Efcient Demodulation in Cooperative Schemes

Using Decode-and-Forward Relays

Tairan Wang
Department of ECE
University of Minnesota
Minneapolis, MN 55455, USA
E-mail: wang0822@umn.edu
Alfonso Cano
Department of TSC
Rey Juan Carlos University
Fuenlabrada, Madrid 28943 Spain
E-mail: alfonso.cano@urjc.es
Georgios B. Giannakis
Department of ECE
University of Minnesota
Minneapolis, MN 55455, USA
E-mail: georgios@ece.umn.edu
AbstractCooperative communication systems using various
relay strategies can achieve spatial diversity gains, enhance
coverage and potentially increase capacity. For the practically
attractive decode-and-forward (DF) relay strategy, we derive an
efcient demodulator at the destination in the form of a weighted
combiner. The weights are selected adaptively to account for the
quality of both source-relay-destination and source-destination
links. Analysis proves that the novel demodulator can achieve the
maximum possible diversity, regardless of the underlying constel-
lation. Its error performance tightly bounds that of maximum-
likelihood (ML) demodulation which provably quanties the
diversity gain of ML detection with DF relaying. Simulations
corroborate our theoretical analyses and compare performance
of the novel decoder with existing diversity-achieving strategies
including analog amplify-and-forward and selective-relaying.
I. INTRODUCTION
The proliferation of wireless terminals has naturally led
to cooperative (a.k.a. relay) links whereby active terminals
benet from surrounding idle ones [7], [8], [5], [3], [1].
As with multi-input multi-output (MIMO) systems, where
multiple collocated antennas are deployed at the transmit- or
receive-ends, a main objective with single-antenna cooperating
terminals is also to enable spatial diversity. In such cooperative
links, the message sent by the source arrives at the destination
through diverse paths: one directly from the active (source)
node and others through relay nodes. Performance will thus
depend on the number of cooperating relay nodes as well as
the processing operations at both relays and destination. If
properly designed, cooperative networks can achieve diversity
order up to the number of diverse paths (what we henceforth
refer to as full diversity). To put our contribution in context,
we next review existing relay strategies, detectors and their
error performance in terms of diversity.
If relays can afford analog processing, they can amplify-
and-forward (AF) the source waveform to the destination.
Unfortunately, analog AF transceivers require expensive RF
chains to mitigate the coupling effects present. This motivates
digital processing at relay nodes to sample and store the source
waveform digitally before retransmission. Because such relays
forward the decoded message to the destination, they are
known as decode-and-forward (DF) relays. A third option is
to have relays forward only those correctly decoded messages,
in which case we call it selective-relaying (SR). Use of
SR presumes incorporation of e.g., cyclic redundancy check
(CRC) codes from a higher layer in order to detect errors.
When it comes to performance analysis, the symbol-error-
probability (SEP) for general AF links has been reported in [6],
where it is proved that full diversity is achievable with AF; see
also [4]. In SR-links, a weighted superposition of the source
and relay signals arriving at the destination can be formed
using maximum-ratio-combining (MRC), which also collects
the maximum available diversity order. Unfortunately, with DF
relays MRC does not offer a full diversity achieving receiver
[3]. To the best of our knowledge, no efcient demodulation
technique is available which at affordable complexity can
provably collect full diversity, regardless of the constellation,
when using the most practical relay strategy, namely DF.
For simplicity and due to space limitations, we will focus
on uncoded DF-streams and symbol-by-symbol demodulators;
i.e., in the remainder of this paper, DF will refer to uncoded-
DF. Our goal is to show that, through a suitable demodulator,
full diversity can be achieved with DF links. This result
was also alluded to in [8], where a maximum-likelihood
(ML) optimal detector has been presented only for binary
phase shift keying (BPSK). As recognized in [8], performance
analysis of ML is quite complicated, which prevents one from
any quantitative diversity assessment, especially for general
constellations. For this reason, a suboptimum combiner termed
-MRC was derived in [8]. Such a combiner facilitates perfor-
mance analysis and closed-form expressions for the probability
of error. Simulations illustrate that -MRC performs close
to ML, but since the combiner parameter has not been
specied analytically, full-diversity claims cannot be proved.
In [1], a piece-wise linear (PL) near-ML decoder has been
derived for coherent and non-coherent demodulation of bi-
nary modulations. Exploiting the average bit-error probability
(BEP) of the source-relay link that can be made available to
the destination, this PL approximation leads to closed-form
bounds on error performance, which still cannot provide full
diversity in general.
Our idea in this paper is to exploit knowledge of the
instantaneous BEP (per fading realization) of the source-relay
link at the destination, and derive a novel combiner capable
of collecting full diversity with DF relaying of any coherent
modulations. Such a combiner that we term cooperative MRC
Fig. 1. Block model for a single-relay cooperative system.
(C-MRC), will be shown to provide a tight lower bound on
the performance of ML detection; meaning that full diversity
claims we will establish for this combiner carry over to the
ML case too. Unlike ML and PL demodulators, it will turn
out that C-MRC offers a high-performance demodulator with
low-complexity regardless of the underlying constellation.
Notation: ()

denotes conjugation; CN(0,

2
) denotes the
circular symmetric complex Gaussian distribution with zero
mean and variance
2
; Re{z} denotes the real part of a
complex number z; for a random variable , p() denotes its
probability density function and = E{} denotes its mean.
II. EFFICIENT DEMODULATION OF DF
A. System Model
With reference to Fig. 1, let us consider the simplest model
in which only one relay (R) helps the source (S) to reach the
destination (D). The source S broadcasts symbols to D that
are also received by R, which forwards the decoded symbols to
D; because signals from S and R arrive through two different
paths, if fading is present, one can (at least in principle) design
a detector capable of collecting diversity up to order two.
In practice, mobile terminals cannot transmit and receive at
the same time and over the same frequency band, so S and R
must transmit over orthogonal channels. In this paper, we use
a time division duplex (TDD) mode, where data transmission
consists of two slots. In Slot I, S broadcasts symbol x with
average power P
x
. The received symbols at R and D are
y
SR
= h
SR
x +z
SR
, (1)
y
SD
= h
SD
x +z
SD
, (2)
where h
SR
and h
SD
denote the fading coefcients from S to R
and D, modelled as h
SR
CN(0,
2
SR
), h
SD
CN(0,
2
SD
),
with
2
SR
:= E{|h
SR
|
2
} and
2
SD
:= E{|h
SD
|
2
}, respec-
tively. Without loss of generality, we assume that noise terms
z
SR
and z
SD
have equal variances N
0
and are modelled
as: z
SR
CN(0, N
0
), z
SD
CN(0, N
0
). At R, coherent
ML demodulation is performed, the detected symbol is re-
modulated, and subsequently transmitted during Slot II with
the same average power P
x
. The received symbol at D is
y
RD
= h
RD
x +z
RD
, (3)
where x is the re-modulated symbol, h
RD
is the channel
coefcient from R to D, h
RD
CN(0,
2
RD
) with
2
RD
:=
E{|h
RD
|
2
}, and z
RD
CN(0, N
0
) is the noise term at D.
B. Relay Link Analysis
Let us dene the instantaneous signal-to-noise-ratio (SNR)
at links SR, RD and SD as
SR
:= |h
SR
|
2
,
RD
:=
|h
RD
|
2
, and
SD
:= |h
SD
|
2
, respectively, with = P
x
/N
0
denoting average SNR. Notice that S R D is a two-hop
channel, whose end-to-end BEP is given by [2]
P
b
eq
(
SR
,
RD
)
= [1P
b
SR
(
SR
)]P
b
RD
(
RD
)+[1P
b
RD
(
RD
)]P
b
SR
(
SR
),
where P
b
SR
(
SR
) and P
b
RD
(
RD
) are the conditional BEPs
at both hops which we assume available at D. One can
think of this BEP as the error probability corresponding to
an equivalent S R D link whose instantaneous output
SNR
eq
is dened as

eq
:=
1

_
Q
1
[P
b
eq
(
SR
,
RD
)]
_
2
, (4)
where Q(x) := (1/

2)
_

x
exp(t
2
/2)dt, and is a con-
stant depending on the underlying constellation; e.g., = 2
for BPSK. This equivalent one-hop SNR
eq
jointly accounts
for the quality of both S R and RD links and will guide
the design of our novel demodulator. To this end, we establish
the following property for
eq
:
1
Property 1 Upon dening
min
:= min{
RD
,
SR
}, it holds
that
eq
in (4) is bounded by:

min

3.24

<
eq

min
. (5)
Property 1 upper-bounds the end-to-end equivalent SNR
by the minimum of its single-hop SNRs. This is intuitively
expected, because the BEP over the aggregate S RD link
cannot exceed that of RD or S R. On the other hand, the
lower bound in (5) implies that for a relatively large
min
, the
constant 3.24/ can be negligible, showing that indeed
min
can offer a tight approximation to
eq
.
As recognized by [6] and [10] in the context of AF,
knowledge of the SR link quality (
SR
) at the destination is
possible by sending pilot symbols through R. In regenerative
schemes such as DF, one can acquire the S R link quality
by sending a pilot from S for the relay to estimate the S R
channel, and forward it to the destination via a second pilot
whose power is scaled according to the estimated channel
coefcient. At the receiver, as in AF, one again recovers the
product of the two S R and R D fading coefcients.
Notice that although knowledge of the S R link quality
at the D requires one more pilot symbol, the price paid in
spectral efciency by other regenerative strategies is steeper;
e.g., SR requires bandwidth consuming CRC codes to ensure
perfect error detection at the relay.
1
Omitted due to space limitations, proofs for all the properties and propo-
sitions in this paper can be found in [9].
C. Cooperative MRC
Consider combining the received y
SD
and y
RD
at the
destination to obtain:
x=arg min
xAx
|w
SD
y
SD
+w
RD
y
RD
(w
SD
h
SD
+w
RD
h
RD
)x|
2
,
(6)
where |A
x
| = M denotes the cardinality (size) of the M-
ary constellation, and w
SD
and w
RD
weights are functions
of h
SD
, h
SR
, and h
RD
. In a collocated multi-antenna setup,
MRC employs weights w
SD
= h

SD
and w
RD
= h

RD
, and
is known to maximize the SNR at the combiner output. This
would also be the optimal choice in our context if x = x.
However, since the fading link S R causes detection errors
at the relay, performance of the standard MRC is far from
being optimal.
Motivated by this, we x w
SD
= h

SD
to maximize
SD
,
and seek a weight w
RD
to maximize the equivalent SNR

eq
in the link S R D, instead of R D alone. These
considerations lead to the choice
w
RD
(h
SR
, h
RD
) =

eq

RD
h

RD
. (7)
Combiner (6) with weights w
SD
= h

SD
and w
RD
as in
(7) constitutes what we term cooperative MRC (C-MRC).
Our C-MRC is reminiscent of the -MRC in [8], where
w
RD
= h

RD
with 0 1. Recall though that in lieu of a
closed-form, the optimum in [8] is found through numerical
search. Our closed-form in (7) will allow for analytical BEP
evaluation. Interestingly, our simulations will show that (7) is
intimately close to the optimum obtained by [8].
Relative to ML, the C-MRC dened by (6) and (7) is sub-
optimum. To conrm the differences, it sufces to write down
the ML coherent detector which for BPSK takes a relatively
simple form as
x
ML
= arg max
xA
x
_
1 P
SR
(
SR
)
2N
0
exp[
|y
SD
h
SD
x|
2
+|y
RD
h
RD
x|
2
2N
0
]
+
P
SR
(
SR
)
2N
0
exp[
|y
SD
h
SD
x|
2
+|y
RD
+h
RD
x|
2
2N
0
]
_
,(8)
where |A
x
| = 2. An implementation of the BPSK demodulator
in (8) can be found in [8]. Clearly, implementing (8) for
higher-order constellations becomes prohibitively complex.
Moreover, asymptotic analysis of (8) is very complicated,
which prevents one from assessing ML performance [1], [8].
Assuming that the average SNR of the S R link is available
at the destination, the PL-ML approximation was advocated
in [1] to overcome this problem.
III. PERFORMANCE ANALYSIS OF DF WITH C-MRC
A. BEP of DF using BPSK
We rst analyze its performance for BPSK. Extensions to
any general constellation will be discussed in Subsection III-
C. With BPSK, x at the relay can only take one of two values:
x = x, or, x = x. For each value, the C-MRC output is:
y
D
= w
RD
y
RD
+w
SD
y
SD
=
_
(w
SD
h
SD
+w
RD
h
RD
)x+w
RD
z
RD
+w
SD
z
SD
, if x=x,
(w
SD
h
SD
w
RD
h
RD
)x+w
RD
z
RD
+w
SD
z
SD
, if x=x.
Since BPSK is real-valued, it sufces to consider only the real
part y = Re{y
D
}, which is a real Gaussian random variable
with zero mean and variance
2
= (|w
RD
|
2
+|w
SD
|
2
)N
0
/2.
Plugging w
SD
= h

SD
and (7), the BEP can be expressed in
terms of the Q-function as:
P
b
(
SR
,
SD
,
RD
) =[1P
b
SR
(
SR
)]Q
_
_

2(
SD
+
eq
)
_

SD
+
2
eq
/
RD
_
_
+P
b
SR
(
SR
)Q
_
_

2(
SD

eq
)
_

SD
+
2
eq
/
RD
_
_
. (9)
Taking expected value of (9) over the instantaneous SNRs
yields the average BEP, which we will analyze in the ensuing
subsection for sufciently large SNR values.
B. Diversity Analysis of DF
Diversity gain (diversity order) G
d
is dened as the negative
exponent in the average BEP when the average SNR tends to
innity, that is:
P
b

(G
c
)
G
d
, (10)
where G
c
is the coding gain. Our goal in this subsection is to
assess the diversity order of C-MRC by establishing bounds
on the average BEP obtained from (9). To this end, we rst
notice that
eq

RD
, which upper-bounds (9) as follows:
P
b
(
SR
,
SD
,
RD
)
Q
_
_
2(
SD
+
eq
)
_
+Q[
_
2
SR
]Q
_

2(
SD

eq
)

SD
+
eq
_
. (11)
Because a sum is dominated by the term with the low-
est diversity exponent, we need to prove that both terms
in the right-hand side of (11) decay with the same
exponent (diversity order), which here equals two. Be-
cause these two terms effect diversity through distinct
means, we will analyze them separately. Let us start by
dening P
b
1
(
SR
,
SD
,
RD
) := Q
__
2(
SD
+
eq
)

, and
(
SR
,
SD
,
RD
) := (
2
SR
,
2
SD
,
2
RD
). Next, we invoke
Property 1, use the Chernoff bound and take expectation over
the three instantaneous SNRs to bound P
b
1
as:
P
b
1

1
2
exp(1.62)(
SR
+
RD
)
(
SD
+ 1)(
SR

RD
+
SR
+
RD
)

(k
1
)
2
, (12)
where k
1
is a constant which depends on
2
SR
,
2
SD
,
2
RD
.
To appreciate (12), recall that (k
1
)
2
is present whenever
the relay is forwarding the correct symbol ( x = x), which
corresponds to the co-located multi-antenna scenario, that is
capable of collecting full diversity.
Fig. 2. Worst case in general modulation.
Turning back our attention to (11), let us dene the second
summand in the bound as
P
b
2
(
SR
,
SD
,
RD
):= Q[
_
2
SR
]Q
_

2(
SD

eq
)

SD
+
eq
_
. (13)
The next proposition upper bounds P
b
2
.
Proposition 1 The expectation E{P
b
2
(
SR
,
SD
,
RD
)} := P
b
2
can be bounded by a term

P
b
2
, which decays with exponent
equal two; i.e. with k
2
denoting a constant, we have
P
b
2


P
b
2

(k
2
)
2
, (14)
Eq.(14) together with (12) establish that C-MRC is full diver-
sity achieving with DF, when BPSK is used.
One can gain insight about Proposition 1 by inspecting (13),
where
eq
is always smaller than
SR
. An increase in
eq
implies an increase of the second Q-function factor, which is
mitigated by a decrease in the rst Q-function factor in (13).
This trade-off suggests that an optimal choice of
eq
may be
possible in C-MRC to jointly account for the quality of both
S R and R D links.
C. Performance Bounds for General Constellations
Following steps similar to Section III-B, we analyze here
the SEP of C-MRC for higher-order constellations. Using the
same notational conventions, the SEP can be expressed as
the superposition of two terms, now denoted as P
s
1
and P
s
2
.
Dening the SEP from S to R as P
s
SR
(
SR
), P
s
1
corresponds
to the case where, with probability (1 P
s
SR
(
SR
)), we are
combining identical symbols arriving from S and R, whose
performance analysis is the same as a co-located antenna
system and the associated error probability is known to decay
with an exponent of order two. To ensure full diversity, we also
need to show that P
s
2
decays with the same exponent when R
is sending an erroneously decoded symbol. To prove the latter,
we again rely on bounds of P
s
2
. With d
min
(d
max
) denoting
the minimum (maximum) Euclidean distance between points
in the constellation, P
s
2
can be bounded by the worst case
which corresponds to the decoded symbol at R being at
distance d
max
from the actual symbol sent from S. In such a
case, integration of the associated probability density functions
0 5 10 15 20 25
10
6
10
5
10
4
10
3
10
2
10
1
(dB)
B
E
P
(+30dB, , ) C-MRC
(, +30dB, ) C-MRC
(, , ) C-MRC
(+30dB, , ) -MRC
(, +30dB, ) -MRC
(, , ) -MRC
(+30dB, , ) ML
(, +30dB, ) ML
(, , ) ML
Fig. 3. BEP comparison using C-MRC, -MRC or ML with BPSK.
(pdfs) can be simplied by reducing the decision region
to a square inscribed within the circle of radius d
min
, as
shown in Fig. 2. Invoking the union bound, we can write
P
s
2
2P
s
2,1D
, where P
s
2,1D
is the error probability across
one-dimension. Furthermore, knowing the BEP conditioned on
the instantaneous
SR
for a general M-ary modulation, call
it P
b
SR
, we can readily upper bound the corresponding SEP
as: P
s
SR
(
SR
) (log
2
M)P
b
SR
(
SR
) = (log
2
M)Q[

SR
].
Finally, we can arrive at the upper bound:
P
s
2
(
SR
,
SD
,
RD
)
4(log
2
M)Q[

SR
]Q
_

2(
SD

eq
)
_
(
SD
+
eq
)
_
, (15)
where :=

d
max
/

d
min
1,

d
min
:= d
min
/(2

2),

d
max
:=
d
max


d
min
, and the average transmit-SNR has also been
bounded by

d
2
min
/N
0
.
We now observe that (15) has form identical (within a scale)
to P
b
2
(
SR
,
SD
,
RD
) in Proposition 1. Because high SNR
behavior is not affected by the constant , our claims in
Proposition 1 are still valid for (15), which establishes that
our C-MRC demodulator enjoys full diversity in DF-based
cooperation, regardless of the underlying constellation.
IV. NUMERICAL RESULTS AND SIMULATIONS
This section presents simulated BEP results in single-relay
scenario for practical SNR values. Unless otherwise stated, the
adopted modulation is BPSK;
eq
is dened as in (4) and nodes
transmit with the same power P
x
, resulting in an average input
SNR = P
x
/N
0
.
With reference to Fig. 1, we consider representative attenua-
tion levels that correspond to those in which R is located either
close to the source, close to the destination, or, equi-distant
from both; the corresponding average output SNRs (
SR
,
RD
,

SD
) in logarithmic-scale are ( +30dB, , ), ( , +30dB, ),
and ( , , ), respectively.
Fig. 3 compares bounds associated with C-MRC, -MRC
0 5 10 15 20 25
10
6
10
5
10
4
10
3
10
2
10
1
10
0
(dB)
B
E
P
(+30dB, , ) DF
(, +30dB, ) DF
(, , ) DF
(+30dB, , ) AF
(, +30dB, ) AF
(, , ) AF
(+30dB, , ) SR
(, +30dB, ) SR
(, , ) SR
Fig. 4. BEP comparison using DF-based C-MRC vs. AF vs. SR (block
length=100), with BPSK.
[8] and the ML detector in (8). As expected, C-MRC tightly
bounds performance of the rest, which as a by-product demon-
strates the performance claims not proven in [8].
We next compare DF with alternative diversity-achieving
strategies, namely AF and SR, in Fig. 4. Based on analog
processing, AF performs best among the three protocols. When
it comes to regenerative protocols, the proposed DF decoder
outperforms SR for block length = 100 (recall that forwarding
in SR takes places only if there is no error in the entire block).
This can be explained if one considers that, although both DF-
based C-MRC and SR are adaptive protocols effecting full
diversity, the SR one is based on hard decisions at R, while
DF exploits soft information from R and adaptively combines
it at the destination.
To check modulations higher than BPSK, we tested QPSK
and compared C-MRC with traditional MRC, which is the only
available demodulator for higher constellations. As shown in
Fig. 5, diversity loss is particularly evident for low-quality
S R links, which justies the importance of link quality
information which C-MRC exploits to collect full diversity.
V. CONCLUSIONS
We developed a high-performance C-MRC demodulator
when cooperating relays utilize the practical DF strategy.
We proved that full diversity gains can be achieved with C-
MRC regardless of the underlying constellation. Simulations
illustrated that our C-MRC performs surprisingly close to ML
while it is computationally much simpler than ML, indepen-
dent of the constellation used. C-MRC adapts its structure
depending on knowledge of the S R link quality, which
is easy to acquire through training. Relative to competing
alternatives, C-MRC with DF relays outperforms existing link-
adaptive regenerative strategies and comes very close to AF,
which is certainly more expensive to implement in practice.
0 5 10 15 20 25
10
6
10
5
10
4
10
3
10
2
10
1
10
0
(dB)
B
E
P
(+30dB, , ) C-MRC
(, +30dB, ) C-MRC
(, , ) C-MRC
(+30dB, , ) MRC
(, +30dB, ) MRC
(, , ) MRC
Fig. 5. BEP comparison using C-MRC vs. MRC (QPSK modulation).
ACKNOWLEDGMENTS
This work was supported through collaborative participation in the Com-
munications and Networks Consortium sponsored by the U.S. Army Research
Laboratory under the Collaborative Technology Alliance Program, Coopera-
tive Agreement DAAD19-01-2-0011. The U.S. Government is authorized to
reproduce and distribute reprints for Government purposes notwithstanding
any copyright notation thereon. The work of the second author was supported
by the Spanish Government TEC2005-06766-C03-01/TCM.
REFERENCES
[1] D. Chen and J. N. Laneman, Modulation and Demodulation
for Cooperative Diversity in Wireless Systems, IEEE Trans.
on Wireless Commun., 2006 (to appear); downloadable from
http://www.nd.edu/ jnl/pubs/twc2004a.pdf.
[2] M. O. Hasna and M.-S. Alouini, End-to-End Performance of Trans-
mission Systems With Relays Over Rayleigh-Fading Channels, IEEE
Trans. on Wireless Commun., vol. 2, no. 6, pp. 1126-1131, Nov. 2003.
[3] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity
in wireless networks: Efcient protocols and outage behavior, IEEE
Trans. on Inform. Theory, vol. 50, no. 12, pp. 3062-3080, Dec. 2004.
[4] W. Mo and Z. Wang, Average symbol error probability and outage
probability analysis for general cooperative diversity system at high
signal to noise ratio, In Conference on Information Sciences and
Systems, Princeton, NJ, Mar. 2004.
[5] R. Pabst, B. H. Walke, D. C. Schultz, P. Herhold, H. Yanikomeroglu,
S. Mukherjee, H. Viswanathan, M. Lott, W.Zirwas, M. Dohler, H.
Aghvami, D. D. Falconer, and G. P. Fettweis, Relay-based deployment
concepts for wireless and mobile broadband cellular radio, IEEE
Communications Magazine, vol. 42, no. 9, pp. 80-89, Sep. 2004.
[6] A. Ribeiro, X. Cai and G. B. Giannakis, Symbol error probabilities for
general cooperative links, IEEE Trans. on Wireless Commun., vol. 4,
no. 3, pp. 1264-1273, May 2005.
[7] A. Sendonaris, E. Erkip, and B. Aazhang, User Cooperation Diversity-
Part I: System Description, IEEE Trans. on Commun., vol. 51, no. 11,
pp. 1927-1938, Nov. 2003.
[8] A. Sendonaris, E. Erkip, and B. Aazhang, User Cooperation Diversity-
Part II: Implementation aspects and performance analysis, IEEE Trans.
on Commun., vol. 51, no. 11, pp. 1939-1948, Nov. 2003.
[9] T. Wang, A. Cano, G. B. Giannakis, and N. Laneman, High-
performance cooperative demodulation with decode-and-forward re-
lays, IEEE Trans. on Wireless Commun., submitted Aug. 2005.
[10] E. Zimmermann, P. Herhold, and G. Fettweis, On the performance of
cooperative diversity protocols in practical wireless systems, Proc. of
58th Vehicular Technology Conference, vol. 4, pp. 2212-2216, Oct. 2003.