Vous êtes sur la page 1sur 8

On the Performance of Cooperative

Amplify-and-Forward Relay Networks


Patrick Herhold, Ernesto Zimmermann, Gerhard Fettweis
Technische Universit at Dresden Vodafone Chair Mobile Communications Systems herhold@ifn.et.tu-dresden.de
Abstract
This paper provides a general analysis of cooperative amplify-and-forward relay networks and gives a unied and
comparative view of various proposed systems. The studied cooperative relay networks offer spatial diversity gains
that conventional multi-antenna systems often fail to achieve due to limitations of the feasible number of antennas per
terminal and correlated propagation. We rst present a general analysis for the case of multiple antennas per node,
which is subsequently used to investigate the relative attractiveness of cooperative schemes for various parameters
that include SNR, spectral efciency, and spatial correlation. We conclude that such virtual antenna arrays are an
attractive option to overcome the drawbacks of conventional relaying systems.
1 Introduction
The exploitation of diversity is often considered to be
one of the last frontiers in wireless communications.
This motivated for strong research efforts in the area of
multi-antenna techniques. The power of such multiple-
input multiple-output (MIMO) systems results from
beamforming gains and diversity gains. Achieving the
latter calls for uncorrelated propagation, which usu-
ally necessitates a spatial separation of the individual
antenna elements that is hardly feasible in small-size
terminals.
In a different context, relaying is often regarded as a
means of improving the performance of infrastructure-
based networks by increasing their coverage. However,
the reduced end-to-end pathlosses come at the cost of
an inherent rate increase and the repetition-coded nature
of relaying systems. Yet, relaying is a viable option for
infrastructure-based networks, and it is a basic means
for service provisioning in mobile ad-hoc networks.
Cooperative relaying brings together the worlds of
MIMO systems and relaying. By allowing cooperation
of mobile terminals, virtual antenna arrays [1], [2] can
be built that overcome the above described drawbacks
of correlation and spatial limitations. In general, co-
operative networks employ relay stations that receive
signals from a source and resend a processed version
of these signals to the intended destination. The destina-
tion node combines the signals from source and relays,
thereby exploiting useful information that conventional
relay systems partially consider as interference.
Relay processing can be classied as either decode-
and-forward or amplify-and-forward. In this paper, the
class of amplify-and-forward (AF) relay networks is
examined in the context of multi-antenna systems. We
This work was supported by Vodafone Group Research and De-
velopment.
investigate the inuence of parameters like SNR, spec-
tral efciency, and correlation on the performance of
cooperative schemes, and compare this to conventional
multi-antenna systems.
We start by reviewing related work. Following a
system description in section 3, we perform a unify-
ing analysis of various proposed systems. Results are
presented in section 5; the paper concludes in section 6.
2 Previous Work
In their ground breaking paper [1], Laneman et al.
provide a fundamental analysis of cooperative relaying
systems, thereby showing that AF schemes provide full
diversity gains. The tradeoff between diversity gains on
one side and repetition coding and rate increase on the
other side is analyzed. Although the work focuses on
single-antenna terminals, this tradeoff is equally appli-
cable for distributed multi-antenna systems. Gastpar et
al. [3] study the performance limits of a SISO link that
is assisted by an innite number of cooperating relays.
A more pragmatic approach is taken by Dohler et
al. [2], [4] in their work on virtual antenna arrays.
Neighboring terminals cooperate by forming a virtual
array; space-time block codes provide the required
orthogonality of superimposed signals. Anghel et al. [5]
take a similar approach to discuss the possibility of
having two relays assist a source-destination pair by
using Alamoutis transmit diversity technique [6] in
a distributive manner. More recently, it is studied by
Wittneben and Rankov [7] how relay nodes can help
improve MIMO transmission by acting as active scat-
terers that increase the rank of the MIMO channel.
Our aim is to provide a unifying view by compara-
tively examining the inuence of crucial parameters.
Towards this end, we rst establish a common system
model.
Source:
m
Tx
antennas
Destination:
m
Rx
antennas
M relays, each
with m
rel
antennas
...
Fig. 1. Network conguration for multiple antennas per terminal.
3 System Description
We study a general cooperative scheme as depicted in
Fig. 1, where we allow all terminals to be equipped with
multiple antennas. More specically, the source node
employs m
Tx
antennas, and the destination receives
using m
Rx
antennas. This communication is assisted
by M relays, each having m
rel
antennas. We will use
the parameter set p = {M, m
Tx
, m
rel
, m
Rx
} to describe
specic congurations.
3.1 System Model
General Protocol: Communication takes place in two
phases. In phase one, the source transmits signals
intended for the destination in a broadcast manner. All
relays and the destination receive faded noisy versions
of these signals. In the second phase, the relays retrans-
mit a processed version of their received signals to the
destination, and the destination combines the signals
received in the two phases.
Orthogonality Constraint: In contrast to conventional
RF repeaters, small-sized terminals are not capable of
receiving and transmitting simultaneously at the same
frequency due to hardware imperfections (e.g. oscil-
lations). Hence, relays must be assigned orthogonal
resources for their transmit and receive paths. For
purposes of exposition, we assume that such separation
is performed in the time domain; another approach
would be to use frequency relaying where the relays
receive and transmit in different frequency bands.
While it is of no concern for our theoretic analysis
which of these methods is chosen, it is crucial to note
the important consequences of assigning orthogonal
sub-channels. Consider time-division relaying as an
example. In direct transmission, the source transmits
K symbols in a period T to the destination. For relay
transmission that achieves the same end-to-end delay,
the source must transmit its K symbols in a period
of T/2, which is followed by the relays transmission
of these K symbols in the second half period. As a
consequence, the spectral efciency of the individual
links must at least be doubled compared to that of the
direct system. This holds in the same way for frequency
relaying: there, available bandwidth instead of time is
shared among the two phases. The impact of such a
rate increase is often neglected [2], [4], [7].
Channel Model: The frequency at fading nature
of the channel is captured by coefcients h drawn
from a circularly complex Gaussian distribution: h
CN(0,
2
), where
2
is the variance of the random
process. The magnitudes |h| follow a Rayleigh distri-
bution, and the channel powers |h|
2
are exponentially
distributed with parameter
2
. This is a widely ac-
cepted channel model; see [1], [4], [7].
Energy Normalization: In order to provide a fair
comparison, it is crucial that the total consumed en-
ergy of the relay system does not exceed that of the
corresponding direct system. We note that this is a
strict and conservative constraint; allowing the relays
to add additional power can then only increase the
attractiveness of cooperation. In our continuous-time
channel model the source transmits with power P over
a period T in the case of direct transmission. For relay
transmission, the source rst broadcasts with power P
over a period of T/2; in the second phase, the relays
use again a maximum total power of P over T/2.
Operation of Relays: Relays can enhance the system
performance by acting in two different ways:
1) Amplifying relays: Relays that simply amplify
their received signals essentially act as active re-
ectors. This provides diversity gains, and can be
understood as enhancing the rank of the MIMO
channel matrix. Most of the proposed systems
employ such amplifying relays; we will analyze
this case in Section 4.1.
2) Coding relays: Relays can employ orthogonal
coding techniques to enhance the diversity order
of the system. An example is the distributed
Alamouti method proposed in [5], where two
relays retransmit in an amplify-and-forward man-
ner, and one of the relays additionally conjugates
and negates the signal.
3.2 Studied Systems
We use the set p = {M, m
Tx
, m
rel
, m
Rx
} to describe
the specic systems under investigation. Our ref-
erence systems are the conventional direct SISO
and m
Rx
m
Tx
MIMO
1
systems (p = {0, 1, 0, 1} and
p = {0, m
Tx
, 0, m
Rx
}). We compare these with systems
that employ a single antenna at relays and destination,
since we assume that spatial limitations prevent from
using multiple antennas at these nodes.
In particular, we focus on the single-antenna cooper-
ative system studied by Laneman et al. [1], in which a
single relay assists a SISO link. See Fig. 2 for this
p = {1, 1, 1, 1} system. The idea of virtual antenna
arrays [2] with two transmit antennas at the source
and a single antenna at relay and destination is denoted
by p = {1, 2, 1, 1}. Finally, we study the distributed
1
In this paper, we refer to a MIMO system as one that provides
diversity gains through space-time coding rather than data rate
enhancements through spanning parallel channels.
{1,1,1,1} system [1] {1,2,1,1} system [2] {2,1,1,1} system [5]
Fig. 2. Studied systems. They are described by the parameter set
p = {M, m
Tx
, m
rel
, m
Rx
}. See also Table I.
Alamouti system with two relays p = {2, 1, 1, 1} [5].
See Table I for a summary.
4 Unifying Analysis
In general, a signal vector x is transmitted from the
source; these signals undergo fading on their way to
the destination. In addition, the signals pass through a
relay-amplied double-Rayleigh fading channel. These
effects can be described by a signal-propagation matrix
A. We observe that there are noise sources at the relays
and the destination receiver, which are summarized in
the noise vector n. While the destination noises occur
at the nal receiver only, the relay noises are amplied
and transmitted over fading channels to the destination;
we therefore use a matrix B to model this fact. To
summarize, we have y = Ax +Bn.
We now aim at deriving the components of this
equation for the systems of interest.
4.1 Amplifying Relays
4.1.1 General Analysis
We perform a similar derivation as in [7] to analyze
the case in which relays simply amplify-and-forward
their received signals. In phase 1, the source transmits
a signal vector x C
m
Tx
1
from its array. The direct
propagation to the destination in the rst phase is
conventionally described by a matrix H
(1,D)
C
m
Rx
m
Tx
and a noise vector n
(1,D)
C
m
Rx
1
at the destination, so
that the destinations received signal becomes
y
(1)
= H
(1,D)
x +n
(1,D)
. (1)
Similarly, in this phase each of the M relays receives
r
(1)
l
= H
(1)
l
x +n
(1,R)
l
, (1 l M) , (2)
where H
(1)
l
C
m
rel
m
Tx
is the MIMO matrix mapping
the transmitted signal x to the receive antennas of relay
l, and n
(1,R)
l
is the noise vector at the receive antennas
of this relay. To obtain a more compact description, we
form the compound signal and noise row vectors
r
(1)
=
_
r
(1)
1
T
, . . . , r
(1)
M
T

T
, r
(1)
C
Mm
rel
1
,
n
(1,R)
=
_
n
(1,R)
1
T
, . . . , n
(1,R)
M
T

T
, n
(1,R)
C
Mm
rel
1
,
and a block channel matrix
H
(1)
=
_
H
(1)
1
T
, . . . , H
(1)
M
T

T
, H
(1)
C
Mm
rel
m
Tx
,
where []
T
denotes the transpose of a matrix. Using this
notation, we can describe the communication from the
TABLE I
SUMMARY OF PROPOSED AF SCHEMES. THE BOUND EXPRESSION
(m
1
, m
2
) DEPICTS A MIMO SYSTEM WITH m
1
TRANSMIT AND
m
2
RECEIVE ANTENNAS.
p Bound Remark From
{0, 1, 0, 1} (1, 1) SISO system
{0, m
Tx
, 0, m
Rx
} (m
Tx
, m
Rx
) MIMO system
{1, 1, 1, 1} (2,1) Single relay [1]
{1, 2, 1, 1} (3,1) Two tx antennas [2]
{2, 1, 1, 1} (3,1) Two relays [5]
{M, m
Tx
, 1, m
Rx
} (m
Tx
, m
Rx
) Rank improvement [7]
sources transmit array to the virtual array spanned by
the M m
rel
relay antennas, taking place in the rst
phase, by
r
(1)
= H
(1)
x +n
(1,R)
. (3)
In phase 2, the relays retransmit an amplied ver-
sion Gr
(1)
of their received signals to the destination;
G C
Mm
rel
Mm
rel
is the diagonal gain matrix that is
conditioned so as to obey the power constraints. These
signals undergo fading from the relay antennas to the
destination; we denote this by a block channel matrix
H
(2)
C
m
Rx
Mm
rel
. The destination receives
y
(2)
= H
(2)
Gr
(1)
+n
(2,D)
= H
(2)
GH
(1)
x +H
(2)
Gn
(1,R)
+n
(2,D)
= H
(2)
GH
(1)
x + [(H
(2)
G) I]
_
n
(1,R)
n
(2,D)
_
. (4)
Note that (1) and (4) describe the signals received
at the destination in the two phases. For purposes of
further exposition, we combine both phases, obtaining
y =
_
y
(1)
y
(2)
_
=
_
H
(1,D)
H
(2)
GH
(1)
_
. .
AC
2m
Rx
m
Tx
x+
_
I 0 0
0 H
(2)
G I
_
. .
BC
2m
Rx
(2m
Rx
+Mm
rel
)
_
_
n
(1,D)
n
(1,R)
n
(2,D)
_
_
.
(5)
This form describes the complete AF protocol over
Rayleigh-fading channels for amplifying relays. It re-
mains to discuss the relay power allocation. From (2)
we note that the total power received by relay l can be
computed as the sum of the powers of the faded signals
and the noise powers, i.e.,
P
l
=
P
m
Tx
m
rel

i=1
m
Tx

k=1

[H
(1)
l
]
i,k

2
+m
rel
N , (6)
where P is the transmission power of the source, and N
is the noise power at the receiver branches of the termi-
nals, which we assume to be equal at all terminals. Note
that we assume equal transmit powers at the antennas
of the source array. A strict and simple interpretation
of the energy constraint is to allow a maximum relay
power of P/M. For this, the amplication factor of
h
1,R
h
2,R
h
1,D
h
2,D
h
R,D

Fig. 3. Example {1, 2, 1, 1} system with channel coefcients.


relay l becomes

l
=

P/M
P
l
. (7)
These values
l
form the diagonal gain matrix G.
4.1.2 Example
To illustrate the above analysis, consider the example
system depicted in Fig. 3. The MIMO channel matrix
from source to destination is H
(1,D)
= [h
1,D
h
2,D
]. Since
there is only one relay, the block channel matrix from
source to the relay becomes H
(1)
= [h
1,R
h
2,R
]. Even
simpler, the gain matrix is the amplication factor of
the relay, G = , and the channel matrix from relay
to destination reduces to H
(2)
= h
R,D
. Equation (5) then
reads
y =
_
h
1,D
h
2,D
h
R,D
h
1,D
h
R,D
h
2,D
_
x + (8)
_
1 0 0
0 h
R,D
1
_
_
_
n
(1)
D
n
(1)
R
n
(2)
D
_
_
.
The noise vector contains the noises at relay and desti-
nation in the rst phase, and the noise at the destination
in the second phase. The amplication factor can be
determined from (7) using the power received at the
relay:
P
1
=
P
2
_
|h
1,R
|
2
+|h
2,R
|
2
_
+N .
The description derived in this section captures the
operation of amplify-and-forward relays. We now turn
to the case in which relays additionally employ a special
form of distributed coding.
4.2 Distributed Alamouti Coding
Further enhancing the diversity order of the system
requires that orthogonal signals be transmitted from
the relay nodes. For amplify-and-forward networks,
transmitting orthogonal waveforms from the relays is a
problem that is difcult (if not impossible) to solve. For
the case of two relays though, a distributed Alamouti
system is feasible; see [5] for details of this p =
{2, 1, 1, 1} system. Focusing on just one symbol period,
we note that we can again divide transmission in two
phases. In the second phase, one of the two relays
retransmits a conjugated and negated version of the
signal it has received from the source, thus imple-
menting a distributed version of Alamoutis space-
time code.
2
In addition, the relays need to reconstruct
the signal phases prior to retransmission. This ensures
separability of the two signals sent from the relays to
the destination. Assuming equal amplied noise powers
at the relays and equal fading coefcients for both relay
paths, the Alamouti system corresponds to a maximum
ratio combiner, which can equivalently be modelled by
3
y =
_
_
y
0
y
1
y
2
_
_
=
_
_
h
0
h
(2)
1

1
h
(1)
1
h
(2)
2

2
h
(1)
2
_
_
. .
A
x+
_
_
1 0 0 0
0
1
h
(2)
1
0 1
0 0
2
h
(2)
2
1
_
_
. .
B
_

_
n
(1,D)
n
(1,R)
1
n
(1,R)
2
n
(2,D)
_

_
,
(9)
where the rst row is the signal received in phase 1,
and rows 2 and 3 describe the signals received by the
destination in the second phase. In this equation, h
0
is the fading coefcient from source to destination in
phase 1, and h
(p)
r
models the fading coefcient to/from
relay r in phase p. The amplication factor of relay r is

r
. The noise terms n
(p,D)
and n
(1,R)
2
, n
(1,R)
2
are those
at the destination in phase p and at the two relays in
phase 1, respectively.
It has been shown in [1] that it is best if the relays
retransmit using the maximum amplication allowed
by the power constraint. We again assume a simple
and conservative power allocation that allows a relay
to transmit at most with power P/M = P/2, so that

r
=

P/2

h
(1)
r

2
P +N
for r {1, 2} . (10)
Note that this is exactly the same power allocation as
described by (6) and (7).
We have obtained a description of the distributed
Alamouti system with two relays. Theoretically, the
above approach can be generalized for M > 2, thereby
eventually leading to the results in [3]. However, lack of
appropriate coding techniques for amplify-and-forward
networks does not motivate for this.
In the following, we assume the p = {2, 1, 1, 1}
system to be implemented as described in this section,
i.e., we enhance diversity by orthogonal coding instead
of having the relays simply amplify.
2
[5] details the protocol for two consecutive symbol periods; due
to symmetry, it sufces to consider just one period in our analysis.
3
The assumptions of having equal fading coefcients on both relay
paths is not fully appropriate, since we assume both paths to be
uncorrelated. Hence, we only provide a bound on the performance
of the system.
4.3 Capacity
So far we have provided a description of cooperative
amplify-and-forward relay networks using the form
y = Ax +Bn. (11)
This has the powerful advantage that we can now
determine the instantaneous capacity of the underlying
system. It is given by the mutual information in the
vectors x and y as
C = log
2
det
_
I +
_
AR
x
A

_ _
BR
n
B

_
1
_
. (12)
We provide a derivation of this result in the ap-
pendix. Its application requires Gaussian inputs x and
knowledge of the channel A at the destination re-
ceiver. In this equation, R
x
= E
_
xx

_
is the co-
variance matrix of the transmitted signal vector, and
R
n
= E
_
nn

_
is the noise covariance matrix. For our
considered systems, we have R
x
= (P/m
Tx
) I
m
Tx
,
and R
n
= N I
n
, where n = 2m
Rx
+M m
rel
.
4.4 Diversity Performance
The outage probability has previously been used in [1]
to describe the performance of cooperative schemes;
we will rely on the excellent notations in this work.
Assume we want to communicate with a certain xed
rate R. An outage occurs if the instantaneous channel
capacity C is insufcient to support this rate R, i.e.,
the outage event is C < R. The outage probability is
the probability of this event, p
out
= Pr[C < R], which
is determined by the statistical properties of the fading
coefcients h, the average signal-to-noise ratio SNR,
and the desired rate R.
We can now conveniently capture the implications
of the orthogonality constraint by noting that we need
to double the rate R, i.e., the spectral efciency, in
the relay case. For cooperative protocols the outage
probability is therefore p
out
AF
= Pr[C < 2R], while we
have the more relaxed condition p
out
Direct
= Pr[C < R]
for conventional direct MIMO and SISO transmission.
C and R are measured in bit/channel use = bit/s/Hz.
Analytical results can be obtained for simple sce-
narios (see [1] for the case of single-antenna terminals
with a single relay); however, such solutions are not
attainable for the general case. We will therefore use
Monte-Carlo simulations of the outage event to numer-
ically evaluate the outage probability using the results
of (5) and (9) in (12).
4.5 Parameters of Interest
The outage probability increases if a higher spectral
efciency R is required. The number of dimensions
of the compound channel, in our case the number of
antennas, determines the diversity order of the system.
For a given rate R, this diversity order denes the slope
with which the outage probability decays as the SNR
increases.
0 5 10 15 20 25 30 35
10
6
10
4
10
2
10
0
Direct {0,1,0,1}
SNR [dB]
O
u
t
a
g
e

p
r
o
b
a
b
i
l
i
t
y



p
o
u
t
{1,1,1,1}
Single relay
{2,1,1,1}
2 relays, single ant
{1,2,1,1}
2 tx ants, single relay
MISO Bound {0,3,0,1}
R=1 bits/s/Hz
Correlation =0
Fig. 4. Outage probability of various cooperative AF schemes as a
function of the SNR. While direct transmission achieves rst-order
diversity only, the p = {1, 1, 1, 1} system proposed by Laneman
et al. achieves second order diversity. The other relaying systems
employ a total of three transmit antennas, therefore they achieve
third-order diversity. The offsets with respect to the MIMO bound
are caused by double transmission rate, repetition coding and noise
amplications. Spectral efciency: R = 1 bit/s/Hz.
An additional requirement for exploiting the potential
benets of multi-antenna systems is a low correlation
of the fading characteristics at neighboring antenna
elements. A high correlation, as it occurs in line-of-
sight cases, reduces the diversity-order of the system
dramatically. One such worst-case scenario is studied in
[7], where the performance of a fully correlated MIMO
system (channel matrix with rank one) is improved by
deploying scattering relays. We take a more general
approach by considering the correlation as a parameter.
Throughout the rest of the paper, we assume channel
state information to be available at the receiver only.
5 Results
In the following we will rst concentrate on the case of
a single antenna at the destination, i.e., m
Rx
= 1. For a
source with a transmit antenna array, this corresponds to
a cellular downlink scenario where transmission from a
base station antenna array is enhanced by cooperating
single-antenna terminals. We start by comparing the
performance of the systems in the absence of spatial
correlation.
5.1 General Performance
Fig. 4 shows the outage probability of the studied
cooperative schemes as a function of the average SNR
for a spectral efciency of R = 1 bit/s/Hz. We see that
cooperative amplify-and-forward networks achieve a di-
versity order according to the total number of antennas
in the system: the {1, 1, 1, 1} system investigated by
Laneman [1] has a single relay helping transmission
from a source with a single antenna, hence it emulates
a (2,1) MIMO system with two transmit antennas.
0 0.2 0.4 0.6 0.8 1
2
0
2
4
6
8
10
12
SISO reference {0,1,0,1}
{1,1,1,1} Single relay
{2,1,1,1} 2 relays, single antenna
{1,2,1,1} 2 tx ant, single relay
MISO bound {0,3,0,1}
Correlation
S
N
R

g
a
i
n

o
v
e
r

d
i
r
e
c
t

S
I
S
O

t
r
a
n
s
m
i
s
s
i
o
n

[
d
B
]

R=1 bit/s/Hz, p
out
=10
2
Fig. 5. Inuence of correlation on the performance of MIMO and
cooperative schemes: SNR gain of these systems with respect to direct
SISO transmission for an outage probability p
out
= 10
2
. For very
high correlation, the performance of the MISO system degrades to
that of the SISO link, while the distributed arrays are less affected.
Both the system with two transmit antennas at the
source {1, 2, 1, 1} [2] and the {2, 1, 1, 1} system [5]
and achieve third order diversity by emulating a (3,1)
MIMO system. The system with two relays suffers from
noise amplication and an additional Rayleigh fading
channel, which causes an offset of approx. 1 dB with
respect to the single-relay system where two antennas
are available at the source node.
5.2 Inuence of Correlation
We now assume the propagation from the transmit array
to the receive antenna at the destination to be correlated.
More precisely, neighboring antenna elements at the
source array are correlated with a correlation coefcient
, where 0 1; yet, propagation to different
nodes is mutually independent. We assess the SNR gain
of both the conventional correlated MISO system and
the cooperative schemes with respect to direct SISO
transmission if an outage probability of p
out
= 10
2
is
to be achieved. Fig. 5 depicts the results.
The performance of the MISO system degrades as the
correlation increases; eventually, it is reduced to that
of the SISO link for line-of-sight conditions ( = 1).
Cooperative schemes are resistant to correlation as
paths to different terminals are mutually independent.
Note that the system with two transmit antennas at
the source {1, 2, 1, 1} exhibits the same performance
as the {1, 1, 1, 1} system for high correlation: in this
case, there is no benet from using an array at the
source, and the single relay improves the diversity order
by one. Irrespective of the correlation, all cooperative
schemes achieve a gain of at least 5 dB over direct SISO
transmission for the specic case of R = 1 bit/s/Hz; the
inuence of R is examined in the following.
2 4 6 8 10 12
10
5
0
5
10
SISO reference {0,1,0,1}
{1,1,1,1}
Single relay
{2,1,1,1}
2 relays, single antenna
{1,2,1,1}
2 tx antennas, single relay
MISO bound {0,3,0,1}
MISO bound {0,2,0,1}
Spectral Efficiency R [bit/s/Hz]
S
N
R

g
a
i
n

o
v
e
r

d
i
r
e
c
t

S
I
S
O

t
r
a
n
s
m
i
s
s
i
o
n

[
d
B
]
p
out
=10
2

Fig. 6. Inuence of spectral efciency (transmission rate): SNR
gain of cooperative systems with respect to direct SISO transmission
for an outage probability p
out
= 10
2
. The effective SNR loss for
high rates demonstrates how these systems suffer from the resulting
double spectral efciency on the relaying links. Correlation: =0.0.
5.3 Inuence of Spectral Efciency
Recall that the orthogonality constraint requires the
assignment of orthogonal resources for the relay trans-
mission, which in turn calls for doubling the spectral
efciency on the individual links in relaying systems.
In addition, relaying systems suffer from the implica-
tions of spatial repetition coding that is employed by
such protocols. The relative attractiveness of relaying
and cooperation is therefore inuenced by converse,
competing trends: diversity gains are traded for a
rate increase and the implications of spatial repetition
coding.
The impact of both disadvantages heavily depends
on the end-to-end spectral efciency R. We there-
fore investigate the SNR gains as a function of R
uncorrelated propagation at an outage probability of
p
out
= 10
2
. The results in Fig. 6 imply a strong
inuence of the spectral efciency, or transmission
rate, on the attractiveness of cooperative schemes: for
high spectral efciencies, direct transmission eventually
becomes favorable. In other words, the closer we ap-
proach channel capacity, the more we suffer from the
double use of the channel that is caused by repetition
coding and doubled transmission rate. To illustrate
this argument, consider the SISO link that operates at
R = 1 bit/s/Hz. From Fig. 4 we see that SNR =20 dB
is required to achieve the desired outage probability of
p
out
= 10
2
. According to Shannon, the mean capacity
for this SNR is C =6.7 bit/s/Hz, so we operate at a
fraction of R/C = 0.15 of channel capacity in the case
of direct transmission. This increases to 2R/C=0.3 in
the cooperative case.
Interestingly, there is one notable exception in terms
of susceptibility of increased transmission rates: the
{1, 2, 1, 1} system represents a compromise between
the conventional MISO system and fully distributed an-
tenna systems. It achieves SNR gains over SISO trans-
mission for all rates. This is justied by the fact that
for this protocol spatial repetition coding contributes
to a smaller extent to the overall redundancy, while
two transmit antennas constantly provide a signicant
degree of diversity.
4
Alternatively, one can argue that
using two transmit antennas at the source helps to
improve the broadcast link, which is of benet for both
the relay and the destination. In contrast, the system
using two relays but just a single source antenna suffers
from the bottleneck of the broadcast transmission.
To summarize, cooperative schemes exhibit draw-
backs if the end-to-end transmission rate increases. For
conventional non-cooperative relaying systems, [8] has
shown a similar tradeoff at system level that leads
to the same conclusion: there, it was shown that the
attractiveness of relaying schemes decreases as the
network load increases.
5.4 Rank Improvement Using Relays
In the following we shift our focus to a different
scenario by studying a 33 MIMO system that suffers
from fully correlated propagation from the source array
to the destination array, denoted by
MIMO
= 1. Single-
antenna relays are placed between source and relay so
as to improve the eigenvalue distribution of the resulting
compound channel matrix. In contrast to [7], we assume
that propagation from the source to the relays and from
the relays to the destination array is also correlated
with a variable correlation coefcient
r
. The rank of
the compound channel matrix A in equation (5) is
improved whenever
r
< 1; the exact rank and therefore
the diversity gains depend on the specic correlation
r
and the number of scattering relays M.
Fig. 7 depicts the SNR gain over conventional
SISO transmission of the rank-improving system for
R = 1 bit/s/Hz as a function of the correlation coef-
cient
r
that affects propagation to and from each of
the relays. The parameter in this plot is the number
of assisting relay stations M. We clearly observe that
relays can help improve the situation; however, com-
pared to the uncorrelated MIMO bound the cooperative
system again suffers from the implications of rate
increase and repetition coding. While the uncorrelated
MIMO system would achieve a gain of more than 20
dB, the cooperative scheme yields approximately 16
dB. This should be compared to the 4.8 dB gain that is
achieved by the fully correlated MIMO system (array
gain provided by three receive antennas).
4
Indeed, this result coincides with what has been shown in [1]:
orthogonality has a much weaker negative impact than repetition
coding. Although the {1, 2, 1, 1} system fully suffers from the
orthogonality constraint, it achieves SNR gains by avoiding the full
implications of repetition coding as it provides a signicant portion
of its redundancy through orthogonal coding from its transmit array.
For full correlation ( = 1), the system eventually degrades to the
{1, 1, 1, 1} system as expected (results not shown).
0 0.2 0.4 0.6 0.8 1
0
5
10
15
20
25
SISO reference {0,1,0,1}
Uncorrelated MIMO bound {0,3,0,3}
MIMO
=0
Correlated MIMO bound {0,3,0,3}
MIMO
=1
M=1
M=3
M=5
Correlation to/from relays
r
S
N
R

g
a
i
n

o
v
e
r

d
i
r
e
c
t

S
I
S
O

t
r
a
n
s
m
i
s
s
i
o
n

[
d
B
]

R=1 bit/s/Hz,
p
out
=10
2
Fig. 7. Inuence of cooperative relays and correlation on the
performance of rank-improving AF schemes: SNR gains with respect
to direct SISO transmission as a function of the correlation of the
relay channels. By acting as active scatterers, relays can improve the
performance of the fully correlated MIMO system by almost 10 dB.
Spectral efciency R = 1 bit/s/Hz.
5.5 Implementation Issues
Assigning orthogonal resources to the receive and trans-
mit sub-channels of a relay station can theoretically be
done in all domains: time, frequency, code, and even
space. However, due to limitations in RF hardware (e.g.,
oscillations), only time- and frequency relaying are
currently considered to be feasible. Realizing amplify-
and-forward schemes in the frequency domain requires
(i) a conversion from one carrier to another in the
relay stations, and (ii) a combining from two different
carriers in the destination. A time-division approach
would require storing or delaying an analog signal in
the relay prior to retransmission, the same holds for
the combining process. Cross-terminal synchronization
is another important aspect.
Finally, it is important to note that our capacity
results hold only when the destination has full knowl-
edge of the effective channel A, while the respective
transmitters do not possess any such channel state
information. The knowledge can be obtained at the
destination receiver by the use of a priori known pilot
sequences that pass through the same channel as the
signals to be detected.
6 Conclusions
We have provided a unied view of various amplify-
and-forward protocols. Under a strict normalization of
energy, bandwidth, and end-to-end information data
rate, we have analyzed their capacity and comparatively
examined their outage probabilities and performance
gains with respect to conventional direct transmission.
All schemes achieve full diversity in the number of
antennas; their relative performance merits are deter-
mined by noise amplication at the relays and the exact
conguration of the distributed array. It was shown
that cooperative schemes can bring diversity benets
to single-antenna terminals: the protocols achieve SNR
gains of 5 to 9 dB for a spectral efciency of 1 bit/s/Hz
at an outage probability of p
out
= 10
2
. For highly
correlated scenarios, these schemes have the power to
outperform conventional MIMO systems.
Most importantly, we have outlined the tradeoff of
the studied static AF protocols: diversity gains come
at the cost of spatial repetition coding and a double
spectral efciency (transmission rate) on the individual
links. For high rates, the achieved diversity gains do not
outweigh the drawbacks caused by the double use of the
channel, so that for operation close to channel capacity
direct communication eventually is more attractive
even given our idealistic assumptions. Hence, for the
studied static schemes, relays are primarily attractive
for low spectral efciencies; there, the emulated MIMO
schemes are used for exploiting diversity rather than
for increasing data rate. By relying only partially on
spatial repetition coding, systems that employ both
physical and distributed arrays represent an interesting
alternative. An example is the {1, 2, 1, 1} system.
Finally, it was shown that scattering relays can
improve MIMO system performance if these are con-
strained by correlated propagation. Using 3 relays to
improve a fully correlated 3 3 system yields gains
of almost 10 dB.
Acknowledgements
The authors gratefully acknowledge the funding re-
ceived from Vodafone Group Research and Develop-
ment. Furthermore, we would like to thank W. Rave
for the valuable discussions and comments.
APPENDIX
MUTUAL INFORMATION
Problem Statement: Assume a complex received signal
vector y C
m
y
1
is given by
y = Ax +Bn,
where is x C
m
x
1
is the complex transmitted signal,
Aand Bare matrices of corresponding dimensions, and
n C
m
y
1
is a zero-mean complex Gaussian noise
vector. We wish to determine the instantaneous capac-
ity C of this system, i.e. the mutual information I(x; y).
General Solution: The mutual information can be
expressed in terms of entropy h,
I(x; y) = h(y) h(y|x)
(a)
= h(y) (h(Ax|x)
. .
=0
+h(Bn|x))
= h(y) h(Bn) , (13)
where (a) follows from the independence of x and n.
The entropy h(y) of a complex Gaussian vector is given
by ([9], p. 231)
h(y) = log
2
((e)
m
y
det(R
y
)) .
In this equation, R
y
is the covariance matrix of the
vector y, given by R
y
= E
_
yy

_
, and a

is the
Hermitian of a. Evaluating the covariance matrix for
the received vector y yields
R
y
= E
_
yy

_
= E
_
(Ax +Bn)(Ax +Bn)

_
(a)
= E
_
Axx

_
+ E
_
Bnn

_
= AR
x
A

+BR
n
B

,
where (a) follows from independence of the signal x
and noise n. E
_
xx

_
= R
x
is the covariance matrix of
the transmitted signal vector; likewise, E
_
nn

_
= R
n
is the noise covariance matrix. Hence, we have
h(y) = log
2
_
(e)
m
y
det
_
AR
x
A

+BR
n
B

_
,
(14)
and, similarly,
h(Bn) = log
2
_
(e)
m
y
det
_
BR
n
B

_
. (15)
Using equations (14) and (15) in (13) results in
I(x; y) =log
2
_
det
_
AR
x
A

+BR
n
B

_
det (BR
n
B

)
_
=log
2
det
_
_
AR
x
A

+BR
n
B

_ _
BR
n
B

_
1
_
=log
2
det
_
I +
_
AR
x
A

_ _
BR
n
B

_
1
_
.
(16)
The conventional MIMO results can easily be obtained
from this relation by noting that in these cases the noise
amplication matrix is an identity matrix, i.e., B = I,
and the noise covariance matrix is a diagonal matrix of
noise powers (R
n
= N I).
REFERENCES
[1] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative
Diversity in Wireless Networks: Efcient Protocols and Outage
Behavior, IEEE Trans. Inform. Theory, Oct. 2003, (accepted
for publication).
[2] M. Dohler, F. Said, and H. Aghvami, Concept of Virtual
Antenna Arrays, in Proc. IEEE Globecom 2002, Taipei, Taiwan,
2002.
[3] M. Gastpar and M. Vetterli, On the Capacity of Wireless
Networks: The Relay Case, in Proc. IEEE INFOCOM 2002,
New York, 2002, vol. 3, pp. 15771586.
[4] M. Dohler, J. Dominguez, and H. Aghvami, Link Capacity of
Virtual Antenna Arrays, in Proc. IEEE Vehic. Techn. Conf.,
Sept. 2002.
[5] P. A. Anghel, G. Leus, and M. Kaveh, Multi-User Space-Time
Coding in Cooperative Networks, in Proc. ICASSP, Hong Kong,
Apr. 2003.
[6] S. M. Alamouti, A Simple Transmit Diversity Technique for
Wireless Communications, IEEE J. Select. Areas Comm., vol.
16, no. 8, pp. 14511458, Oct. 1998.
[7] A. Wittneben and B. Rankov, Impact of Cooperative Relays on
the Capacity of Rank Decient MIMO Channels, in IST Mobile
& Wireless Communications Summit 2003, Aveiro, Portugal, June
2003.
[8] P. Herhold, W. Rave, and G. Fettweis, Relaying in CDMA
Networks: Pathloss Reduction and Transmit Power Savings, in
Proc. 57th IEEE Vehic. Technol. Conf., Jeju, Korea, Apr. 2003.
[9] Thomas M. Cover and Joy A. Thomas, Elements of Information
Theory, John Wiley & Sons, New York, 1991.

Vous aimerez peut-être aussi