2010_Wang_ComMag_02_80

INTRODUCTION

Multiple-input multiple-output (MIMO) is anadvanced technology that can effectively exploitthe spatial domain of mobile fading channels tobring significant performance improvements towireless communication systems. ConventionalMIMO systems, known as point-to-point MIMOor collocated MIMO, require both the transmit-ter and receiver of a communication link to beequipped with multiple antennas. In practice,however, many wireless devices may not be ableto support multiple antennas due to size, cost,and/or hardware limitations. Cooperative MIMO[1], also known as virtual or distributed MIMO,aims to utilize distributed antennas on multipleradio devices to achieve some benefits similar tothose provided by conventional MIMO systems.The basic idea of cooperative MIMO is to groupmultiple devices into virtual antenna arrays

(VAAs) to emulate MIMO communications. Acooperative MIMO transmission involves multi-ple point-to-point radio links, including linkswithin a VAA and links between possibly differ-ent VAAs. For relay-based cooperative MIMOcommunications, there are three main coopera-tive strategies: amplify-and-forward, decode-and-forward, and compress-and-forward techniques.

Previous theoretical studies have revealed thepros and cons of cooperative MIMO comparedto point-to-point MIMO systems [1]. The disad-vantages of cooperative MIMO come from theincreased system complexity and the large signal-ing overhead required for supporting devicecooperation. The advantages of cooperativeMIMO, on the other hand, are due to its capa-bility to improve the capacity, cell edge through-put, coverage, and group mobility of a wirelessnetwork in a cost-effective manner. These advan-tages hinge on the usage of distributed antennas,which can increase the system capacity by de-correlating the MIMO channels and allow thesystem to exploit the benefits of macro-diversityin addition to micro-diversity. In many practicalapplications, such as cellular mobile and wirelessad hoc networks, the advantages of deployingcooperative MIMO technology usually outweighthe disadvantages [2, 3]. Over recent years,cooperative MIMO technologies have beengradually adopted into the mainstream of wire-less communication standards.

In this article we focus on cooperative MIMOin the context of cellular mobile systems. Threetypes of cooperative MIMO schemes have beenproposed for cellular systems: coordinated multi-point transmission (CoMP) [2], fixed relay [3],and mobile relay [3]. Simple illustrations of thesecooperative MIMO schemes are shown in Fig.1a–c. As the most mature cooperative MIMOtechnology, fixed relays have been incorporatedinto the IEEE 802.16j WiMAX standard. Mean-while, the Third Generation Partnership Project(3GPP) is currently evaluating cooperativeMIMO technologies for its fourth-generation(4G) cellular communication standard, LongTerm Evolution-Advanced (LTE-Advanced).

IEEE Communications Magazine • February 201080 0163-6804/10/$25.00 © 2010 IEEE

ABSTRACT

Cooperative multiple-input multiple-outputtechnology allows a wireless network to coordi-nate among distributed antennas and achieveconsiderable performance gains similar to thoseprovided by conventional MIMO systems. Itpromises significant improvements in spectralefficiency and network coverage and is a majorcandidate technology in various standard pro-posals for the fourth-generation wireless com-munication systems. For the design and accurateperformance assessment of cooperative MIMOsystems, realistic cooperative MIMO channelmodels are indispensable. This article providesan overview of the state of the art in cooperativeMIMO channel modeling. We show thatalthough the existing standardized point-to-pointMIMO channel models can be applied to a cer-tain extent to model cooperative MIMO chan-nels, many new challenges remain in cooperativeMIMO channel modeling, such as how to modelmobile-to-mobile channels, and how to charac-terize the heterogeneity and correlation of multi-ple links at the system level appropriately.

LTE TECHNOLOGY UPDATE

Cheng-Xiang Wang and Xuemin Hong, Heriot-Watt University

Xiaohu Ge, Huazhong University of Science and Technology

Xiang Cheng, Heriot-Watt University

Gong Zhang, Huawei Technologies Company Ltd.

John Thompson, The University of Edinburgh

Cooperative MIMO Channel Models:A Survey

WANG LAYOUT 1/19/10 12:00 PM Page 80

Authorized licensed use limited to: Heriot-Watt University. Downloaded on February 7, 2010 at 17:27 from IEEE Xplore. Restrictions apply.

IEEE Communications Magazine • February 2010 81

To accurately evaluate and compare the per-formance of different transmission technologiesfor cooperative MIMO cellular systems, realisticcooperative MIMO channel models are indis-pensable. Earlier theoretical research on cooper-ative MIMO systems utilized simple narrowbandfading channel models such as Rayleigh, Rician,or Nakagami channel models [1]. These analyti-cal channel models are oversimplified and clearlyinappropriate for system-level performance simu-lations. The existing standardized MIMO channelmodels, such as the 3GPP spatial channel model(SCM) [4], the WINNER II channel model [5],and the IEEE 802.16j channel model [6], can beused to simulate individual point-to-point chan-nels. However, since cooperative MIMO involvesmultiple point-to-point links, its channel modelshould consider not only the properties of theindividual links, but also the system-level varia-tions (or heterogeneity) and correlation of multi-ple links in a multicell environment. Also, forcooperative MIMO using mobile relays, realisticmobile-to-mobile (M2M) channel models shouldbe developed for channels between mobile users.

This article provides an overview of the exist-ing cooperative MIMO channel models. Differ-ent types of cooperative MIMO schemes incellular systems are presented in the next section.We then review and compare major standardizedMIMO channel models applicable to cooperativeMIMO systems. Future challenges in cooperativeMIMO channel modeling are subsequently iden-tified. The final section concludes the article.

COOPERATIVE MIMO INCELLULAR SYSTEMS

COMPCoMP [2] is an emerging technique to combatintercell interference and improve cell edge per-formance. The system architecture is illustrated inFig. 1a. The idea is to share data and channelstate information (CSI) among neighboring basestations (BSs) to coordinate their transmissions inthe downlink and jointly process the received sig-nals in the uplink. CoMP techniques can effec-tively turn otherwise harmful intercell interferenceinto useful signals, allowing significant powergain, channel rank advantage, and/or diversitygains to be exploited. The promised advantages ofCoMP rely on a high-speed backbone enablingthe exchange of information (e.g., data, controlinformation, and CSI) between the BSs.

CoMP has been studied intensively in bothacademia and industry over recent years, and isa very strong candidate technology for 4G stan-dards. CoMP systems are only concerned withthe BS to mobile station (MS) channels, whichare fixed-to-mobile (F2M) channels.

FIXED RELAYSAs illustrated in Fig. 1b, fixed relays are low-costand fixed radio infrastructures without wiredbackhaul connections. They store data receivedfrom the BS and forward to the MSs, and viceversa. Fixed relay stations (RSs) typically havesmaller transmission powers and coverage areasthan a BS. They can be deployed strategically andcost effectively in cellular networks to extend cov-

erage, reduce total transmission power, enhancethe capacity of a specific region with high trafficdemands, and/or improve signal reception. Bycombining the signals from the relays and possiblythe source signal from the BS, the MS is able toexploit the inherent diversity of the relay channel.The disadvantages of fixed relays are first theadditional delays introduced in the relaying pro-cess, and second the potentially increased levelsof interference due to frequency reuse at the RSs.

As the most mature cooperative MIMO tech-nology, fixed relay has attracted significant sup-port in major cellular communication standards.The IEEE 802.16j specification has incorporatedfixed relays to enhance WiMAX performance.Fixed relay is also a very strong candidate tech-nology for 4G, with possible mesh extensions ina later standard release. As shown in Fig. 1b,fixed relay systems involve different types oflinks, including BS-MS, BS-RS, RS-RS, and RS-MS links. The BS-RS and RS-RS channels arefixed-to-fixed (F2F) channels, while the BS-MSand RS-MS channels are F2M channels.

Figure 1. Three types of cooperative MIMO schemes in cellular systems: a)CoMP; b) fixed relay; c) mobile relay.

(a)

(b)

(c)

a1

a2

c1

c2



IEEE Communications Magazine • February 201082

MOBILE RELAYS

Mobile relays differ from fixed relays in the sensethat the RSs are mobile and are not deployed asthe infrastructure of a network. Mobile relays aretherefore more flexible in accommodating vary-ing traffic patterns and adapting to differentpropagation environments. For example, when atarget MS temporarily suffers from bad channelconditions or requires relatively high-rate service,its neighboring MSs can help to provide multihopcoverage or increase the data rate by relayinginformation to the target MS. Moreover, mobilerelays enable faster and lower-cost network roll-out. Similar to fixed relays, mobile relays canenlarge the coverage area, reduce the overalltransmit power, and/or increase the capacity atcell edges. On the other hand, due to their oppor-tunistic nature, mobile relays are less reliablethan fixed relays since the network topology ishighly dynamic and unstable.

As shown in Fig. 1c, two types of mobile relaysystems can be distinguished: moving networks andmobile user relays. The moving network employsdedicated RSs on moving vehicles (e.g., trains) toreceive data from the BS and forward to the userMSs onboard, and vice versa. The purpose of themoving network is to improve coverage on thevehicle. The mobile user relay enables distributedMSs to self-organize into a wireless ad hoc net-work, which complements the cellular networkinfrastructure using multihop transmissions. Theo-retical studies have shown that mobile user relayshave a fundamental advantage in that the total net-work capacity, measured as the sum of the through-puts of the users, can scale linearly with the numberof users given sufficient infrastructure supports.Mobile user relays are therefore a desirableenhancement to future cellular systems. However,mobile user relays also face huge challenges inrouting, radio resource management, and interfer-ence management. The major disadvantage ofmobile user relays is that MS batteries can be usedup by relay transmissions even if the user does notuse them. Mobile user relays also complicate thebilling problem (i.e., who shall pay the bill when auser helps other users as a relay).

Currently, moving networks are supported bythe IEEE 802.16j WiMAX standard. A numberof advanced mobile relays concepts are beingevaluated for the 4G standard. Moving networksinvolve the BS-MS, BS-RS, and RS-MS links,where the BS-MS and BS-RS channels are F2Mchannels, while the RS-MS channels are F2F orF2M channels. Mobile user relays involve BS-MS and MS-MS links, where the BS-MS chan-nels are F2M channels, while the MS-MSchannels are M2M channels.

COOPERATIVE MIMOCHANNEL MODELS

To allow accurate assessments and fair compar-isons of different cooperative MIMO technolo-gies, realistic cooperative MIMO channel modelsare indispensable. Although a standardized coop-erative MIMO channel model is not yet available,it can be constructed with reasonable accuracybased on the existing standardized point-to-point

MIMO channel models and recently developedmodels for some scenarios (e.g., M2M models).This section aims to review the existing standard-ized MIMO channel models, compare their prosand cons, and discuss how they can be extendedand applied to cooperative MIMO systems.

Standardized channel models are developedfor different purposes, based on which we canclassify them into two types: system simulationmodels and calibration models. System simula-tion models are intended for accurate perfor-mance assessments of different algorithms andsystems. They seek to represent the real-worldchannels as realistically as possible, with reason-able computational complexity. Calibrationchannel models, on the other hand, are simpli-fied channel models developed for conformancetesting of different products and technologies.Conformance testing specifies a required level ofsystem performance when using the calibrationchannel model and indicates whether a testedsystem fulfills this requirement. In this article weconcentrate on system simulation models.

Prominent standardized MIMO channel mod-els include the COST 259/273 channel model,the 3GPP SCM, the SCM-Extended model, theWINNER II channel model, the IEEE 802.11nchannel model, the Stanford University interim(SUI) channel model, and the IEEE 802.16channel model [7]. Although these models aredeveloped by different research bodies and pro-jects, they are closely related. In the followingwe focus on the three most representativeMIMO channel models:• The SCM [4], which is a cellular MIMO

channel model developed by the 3GPP in2003 for the evaluation of different MIMOschemes in high-speed downlink PacketAccess (HSDPA) systems. The SCM sup-ports a center frequency of 2 GHz and asystem bandwidth of 5 MHz.

• The WINNER II model [5], which wasdeveloped by the IST-WINNER II projectin 2007 and represents the state of the artin wireless channel modeling. It supports acenter frequency range from 2 to 6 GHzand a system bandwidth up to 100 MHz.

• The IEEE 802.16j channel model [6], whichis based on the SUI model and was extend-ed in 2007 to cover relay scenarios in IEEE802.16j WiMAX systems. It supports a cen-ter frequency of 5 GHz and a maximumsystem bandwidth of 20 MHz.

GENERAL FEATURESAll three models employ a tapped-delay-linestructure to construct the wideband channelimpulse response, with each tap representing aresolvable path or a cluster of scatterers with adifferent delay. The numbers of taps used by theSCM and IEEE 802.16j channel model are fixedto 6 and 3, respectively, while the numbers oftaps in the WINNER II channel model can varybetween 4 and 24. Intracluster delay spread isfurther introduced in the WINNER II channelmodel to account for better delay resolution andconsequently broader bandwidths.

Based on the modeling methodology, standard-ized MIMO channel models can be roughly classi-fied into geometry-based stochastic models

Although a

standardized

cooperative MIMO

channel model is not

yet available, it can

be constructed with

a reasonable

accuracy based on

the existing

standardized

point-to-point MIMO

channel models and

recently developed

models for some

scenarios.




(GBSMs) and correlation-based stochastic models(CBSMs). The SCM and Winner II channel mod-els are GBSMs, while the IEEE 802.16j channelmodel is a CBSM. Both modeling approacheshave pros and cons. GBSMs simulate the geomet-rical propagation of multipaths using a large num-ber of random variables as model parameters. Asimplified plot of the SCM is given in Fig. 2,where only one cluster of scatterers is shown as anexample with corresponding random angle param-eters associated with the BS and MS. The detailedexplanations of the angle parameters can be foundin [4, 8] and are neglected here for simplicity. Acluster/tap or resolvable path further consists ofmultiple subpaths (i.e., irresolvable rays) with dif-ferent amplitudes, phases, angles of arrival, andangles of departure. The sum-of-sinusoids methodis used to sum up the irresolvable rays and gener-ate taps. With proper parameterization, GBSMsare usually accurate and flexible for describing dif-ferent scenarios. However, they are complex andcomputationally inefficient. Besides, the spatial-temporal correlation properties are not explicitlyspecified in the model and need to be furtherderived from other parameters [8]. This meansthat theoretic performance studies of MIMO sys-tems cannot be directly linked to the underlyingspatial-temporal correlation properties. On theother hand, CBSMs assume Rayleigh or Riceanfading for each tap, and then carefully introduceautocorrelation and cross-correlation into theMIMO channel matrix via Doppler filtering andspatial filtering. CBSMs are relatively simple andcomputationally efficient. They also explicitlyaddress the correlation properties of MIMO chan-nel matrices, providing a useful reference for sys-tem designers. The drawback of CBSMs comesfrom the oversimplification and consequently anunrealistic representation of link variations in sys-tem-level simulations. In general, GBSMs tend tobe better system simulation models, while CBSMstend to be better calibration models [8].

HETEROGENEITY OF MULTIPLE LINKSConventional cellular systems and CoMP are onlyconcerned with the BS-MS links. In fixed andmobile relay cooperative MIMO systems, morelinks need to be considered: BS-RS, RS-RS, RS-MS, and MS-MS links. Because the BSs, RSs, and

MSs have different heights, densities, movingspeeds, and transmission ranges, different links ina cooperative MIMO system can have distinct sta-tistical properties in terms of the path loss, shad-owing standard deviation, delay spread, Dopplerspread, line-of-sight (LoS) probability, and so on.In other words, a high degree of link heterogene-ity is expected in cooperative MIMO systems.

The SCM, WINNER II, and IEEE 802.16jchannel models all support multiple scenarios toreflect significantly varying channel propertiesacross typical environments. Some scenarios fur-ther differentiate LoS and non-LoS (NLoS) con-ditions, leading to the received signal oftenfollowing Rayleigh and Rice distributions,respectively. The resulting channel models canbe used to characterize the link heterogeneity incooperative MIMO systems.

Multiple Scenarios — The SCM [4] was dedi-cated to outdoor propagation and defined threemain propagation scenarios: suburban macrocell,urban macrocell, and urban microcell. The urbanmicrocell scenario supports both LoS and NLoSconditions. Additionally, it is possible to modifythe urban macrocell scenario by applying a far-scattering cluster or an urban canyon option.

The WINNER II channel model [5] is themost comprehensive channel model and includes13 scenarios:• A1: indoor (small office/residential), A2:

indoor to outdoor• B1: urban microcell, B2: bad urban micro-

cell, B3: indoor hotspot, B4: outdoor toindoor, B5: stationary feeder

• C1: suburban macrocell, C2: urban macro-cell, C3: bad urban macrocell, C4: urbanmacrocell outdoor to indoor

• D1: rural macrocell, D2: moving networksAll 13 scenarios can distinguish LoS and

NLoS conditions. It is worth noting that the B5scenario in WINNER II corresponds to fixedrelays (i.e., stationary feeders). Because fixedrelays can be deployed above the rooftop (ART),below the rooftop (BRT), or at street level, theB5 scenario is further divided into five sub-sce-narios: B5a (LoS, ART to ART), B5b (LoS,street level to street level), B5c (LoS, BRT tostreet level), B5d (NLoS, ART to street level),

Figure 2. BS and MS angle parameters in the 3GPP SCM with one cluster of scatterers [4, 8].

N

N

v

ΩBS

ΩMS

θBS

θMS

θv

θn,m,AoD

θn,m,AoA

Δn,m,AoD

Δn,m,AoA

δn,AoD

δn,AoA

BS array

MS array MS array broadside

BS array broadside MS direction of travel

Cluster n

Subpath m

The SCM, WINNER

II, and IEEE 802.16j

channel models all

support multiple

scenarios to reflect

significantly varying

channel properties

across typical

environments.




and B5f (LoS/NLoS, ART to ART/BRT). More-over, the D2 channel corresponds to the movingnetwork scenario in mobile relay networks.

IEEE 802.16j presents nine types of channelmodels, Types A–H and Type J [6], commonlyexperienced in WiMAX multihop relay net-works. Similar to the B5 sub-scenarios in WIN-NER II, the IEEE 802.16j distinguishes twodifferent locations of BSs and RSs: ART andBRT. Three terrain features are defined byIEEE 802.16j: hilly terrain with moderate toheavy tree densities, intermediate path loss con-dition, and flat terrain with light tree densities.Types A, B, C, F, and G support both LoS andNLoS conditions, Types D and H support onlythe LoS condition, while Types E and J supportonly the NLoS condition.

The different scenarios described above canbe adopted to simulate most of the links in coop-erative MIMO systems. In [6] four relay usagemodels or application scenarios are defined:fixed infrastructure, in-building coverage, tempo-rary coverage, and coverage on a mobile vehicle.A mapping of the nine types of channel modelsto the four relay usage models was provided in[6]. By analogy, we can selectively apply the 13scenarios defined in the WINNER II channelmodel to construct a cooperative MIMO chan-nel model. An example is given in Table 1.

LoS Probability and Scenario Transition —The probability that LoS or NLoS links occur ina system depends on various environment fac-tors (e.g., terrain features and distance). Proba-bilistic models are employed by the SCM,WINNER II, and IEEE 802.16j channel modelsto describe the LoS probability. In the SCM anLoS model is defined for the urban microcell

scenario only and not for the suburban or urbanmacrocell cases due to low probabilities of LoSoccurrence. The LoS probability in the microcellscenario decreases linearly with the propagationdistance and reduces to 0 at a distance of 300 m[4]. The WINNER II channel model uses a setof scenario-dependent expressions for the LoSprobability, given that neither the transmitternor the receiver is inside a building. Theseexpressions are simple functions of a few param-eters, including the horizontal distance of thepropagation path, BS height, MS height, as wellas parameters that characterize the radio envi-ronment [5]. The IEEE 802.16j channel modeldefines two LoS models for Type F and Type Gscenarios. The LoS probability is expressed as anonlinear function of the transmitter-receiverdistance with cutoff points [6].

The time evolution behavior of mobile fadingchannels, including transitions between differentscenarios and LoS/NLoS conditions, may have aconsiderable impact on system performance. TheSCM supports the time evolution feature basedon a concept called drops. A drop correspondsto a local stationary interval, during which thechannel undergoes fast fading but the large-scaleparameters do not change significantly. A quasi-stationary framework is used which assumes thatparameters in consecutive drops are indepen-dent [4]. The WINNER II channel model alsouses a concept called channel segments, similar tothe drops concept. A non-stationary frameworkis used to support smooth channel evolution.Transitions from an old segment to a new seg-ment are carried out by linearly decreasing thepowers of clusters in the old segment andincreasing the powers of clusters in the new seg-ment so that the old clusters are replaced one by

Table 1. A cooperative MIMO channel model based on the WINNER II channel model.

Cooperative MIMO scheme Link type Description Recommended scenario

CoMP BS-MSMS indoor C2 (NLoS)

MS outdoor C2 (LoS)

Fixed relay BS-MS Indoor or outdoor C2

BS-RS Various RS locations B5 (LoS/NLoS)

RS-RS Various RS locations B5 (LoS/NLoS)

RS-MS

Indoor-to-indoor A1 (LoS/NLoS)

Indoor-to-outdoor A2 (NLoS)

Outdoor-to-outdoor B1 (LoS/NLoS)

Moving network

BS-MS Indoor C2 (NLoS)

BS-RS LoS for RS C2 (LoS)

RS-MS Indoor B3 (LoS/NLoS)

Mobile user relayBS-MS Indoor or outdoor C2

MS-MS LoS B5b (not aN M2M scenario)

The probability that

LoS or NLoS links

occur in a system

depends on various

environment factors,

e.g., terrain features

and distance.

Probabilistic models

are employed by the

SCM, WINNER II,

and IEEE 802.16j

channel models to

describe the LoS

probability.




one by the new clusters [5]. In the SCM andWINNER II channel models, the scenario tran-sition and LoS/NLoS transition can be simulatedby changing the scenarios and LoS/NLoS condi-tions in consecutive drops. The IEEE 802.16jmodel does not explicitly address the time evolu-tion behavior of channels.

CORRELATION OF MULTIPLE LINKSDue to the environment similarity arising fromcommon shadowing objects and scatterers con-tributing to different links, the parametersdescribing different links may exhibit certaindependence/correlation from the system point ofview. The system-level correlation is closely relat-ed to the deployment assumptions such as theheights, densities, and distances of the transmit-ters and receivers. At the system level, two typesof correlation can be identified: intrasite correla-tion and intersite correlation. A site is a highlyelevated radio station, including all BSs and ARTRSs. Intrasite correlation refers to the correla-tion between MSs connected to a site (e.g., thecorrelation between links c1 and c2 in Fig. 1c).Intersite correlation refers to the correlation oflinks from a single MS to multiple sites (e.g., thecorrelation between links a1 and a2 in Fig. 1a).

Both intersite and intrasite correlations need tobe considered properly in the cooperative MIMOchannel model to allow accurate performanceevaluation of cooperative MIMO systems.

The correlation of lognormal shadow fading(SF) is perhaps the most important system-levelcorrelation because it directly influences themacro-diversity gain, which is a major benefitpromised by cooperative MIMO systems. TheSCM, WINNER II, and IEEE 802.16j modelshave all considered SF correlation. In the SCMthe intrasite and intersite SF correlations are fixedto 0 and 0.5, respectively. In the WINNER IImodel the intrasite SF correlation is a cut-offexponential decaying function of the distancebetween the two MSs, while the intersite SF corre-lation is fixed to 0. The IEEE 802.16j model usesa distance-dependent exponential decaying func-tion to describe the intrasite SF correlation (calledautocorrelation in [6]), and a distance-and-angle-dependent function to describe the intersite SFcorrelation (called cross-correlation in [6]).

Besides the SF, other large-scale parameters(LSPs) such as delay spread (DS) and azimuthspread (AS) may also display intrasite correla-tion. The pair-wise intrasite correlations of SF,DS, and AS are determined by fixed values in the

Table 2. Major comparisons of the SCM, WINNER II, and IEEE 802.16j channel models.

SCM [4] WINNER II [5] IEEE 802.16j [6]

General model features

Channel model type GBSM GBSM CBSM

Maximum bandwidth 5 MHz 100 MHz 20 MHz

Carrier frequency 2 GHz 2–6 GHz 5 GHz

Number of taps 6 4–24 3

Intercluster delay spread No Yes No

Heterogeneity of multiple links

Number of scenarios 3 13 9

(i) Outdoor-to-indoor No Yes Yes

(ii) Fixed relay No Yes Yes

(iii) Moving network No Yes Yes

(iv) Mobile user relay No No No

LoS probability model Urban microcell only Scenario-dependent Scenario-dependent

Scenario transition Quasi-stationary Non-stationary No

Correlation of multiple links

Intrasite SF correlation 0 Distance-dependent Distance-dependent

Intersite SF correlation 0.5 0 Distance-and-angle-dependent

Correlation of other LSPs Fixed values Distance-dependent Not considered

In the SCM and

WINNER II channel

models, the scenario

transition and

LoS/NLoS transition

can be simulated by

changing the

scenarios and

LoS/NLoS conditions

in consecutive drops.

The IEEE 802.16j

model does not

explicitly address the

time evolution

behavior of channels.




SCM and by distance-dependent functions in theWINNER II model. The correlation of DS andAS is not considered in the IEEE 802.16j model.

The major comparisons of the SCM, WIN-NER II, and IEEE 802.16j channel models aresummarized in Table 2.

CHALLENGES IN COOPERATIVEMIMO CHANNEL MODELING

While many issues in cooperative MIMO chan-nel modeling have been covered by the existingstandardized point-to-point MIMO channelmodels, there are still some challenges thatremain to be addressed.

The first challenge is how to implement a chan-nel model that can run multiple scenarios simulta-neously. As shown in Table 2, cooperative MIMOchannels should rely on different scenarios tocharacterize different links. Providing a pool of 13scenarios, the WINNER II channel model is obvi-ously the best existing channel model to use. How-ever, the current implementation of the WINNERII channel model does not allow simultaneoussimulation of multiple links in different scenarios.Instead, to evaluate the performance of a networkcontaining different scenarios, multiple dropsshould be run, one for each scenario, and mergedafterward. This approach can be justified for con-ventional point-to-point MIMO systems, where anend-to-end transmission involves only one link.However, this approach cannot be used in cooper-ative MIMO systems, where an end-to-end trans-mission involves multiple heterogeneous linkssimultaneously.

Another important issue to address is the corre-lation model for multiple links at the system level.Table 2 has summarized the correlation models forLSPs used in standardized MIMO channel models.We can see that these correlation models are notconsistent and even contradict each other. Forexample, the SCM treats the intrasite SF correla-tion as negligible and the intersite SF correlation assignificant, while the WINNER II channel modelconsiders the opposite. It is therefore desirable to

develop a unified correlation model for LSPs,incorporating a new realistic SF correlation model,such as the one in [9]. Besides the LSP correlation,the small-scale fading of multiple links in coopera-tive MIMO systems may also be correlated, evenwith largely separated antennas. A simple GBSMcan be used to gauge the extent of small-scale fad-ing correlation. As illustrated in Fig. 3, two links areestablished from an MS to BS1 and BS2 in a CoMPsystem. Based on the classic one-ring model, scat-terers are assumed to be distributed on a circle ofradius R centered at the MS, following the well-known von-Mises distribution. This cluster of scat-terers contributes to a tap in the channel impulseresponse. Without loss of generality, we assumethat the carrier frequency is 5 GHz, the clustermean angle of arrival is 90° and the cluster azimuthspread at arrival is 22°, corresponding to the urbanmicro-cell scenario with the NLoS condition in theWINNER II channel model [5]. The MS-BS1 andMS-BS2 distances are denoted D1 and D2, respec-tively. The angle between the two links is denotedθ. Assuming D1 = 500 m and R = 30 m, the small-scale fading correlation of the two links is calculat-ed numerically as a function of D2 and θ. Theabsolute values of the resulting correlation coeffi-cient are shown in Fig. 4. We can see that althoughthe two BSs are far apart, high correlations betweenthe two links can occur at certain distances and cer-tain values of θ (e.g., when θ is small). Such small-scale fading correlation should not be neglected,and new small-scale fading correlation models forcooperative MIMO systems are needed.

The existing scenarios defined in the stan-dardized MIMO channel models need to beexpanded to include the M2M scenario. Asshown in Fig. 1c, the M2M channel is an impor-tant type of channel in mobile user relay sys-tems. In M2M systems both the transmitter andreceiver are MSs in motion, often equipped withlow-elevation antennas. M2M channels thereforediffer significantly from conventional F2M chan-nels, where only the MS is moving while the BSis fixed, often with high-elevation antennas. Forexample, the Doppler power spectrum density inM2M channels may exhibit different shapes, thedensity of mobile users has a great impact on theunderlying M2M channel statistics, and M2Mchannels are in most cases statistically non-sta-tionary, since the stationarity conditions pertainto a much shorter time period than in F2Mchannels [10]. For the state of the art and futurechallenges in M2M channel modeling and mea-surements, we refer the interested readers to[10, references therein]. Currently, none of theexisting standardized point-to-point MIMOchannel models includes M2M scenarios.

CONCLUSIONSThis article provides a brief survey of coopera-tive MIMO channel models applicable to cellu-lar mobile networks. First, we have brieflyintroduced three different cooperative MIMOschemes and illustrated the corresponding prop-agation scenarios. We have then reviewed theexisting standardized point-to-point MIMOchannel models, compared their pros and cons,and discussed how they can be extended andapplied to cooperative MIMO systems. New

Figure 3. A one-ring GBSM for a CoMP system.

R

BS2

BS1

D2

D1

MS

θ




challenges in cooperative MIMO channel model-ing have been identified and discussed, such ashow to develop realistic M2M channel modelsand how to characterize heterogeneity and cor-relation of multiple links in multi-cell environ-ments. Our discussions will hopefully inspirefuture efforts to develop standardized and morerealistic cooperative MIMO channel models.

ACKNOWLEDGMENTThe authors acknowledge the support from theRCUK for the UK-China Science Bridges Project:R&D on (B)4G Wireless Mobile Communica-tions. C.-X. Wang, X. Hong, X. Cheng, and J. S.Thompson acknowledge the support from theScottish Funding Council for the Joint ResearchInstitute in Signal and Image Processing betweenthe University of Edinburgh and Heriot-Watt Uni-versity, which is a part of the Edinburgh ResearchPartnership in Engineering and Mathematics(ERPem). X. Ge also acknowledges the supportfrom the National Natural Science Foundation ofChina (NSFC) (contract/grant number 60872007)and National 863 High Technology Program ofChina (contract/grant number 2009AA01Z239).

REFERENCES[1] Y. Fan and J. S. Thompson, “MIMO Configurations for

Relay Channels: Theory and Practice,” IEEE Trans. Wire-less Commun., vol. 6, no. 5, May 2007, pp. 1774–86.

[2] M. K. Karakayali, G. J. Foschini, and R. A. Valenzuela,“Network Coordination for Spectrally Efficient Commu-nications in Cellular Systems,” IEEE Commun. Mag., vol.44, no. 8, Aug. 2006, pp. 56–61.

[3] D. Soldani and S. Dixit, “Wireless Relays for BroadbandAccess,” IEEE Commun. Mag., vol. 46, no. 3, Mar.2008, pp. 58–66.

[4] 3GPP TR 25.996, “Spatial Channel Model for MultipleInput Multiple Output (MIMO) Simulations (Rel. 6),”Sept. 2003.

[5] P. Kyosti et al., “WINNER II Channel Models,” IST-WIN-NER II D1.1.2, Nov. 2007.

[6] G. Senarth et al., “Multi-hop Relay System EvaluationMethodology (Channel Model and Performance Met-ric),” IEEE 802.16j-06/013r3, Feb. 2007.

[7] P. Almers et al., “Survey of Channel and Radio Propaga-tion Models for Wireless MIMO Systems,” EURASIP J.Wireless Commun. Net., vol. 2007, Article ID: 19070,2009.

[8] C.-X. Wang et al., “Spatial Temporal Correlation Proper-ties of the 3GPP Spatial Channel Model and the Kro-necker MIMO Channel Model,” EURASIP J. WirelessCommun. Net., vol. 2007, Article ID: 39871, 2007.

[9] P. Agrawal and N. Patwari, “Correlated Link ShadowFading in Multihop Wireless Networks,” IEEE Trans.Wireless Commun., vol. 8, no. 8, Aug. 2009, pp.4024–36.

[10] C.-X. Wang, X. Cheng, and D. I. Laurenson, “Vehicle-to-Vehicle Channel Modeling and Measurements:Recent Advances and Future Challenges,” IEEE Com-mun. Mag., vol. 47, no. 11, Nov. 2009, pp. 96–103.

BIOGRAPHIESCHENG-XIANG WANG ([email protected]) receivedhis Ph.D. degree from Aalborg University, Denmark, in2004. He joined Heriot-Watt University, United Kingdom,as a lecturer in 2005 and became a reader in August 2009.His research interests include wireless channel modelingand simulation, cognitive radio networks, vehicular com-munication networks, cooperative communications, greencommunications, MIMO, and (beyond) 4G. He served or isserving as an editor or guest editor for several internationaljournals including IEEE Transactions on Wireless Communi-cations. He has published one book chapter and over 120papers in journals and conferences.

XUEMIN HONG ([email protected]) received his Ph.D. degreein 2008 from Heriot-Watt University, where he is currentlya postdoctoral research associate. Previously he was a post-

doctoral fellow at the University of Waterloo, Canada. Hisresearch interests include (beyond) 4G systems, cognitiveradio networks, cooperative communications, and wirelesschannel modeling. He has published one book chapter and17 papers in journals and conferences. He is a publicitychair for CMC 2010 and has served as a TPC member forseveral conferences.

XIAOHU GE ([email protected]) received his Ph.D.degree in communication and information engineeringfrom Huazhong University of Science and Technology(HUST), China, in 2003. He has been working at HUST since2005 and is currently an associate professor. Prior to that,he worked as an assistant researcher at Ajou University,South Korea, and Politecnico Di Torino, Italy. His researchinterests include mobile communications, traffic modeling,and interference modeling in wireless networks.

XIANG CHENG ([email protected]) received his B.Sc. andM.Eng. degrees in communication and information systemsfrom Shandong University, China, in 2003 and 2006,respectively, and his joint Ph.D. degree from Heriot-WattUniversity and the University of Edinburgh, United King-dom, in 2009. His current research interests include mobilepropagation channel modeling and simulation, multipleantenna technologies, mobile-to-mobile communications,and cooperative communications. He has published approx-imately 30 journal and conference papers. He has served asa TPC member for several international conferences.

GONG ZHANG ([email protected]) received his M.Sc.degree in electrical engineering from Harbin Institute ofTechnology. He has been with Huawei Technologies since1998. He joined the UMTS product unit in 1999, becamethe product development team leader in the smart termi-nal unit in 2002, and was system architect in charge ofmobile TV solutions in 2005, in charge of the LTE-SAE pro-gram in 2007, and in charge of the interference coordina-tion and cooperation project in 2008. He is now head ofthe Network Architecture Department in Huawei CorporateResearch.

JOHN THOMPSON ([email protected]) received hisB.Eng. and Ph.D. degrees from the University of Edinburghin 1992 and 1996, respectively. He has been working atthe University of Edinburgh, first as a postdoctoralresearcher, then a lecturer, and now a reader since October2005. His research interests include signal processing algo-rithms for wireless systems, antenna array techniques, andmultihop wireless communications. He has publishedapproximately 150 papers, a few book chapters, and anundergraduate textbook on digital signal processing. He iscurrently Editor-in-Chief of the IET Signal Processing journaland was a Technical Program Co-Chair for IEEE ICC 2007.

Figure 4. Absolute values of small-scale fading correlation coefficients of the twolinks in a CoMP system as a function of θ and D2 (D1 = 500 m, R = 30 m).

Angle θ between two links20 0

50

0

Dis

tanc

e D

2 be

twee

n th

e M

S an

d BS

2

100

150

200

250

300

350

400

450

500

0.1

0

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

40 60 80 100 120 140 160 180

WANG LAYOUT 1/19/10 12:00 PM 7


2010_Wang_ComMag_02_80

Documents

Transcript of 2010_Wang_ComMag_02_80