IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

Energy- and Cost-Efficient Mobile Communication

using Multi-Cell MIMO and Relaying

Peter Rost, Member, IEEE, Gerhard Fettweis, IEEE Fellow, and J. Nicholas Laneman, Senior Member, IEEE

ABSTRACT In this paper, relaying and multi-cell MIMO trans-mission are investigated as approaches for improving resource reuse
and more flexible organization of cellular networks. The analysis focuses on approaches for future cellular systems, which jointly
exploit relaying and multi-cell MIMO transmis-sion. Possible candidate approaches are identified, simplified for practical applications
and evaluated using a system-level model from the European research project WINNER. Their achievable throughput is analyzed under
practical constraints using three different normalization approaches: cost-normalization, energy-normalization, and joint cost-energynormalization.
It can be shown that the combined approach of relaying and multi-cell MIMO provides significant gains for the uplink com-munication. The
selected approach exploits cooperative multi-cell MIMO processing between base stations and relay nodes, and uses a resource coordination
technique on the links between relay nodes and user terminals. If a relay-based deployment is subject to a cost- and energy-normalization,
multi-cell MIMO outper-forms relaying with respect to achievable downlink throughput.

INDEX TERMS Relaying, MIMO, cellular networks, half-duplex, cost-benefit trade-off, wide-area coverage scenario

I. INTRODUCTION
A. Motivation
Among the most serious challenges in current and future mobile communication systems are the mitigation and avoid-ance of
inter-cell interference. Future mobile communication systems are likely to use multi-cell multiple-input multiple-output (MIMO)
approaches [1] in which multiple base stations (BSs) cooperatively serve user terminals (UTs). In multi-cell MIMO, techniques
are applied that were introduced in the context of the MIMO broadcast channel (BC) [2], [3] and the MIMO multi-access channel
[4], [5]. Furthermore, additional relay nodes (RNs) will help to provide high data rates in otherwise shadowed areas as well as to
improve the channel conditions at cell borders [6], [7]. Although relay protocols are well investigated on a link level, i. e. a
plethora of different protocols has been presented in the literature, there is little work on system-level evaluation of relay
deployments. System-level evaluation of relaying and normalization of ob-tained results constitute the first major problems
treated in
P. Rost is with NEC Laboratories Europe, Heidelberg, Germany ([email protected]), G. Fettweis is with the Vodafone Chair for Mobile Communications
Systems, Technische Universitat Dresden (TUD), Dresden, Germany ([email protected]). J. N. Laneman is with the University of Notre Dame (IN), USA
([email protected]). Part of this work has been performed in the framework of the IST project IST-4-027756 WINNER II, which has been partly funded by the European
Union, and the Celtic project CP5-026 WINNER+. The authors would like to acknowledge the contributions of their colleagues in WINNER II and WINNER+, although
the views expressed are those of the authors and do not necessarily represent the project. J. N. Lanemans work has been supported in part by NSF Grant CNS0626595.

this paper as relaying not only effects the system performance but also, among others, the deployment costs and energy
consumption. Furthermore, this work deals with the integration of multi-cell MIMO and relaying as complementary instead
of competing technologies.
The integration of multi-cell MIMO and relaying as com-plementary technologies is of particular interest as multi-cell
MIMO is able to combat inter-cell interference [8] but not the decreasing SNR due to path-loss. On the other hand,
relaying can alleviate the effects of path-loss but does not provide the same interference-cancellation opportunities as
multi-cell MIMO. Relaying further requires a high-data rate feeder-link towards the assigned BS, which makes multi-cell
MIMO a reasonable choice for the feeder-link. Compared to relaying, multi-cell MIMO cannot increase the spatial reuse
within the cell without deploying additional BSs and connecting these BSs to the backhaul.
The opportunities to increase the achievable cell throughput by combining multi-cell MIMO and relaying are imminent,
however, it is not intuitively clear whether relaying is also more cost-efficient and energy-efficient than multi-cell MIMO. As an
alternative to relaying, one could also increase the density of BSs while achieving the same performance at possibly the same
costs. By contrast to existing work that only regards the achievable throughput, this paper presents a strategy to normalize relaybased and conventional systems in order to provide a fair comparison of both.

B. Contribution
In this paper, we explore novel system architectures based upon the transmission schemes introduced in [9], where we
conducted a link-level analysis of the two-path relay-assisted interference channel. The introduced schemes are able to combine
the benefits of multi-cell MIMO and relaying. Hence, a system applying these schemes is able to combat inter-cell interference
(using multi-cell MIMO) as well as to alleviate path-loss effects (using multi-hop transmission). We further propose and discuss
simplifications of the schemes introduced in [9]. The resulting protocols are less complex and induce less signaling overhead,
which makes them suitable for practical application in future systems. A thorough system-level analy-sis evaluates these
simplifications regarding their influence on the achievable user throughput.
Using a system-level analysis of a next-generation mobile communication network, we compare the uplink and downlink
performance of a conventional system with frequency reuse 3, a system using multi-cell MIMO only, a system only applying
relaying, and a system in which both multi-cell MIMO and

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

relaying are combined. This analysis employs the channel and system model of the European research project
WINNER [10], which was also used for the IMT-Advanced evaluation [11] and therefore is widely accepted and
1
publicly available [12], [13] .
Previous work on the evaluation of relaying-schemes mostly focuses on the achieved performance, which ignores the fact that
RNs may imply additional costs as well as an increased energy consumption. Both are serious aspects for the deploy-ment of
cellular systems and require a trade-off analysis with the achieved performance benefits. Hence, the system-level analysis
carried out in this paper applies two novel ways of normalizing its results. At first, the analysis is conducted under the constraint
that the compared systems must cause the same overall costs, i. e. capital and operational expenditures (CAPEX and OPEX).
Secondly, the analysis is subject to the constraint that the overall energy-consumption is normalized. As a result of this analysis,
basic guidelines for the design and operation of relay nodes are given, i. e. given the applied system model RNs should not cost
more than about 10-20% of a BS and the average transmit power of RNs should be about 28 dBm. Even though these
parameters reflect the system and channel model choice, the results are of use for practical applications due to the realistic and
widely accepted system and channel model. Detailed description of the underlying assumptions will be given in the course of the
following sections.

C. Related Work
So far only [14] considered a setup with two cooperating multi-hop paths. However, [14] focused on a linear Wynermodel with users cooperating over a limited-capacity link. Recently, the interference channel with a single RN [15]
attracted more attention and has been investigated in [16], [17]. By contrast to our work, the single-relay interference
channel models a RN that is placed on the cell boundary and forwards traffic originating from two different cells. This case
is not considered in our analysis but only deployments where both multi-hop paths are associated to different cells.
Multi-cell MIMO may refer to two different cases [18]. In the first case, channel state information (CSI) as well as user data are
jointly processed at physically separated base-stations, and in the second case, user data is separately processed at each basestation but exploiting joint CSI. Throughout this paper, we focus on the former case and refer to it as multi-cell MIMO. The latter
case refers for instance to coordinated beamforming, which is not part of this paper. A detailed introduction and analysis of both
cases has been given recently in [19]. Furthermore, multi-cell MIMO also received more attention by standardization bodies, e.
g., [20], [21].

In [22], Werner et al. compared relay-based deployments with conventional systems on the basis of a performance
normalization. An indifference curve compares the required deployment density of BSs and RNs to provide the same
performance as a reference system without relays. Such a
1

All referred WINNER documents are publicly available on http://www.ist-winner.org and all ITU-R documents are available on http://www.itu.int.

BS
uBS BS

BS
uBS BS

u
u

u
u

BS

uBS BS
u
u u

BS
uBS BS

u
u

BS

BS
uBS BS

BS
uBS BS

uBS BS
u
u u
Fig. 1. Reference scenario considered in this paper with one tier of interfering sites. Triangles indicate relay nodes and arrows the main lobe
direction in each cell.

tool provides the means to identify the potential cost-savings of relay-enhanced cells. A thorough overview of cooperative
networks including relay-networks has been recently given in [23]. In [24], opportunistic cooperative networking has been
discussed and optimal strategies are derived for the decision if a relay is considered for a multi-hop link. Furthermore, [25]
discusses open problems in the context of energy-efficient wireless communication for which multi-antenna techniques and
relaying play an imortant role. Also more recently, [26] discusses energy-efficient relaying and shows how energy-harvesting
affects the performance of relaying networks. Fi-nally, [27] considered the trade-off of delay and an arbitrary cost-measure such
as energy.

D. Outline
Section II briefly introduces the considered multi-cell system architecture and discusses the applied transmission schemes
from a conceptual point of view. We then proceed in Section III with a detailed introduction of three different methods to
normalize the comparison of a relay-based system and a conventional system. Section IV then introduces the compared
protocols in more detail. The performance results based on the introduced normalization are discussed in Section V and the
paper is concluded in Section VI.

II. SYSTEM ARCHITECTURE AND DESIGN GOALS

This paper provides a system-level evaluation for a typical multi-cell network with hexagonal structure as
illustrated in Fig. 1. Sites are uniformly placed and each is equipped with three BS antenna arrays (indicated with
BS) serving three adjacent sectors (with main lobe directions indicated by arrows) and two stationary RNs per
sector (indicated by triangles).
In 3G and 4G networks, the site density must be increased compared to 2G networks in order to satisfy the demands of high
data rates. Additional BSs improve an operators revenue

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

as they provide additional capacity and coverage, but on the other hand their deployment necessitates significant CAPEX and
OPEX. RNs have the potential to reduce the CAPEX and OPEX as they require less space (and therefore lower lease costs),
they do not need active air conditioning, and they consume less energy than a BS. On the other hand, RN deployments are
denser than BS deployments due to the constraint that RNs cannot concurrently transmit and receive on the same timefrequency resource (half-duplex constraint) as well as the constraint of having a wireless feeder-link with finite data-rate towards
the assigned BS. In this paper, we focus on non-cooperative decode-and-forward RNs.
Hence, there is the demand for a framework to as-sess the cost-benefit trade-off, which relates the required OPEX/CAPEX of
a system to the systems performance. The goal of this paper is to define such a framework, which allows us to find low-complex
and low-cost transmission schemes. The basic concept of the evaluated transmission schemes and the arising requirements on
the system architecture are introduced in the following of this manuscript.

The reference scheme of this paper is a conventional system, which applies frequency reuse 3, i. e. each BS per site
has an orthogonal resource pool of equal size. Furthermore, each UT is assigned with the BS that has the lowest pathloss towards this UT. Such a conventional approach has two major drawbacks. Firstly, a low resource utilization as each
resource is only used by approximately one-third of the UTs and secondly, a large number of the UTs experience a high
path-loss.
The problem of low resource utilization is addressed by multi-cell MIMO where all BSs make use of all available
resources to achieve frequency reuse 1. In order to allevi-ate inter-cell interference, BSs either jointly process receive-and
transmit-signals or they coordinate their transmissions. However, this requires a high-capacity and very-low-latency
backhaul-connection. These requirements might not be satis-fied if cooperating BSs are located at different sites. Even
though multi-cell MIMO is able to mitigate the effects of inter-cell interference, it still has to cope with path-loss effects. In
this paper, we present scenarios where multi-cell MIMO is applicable and limitations on the backhaul-connection can be
mitigated.
By contrast to multi-cell MIMO, relaying provides ad-ditional radio access-points in order to increase the spatial resource utilization
(frequency reuse) and to allay the effects of path-loss. However, even though practical RNs improve the radio channel conditions they
are subject to the half-duplex constraint and there is still the need for intra-cell and inter-cell interference mitigation and avoidance. This
can be achieved by combining the benefits of both, multi-cell MIMO and relaying. If multi-cell MIMO is applied on the link between RNs
and BSs, the effects of the half-duplex constraint can be extenuated due to the expedient channel conditions, i. e. no mobility and less
multi-path components. We can further apply multi-cell MIMO towards UTs at the border between two BSs located at the same site
where an almost unlimited connection between BSs exists. Relays then serve users at the cell border between different sites in order to
alleviate the path-loss effects.

III. NORMALIZATION OF SYSTEM EVALUATION

Comparing strategies in a mobile communication system can be done in many different ways such as comparing the sumthroughput, the worst-user performance, average delay, or blocking-rate. Up to now, most work, however, only applied
performance-criteria to compare different strategies for cellular networks. Out of economical and ecological reasons, this focus
recently changed and it is more than ever of particular interest to take the CAPEX and OPEX as well as the systems energy
consumption into account.

This section introduces a novel framework that allows for an evaluation of the cost-benefit-trade-off as well as energybenefit-trade-off of relay-based systems. Hence, it is able to relate the potential performance benefits of relaying to the
possibly increased deployment costs as well as energy consumption. In order to allow for a fair comparison of multi-cell
MIMO and a system employing RNs, we present a model to normalize costs and energy consumption. Even though this
work focuses on a macro-cellular deployment, the introduced model is simple and flexible enough to be applicable to a
variety of other scenarios such as micro-cellular and indoor deployments.

A. Nomenclature
2

Throughout this paper, x CN (0, ) denotes a circularly symmetric i.i.d. Gaussian random process with each element having
2
zero mean and variance . We will use non-italic lowercase letters x to denote random variables and italic letters (N and n) to
denote scalar values. Matrices are denoted by bold uppercase letters, e. g., H denotes the compound channel matrix with
dimension NRx NTx where NRx is the sum of receive antennas and NTx is the sum of transmit antennas. Vectors are denoted by
T
H
x, the transpose of a vector or matrix is denoted by x , the Hermitian transpose is denoted by x , and the trace of matrix X is
denoted by tr(X). We further use the capacity function C(x) = log(1 + x) with logarithms taken to the base of 2.

B. Cost-Normalization
In order to normalize the performance evaluation, we com-pare the throughput of a conventional and relay-based
system while keeping the OPEX/CAPEX constant. This implies that the relay-based deployment must use a lower BS
density. In the following, we use CSite to denote the costs of one site and CRelay to denote the costs of one relay node,
which includes hardware costs, rental costs, average costs for power supply as well as backhaul costs (in the case of a
site). In order to present results independent of actual budget figures, we conduct our analysis solely based on the relative
relay costs RN = CRelay/CSite for which [28] estimated a typical value, i. e. RN 0.2. These relative relay costs can capture
both fixed and running costs while the running costs refer to a specific period of operation.
Assuming a regular BS distribution with an inter-site dis-tance dref as shown in Fig. 1. The area covered by nx BSs in x-direction and ny BSs in y-direction is given by

A = (nxdref ) ny

_
/2dref .

(1)

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

The overall costs for this deployment are n xny CSite, hence and similarly to (3) for the relay-based deployment by
the costs normalized over the covered area are given by
Psite (PTx,site) + NsumPRN
(PTx,RN)
nxny CSite
ref = (nxdref ) ny
which is used as reference value for

3/2dref

RN =

CSite

2
3 /2d

ref

(10)

= 3 /2dref ,
(2) In order to fulfill the energy-normalization constraint ref =
the relay-based deploy- RN, we express the overall power consumption at one site as

ment. With NRelay RNs deployed at each site, the costs nor- a function of the RN transmit power:
malized over the covered area by the relay-based deployment
Psite (PTx,site) = Psite (PTx,ref ) NRelayPRN (PTx,RN) .
are given by

(11)
Using the definition of the overall power consumption in
.
(3) (5), the average transmit power of a BS antenna is given as
function of the RN transmit power:
Normalizing the costs for both deployments requires finding
NRelay
an inter-site distance dis (RN) such that ref = RN, which
PTx,site = PTx,ref
PRN (PTx,RN) . (12)

Tx,site
is fulfilled by
Note that we could equally compute the required transmitCSite + NRelayCRelay
2
RN =
3/2dis

p
dis () = dref 1 + RNNRelay.
C. Energy-Normalization

(4) power per RN for a given transmit-power per site.

The normalization of the systems deployment costs does not necessarily imply normalized energy consumption at BSs and
RNs. Although the cost-normalization approach decreases the BS density to achieve normalized deployments, the energynormalization approach appropriately adjusts the RN and BS power such that the overall energy consumption is the same. This
work considers only a downlink-based energy normalization, where the transmit-energy dominates compared to an uplink-based
energy normalization where only signal-processing energy needs to be considered. The overall power consumption of a radio
access point (RAP) can be modeled as the sum of the linearly weighted transmit power P Tx and a constant part Pconst, which is
independent of the transmit

power [29]. Let Psite denote the sum-power consumption

at a site and PRN denote the sum-power consumption at
a RN. Using the factors Tx,site and Tx,RN, which weight how the transmission power contributes to the overall
power consumption, the sum-power consumption at RNs and BSs is given by

D. Joint Energy and Cost-Normalization

Our third approach to assessing the performance of a relay-based deployment is based on a joint energy- and costnormalization where the density of the relay-based deployment is increased in order to normalize the costs, and the energy per
RAP is appropriately adjusted to normalize the energy consumption. Due to the decreased BS density, the energy-consumption
per area-element is given by
RN ()

Psite (PTx,site) + NsumPRN (PTx,RN)

which results in

3/2 (dis

())2

. (13)

We again normalize the energy-consumption by solving the equation RN () = ref ,

1
_ dis () 2
PTx,site =
dref _ Tx,site Psite (PTx,ref ) Pconst,site
_

NRelayPRN (PTx,RN) ,

(14)

with

dis ()

p
1 + RNNRelay 1.
dref =
Psite (PTx)
=
Tx,sitePTx + Pconst,site
PRN (PTx)
=
Tx,RNPTx + Pconst,RN.

(15)
(5) This implies that for increasing deployment costs the density
of radio access points decreases and the power per node in(6)
creases in order to normalize the energy. If the operational exAll quantities depend on the number of BSs per site, the penditures consider the costs for energy supply (RN as a funcnumber of power amplifiers, the power amplifier efficiency, tion of PTx,site), we have to iteratively solve for an energythe cooling power, and the battery backup. Using the reference normalization that satisfies the given cost-normalization. The
values given in [29], we can apply the following values:
solution to this problem mainly depends on how energy-costs
contribute to the OPEX.
Tx,site = 44
Pconst,site = 59 dBm
(7)
Pconst,RN = 47 dBm.
Tx,RN = 4
(8)
IV. EVALUATED PROTOCOLS
This section details the system model of the two-path relayThe significant difference between a RN and a site results from
interference channel, which is used to describe the mathemathaving 12 antennas at one site and only one antenna per RN.
ical basis of the protocols, which are compared in Section V
Again we assume a regular distribution of BSs with inter-site
based on the previously described normalization methods. This
distance dref where in the site-only deployment a BS antenna model is used to comprehensively introduce the structure of the
transmits with average power P Tx,ref . Similarly to (2), we can protocols and the achievable rates of the individual strategies.
express the power-density over the covered area by
In the last part of this section, we detail how these protocols
Psite (PTx,ref )
are applied to a multi-cell multi-user system, which is used
ref

3/2dref

(9) for our performance evaluation in Section V.

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

x1

C1,2/C2,1

2 x2

y3 : x3 y5

4
(a) Abstract model

BS1 BS2

6 y4 : x4 y6

with the compound message w BS

= (w w )

1 2

CN (0, Rww), the compound channel output vector yUT CN (0, Ryy ) with dimension NRx 1 and co-variance Ryy , the compound
channel matrix HBS with dimension NRx NTx, the pre-coding matrix VBS with dimension NRx 2, which is applied across both BSs
2
and guarantees the channel input covariance matrix P BS, and the receiver noise nUT with covariance Rnn = nn I.

Our analysis uses a multi-cell MIMO approach based on the transmit Wiener filter [30], which is the linear filter minimizing
the minimum mean-square error (MMSE). We apply a linear transmit filter as it is less complex than non-linear methods such
as Dirty-Paper Coding [31] and it is more robust to imperfect channel knowledge [32]. Recently, results of an experimental
multi-cell MIMO setup with a linear Wiener filter have been presented in [33]. Distributed multi-cell MIMO transmission
underlies a per-antenna-group power constraint [34], which is enforced in time-domain and implies that the average
consumed power may not exceed a certain value.

UE1

RN1

RN2

UE2

(b) Relay-Interference channel in the context of mobile communication systems

Fig. 2. An abstract model of the two-path relay-interference channel as well as how it applies to mobile communication systems

A. Two-Path Relay Interference Channel

Consider Fig. 2(a) which shows the relay interference-channel consisting of two BSs, each serving one UT supported by two
RNs. One possible scenario that is represented by this channel is illustrated in Fig. 2(b). We further assume single-antenna
nodes for the protocol description in this section even though the analysis in Section V considers a multiple antenna case. In the
following, we use this model to describe the already introduced transmission schemes with the means of a mathematical
framework. Please note that the transmission schemes are not limited to two nodes and one antenna but can be applied to
multiple paths and multiple antennas.

In the two-path relay interference channel as shown in Fig. 2(a), BSs 1 and 2 are connected by backhaul links with
capacities C1,2 and C2,1, which allow for multi-cell MIMO cooperation. Both BSs serve UTs 5 and 6, and each path is
supported by a RN, i. e., nodes 3 and 4. In the following, the channel input at node k [1; 6] is given by xk CN (0, Pk).
The channel output is given by yk =

with hk,k denoting the channel coefk 6=k hk,k xk + nk

ficient between nodes

and

, and additive white Gaussian

k
k2
noise (AWGN) nk CN (0, nn,k). Unless otherwise noted
2
2
we assume that

nn,k

for all k [3; 6].

nn

B. Multi-cell MIMO approach

Consider Fig. 2(a) and assume that both relays remain silent, i. e. x 3 = x4 = 0. Then, the downlink channel output at
each UT in an AWGN system model can be given by

y UT =

H V
BS

BSwBS

+ nUT,

(16)

Since BSs are physically separated and use different power-supply, it is not possible to argue that power may be shifted from one BS to
another in order to fulfill an average sum-power constraint. However, for a large number of subcarriers (our analysis considers 2048
subcarriers) and over multiple symbols, the inherent diversity implies that the probability for violating the per-antenna-group power
constraint is sufficiently low. Hence, this work does not consider an instantaneous per-antenna-group power constraint for the resulting
transmit filter but only ensures that the average sum-power constraint over all antennas is satisfied. Using the previously introduced
notation,

the Wiener filter can be expressed by [30]

tr(Rnn)
_
_
H
VBS = HBS HBS + tr(PBS) I
with the power normalization factor

tr

u
t

tr(RNN )

HBS

(17)

tr(PBS)

=
u

HBS +
BS

tr(

H
H

BS)

ww BS

BS

_
_

(18) For a given covariance matrix Rww, the achievable rate region
RTxWF (HBS) for multi-cell MIMO using the transmit Wiener filter is the set of all rate pairs (R1, R2) RTxWF (HBS) that satisfy
i [1; 2] : Ri < C

[H
_
_

[Ryy ]
_

BS

BS

[H
_

i,i

BS

i,i

_
_

VBS]
_

[Rww]

i,i _

i,i

[Rww]

i,i

(19) The rate region given in (19) is the BC rate region of a MIMO
system in which all BSs form one virtual antenna array and apply a transmit Wiener filter. We introduced the rate region using
only two user terminals, however, it can be extended to NRx > 2 and NTx > 2 by extending the system model in (16) accordingly.
The uplink of the considered system can be expressed by
H

y BS = HBS xUT + nBS

(20)

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

with channel input xUT = (x5x6) . Again we consider an approach, which is of interest for practical implementations. The BS uses the QR
H

decomposition H BS = QHe into the unitary matrix Q and upper triangular matrix H e . After multiplying the channel output yBS with Q , the
effective channel is given by He . Due to the fact that matrix He is an upper triangular matrix, any user i is only interfered by users j > i (all
other users have channel gain 0). The remaining interference is removed via successive interference cancellation (SIC) such that the
rate for each terminal i is given by [35]

[
i [1; 2] : Ri < C

_
_

]i,i
2

Pi
.

nn

(21)

Throughout this work, we assume perfect SIC not taking account effects that result from imperfect channel
estimation and decoding errors.
C. Two-Stage Multi-cell MIMO and Relaying
This paper compares multi-cell MIMO with a relaying protocol that uses multi-cell MIMO in a first phase between BSs
and RNs, and in a second phase it employs Han-Kobayashi (HK) super-position coding [36] to coordinate the resources
between RNs and UTs. In the case of our model, relay 3 divides its messages into two parts w (3,5) and w(3,D) with power
assignment P(3,5) and P(3,D) (similarly relay 4
uses w(4,6) and w(4,D)). While the first part w(3,5) is only decoded by the assigned UT 5, w(3,D) is decoded by both UTs

D = {5, 6}. Message w(3,D) reduces the interference-level for w(3,5) and is called a common message. Both common messages and the
private message are jointly decoded by the
respective receiver. Let P(i,i+2)+P(i,D) = Pi, then the channel input at terminal i is given by

q
xi =

q
P(i,i+2)w(i,i+2) + P(i,D)w(i,D).

The corresponding rate region RHK(HRN), with HRN denot-ing the channel gain between RNs and UTs, has been
derived in [36].
Consider the downlink transmission illustrated in Fig. 2(a) and divide it into two phases: in phase 1 both BSs transmit using
fraction 0 < t1 < 1 of resources, and in phase 2 both RNs transmit in the remaining fraction t2 = 1t1 of resources. Using this setup,
the achievable rate region is the convex hull

Co

t1 ,t2

t1+t2=1 t1RTxWF ( BS) t2RHK ( RN)

, (22)

:[
where t1RTxWF (HBS) denotes a scaling of all rates in RTxWF (HBS) by t1 (and similarly for RHK).

HK coding requires a complex optimization of the power assignment at the transmitter and a complex joint decoder at the
receiver. Etkin et al. derived in [37], [38] an approach that has been shown to be within 1 bpcu of channel capacity while being
much simpler than the HK coding approach. Their approach aligns the interference power caused by each

transmitter with the noise power at the non-intended receiver. Specifically, if the channel between RNs and UTs is given by

1 b

HRN =
a 1 _
(23)
both terminals choose according to the Etkin-Tse-Wang (ETW) approach the private message power such that
2

|a| P(3,5) nn, 6 and |b| P(4,6) nn, 5

(24)

holds. Compared to the HK approach, each terminal at first decodes both common messages jointly and then each one decodes
its own private message. Coefficients a and b reflect the cross-coupling of both communication pairs, which in a mobile
2

communication system are usually constrained by 0 < |a| , |b| < 1, as terminals are assigned based on the lowest path loss to the
serving base station. The achievable rate region RETW(HRN) of the ETW approach has been derived in [37], [38] and is omitted
here for brevity.
Signaling the individual power assignments and joint decod-ing are complex and overhead-intensive operations. Empirical
observations in [28] show that a floating power assignment
0 < P(i,D)/Pi < 1 is used in very few cases where the interference is not weak enough to be ignored but also not
strong enough to dominate the transmission. Hence, we further simplify the ETW approach by applying the power assignment
2

(
P(3,D) =

P
0

/a 2

NN,6

3/2

otherwise,

(25)

which implies that we only use either common or private messages. This reduces the signaling overhead for the
power level to 1 Bit.
D. An Integrated Approach
In mobile communication systems applying both multi-cell MIMO and relaying, it is preferable that each BS serves those UTs
that are located in its main lobe direction or with LOS towards the BS. In these cases, the maximum rate is more likely to be
limited by the available modulation and coding schemes (MCS) than the achievable signal-to-interference-and-noise ratio (SINR).
Since relaying suffers from a half-duplex loss, direct transmission from BSs to UTs is likely to outperform relaying. On the other
hand, UTs at the cell edge between two sites or with a very good link towards the next RN should be served using relaying.
Since it is very complex and unreliable to predict whether a UT should prefer using a BS or a RN, we show in Section V results
for an integrated approach where UTs are assigned to a RN or BS based on the path loss, i. e., each UT measures the effective
path-loss incorporating transmit-power, actual path-loss, and shadowing, and then selects the radio access-point (either RN or
BS) with the lowest effective path-loss.

E. Application to Multi-Cell Multi-User Scenario

The previous part introduced different protocols based on the two-path relay-interference channel. In order to evaluate these
protocols in the context of cellular networks, this section

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

discusses their application to a mobile communication system while the details regarding scheduler, channel model,
and system parameters are given in Section V. Our work considers a hexagonal grid as illustrated in Fig. 1 with
three BSs per site serving three adjacent sectors and two stationary RNs per sector.
The first protocol is a conventional approach with frequency reuse 3 where the available resources are divided into three
orthogonal parts each assigned to exactly one BS per site. Each BS is equipped with four antennas and each UT with one
antenna, hence, up to four UTs are served on the same time-frequency resource block, which is also called chunk [12]. Our
results consider full interference from all BSs (downlink) and UTs (uplink). Accordingly, the channel output term in (19) and noise
term in (21) include the inter-cell interference. We further assume that in (19) Rww = I (downlink messages with equal power) and
in (21) all UTs transmit with maximum power. Furthermore, we assume that the BS has no knowledge of the average downlink
interference caused by other cells, which is therefore not considered in the design of the Wiener filter (17).
Multi-cell MIMO is applied to at most three cooperating BSs, motivated by the facts that each site is equipped with three BSs
and multi-cell MIMO applied at one site appears more realistic than across separate sites. Hence, with three cooperating BSs at
most 12 UTs are cooperatively served and the resulting compound channel matrix H BS is used in (16) to determine the
achievable rates. We further consider frequency reuse 1 and therefore all BSs and UTs not involved in a specific multi-cell MIMO
cooperation are assumed to be interfering. The selection of UTs is done iteratively by choosing at first four UTs at the first BS
and then selecting their strongest interferer-BSs, which then choose the UTs for which the other two BSs represent the strongest
interferers.
Our paper further evaluates the performance of a relaying protocol where multi-cell MIMO is used for the link between RNs
and BSs. Again, the same algorithms as on the BS-UT links are applied and full interference is considered. BS-RN and RN-UT
links share the same spectrum and BS-RN resources are assigned first based on the expected throughput as well as the databuffers at each RN. On the RN-UT links always two RNs and two UTs cooperate. Again, those pairs are assigned iteratively, i. e.,
a RN selects a UT and then the RN causing the highest interference. The interfering RN then selects a UT for which the first RN
is the highest interferer. As a fast-fading link adaptation using ETW coding appears unre-alistic due to the immense signaling
overhead between RNs, we utilize only the long-term statistics of HRN at RNs and UTs instead. Furthermore, [28] showed that
using fast-fading information provides only marginal performance benefits. This simplified ETW approach based on long-term
statistics of HRN has been applied to obtain the results in Section V. Our analysis in Section V uses the maximum common rate
achievable in the rate region given by RETW(HRN).
Finally, we will present results for an integrated approach where UTs are assigned to either a BS or a RN based on the experienced
effective path-loss as explained in the last section. In this case, all resources are divided into those used for BS-

User density
Average no. users/cell
Channel models
Number
of antennas
BS/RN/UT
BS transmit power
RN transmit power
UT transmit power
Noise figure
Noise power spectral
density
FFT-Size
Carrier frequency
System bandwidth
OFDM-Symbol duration
Superframe duration
Guard interval
Used subcarriers
Channel state information

90 per km

26
as defined in [39]
4/1/1
46 DBM
37 DBM
24 DBM
7 DB
174 DBM/HZ
2048
3.95 GHZ
100 MHZ
20.48 S
5.89 MS
2.00 S
[920; 920] \ {0}
assumed to be perfectly known at transmitter and receiver

Channel models accord- BS to RN: B5a

BS to UT: C2
ing to [40]
RN to UT: C2
TABLE I
PARAMETERS OF THE EVALUATED SYSTEM MODEL, CHOSEN ACCORDING
TO THE WINNER SPECIFICATIONS [12]

UT, BS-RN, and RN-UT links.

V. RESULTS
This section applies the previously introduced framework for normalizing the system level comparison of conventional
deployments and relay-based deployments. In the course of our evaluation, we are discussing both uplink and downlink
performance and show how different system parameters such as relative deployment cost or transmit power affect the
performance of relaying-protocols.

A. Reference Scenario
Our system-level results are based on the channel and system models defined by the European research project WIN-NER
[10], [12]. Our paper considers results for one central site surrounded by two tiers with a total of 18 interfering sites as illustrated
in Fig. 1. The distance between two adjacent sites is given by d ref = 1000 m. UTs are uniformly distributed and assigned to the
1
BS with the lowest path-loss. Furthermore, RNs are positioned as shown in Fig. 1 and at distance /3dref .
We apply Orthogonal Frequency Division Multiplexing (OFDM) [41], [42] at a carrier frequency f c = 3.95 GHz with bandwidth
Bw = 100 MHz and Nc = 2048 subcarriers. Time division duplex (TDD) is used to separate uplink (UL) and downlink (DL)
resources. All further air interface parameters are listed in Table I and are described in further detail in [12]. The WINNER system
offers a variety of channel models suitable for each scenario. Each consists of a model for the probability that a strong line-ofsight (LOS) link exists, the path-loss model, and the power delay profile, which models the small scale fading assuming each
channel-tap is Gaus-sian distributed. We further assume block fading where each

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

channel realization is independently generated. Table I lists all channel models, which are used in our analysis.
Furthermore, we assume Gaussian alphabets, perfect rate adaptation, and infinite blocklengths in order for the informationtheoretic rate regions to be relevant and to have an upper bound on the achievable rates. Throughout the following discussion
we assume perfect channel estimation and that the system is perfectly synchronized. This paper considers neither automatic
repeat-request (ARQ), which is not necessary due to the perfect link adaptation, nor quality-of-service (QoS), since all users are
assumed to have equal priority. The considered scheduler assumes full queues for each user, which implies that all available
resources are fully exploited. We address the problem of an in-band feederlink between BSs and RNs, which might result in
empty relay-buffers due to unexpected high throughput on the links between a RN and its assigned UTs.

B. Simulation Methodology
In order to evaluate the different protocols, we use a snap-shot based system-level simulation, which does not explicitly model
user mobility but models the effects of mobility in the channel instead. Users are placed randomly, and for each snapshot 16
frames, each consisting of 15 OFDM symbols, are simulated before the next snapshot is drawn. For each frame an independent
instance of the small scale fading process is drawn and the user scheduling is performed. We do not apply a wrap-around model
but evaluate only the performance of the central site in order to avoid edge-effects.
In order to allow for comparability of the results, we apply a fair resource-scheduler (round-robin) based on resource-blocks of
size 15 OFDM-symbols 8 subcarriers (which is referred to as a chunk). The obtained results would differ quantitatively to a
proportional fair or equal-rate scheduler but the qualitative results remain the same. Furthermore, the focus of this analysis is on
the normalization methods rather than the actual scheduler implementation. Our evaluation uses two measures, i. e. average
throughput = Eu,t {(u, t)} over all users u and time-instances t as well as the 5% quantile of throughput 5% defined as Pr {Et {(u,
t)} 5%} 5%.

C. Downlink Performance Evaluation

Fig. 3 shows the downlink performance results for full cooperation and limited cooperation. Full cooperation refers to the case
that physically separated BSs jointly apply multi-cell MIMO and relay nodes use the previously introduced ETW approach. By
contrast, limited cooperation refers to the case that only BSs at the same site are cooperating (no inter-site backhaul) and relay
nodes use the simplification in (25), i. e., no super-position coding is applied.
Examining Fig. 3 shows, with regard to , a performance advantage of multi-cell MIMO over both relaying protocols and a quite
significant performance gain over a conventional system. Fig. 3 further shows the performance results for limited cooperation
(solid lines) where the BS cooperation is limited to the same site and excludes inter-site cooperation. The average throughput
of multi-cell MIMO remains almost

Downlink Throughput [MBit/s]

10

+
+

10

10

+ Multi-Cell MIMO
Conventional
2 Relays, relay-only
2 Relays, mixed

5%

102

0.05

0.10

0.15

0.20

0.25

RN to BS Cost Ratio RN
Fig. 3. Cost-Benefit trade-off with normalized deployment costs. Dashed lines indicate full cooperation and solid lines indicate limited cooperation.

unaffected by this limitation but the 5% quantile throughput 5% decreases by about 50%, which suggests that users at the cell
edge between two sites especially benefit from an inter-site BS cooperation. We also apply this limitation to the link between BS
and RN of both relay-based protocols, which is indicated by solid lines in Fig. 3. In contrast to multi-cell MIMO, this limitation has
little effect on the protocol performance, which suggests that users at the cell edge between two sites are served by relay nodes
instead of BSs. However, only the integrated approach (indicated by mixed in Fig. 3) can slightly improve 5% for RN < 0.15
compared to multi-cell MIMO, which implies that in most cases multi-cell MIMO is preferable over relaying for the downlink.
Among other reasons, these figures result from the orthogonality constraint at RNs, the requirement to use the same resources
for BS-RN links and RN-UT links, and the fact that in the considered scenario each BS is equipped with four antennas and
therefore serves up to four UTs. By contrast, each BS serves only two RNs per cell on the same time-frequency resource.
However, even though an increased number of RNs improves the multiplexing gain on the BS-RN links, it also increases the
deployment costs.
Now we examine the results for the energy-normalization in Fig. 4. Although for cost-normalized deployments relaying was
partly able to meet the performance figures of multi-cell MIMO, it is now outperformed for all relay transmit powers by at least
25%. The maximum average throughput and 5% quantile throughput drop by about 25% and 70% compared to the scenario
without energy-normalization. The best average throughput is achieved for relay transmit power 28 dBm and BS transmit power
24.5 dBm (per antenna) as most users are assigned to RNs and those assigned to BSs have very good channel conditions. This
result will certainly change if the constant energy consumption Pconst,RN at a RN is reduced and the BS transmit power is
increased.
Finally, Fig. 5 shows the results if both overall cost and the

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

10

10
+

10

10

+ Multi-Cell MIMO

5%

Conventi onal

2 Relays, relay-only
2 Relays, mixed

102

24

26
28
Relay Node Power in dBm

30

Uplink Throughput [MBit/s]

Downlink Throughput [MBit/s]

10

101

102
103
104

+
+

+ Multi-Cell MIMO
Conventional
2 Relays, relay-only
2 Relays, mixed
0

0.05

0.10

0.15

5%

0.20

0.25

RN to BS Cost Ratio RN
Fig. 4. Energy-Benefit downlink trade-off with normalized energy consump-tion. Dashed lines indicate full cooperation and solid lines indicate
limited cooperation.

Fig. 6. Cost-Benefit trade-off with normalized deployment costs. Dashed lines indicate full cooperation and solid lines indicate limited cooperation.

Downlink Throughput [MBit/s]

10

+
+

10

101

+ Multi-Cell MIMO
Conventional
2 Relays, relay-only
2 Relays, mixed

5%

10

0.05

0.10

0.15

0.20

0.25

RN to BS Cost Ratio RN
Fig. 5. Cost-Benefit downlink trade-off with normalized energy and costs.

energy consumption are normalized as described in Section III-D. For each relative cost-factor RN, Fig. 5 only plots the
maximum achievable performance (maximized over all possible PTx,RN and PTx,site). In all cases the maximum is again
achieved by PTx,RN = 28 dBm. We can observe a significant performance loss of relaying compared to multi-cell MIMO
due to the normalized energy. Compared to the cost-normalized case, the performance drops by almost 30%. The
integrated approach and the relaying-only approach achieve comparable results as a result of the low transmission power
at the BS.

D. Uplink Performance Evaluation

We now explore the uplink-performance, which is shown in Fig. 6 for a cost-normalized evaluation. By contrast to

the downlink results, the integrated approach significantly improves the average and 5% quantile throughput. However, again the
5% of the relay-only strategy drops severely, which is caused by those users located close to the BS. The increased RAP
density implies a lower average path loss of mobile terminals to their assigned RAP, which increases the average throughput
performance if relay nodes are deployed. We can further see from Fig. 6 that relaying especially improves the 5% quantile
throughput as the path-loss of cell-edge users is significantly reduced by relaying. This indicates that relaying is a reasonable
choice to improve the cell-edge performance. Again, limited cooperation has almost no effect on multi-cell MIMO and relaying,
which suggests that the system could easily waive an inter-site cooperation in order to reduce the deployment and infrastructure
costs.

VI. CONCLUSIONS
In this paper, we compared multi-cell MIMO and relaying at a system level on the basis of three different normalization criteria.
Given the macro-cellular scenario as previously in-troduced, relaying provides significant performance gains for uplink
transmission even if simplifications such as limited BS cooperation and no super-position coding are used. On the other hand, in
the given scenario and downlink, multi-cell MIMO achieves the best performance. Thus, our results suggest the use of relaying in
the uplink where UTs suffer from high path loss and low transmission power, and multi-cell MIMO in the downlink.
This paper applied the macro-cellular scenario as it is widely accepted and used in the IMT-Advanced evaluation process [11].
Nonetheless, there are different and more specific scenarios such as the Manhattan scenario where the con-clusions may
change due to the very specific requirements following from the scenarios deployment and channel model. For instance, in a
very dense deployment with a very high ratio of indoor coverage, relaying may further improve the uplink

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS

and downlink performance benefit over multi-cell MIMO [43], [44]. In this case, a provider is forced to deploy additional
indoor base stations to offer an acceptable service quality and therefore relays become even more beneficial. A detailed
analysis of the introduced normalization criteria in the context of a micro/femto-cellular (Manhattan) scenario will be part of
our future research.

REFERENCES
[1] S. Shamai and B. Zaidel, Enhancing the cellular downlink capacity via co-processing at the transmitting end, in IEEE Vehicular Technology Conference (VTC), vol.
3, Rhodes, Greece, May 2001, pp. 17451749.

[2] S. Vishwanath, N. Jindal, and A. Goldsmith, Duality, achievable rates, and sum-rate capacity of Gaussian MIMO broadcast channels, IEEE
Transactions on Information Theory, vol. 49, no. 10, pp. 26582668, October 2003.
[3] H. Weingarten, Y. Steinberg, and S. Shamai, The capacity region of the Gaussian MIMO broadcast channel, in IEEE International Symposium on Information
Theory, Chicago (IL), USA, June 2004, p. 174.

[4] R. Ahlswede, Multi-way communication channels, in International Symposium on Information Theory, Tsakkadsor, Armenian SSR, 1971, pp.
2352.
[5] H. Liao, Multiple access channels, Ph.D. dissertation, Department of Electrical Engineering, University of Hawaii, Honolulu, 1972.
[6] T. Cover and A. E. Gamal, Capacity theorems for the relay channel,

IEEE Transactions on Information Theory, vol. 25, no. 5, pp. 572584, September 1979.
[7] R. Pabst, B. Walke, D. Schultz, P. Herhold, H. Yanikomeroglu,
S. Mukherjee, H. Viswanath, M. Lott, W. Zirwas, M. Dohler, H. Agh-vami, D. Falconer, and G. Fettweis, Relay-based deployment concepts for
wireless and mobile broadband radio, IEEE Communications Mag-azine, vol. 42, no. 9, pp. 8089, September 2004.
[8] D. Gesbert, S. Hanly, H. Huang, S. Shitz, O. Simeone, and W. Yu, Multi-cell MIMO coopeative networks: A new look at interference,

IEEE Journal on Selected Areas in Communications, vol. 28, no. 9, December 2010.
[9] P. Rost, G. Fettweis, and J. Laneman, Opportunities, constraints, and benefits of relaying in the presence of interference, in IEEE
International Conference on Communications, Dresden, Germany, June 2009.
[10] WINNER, IST-Winner, http://www.ist-winner.org, January 2009.
[11] ITU-R, Requirements, evaluation criteria and submission templates for the development of IMT-Advanced, International
Telecommunication Union (ITU), Tech. Rep. M.2133, 2008.
[12] IST-4-027756 WINNER II, D6.13.7 Test scenarios and calibration issue 2, December 2006.
[13] ITU-R, Guidelines for evaluation of radio interface technologies for IMT-Advanced, International Telecommunication Union (ITU), Tech.
Rep. M.2135, 2009.
[14] S. Shamai, O. Somekh, O. Simeone, A. Sanderovich, B. Zaidel, and

[15]
[16]
[17]
[18]
[19]
[20]
[21]

H. Poor, Cooperative multi-cell networks: Impact of limited-capacity backhaul and inter-users links, in Joint Workshop on Coding and
Communications, Durnstein, Austria, October 2007.
O. Sahin and E. Erkip, Achievable rates for the Gaussian interference relay channel, in IEEE Global Communications Conference, Washington D.C., USA, December 2007.
O. Sahin, E. Erkip, and O. Simeone, Interference channel with a relay: Models, relaying strategies, bounds, in UCSD ITA Workshop, San
Diego (CA), USA, February 2009.
I. Maric, R. Dabora, and A. Goldsmith, On the capacity of the interfer-ence channel with a relay, in International Symposium on Information
Theory, Toronto, Canada, July 2008.
R. Irmer, H. Droste, P. Marsch, M. Grieger, G. Fettweis, S. Brueck, H.-P. Mayer, L. Thiele, and V. Jungnickel, Coordinated multipoint:
Concepts, performance, and field trial results, IEEE Communications Magazine, vol. 49, no. 2, February 2011.
P. Marsch and G. Fettweis, Eds., Coordinated Multi-Point in Mobile Communications: From Theory to Practice. Cambridge University Press,
2011.
3GPP, Work Item: Coordinated Multi-Point Operation for LTE, 3GPP, Tech. Rep., March 2012.
IEEE Computer Society, IEEE standard for local and metropolitan area networks - part 16: Air interface for broadband wireless access
systems, IEEE, Tech. Rep., May 2011.

10

[22] M. Werner, P. Moberg, and P. Skillermark, Cost assessment of radio access network deployments with relay nodes, in IST Mobile & Wireless
Communications Summit, Stockholm, Sweden, June 2008.

[23] E. Hossain, D. I. Kim, and V. K. Bhargava, Eds., Cooperative Cellular Wireless Networks. Cambridge University Press, 2011.
[24] X. Gong, C. Thejaswi, J. Zhang, and H. Poor, Opportunistic cooperative networking: To relay or not to relay? IEEE Journal on Selected Areas
in Communications, vol. 30, no. 2, February 2012.
[25] Z. Xu, C. Xiong, C. Yang, S. Zhang, Y. Chen, and S. Xu, Energy-efficient wireless communications: Tutorial, survey, and open issues,

IEEE Wireless Communications Magazine, vol. 18, no. 6, December 2011.

[26] L. Huijiang, N. Jaggi, and B. Sikdar, Relay scheduling for cooperative communications in sensor networks with energy harvesting, IEEE
Transactions on Wireless Communications, vol. 10, no. 9, September 2011.
[27] E. Ciftcioglu, Y. Sagduyu, R. Berry, and A. Yener, Cost-delay tradeoffs for two-way relay networks, IEEE Transactions on Wireless Communications, vol. 10, no. 12, December 2011.
[28] P. Rost, Opportunities, benefits, and constraints of relaying in mobile communication systems, Ph.D. dissertation, Technische Universitat
Dresden, Dresden, Germany, 2009.
[29] F. Richter, A. Fehske, and G. Fettweis, Traffic demand and energy efficiency in heterogeneous cellular mobile radio networks, in IEEE Vehicular
Technology Conference (VTC), Taipeh, Taiwan, May 2010.

[30] M. Joham, W. Utschik, and J. Nossek, Linear transmit processing in MIMO communications systems, IEEE Transactions on Signal
Processing, no. 8, pp. 27002712, August 2005.
[31] M. Costa, Writing on dirty paper, IEEE Transactions on Information Theory, vol. IT-29, no. 3, pp. 439441, May 1983.
[32] P. Marsch and G. Fettweis, On downlink network MIMO under a constrained backhaul and imperfect channel knowledge, in IEEE Global Communications
Conference, Honolulu (HI), USA, December 2009.

[33] J. Holfeld, V. Kotzsch, and G. Fettweis, Order-recursive precoding for cooperative multi-point transmission, in 2010 International ITG/IEEE Workshop
on Smart Antennas, Bremen, Germany, February 2010.

[34] W. Yu and T. Lan, Transmitter optimization for the multi-antenna downlink with per-antenna power constraints, IEEE Transactions on Signal
Processing, vol. 55, no. 6, pp. 26462660, June 2007.
[35] D. Wubben, J. Rinas, R. Bohnke, V. Kuhn, and K. Kammeyer, Efficient algorithm for detecting layered space-time codes, in International
ITG Conference on Source and Channel Coding, Berlin, Germany, January 2002.
[36] T. Han and K. Kobayashi, A new achievable rate region for the interference channel, IEEE Transactions on Information Theory, vol. IT-27, no.
1, pp. 4960, January 1981.
[37] R. Etkin, D. Tse, and H. Wang, Gaussian interference channel capacity to within one bit: the symmetric case, in IEEE Information Theory Workshop,
Chengdu, China, October 2006, pp. 601605.

[38] , Gaussian interference channel capacity to within one bit, IEEE Transactions on Information Theory, vol. 54, no. 12, pp. 55345562,
December 2008.
[39] IST-4-027756 WINNER II, D1.1.1 WINNER II Interim channel mod-els, November 2006.
[40] , D1.1.2 WINNER II channel models, September 2007.
[41] S. Weinstein and P. Ebert, Data transmission by frequency-division multiplexing using the discrete fourier transform, IEEE Transactions on
Communications, vol. 19, no. 5, pp. 628634, October 1971.
[42] L. Cimini, Analysis and simulation of a digital mobile channel using orthogonal frequency division multiplexing, IEEE Transactions on Information
Theory, vol. 33, no. 7, pp. 665675, July 1985.

[43] CELTIC / CP5-026 WINNER+, Enabling Techniques for LTE-A and beyond, July 2010.
[44] G. Fettweis, J. Holfeld, V. Kotzsch, P. Marsch, E. Ohlmer, Z. Rong, and P. Rost, Field trial results for LTE-advanced concepts, in IEEE
Intnl. Conf. on Acoustics, Speech and Signal Processing (ICASSP), Dallas (TX), USA, March 2010.