A study on a Cooperative Relay

using Joint Network and Channel Coding

in Simple Triangular Network Topology
Masatoshi MIZOBUCHI

, Shinsuke IBI

, and Seiichi SAMPEI

Graduate School of Engineering, Osaka University, Yamada-oka 2-1, Suita-shi, 5650871, Japan
Email:

[email protected],

{ibi,sampei}@comm.eng.osaka-u.ac.jp
AbstractThis paper proposes a cooperative relay transmis-
sion exploiting a concatenated structure composed of a simple
exclusive OR (XOR) network code and channel codes. A main
contribution of this paper is to clarify the joint network-and-
channel coding combined with an iterative decoding is effective
in providing highly reliable wireless communications even in
the case where distances from one destination node to two
source nodes are not the same in the case of triangular topology
network. In this case, it is subject to imbalanced signal powers
at the destination node, resulting in imbalanced link quality.
In this paper, we focus on an analogy between the triangular
topology network and a quaternary PSK (QPSK) symbol with a
rectangular shape, i.e., imbalanced the in-phase and quadrature-
phase (I-Q). After demonstrating that the rectangular QPSK
has a potential to detect signals with high accuracy by iterative
decoding, a conguration of the cooperative relay to be proposed
in this paper is revealed. Computer simulation conrms that both
the rectangular QPSK and the cooperation are effective to solve
the imbalanced problem.
I. INTRODUCTION
A multi-hop relay transmission resulting from the collection
of wireless mobile nodes is well-known as one of the key
techniques that promise to improve a link quality between
source and a nal destination node [1], [2]: the source and
destination nodes can directly communicate only when the
received signal power at the destination node is guaranteed to
be sufciently high. Otherwise, data packets to be conveyed
are relayed from the source to the nal destination via relay
nodes. This rule implies that more amount of radio resource is
consumed by one information packet in the relay transmission
than in the direct transmission, resulting in lower spectrum
efciency [3], [4].
Aiming at improving the efciency, network coding has
attracted much attention on its special benets, i.e., multi-
plex of information packets without sacricing the spectral
efciency and power efciency so much while obtaining path
diversity effects over wireless channels [5], [6]. In order to
achieve high reliability of signal detection, joint channel-and-
network coding is also studied in two-way relay and multiple-
access relay channels [7], [8]. In such joint coding, an iterative
decoding based on turbo principle is a key technique to open
vistas for the coding design [9], [10].
In this paper, we propose a joint channel-and-network
coding combined with an iterative decoding in simple relay
environments (triangular topology) where two source nodes
communicate with one destination node. One source (U
1
) node
sends its own information packet, and the other source node
(U
2
) sends XORed packet comprised of its own and U
1
s
packets as a network coded packet. At this stage, the network
code constructs cooperative relay partnership between U
1
and
U
2
. Therefore, we call this network coding functionality co-
operative relay. The destination (D) node detects two packets
using an iterative decoder. A main contribution of this paper
is, based on the above mentioned techniques, demonstrating a
cooperative networking methodology having high spectral and
power efciency as well as high exibility against geometric
node location.
Honestly, the idea of the proposed channel and network
coding comes from a study on the iterative detection for
QPSK symbols with rectangular shape. The study tells us
that natural binary (NB) coded QPSK allows us to improve
detection quality in comparison to Gray coded QPSK when bit
interleaved coded modulation with iterative detection (BICM-
ID) is adopted. The cooperative relay exploits the nature of NB
coded QPSK by focusing on an analogy between the in-phase
and quadrature-phase signals of the rectangular QPSK, and
the direct and cooperated received signals in the cooperated
triangular network topology.
The rest of this paper is organized as follows. We study the
nature of BICM-ID for NB coded QPSK before a proposal of
the cooperative relay in Sect. II. This is because the study
tells us evocative facts. Then, the analogy is discussed in
Sect. III. Based on the discussion, cooperative coding and
iterative decoding are proposed and evaluated in Sects. IV
and V, respectively. Finally, this paper comes to conclusions
in Sect. VI
II. BICM-ID FOR RECTANGULAR QPSK
To provide deep understanding of the cooperative relay
system to be discussed in this paper, let us study transmission
performances for BICM-ID with rectangular QPSK signaling.
Figure 1 shows a conguration of the transceiver to be consid-
ered in this section. At the transmitter, two information packets
comprised of d
i
(k

) (i = 1, 2) are independently encoded by

channel encoders. Then, the coded bits are interleaved to form
c
i
(k). The coded bits c
1
(k) and c
2
(k) are mapped onto a
transmitted symbol at a QPSK symbol mapper (modulator).
Channel
Encoder
1
Channel
Encoder
2
c1(k)
c2(k)
d (k)
1
d (k) 2
Symbol
Mapper
1
s(k)
Symbol
Demapper
Channel
Decoder
1
Channel
Decoder
2
d (k) 2
d (k) 1
1

2
dem
dec
-1
-1
r(k)
A
W
G
N
[c1(k)]
dec
[c2(k)]
[c1(k)]
dem
[c2(k)]
QPSK
QPSK
Transmitter
Receiver

Fig. 1. Conguration of transceiver for BICM-ID with multi-level QPSK.

c1(k)
c2(k)
I( )
j
BPSK
BPSK
Q( )
[A] Gray Coded.
Qch
Ich
S
[C] Constellations in a rectangle shape.
S
E

=
2
cos ) (

S
E

=
2
sin ) (

S
E
I
Q
0 S
S S 1
2
3
S
q
S
S
S
S
C
gray
q,1
0
1
2
3
0 0
0
0
1
1
1 1
[D] Value of coded bits for each constellation.
C
q,2
C
nat
q,1
C
q,2
S
S
S
S
S
q
0
0
0
1
1
0
1
0
1
1
2
3
c1(k)
c2(k) j XOR
BPSK
BPSK Q( )
I ( )
[B] NB Coded.
gray nat

Fig. 2. Symbol mapping rule for QPSK constellation.

The mapping rule for the rectangular QPSK signaling is spec-
ied in Fig. 2. In the gure, [A] and [B] show congurations
of the symbol mapper for Gray and NB coded constellations,
respectively.
In the Gray coded case, after the baseband signals for binary
phase shift keying (BPSK) is generated in each stream based
on the input coded bit sequence c
i
(k), each signal is weighted
by a coefcient
I
() or
Q
() to allow rectangular shape
of QPSK signaling. The coefcients for the in-phase and
quadrature-phase are given by

I
() =

E
s
sin
(

2
)
,
Q
() =

E
s
cos
(

2
)
, (1)
where E
s
denotes transmit energy per symbol, and is an an-
gle between diagonal lines comprised of four constituent con-
stellation points. Eventually, the points S {S
0
, S
1
, S
2
, S
3
}
for the rectangular QPSK signaling are depicted in Fig. [C]. In
the NB coded case, the in-phase component in the transmitted
symbol is determined in the same manner with the Gray coded
case. On the other hand, the quadrature-phase component is
generated from the exclusive OR (XOR) of c
1
(k) and c
2
(k).
The mapping rules for Gray and NB coded QPSK signaling
are summarized in [D]. In the Gray coded case, S
q
is generated
when c
1
(k) and c
2
(k) are equal to C
gray
q,1
(k) and C
gray
q,2
(k). In
the NB coded case, on the other hand, S
q
is generated when
c
1
(k) and c
2
(k) are C
nat
q,1
(k) and C
nat
q,2
(k).
In general, Gray coded QPSK is widely adopted in wireless
communication systems. This is because the maximum Ham-
ming distance between the nearest adjacent symbols is half
compared with NB coded QPSK. However, when we carefully
observe the NB coded QPSK, we can nd that its quadrature-
phase components can be regarded as the channel coded bits
with coding rate one thanks to XOR process. This feature
brings us to a concept that an appropriate decoding algorithm
would help us in obtaining higher coding gain.
Now, review the Fig. 1. The conguration of the transmitter
can be regarded as a serial concatenated structure between
channel encoders and a symbol mapper. Recently, an iterative
detection is well-known as a powerful decoding algorithm for
such a concatenated structure [11]. Therefore, the receiver em-
ploys an iterative decoding as shown in the gure. Assuming
additive white Gaussian noise (AWGN) channels, the received
symbol r(k) is given by
r(k) = s(k) + (k) (2)
where (k) denotes AWGN, which is subject to zero-mean
and complex variance N
0
. The iterative detection is capable of
improving accuracy of the detection by exchanging extrinsic
Log-Likelihood Ratios (LLRs) between a symbol demapper
and channel decoders with soft-input soft-output (SSfo) func-
tion [11]. At the rst, extrinsic LLRs are fed back from the
decoder output dened by

dec
[c
i
(k)] = ln
Pr [c
i
(k) = 1]
Pr [c
i
(k) = 0]
, (i = 1, 2). (3)
Note that each
dec
[c
i
(k)] is zero at the rst iteration because
the a-priori knowledge about c
i
(k) is unavailable. Then, a-
priori probability for symbols Pr[s(k) = S
q
] is given by
Pr [s(k) = S
q
] =
2

i=1
Pr [c
i
(k) = C
q,i
] (4)
where C
q,i
is C
gray
q,i
when the Gray coded QPSK is adopted, or
C
nat
q,i
when the NB coded QPSK is done. A-priori probability
for coded bits c
i
(k) is dened by
Pr [c
i
(k) = C
q,i
] =f(
dec
[c
i
(k)], C
q,i
) (5)
where
f(x, y) =
1
2
[1 + (2y 1)tanh
(
x
2
)
]. (6)
With the aid of a-priori knowledge, a-posteriori proba-
bility for coded bits c
i
(k) is given by a marginalization of
probabilities based on Bayes theorem, expressed as
Pr [c
i
(k)|r(k)]

Sq(S|ci(k)=Cq,i)
p(r(k)|s(k) = S
q
)Pr [s(k) = S
q
] . (7)
In the AWGN channel, probability density function of the
received symbol conditioned by S
q
, is given by
p(r(k)|s(k)=S
q
) =
1
N
0
exp
(

|r(k) S
q
|
2
N
0
)
. (8)
Based on Eq. (7), extrinsic LLR for the demapper output is
dened by
TABLE I
SIMULATION PARAMETERS.
Modulation QPSK
Channel Coding
Non-Systematic Convolutional
(Coding rate 1/2)
(Constraint length = 4)
Decoder
Max-Log-MAP
with Jacobian logarithm [12]
packet length 512 information bits/packet
Interleaver Random
Channel model AWGN
Number of itereations 8

dem
[c
i
(k)] = ln
Pr [r(k)|c
i
(k) = 1]
Pr [r(k)|c
i
(k) = 0]
= ln
Pr [c
i
(k) = 1|r(k)]
Pr [c
i
(k) = 0|r(k)]

dec
[c
i
(k)]. (9)
For the sake of error corrections, LLR for demapper output

dem
[c
i
(k)] is delivered to SSfo channel decoder via a
deinterleaver. SSfo channel decoders yield extrinsic LLRs

dec
[c
i
(k)] with much higher reliability by exploiting redun-
dancy of the channel code. Then, the improved LLRs are fed
back to the demapper via an interleaver. After this process
is iterated several times, the SSfo channel decoder detects
information bits d
i
(k

).
Packet error rate (PER) performances for the BICM-ID
with rectangular QPSK signaling are shown in Fig. 3. The
simulation conditions are summarized in Tab. I. In the gure,
(a) and (b) show PER performances for information packets
comprised of d
1
(k

) and d
2
(k

), which are simultaneously sent

over the in- and quadrature-phases, respectively. Note that two
packets are embedded in one symbol stream.
In the Gray coded case, wider angle causes larger imbal-
ance between PERs for d
1
(k

) and d
2
(k

) due to difference
of their Euclidian distances dened by . In an extreme case
where is 0 [rad], the Euclidian distance for the in-phase
component becomes 0, resulting in PER=1 for the packet
d
1
(k

). In the NB coded case, on the other hand, the angle

wider than approximately 0.25 [rad] promises low PER even
for the packet d
1
(k

) sent over the in-phase due to the iterative

decoding process. This means that BICM-ID with NB coded
rectangular QPSK is capable of obtaining signicant coding
gain with the aid of XOR in the quadrature-phase.
III. RELATIONSHIP BETWEEN TRIANGULAR NETWORK
TOPOLOGY AND RECTANGULAR QPSK SIGNALING
Let us assume a node topology shown on the left top in
Fig. 4, in which two user nodes (U
i
, i = 1, 2) independently
transmit information packets to a destination (D) node in
different time phase. A packet comprised of information bit
d
i
(k

) is encoded to form c
i
(k

), c
i
(k

) is transformed into
transmit symbols s
i
(k) by interleaver and symbol mapper. For
simplicity of analysis, we assume here an equivalent low-pass
signal model dened by
r
1
(k) =

gs
1
(k) +
1
(k), (10)
r
2
(k) = s
2
(k) +
2
(k), (11)
where r
i
(k) is the received symbol from node i, and
i
(k) is
a Gaussian noise term included in the received signal, which
0 0.1 0.2 0.3 0.4 0.5
10
4
10
3
10
2
10
1
10
0
angle[ rad]
P
E
R


Nat. (2.0)
Nat. (3.0)
Nat. (4.0)
Gray (2.0)
Gray (3.0)
Gray (4.0)
(a) Information packet for d
1
(k

).
0 0.1 0.2 0.3 0.4 0.5
10
4
10
3
10
2
10
1
10
0
angle[ rad]
P
E
R


Nat. (2.0)
Nat. (3.0)
Nat. (4.0)
Gray (2.0)
Gray (3.0)
Gray (4.0)
(b) Information packet for d
2
(k

).
Fig. 3. PER for angle of each constellation: () is received Es/N
0
dB.
is subject to zero-mean and complex variance N
0
. We will
assume that the received signal from node i (i = 1, 2) is
received in the i-th phase. The variable g denotes the power
gain for the link U
1
-D relative to that for the link U
2
-D. In
this case, LLRs yielded by demapper for c
1
(k

) and c
2
(k

)
at the D node are characterized by different variances. The
most important aspect is that distributions of the LLRs have an
analogy with that for Gray coded rectangular QPSK signaling
as shown in Fig. 4. The deferent variances for the triangular-
network and the rectangular QPSK signaling are caused by
the imbalanced power gain g and the angle , respectively.
In Sect. 2, we have demonstrated that the NB coded
rectangular QPSK is capable of improving PER performances.
As shown in Fig.2 [B], the XORed packet is sent over via the
quadrature-phase. This fact suggests that if U
2
takes XOR of
correctly detected U
1
packet and its own packet to create U
2
s
packet in a cooperative manner, the D node can regard the
directly received signal from U
1
as the in-phase signal and
the XORed U
2
s packet as the quadrature-phase signal in the
rectangular QPSK signal, thereby the same process could be
applicable to the decoding process of U
1
and U
2
signals. In
the next section, a methodology based on this concept will be
explained.
IV. COOPERATIVE RELAY WITH ITERATIVE DECODING
In the triangular-network topology depicted in Fig. 4, it is
obvious that a link quality for U
1
-D is worse than that for
U
2
-D due to its longer distance discussed in the previous
section. To overcome this disadvantage, it is effective to relay
U
2
U
1
D
Qch
Ich
S
0
S
S S 1
2
3
index index
LLR of c2 or Q-ch LLR of c
1
or I-ch
System Model
rectangular QPSK
c1
c
2
Imbalance
Equivalence

Fig. 4. Relationship between triangular network-topology and Gray coded

rectangular QPSK.
the packets sent from U
1
at U
2
[3], [4]. However, three time
phases are necessary for the initial U
1
, U
2
, and U
1
relayed
packet transmissions. In this paper, assuming U
2
has perfect
knowledge about information d
1
(k

) sent form U
1
, XOR
operation between d
1
(k

) and d
2
(k

) is adopted to multiplex
two information packets. As the result, the transmisson can
be completed in two time phases. Fortunately, the XOR
is absolutely the same as the operation in the quadrature-
phase for NB coded rectangular QPSK signaling. One more
advantage is that such an XOR introduced relaying does not
require any extra power even if two streams are multiplexed.
This analogy indicates that the link quality for U
1
-D can be
improved with the aid of knowledge on d
2
(k

) by iterative
decoding at the D node. Thus, we call the XOR operation
cooperative coding.
A conguration of a transceiver with the cooperative coging
is shown in Fig. 5.
In the U
1
node, after packet d
1
(k

) is encoded to form
coded bits c
i
(k

), interleaved to obtain b
1
(k

), and mapped
onto symbols s
1
(k), the signal is transmitted to D, which is
also received at U
2
. In the U
2
node, when the U
2
identies
that U
1
s packet is not correctly detected at the D, the
packets d
1
(k

) and d
2
(k

) are encoded, interleaved, XORed to

construct cooperative coded bits b
2
(k

), mapped onto symbols

s
2
(k), and then sent to the D node.
In the D node, after receiving r
1
(k) and r
2
(k), demodulator
yeilds extrinsic LLRs given by

dem
[b
i
(k

)]=ln
Pr[b
i
(k

)=1|r
i
(k)]
Pr[b
i
(k

)=0|r
i
(k)]
=ln
Pr[r
i
(k)|b
i
(k

)=1]
Pr[r
i
(k)|b
i
(k

)=0]
.(12)
Note that feedback LLRs form channel decoders are not
exploited at the demapper in this conguration. Thus, this con-
guration is not BICM-ID but BICM. The LLR is calculated
by the following iterative BICM decoder
1
.
Now, instead of s(k) in Sect. II, let v(k) be a virtual QPSK
symbol which is comprised of b
1
(k

) and b
2
(k

) with Gray
coded mapping rule. In this case, a-posteriori probability for
coded bits c
i
(k

) is given by a marginalization of probabilities

based on Bayes theorem like Eq. (7), expressed as
1
When BICM-ID is applied, an optimization of mapping rule for constel-
lations should be analyzed. The optimization would be capable of further
improving the transmission quality.
Channel
Encoder
1
Symbol
Mapper
d1(k)
U1 node
c1(k)
s
1
(k)
Channel
Encoder
1
Channel
Encoder
2
d (k)
1
d (k)
2
Symbol
Mapper
Cooperative encoder
U2 node
Symbol
Demapper
Symbol
Demapper
C
o
o
p
e
r
a
t
i
v
e
D
e
c
o
d
e
r
1
1
2
2
Channel
Decoder
Channel
Decoder
d (k)
2
d (k)
1
[b1(k)]
dem
D node
c1(k )
c2(k )
XOR
s
2
(k)
[b2(k )]
dem
[c1 (k )]
coop
[c2(k )]
coop
[c
2
(k )]
dec
[c1(k )]
dec
r
i
(k)
b1(k)
b2(k)
-1
-1
Fig. 5. Congulation of transceiver adopted cooperative coding.
Pr[c
i
(k

)|
dem
[b
1
(k

)],
dem
[b
2
(k

)]]

Sq(S|ci(k

)=C
nat
q,i
)
Pr[
dem
[b
1
(k

)],
dem
[b
2
(k

)]|v(k)=S
q
]
Pr [v(k)=S
q
] , (13)
where
Pr[
dem
[b
1
(k

)],
dem
[b
2
(k

)]|v(k) = S
q
] =
2

i=1
Pr[b
i
(k

) = C
gray
q,i
], (14)
Pr[b
i
(k

) = C
gray
q,i
] = f(
dec
[b
i
(k

)], C
gray
q,i
). (15)
On top of that, a-priori probability Pr[v(k) = S
q
] is given by
Pr [v(k) = S
q
] =
2

i=1
Pr
[
c
i
(k

) = C
nat
q,i
]
, (16)
Pr
[
c
i
(k

) = C
nat
q,i
]
=f(
dec
[c
i
(k

)], C
nat
q,i
). (17)
Note that C
gray
q,i
is used in Eqs. (14) and (15), and C
nat
q,i
is
done in Eqs. (13), (16), and (17). This is because LLRs after
demapper in Eq. (13)
dem
[b
1
(k

)] and
dem
[b
2
(k

)] have
an orthogonal (Gray coded) relationship for a decoding of
cooperative code. Eventually, extrinsic LLR for cooperative
decoder output is given by

coop
[c
i
(k

)] =
ln
Pr
[
c
i
(k

)=1|
dem
[b
1
(k

)],
dem
[b
2
(k

)]
]
Pr [c
i
(k

)=0|
dem
[b
1
(k

)],
dem
[b
2
(k

)]]

dec
[c
i
(k

)]. (18)
Then, the LLR
coop
[c
i
(k

)] is delivered to channel decoder

via a deinterleaver. After this process is iterated several times,
d
i
(k

) is decoded by channel decoder. With this methodology,

even if there exists the received signal level imbalance between
U
1
-D and U
2
-D links, both information d
1
(k

) andf d
2
(k

)
can be detected with high reliability.
V. SIMULATION RESULTS
For evaluations of the cooperative relay using cooperative
coding and iterative decoding techniques, computer simula-
tions are conducted to verify PER versus imbalanced power
gain g. Simulation parameters are the same as Tab. I. In this
simulation, square Gray coded QPSK is adopted in symbol
mapper at each user node for conventional. Figures. 6 (a) and
(b) show PER performances for information packets comprised
of d
1
(k

) and d
2
(k

). In the gure, PER performances in

the cases with and without cooperative relay (w/o Coop, w/
Coop). In the case without the cooperative relay, the nodes
U
1
and U
2
independently send packets only comprised of
d
1
(k

) and d
2
(k

) without any cooperation at the node U

2
.
The performances are depicted in an arbitrary received E
s
/N
0
which is dened by E
s
/N
0
between U
2
and D.
In the case without cooperation shown in Fig. 6 (a), the
lower value of g causes higher PER for d
1
(k

) due to lower
received E
s
/N
0
between U
1
and D. In the case with the coop-
eration, on the other hand, it is obvious that the transmission
quality for d
1
(k

) is signicantly improved. When E

s
/N
0
= 5
[dB] and g > 6 [dB], the PER is around 10
3
, although
the packet is not detectable (PER = 1) for the non-cooperative
case. This observation clearly shows a validity of the proposed
cooperation with iterative decoding.
On the other hand, Fig. 6 (b) tells us a negative point.
When the received signal level of the U
1
is getting lower,
which corresponds to g becoming smaller, it becomes more
difcult to resolve the multiplexed packets on the XOR basis
from U
2
. As a result, when g < 6 [dB], PERs for d
2
(k

)
in the cooperative case becomes worse than that for the non-
cooperative case. However, it is not so serious if we introduce
a rule that each node cooperates only when such a cooperation
is benecial for both nodes. For example, in the Fig. 6 case,
cooperation is benecial when g is higher than 6 [dB].
This rule might sound like strict, which is not really. The
condition for the cooperation (g > 6 [dB]) means that the
distance need not be strictly the same, which indicates that
the proposed cooperative relay with the cooperation rule is
very exible in forming a triangular network topology under
dynamic variation of channel conditions. When there are a lot
of nodes in the wireless environments, it would be much easier
to nd a node that meets requirements for the cooperation.
VI. CONCLUSION
In this paper, a cooperative relay exploiting XORed joint
network-and-channel coding and iterative decoding is pro-
posed for cooperative coding based relay. An iterative de-
coding for a concatenated structure comprised of the joint
network-and-channel coding is capable of obtaining signicant
coding gain in triangular topology network which is subject to
imbalanced power gain between the direct and relayed links.
Computer simulation conrms that the proposed cooperation is
effective to solve the imbalanced power problem. The solution
promises that wireless networking to support such a coopera-
tion can be easily found in the multi-node environments.
ACKNOWLEDGMENT
This research was supported in part by Global COE
(Centers of Excellence) Program of the Ministry of Edu-
10 8 6 4 2 0
10
3
10
2
10
1
10
0
imbalanced power gain g[dB]
P
E
R


w/ Coop(3.0)
w/ Coop(4.0)
w/ Coop(5.0)
w/o Coop(3.0)
w/o Coop(4.0)
w/o Coop(5.0)
(a) Information packet for d
1
(k

).
10 8 6 4 2 0
10
3
10
2
10
1
10
0
imbalance power gain g[dB]
P
E
R


w/ Coop(3.0)
w/ Coop(4.0)
w/ Coop(5.0)
w/o Coop(3.0)
w/o Coop(4.0)
w/o Coop(5.0)
(b) Information packet for d
2
(k

).
Fig. 6. PER for both of the schemes: () is received Es/N
0
dB.
cation, Culture, Sports, Science and Technology, Japan, and
KAKENHI (20360168), Japan.
REFERENCES
[1] T. M. Cover and J. A. Thomas, Elements of Information Theory. New
York: Wiley, 1991.
[2] A. Goldsmith, Wireless Communication. New York; Cambridge Press,
2005.
[3] P. Gupta and P. R. Kumar, The capacity of wireless networks, IEEE
Trans. Inform. Theory, vol. 46, no. 2, pp. 399404, Mar. 2000.
[4] K. Yamamoto and S. Yoshida, Tradeoff between area spectral efciency
and end-to-end throughput in rate-adaptive multihop radio networks,
IEICE Trans. Commun., vol. E88-B, no. 9, pp. 35323540, Sept. 2005.
[5] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung, Network informa-
tion ow, IEEE Trans. Inform. Theory, vol. 46, no. 4, pp. 10241216,
July 2000.
[6] Y. D. Chen, S. Kishore, and J. Li, Wireless diversity through network
coding, in Proc. WCNC 06, vol. 3, Las Vegas, NV, USA, Apr. 2006,
pp. 16811686.
[7] C. Hausl and J. Hagenauer, Iterative network and channel decoding
for the two-way relay channel, in Proc. ICC06, Istanbul, Rurkey, June
2006, pp. 15681573.
[8] C. Hausl and P.Dupraz, Joint network-channel coding for the multiple-
access relay channel, in Proc. SECON06, Reston, VA, USA, Sept.
2006, pp. 817822.
[9] C. Berrou, A. Glavieux, and P. Thitimajshima, Near shannon limit error-
correcting coding and decoding: Turbo-codes (1), in Proc. ICC93,
Geneva, Switzerland, May 1993, pp. 10641070.
[10] S. ten Brink, Convergence behavior of iteratively decoded parallel
concatenatedcodes, IEEE Trans. Commun., vol. 49, no. 10, pp. 1727
1737, Oct. 2001.
[11] T. Matsumoto, S. Ibi, S. Sampei, and R. Thoma, Adaptive transmission
with single-carrier multilevel BICM, Proceedings of the IEEE, vol.
95,no.12, pp. 23542367, Dec. 2007.
[12] P. Robertson, E.Villebrun, and P. Hoeher, A comparison of optimal and
sub-optimal MAP decoding algorithms operating in the log domain, in
Proc. ICC95, Seatle, USA, June 1995, pp. 10091013.