minimizing the totalpower consumption subject to the QoS-threshold

Figure 2. Mobile user mobility model with different classes (categorized by thedifferent radius distances from the BS to its corresponding MUs) denoted bythe three different sizes of boxes.

Cell edge BSBS



constraint of signal-to-interference-plus-noiseratio (SINR). The SINR is calculated for eachtransmission power allocation and is used todetermine the QoS achieved by an MU in themulticell cooperation network. Since the wirelesschannel quality is a time-varying function ofmultiple variables and parameters, we can deriveits corresponding SINR implemented by MU kas follows:

(1)

where gk(i) represents the SINR for mobile userk in cell i, h is the propagation coefficient, x isthe attenuation coefficient, typically rangingfrom 2 to 5, P tran(i) represents the transmitpower of cell i, dk(i) denotes the distance frommobile user k to the BS of cell i, and N0 denotesthe additive white Gaussian noise.

STOCHASTIC OPTIMIZATION FOROPTIMAL TRADE-OFF BETWEEN POWER

SAVING AND QOS PROVISIONINGUsing the system models of MU density, mobili-ty/spatial distribution, and power-profile schedul-ing described in “The System Models” section,for multicell cooperation power scheduling, weformulate the stochastic optimization problemby applying the semi-Markov decision process(SMDP) [15] model to achieve the optimaltrade-off between power saving and QoS provi-sioning for multicell cooperation networks. Thenwe solve the stochastic optimization problem byderiving a set of optimal multilevel powerscheduling policies and system reward functionsthat optimize the trade-off between power savingand QoS provisioning over multicell cooperationnetworks. Finally, we develop a concrete andefficient iteration algorithm to solve the SMDP-based stochastic optimization, which can yieldthe optimal decision/action functions to mini-mize the entire power consumption at BSs whileguaranteeing the QoS provisioning required byMUs over multicell cooperation networks.

THE SMDP MODEL AND ITSSTATE AND ACTION FUNCTIONS

We observe that the optimization of trade-offbetween the power saving at BSs and QoS provi-sioning at MUs over multicell cooperation net-works can be efficiently solved by formulating astochastic optimization problem by using theSMDP model. In particular, we can develop thestochastic reward/utility and constraint functionsto maximize the power saving subject to the con-straints of QoS provisioning. To apply the SMDPmodel, we first need to define its state space andaction functions. The SMDP state of each cell attime t is characterized by the state of user densi-ty, denoted by Ni(t), the state of users’ locationsLi(t), and the state of transmit power of eachcell, denoted by Ei(t). Then the SMDP state ofcell i can be expressed as si(t) = {Ni(t), Li(t),

Ei(t)}, and the state transition probability matrixof the SMDP model is defined as: P i(t) =[zsis¢i(t)] = [Fi(t), Qi(t), Yi(t)], where zsis¢i(t) D=fgihi(t)jUiVi(t)ywi,zi(t), which is the SMDP statetransition probability that the state of cell ichanges from state si to state s¢i.

Let S be the state space of the SMDP modelwhere S D= {D, L, E} and A are the SMDP modelaction space, denoted by A = {a1(t), a2(t), a3(t),a4(t)}, respectively, where a1(t) denotes theaction through which the central cell is turnedoff while the transmit power of neighboring cellsremain unchanged; a2(t) denotes the action thre-ough which the central cell is turned off whileincreasing (within E levels) the transmit powerof two neighboring cells through multilevelpower control to guarantee the QoS of MUs inthe original central cell; a3(t) denotes the actionthrough which the central cell remainsunchanged while the transmit power of twoneighboring cells is increased with multilevelpower control to guarantee the QoS provisioningto mobile users in the central cell; and a4(t)denotes the action through which the central celldecreases (within E levels) its transmit powerthrough multilevel power control due to thereduced MUs’ traffic load while the neighboringcells remain unchanged. Given the currentSMDP state s(t) Œ S, and the selected action a(t)Œ A, the SMDP state transition probability func-tion for the next state s(t + 1) is denoted byPr{s(t + 1)Ôs(t), a(t)}. This function is Marko-

∑γ η

[ ]=

+⎛

⎝

⎜⎜⎜

⎞

⎠

⎟⎟⎟

ξ

= ≠

P

P

ii

d i N j

( )( )

( ) ( )

ktran

k tranj j i

N

01,

Figure 3. Multicell cooperation power scheduling with time-varying mobile-users’ traffic loads in multicell cooperation networks: a) the central cell isturned off while the transmit power of neighboring cells remain unchanged; b)the central cell is turned off while increasing the transmit power of two neigh-boring cells through multilevel power control to guarantee the QoS of MUs inthe original central cell; c) the central cell remains unchanged while increas-ing the transmit power of two neighboring cells with multilevel power controlto guarantee the QoS provisioning to mobile users in the central cell; d) thecentral cell decreases its transmit power through multilevel power control dueto the reduced mobile users’ traffic load while the neighboring cells remainunchanged.

(a) (b)

(c) (d)

Mobile usersBase station



vian because the state transition probability ofthe next state is independent of the previousstates when given the current state.

In our SMDP model, at each decision epoch,the multicell cooperation power scheduling con-trol has to decide how to allocate the transmitpower to each cell according to the changes ofcell state in terms of traffic load and power pro-files. As indicated in Figs. 3a–d, the multicellcooperation with power scheduling is imple-mented by the selected action a(t) Œ A at eachSMDP model decision epoch.

THE SMDP REWARD AND POLICY FUNCTIONS FORMULTICELL COOPERATION POWER SCHEDULING

Since the objective of our proposed scheme is tominimize the total power consumption whileguaranteeing QoS provisioning for MUs, wedefine the system reward function to be thefunction of power consumption for all cells withQoS provisioning constraints. Thus, we candefine the system reward function as follows:

(2)

where P(s(t), a(t)) is the total power at all BSsof the multicell cooperation networks as a func-tion of the SMDP state and SMDP action forthe optimal trade-off between power saving andQoS guarantee for multicell cooperation powerscheduling under the current SMDP state andthe action selected in our multicell cooperationnetworks. Given the current SMDP state and theselected SMDP action, we can derive the powerconsumption P(s(t), a(t)) for our proposed mul-ticell cooperation power scheduling scheme.

The SMDP decision rules specify a mappingfunction d(t): S Æ A, which selects an actionfunction a(t) for given cells’ state at SMDP deci-sion epoch t. Thus, our SMDP policy function,denoted by p D= (d(1), d(2), …, d(t)), is asequence of decision rules that are used at alldecision epochs {1, 2, …, t}. Then we can definethe expected total reward function of the SMDP

model, which is defined as a utility functiongiven in Eq. 3.

THE STOCHASTIC POWER SAVING OPTIMIZATIONAND QOS PROVISIONING CONSTRAINTS

We need to derive an optimal policy, denoted byp*, for the optimal trade-off between power sav-ing and QoS provisioning in multicell coopera-tion networks, which maximizes the expectedtotal reward value under the given QoS-thresh-old SINR constraint on the kth mobile user’swireless channel SINR gk(i) within cell i as fol-lows:

(3)

where up*(s(t)) represents the maximum expect-ed total reward value under the optimal policyp* for the current state s(t), l is the discountfactor, gk(i) represents the SINR of user k in celli (which is defined in Eq. 1), and gth denotes thetargeted QoS-threshold SINR of MUs in a cell.

THE ITERATION ALGORITHM TO IMPLEMENT THESTOCHASTIC POWER-SAVING OPTIMIZATION

We propose to use the iterative algorithm tosolve the optimization problem specified by Eq.3. In particular, we use the iterative algorithm[15] to derive a stationary optimal policy and thecorresponding expected total reward function.The iterative algorithm is described as follows,where e denotes an infinitesimal gap, s Œ {0, 1,2, …} is the iteration steps index, and usp(s(t)) isdefined in Eq. 3.Step 1: Set u0

p(s(t)) = 0 for each state s(t). Speci-fy e > 0 and initialize s to be 0.

Step 2: For each state s(t), compute usp+1(s(t)).Step 3: If Ôusp+1(s(t)) – usp(s(t))Ô < e{(1 – l)/(2l),

then go to Step 4;Else, increase s by 1, then go to Step 2.

Step 4: For each s(t) Œ S, compute the station-ary optimal policy p*.The computational complexity of the itera-

tion algorithm is O(ÔÔAÔÔÔÔSÔÔ2) [15], where ÔÔ.ÔÔ isthe cardinality of a set. While the system statespace can be large, in the real network imple-mentation the computational and time complexi-ty can be significantly reduced because theoptimal transmit-power control policy searchingthrough our proposed iteration algorithm isimplemented by an offline approach. Once theoptimal policy is obtained corresponding to agiven space state through the offline approach, itis stored in an optimal-policy lookup table. Eachentry of the lookup table specifies the optimalaction policy for any given system state, consist-ing of the MU density state, MU location state,and BS transmit power state of each cell. Forthe online operation, at each decision epoch, themulticell cooperation power scheduling con-troller can quickly look up the table to obtainthe optimal action policy corresponding to thecurrent system state and then execute the opti-

R s t a ts t a t

( ( ), ( ))( ( ), ( ))

= 1

P

s t

R s t a t

s t s t a t s t

i

( ( )) max( ( ), ( ))

Pr{ ( 1) ( ), ( )} ( ( 1))

subject to : ( )

a

s t

k th

( 1)*

AS

∑υ

λ

υ

γ γ

=+

⋅ + +

⎧⎨⎪

⎩⎪

⎫⎬⎪

⎭⎪

≥

⎧

⎨⎪⎪

⎩⎪⎪

π

π∈+ ∈

Figure 4. The expected total reward function ups(s) vs. the SMDP optimizationiteration steps s.

Iteration steps σ5

960

950

Expe

cted

tot

al r

ewar

d fu

ncti

on

970

980

990

1000

1010

1020

1030

1040

1050

10 15 20 25 30 35 40 45 50

Our proposed SMDP scheme: execution 1Our proposed SMDP scheme: execution 2



mal decision or the optimal action policy. Thelookup table can readily be generated offline byusing separate and independent high-speed com-puter systems with the super processing capabili-ty and power supply.

PERFORMANCE EVALUATIONS

We conduct numerical analyses to evaluate theperformance of our proposed schemes based onthe SMDP model for optimal trade-off betweenpower saving at base stations and QoS provision-ing to mobile users over multicell cooperationnetworks. The multicell cooperation in ournumerical analyses follows the SMDP model-based stochastic power scheduling by using theaction functions as described in Figs. 3a–d. Usingthe SMDP model-based power scheduling opti-mization and QoS provisioning constraints andsolving for the stochastic optimization problemby the iteration algorithm described in “The Iter-ation Algorithm to Implement the stochasticPower-Saving Optimization” section, we obtain aset of numerical solution results detailed as fol-lows. Our performance evaluation model consistsof the following parameters: the constant trans-mit power of each cell is 500 W, the propagationcoefficient h is –31.54 dB, the discount factor l is0.8, and the attenuation coefficient x is 3.

Figure 4 plots the convergence process of thetwo executions of the expected total reward func-tion up(s(t)) specified in Eq. 3. We can obtain thefollowing observations from Fig. 4. First, Fig. 4shows that our SMDP-based stochastic controlprocess for optimal multicell cooperation powerscheduling is stable as it quickly converges to itsequilibrium state from any initial state. Second,Fig. 4 confirms that there is always the optimalsolution for our SMDP-based stochastic powerscheduling. Finally, Fig. 4 shows that the multipleexecutions (we only plot two of them in Fig. 4 forlack of space) of the iteration algorithm for solv-ing SMDP-based stochastic optimization con-verge to the unique optimal solution regardlessof the initial states.1 The above observationsimply that our proposed SMDP-based stochasticoptimization and its corresponding iteration algo-rithm are feasible and valid.

Figure 5 plots the power consumption func-tions against the MUs’ traffic load under threedifferent multicell cooperation power schedulingschemes, including our proposed SMDP scheme,the random scheme, and the static scheme,respectively. As shown in Fig. 5, although theaverage power consumption of these threeschemes all increase as MUs’ traffic load increas-es, our proposed SMDP scheme significantlyoutperforms the other two schemes, especiallywhen the traffic load is heavy. This is becauseour proposed SMDP scheme consumes muchless power than the other two schemes due tomulticell cooperation power scheduling. In con-trast, the static scheme statically allocates themaximum power specified by cell capacity, andthe random scheme’s average power allocation isalso much larger than that of our proposedSMDP scheme as multicell cooperation is notemployed.

Under the given power consumption dynam-ics and conditions obtained in Fig. 5, Fig. 6 plots

the implemented average SINR functions againstMUs’ traffic load also for the same three powerscheduling schemes. We observe that as theMUs’ traffic load and power consumptionincrease, the average SINR implemented by ourproposed SMDP-based scheme increases mono-tonically and is always much larger than therequired QoS-threshold SINR, as shown in Fig.6, which implies that our proposed SMDPscheme can guarantee much better QoS provi-sioning with the least power consumption, asshown in the corresponding Fig. 5 in multicellcooperation networks. Figure 6 also indicatesthat the implemented average SINR by our pro-posed SMDP scheme is much higher than thatby the static scheme, and thus, our SMDPscheme can provide much better QoS provision-ing than the static scheme. Figure 6 shows thatthe SINR implemented by our SMDP schemeand the static scheme can both guarantee therequired QoS provisioning, because they areboth larger than QoS threshold SINR as shownin Fig. 6, but the power consumed by the staticscheme is much larger than that consumed byour proposed SMDP scheme, as shown in thecorresponding Fig. 5. On the other hand, Fig. 6also shows that the average SINR implementedby the random scheme cannot guarantee theQoS provisioning requirement because its imple-mented average SINR is often lower than theQoS-threshold SINR, as shown in Fig. 6.

CONCLUSIONS

We have proposed the semi-Markov decisionprocess (SMDP) model-based stochastic opti-mization scheme for optimal trade-off betweenpower saving and QoS provisioning over multi-cell cooperation networks. Our optimizationobjective is to minimize the power consumptionat base stations while guaranteeing QoS provi-sioning to mobile users by using multicell coop-eration power scheduling. Applying thefinite-state Markov-chain techniques, we havedeveloped a set of models to characterize mobileuser density, mobility, multicell cooperationpower-profile scheduling, and QoS provisioning.

Figure 5. Power consumption against mobile users’ traffic load in multicellcooperation networks.

Mobile users’ traffic load in multicell cooperation networks0.20.1

1000

500

Pow

er c

onsu

mpt

ion

(W)

1500

2000

2500

3000

0.3 0.4 0.5 0.6 0.7 0.8 0.9

Our proposed SMDP schemeStatic schemeRandom scheme

1 We only plot two execu-tions in Fig. 4 for lack ofspace. In fact, we get thenumerical results for morethan two executions,which all converge to thesame unique optimalsolution regardless of theinitial states.



Based on these models, we have formulated theSMDP-based stochastic optimization problem tooptimize the trade-off between the power savingat BSs and QoS provisioning to mobile usersover multicell cooperation networks. We havealso developed an efficient iteration algorithm toimplement our proposed SMDP-based optimiza-tion scheme. Finally, we have conducted thenumerical analyses to evaluate our proposedschemes, which show that our proposed SMDP-based stochastic control process converges to theunique optimal solution and significantly outper-forms the other schemes in terms of minimizingthe power consumption at BSs while guarantee-ing the QoS provisioning to mobile users.

REFERENCES[1] 3GPP TR 32.826, “Telecommunication Management,

Study on Energy Savings Management (ESM),” Rel-10,Mar. 2010, http://www.3gpp.org/ftp/Specs/html-info/32826.htm.

[2] 3GPP TS 25.214, “Universal Mobile TelecommunicationsSystem (UMTS), Physical Layer Procedures (FDD),” Rel-10, 2010.

[3] IEEE Std. 802.16-2004, “Local and Metropolitan AreaNetworks, Part 16: Air Interface for Fixed BroadbandWireless Access Systems,” Oct. 2004.

[4] X. Zhang et al., “Cross-Layer Based Modeling for Quality ofService Guarantees in Mobile Wireless Networks,” IEEECommun. Mag., vol. 44, no. 1, Jan. 2006, pp. 100–06.

[5] J. Tang and X. Zhang, “Quality-of-Service Driven Powerand Rate Adaptation for Multichannel CommunicationsOver Wireless Links,” IEEE Trans. Wireless Commun.,vol. 6, no. 1, Dec. 2007, pp. 4349–60.

[6] ITU-T Focus Group on Future Networks, FG-FN OD-66,“Draft Deliverable on Overview of Energy Saving ofNetworks,” Oct. 2010.

[7] P. Marsch and G. Fettweis, “Uplink CoMP Under a Con-strained Backhaul and Imperfect Channel Knowledge,” IEEETrans. Wireless Commun., vol. 10, 2011, pp. 1730–42.

[8] S. Han et al., “On the Energy Efficiency of Base StationSleeping with Multicell Cooperative Transmission,”Proc. IEEE PIMRC, 22th IEEE Int’l. Symp. Digital ObjectIdentifier, 2011, pp. 1536–40.

[9] B. Rengarajan, G. Rizzo, and M. A. Marsan, “Bounds onQoS-Constrained Energy Savings in Cellular Access Net-works with Sleep Modes,” IEEE Proc. 23th Int’l. Tele-

traffic Congress, 2011, pp. 47–54.[10] S.-E. Elayoubi, L. Saker, and T. Chahed, “Optimal Con-

trol for Base Station Sleep Mode in Energy EfficientRadio Access Networks,” Proc. IEEE INFOCOM, 2011,pp. 106–10.

[11] K. Dufkova et al., “Energy Savings for Cellular Networkwith Evaluation of Impact on Data Traffic Perfor-mance,” Proc. Euro. Wireless, 2010, pp. 916–23.

[12] M. Anderberg, Cluster Analysis for Applications, Aca-demic Press, 1973.

[13] A. Jain and R. Dubes, Algorithms for Clustering Data,Prentice Hall, 1988.

[14] I. Humar et al., “Energy Saving Modeling and PerformanceAnalysis in Multi-Power-State Base Stations Systems,” Proc.GreenCom, IEEE/ACM Int’l. Conf. Cyber, Physical and SocialComputing (CPSCom), 2010, pp. 474–78.

[15] M. Puterman, Markov Decision Processes: DiscreteStochastic Dynamic Programming, Wiley, 1994.

BIOGRAPHIESXI ZHANG [S’89, SM’98] ([email protected]) receivedhis Ph.D. degree in electrical engineering and computerscience (electrical engineering-systems) from the Universi-ty of Michigan, Ann Arbor. He is currently an associateprofessor and the founding director of the Networkingand Information Systems Laboratory, Department of Elec-trical and Computer Engineering, Texas A&M University,College Station. He was a research fellow with the Schoolof Electrical Engineering, University of Technology, Syd-ney, Australia, and the Department of Electrical and Com-puter Engineering, James Cook University, Australia. Hewas with the Networks and Distributed Systems ResearchDepartment, AT&T Bell Laboratories, Murray Hill, New Jer-sey, and AT&T Laboratories Research, Florham Park, NewJersey, in 1997. He has published more than 230 researchpapers on wireless networks and communications sys-tems, network protocol design and modeling, statisticalcommunications, random signal processing, informationtheory, and control theory and systems. He received theU.S. National Science Foundation CAREER Award in 2004for his research in the areas of mobile wireless and multi-cast networking and systems. He is an IEEE DistinguishedLecturer. He received Best Paper Awards at IEEE GLOBE-COM 2007, IEEE GLOBECOM 2009, and IEEE WCNC 2010,respectively. He also received a TEES Select Young FacultyAward for Excellence in Research Performance from theDwight Look College of Engineering at Texas A&M Univer-sity, College Station, in 2006. He is serving or has servedas an Editor for IEEE Transactions on Communications,IEEE Transactions on Wireless Communications, and IEEETransactions on Vehicular Technology, twice as a GuestEditor for IEEE Journal on Selected Areas in Communica-tions for two special issues on “Broadband Wireless Com-munications for High Speed Vehicles” and “WirelessVideo Transmissions,” an Associate Editor for IEEE Com-munications Letters, a Guest Editor for IEEE Communica-tions Magazine and IEEE Wireless Communications, anEditor for Wiley’s Journal on Wireless Communicationsand Mobile Computing, Journal of Computer Systems,Networking, and Communications, and Wiley’s Journal onSecurity and Communications Networks, and an Area Edi-tor for Elsevier’s Journal on Computer Communications,among others. He is serving or has served as TPC Chairfor IEEE GLOBECOM 2011, TPC Vice-Chair IEEE INFOCOM2010, TPC Area Chair for IEEE INFOCOM 2012,Panel/Demo/Poster Chair for ACM MobiCom 2011, Gener-al Vice-Chair for IEEE WCNC 2013, Panel/Demo/PosterChair for ACM MobiCom 2011, and TPC/General Chair fornumerous other IEEE/ACM conferences, symposia, andworkshops.

PING WANG received her B.S. degree in communicationengineering from Henan University, China, in 2007. She iscurrently a Ph.D. student studying in the joint Ph.D. pro-gram between the Department of Electrical and ComputerEngineering, Texas A&M University, and the School of Elec-tronic Engineering, Beijing University of Posts and Telecom-munications, China. She has published 11 papers in IEEEjournals and conference proceedings. Her research interestsare in wireless information and communication theory,underwater wireless networks, cooperative relay networks,QoS provisioning, and wireless resource management.

Figure 6. The implemented average SINR of different schemes against themobile-users' traffic load in the multicell cooperation networks.

Mobile users’ traffic load in multicell cooperation networks0.2

13

12

Impl

emen

ted

aver

age

SIN

R (d

B)

14

15

16

17

18

19

20

21

22

0.3 0.4 0.5 0.6 0.7 0.8 0.9

Our proposed SMDP schemeStatic schemeRandom schemeQoS-threshold SINR



http://www.3gpp.org/ftp/Specs/html-info/32826.htm

COMMUNICATIONSTECHNOLOGY

Visit www.wiley.com and quote code VB781* when you order

20% Discount On These New Books

www.twitter.com/WBComms

Sign up for email alerts. Visit:www.wiley.com/commstech

*Discount valid until 17/03/13

WileyCommunicationsTechnology

Connect with us:

13-512

07

IENYCM3225.indd 1IENYCM3225.indd 1 1/18/13 10:18 PM1/18/13 10:18 PM

http://www.wiley.com

http://www.twitter.com/WBComms

http://www.wiley.com/commstech

U L T R A - H I G H E F F I C I E N C Y RF PLATFORMS

CA

RR

IE

RA

GG

RE

GAT

ION O P T I M I Z E D

F O R U S E W I T HMULTIPLE BASEBANDS

RF PLATFORM COVERS ALL FD-LTE, TD-LTE, HSPA, CDMA EV-DO, AND EDGE FREQUENCY BANDS IN COMMERCIAL U S E TO DAY, F R O M B A N D S 1 - 41

LOW INSERTION LOSS PRIMARY & DIVERSITY T X / R X P A T H SLT

ECOMPLETE RF SOLUTIONS—INCLUDING POWER AMPLIFIER, SWITCH, FILTER, AND ANTENNA TUNING CONTENT—FOR NEXT-GENERATION LTE SMARTPHONES AND TABLETS

NEXT-GENERATION A N T E N N A T U N I N G , ENABLING SMALLER, HIGH PERFORMANCE ANTENNAS

S C A L A B I L I T Y

E N V E L O P ET R A C K I N G

IENYCM321indd 1IENYCM3214_402162 1/8/13 8:42 PM1/8/13 8:42 PM

ieee wireless magazine

Documents

edition of lte

ieee wireless communications

g wireless

wireless networkspaul

advanced testing

evolution ofcoordination

wide area systems coordination

ieee computer