3500

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

Relay Selection with ANC and TDBC Protocols in

Bidirectional Relay Networks
MinChul Ju and Il-Min Kim, Senior Member, IEEE
AbstractWe study relay selection (RS) with the analog
network coding (ANC) and time division broadcast (TDBC),
which are two major amplify-and-forward (AF)-based protocols
in bidirectional relay networks. We consider a bidirectional
network consisting of two different end-sources and multiple
relays, where each terminal has a single antenna and operates
in a half-duplex mode. In this network, a single best relay is
selected depending on channel conditions to help bidirectional
communication between the two end-sources. Specifically, we first
consider RS schemes for the ANC and TDBC protocols based
on a max-min criterion to minimize the outage probabilities.
Then, for the RS in the ANC protocol, we derive a closed-form
expression of the outage probability; for the RS in the TDBC
protocol, we derive a one-integral form of the outage probability
and its lower bound in closed-form. Numerical results confirm
that the closed-form expression of the ANC protocol and the
one-integral form of the TDBC protocol are very accurate, and
that the closed-form lower bound of the TDBC protocol is also
tight.
Index TermsAmplify-and-forward (AF), analog network coding (ANC), bidirectional relay network, outage probability, relay
selection (RS), time division broadcast (TDBC).

I. I NTRODUCTION

ELAY communication is an effective method to attain

broader coverage range and to mitigate channel impairments by relaying signals in a network [1], [2]. One of
the most well-known relaying strategies is the amplify-andforward (AF) relaying where each relay terminal samples,
amplifies, and retransmits the incoming signals. In many
applications of relay networks, two different end-sources may
need to exchange information with the help of relays. Since
this bidirectional communication can increase bandwidth efficiency, it has recently received considerable attention [3].
Many researchers have studied two major AF-based protocols in bidirectional relay networks: the analog network
coding (ANC) protocol [4][9] and the time division broadcast
(TDBC) protocol [10][12]. The ANC protocol requires two
time slots to exchange information between the two endsources. Under a half-duplex constraint, however, the ANC
Paper approved by J. Wang, the Editor for Wireless Spread Spectrum of
the IEEE Communications Society. Manuscript received September 28, 2009;
revised February 4, 2010, April 5, 2010, and June 16, 2010.
This research was supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC), and by the Ubiquitous
Computing and Network (UCN) Project, Knowledge and Economy Frontier
R&D Program of the Ministry of Knowledge Economy (MKE) in Korea as
a result of UCNs subproject 10C2-C2-12T. This paper was presented in part
at the 25th Biennial Symposium on Communications, Kingston, May 2010.
M. Ju was with the Department of Electrical and Computer Engineering,
Queens University, Kingston, Ontario, Canada. He is now with Korea Electronics Technology Institute (KETI), Seoul, Korea (e-mail: mcju@keti.re.kr).
I.-M. Kim is with the Department of Electrical and Computer
Engineering, Queens University, Kingston, Ontario, Canada (e-mail:
ilmin.kim@queensu.ca).
Digital Object Identifier 10.1109/TCOMM.2010.101210.090585

cannot exploit the direct-path between the two end-sources

even if such direct-path physically exists. On the other hand,
the TDBC protocol can utilize the direct-path even under a
half-duplex constraint. However, the TDBC needs three time
slots for the exchange of information. Therefore, the ANC can
increase bandwidth efficiency, while the TDBC can improve
reliability due to the utilization of the direct-path. For instance,
in a bidirectional relay network with a single relay where the
direct-path between two end-sources exists, the diversity order
of the ANC protocol is still one due to a half-duplex constraint,
whereas that of the TDBC protocol is two.
Relay selection (RS) in relay networks has also received
considerable attention in the literature [13][19] because it can
enhance system performance with simple hardware. Bletsas
et al. proposed opportunistic relaying [13] and showed that
opportunistic AF relaying was outage-optimal among singleRS schemes [14]. Zhao et al. showed that the AF relaying with
RS provided lower outage probability and higher throughput
than the AF relaying without RS except when signal-to-noise
ratio (SNR) was very low [15]. Jing et al. derived the diversity
order of many single-RS schemes and proposed several SNRsuboptimal multi-RS schemes [19]. Recently, some works
studied RS based on the max-min criterion in the presence
of interference [20], [21]. However, these works have focused
on RS in unidirectional networks.
Very recently, Hwang et al. considered RS in bidirectional
networks [22]. In [22], the authors proposed a criterion for RS
to maximize the instantaneous sum-rate of two opposite traffic
flows in the ANC protocol, and they derived the performance
bounds in terms of the average sum-rate, average symbolerror rate, and outage probability. This sum-rate maximization
criterion is useful to maximize the whole system capacity. On
the other hand, in a multiuser system, the whole system is in
outage if any user is in outage [23, eq. (25)], [24]. In order
to minimize the outage of the multiuser system, therefore, the
minimum mutual information among all the users must be
maximized. In the ANC and TDBC protocols, there are two
opposite traffic flows from two different users in the same
channel; therefore, the two protocols can be considered as
multiuser systems, specifically two-user systems. To minimize
the outage of the ANC and TDBC protocols, therefore, the
minimum mutual information of two opposite traffic flows
must be maximized. This problem can be formulated as a maxmin problem. Consequently, when RS is combined with ANC
or TDBC, it is very useful to adopt the max-min criterion.
To the best of our knowledge, however, there has been no
work in which the max-min criterion was applied to RS in
the ANC and TDBC protocols and the outage performance
was analyzed based on this criterion. This has motivated our
work. In this paper, we consider RS for the ANC and TDBC

c 2010 IEEE
0090-6778/10$25.00

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

protocols in a bidirectional relay network consisting of two

different end-sources and multiple relays.
The contributions of this paper can be summarized as
follows.
We propose RS schemes for the ANC and TDBC protocols to minimize the outage probabilities. Specifically, we
adopt the max-min criterion to maximize the minimum
mutual information of two opposite traffic flows.
For the RS in the ANC protocol, we derive closed-form
outage probability.
For the RS in the TDBC protocol, we derive outage
probability in a one-integral form and its lower bound
in closed-form.
In particular, our work is different from [22]. In [22], a
single best relay was selected such that the sum of two mutual
information of two opposite traffic flows was maximized,
whereas in our work, a single best relay is selected such
that the minimum mutual information of two opposite traffic
flows is maximized. Note that the sum-rate maximization
criterion is useful when one focuses on the total traffic flows of
bidirectional communication, whereas the max-min criterion
is useful when one focuses on the individual traffic flow of
bidirectional communication. As a result, the two criterions
complement each other, not exclude each other.
The remainder of this paper is organized as follows. In
Section II, we describe the system model. In Section III, we
consider RS schemes based on the max-min criterion for the
ANC and TDBC protocols. In Section IV, we derive the outage
probability of the RS in the ANC protocol. In Section V,
we derive the outage probability of the RS in the TDBC
protocol. In Section VI, we present simulation results. Finally,
conclusions are drawn in Section VII.
Notation: We use := to denote that , by definition,
equals , and we use =: to denote that , by definition,
equals . Also, for a random variable , () denotes its
probability distribution function (PDF), and (), its cumulative distribution function (CDF). Finally, (, )
indicates that is a circularly symmetric complex-valued
Gaussian random variable with mean and variance .

3501

Rayleigh fading channel.1 Also, we can accommodate pathloss effect by choosing appropriate values for 0 , 1, , and
2, . Specifically, let 1 , denote the distance between S1
and R , and 2 , , the distance between S2 and R , both of
which are normalized by the distance between S1 and S2 .
Therefore, we have 1 , + 2 , = 1. Furthermore, we set
the path-loss exponent as four to model radio propagation in
urban areas [26]. As a result, we set 0 = 1, 1, = 4
1 , ,
4
and 2, = (1 1 , ) where = 1, , .
In the ANC protocol, there are two opposite traffic flows:
one is from S1 via R to S2 , and the other is S2 via R to
S1 . For the first traffic flow, R and S2 are the receivers.
Thus, it is reasonable to assume that R knows 1, , and
S2 knows {1, , 2, : = 1, , } as in the traditional
unidirectional relay network [27][29].2 Similarly, considering
the second traffic flow, it is reasonable to assume that R
knows 2, , and S1 knows {1, , 2, : = 1, , }. In
total, we assume that both end-sources know all the channel
coefficients {1, , 2, : = 1, , }, and that each relay R
knows the channel coefficients 1, and 2, . In a similar way,
for the TDBC protocol, we assume that both end-sources know
all the channel coefficients {0 , 1, , 2, : = 1, , }, and
that each relay R knows the channel coefficients 1, and
2, . Compared with the ANC, the TDBC needs additional
channel information on 0 at both end-sources because the
TDBC utilizes the direct-path between the two end-sources,
while the ANC cannot utilize the direct-path due to the halfduplex constraint even if such direct-path physically exists.
The channel information assumption adopted in this paper
has been made in almost all previous works3 on AF-based
bidirectional protocols [4][9], [12].4
A. ANC Protocol with Multiple Relays

II. S YSTEM M ODEL

The ANC protocol has widely been studied [4][9], [22].

The system model for the ANC protocol with multiple relays
is depicted in Fig. 1(a). In this protocol, at the first time slot,
S1 and S2 transmit simultaneously their symbols, 1 and 2 ,
over the multiple access channel, and the signal ANC received
by the -th relay R is given by

ANC = 1, 1 + 2, 2 + ANC
,
(1)

Consider a bidirectional relay network consisting of two

different end-sources and relays, where each terminal has
a single antenna and operates in a half-duplex mode, and
the relays adopt the AF protocol. We use S1 , S2 , and R
to denote the first source, the second source, and the -th
relay for = 1, , , respectively. We let 1 and 2 denote
the information-bearing symbols transmitted from S1 and S2 ,
respectively, and both symbols have unit average power. The
complex channel coefficient between S1 and S2 is denoted by
0 ; the complex channel coefficient between S1 and R is denoted by 1, ; and the complex channel coefficient between S2
and R is denoted by 2, . We assume that all the channels are
reciprocal and they are modeled as follows: 0 (0, 0 ),
1, (0, 1, ), and 2, (0, 2, ). Also, all the
channel coefficients are assumed to be fixed over two time
slots for the ANC protocol and three time slots for the TDBC
protocol. That is, our channel model is the frequency-flat

1 Also, it is an interesting topic to investigate a system for frequencyselective Rayleigh/Rician/Nakagami fading channels with timing errors [25].
2 Note that R is the receiver associated only with the path from S to R ;
1

thus, R knows 1, . On the other hand, S2 is the receiver associated with

all the pathes from S1 via {R }
=1 to S2 ; thus, S2 knows {1, , 2, : =
1, , }.
3 In some works on AF-based bidirectional protocols, even a stronger
channel assumption has been made. In [30], [31], it is assumed that each
relay R knows all the channel coefficients even for other relays. That is,
each R knows {1, , 2, : = 1, , }. In this case, the relays may use
beamforming techniques, but under the assumption of ideal frequency/time
synchronization across all the relays, which does not need to be assumed in
the RS.
4 Due to the higher signaling overhead, the perfect instantaneous channel
knowledge of all the links might be hard for a system with a large number
of relays. In practice, however, the number of relays should be restricted
to a small number due to some reasons: the nature of indoor home/office
applications may need only a small number of relays; timing, frequency, and
phase synchronizations become very difficult for a large number of relays; and
the complexity of protocol becomes very high for a large number of relays.
That is, assuming a small number of relays is practical and reasonable. As a
result, the burden on total signaling overhead may not be too much.

3502

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

Since S knows its own symbol and all thechannel coefficients, it can remove the self-interference, i.e. 21, 1

ANC
for S1 and 22, 2 for S2 , from ,
in (2). Therefore,
ANC
and
S1 and S2 can obtain new interference-free signals 1,
ANC
2, , respectively, as follows:

ANC
ANC
1,
= 1, 2, 2 + 1, ANC
+ 1,
,

(5)

ANC
ANC
ANC
+ 2, .
2, = 1, 2, 1 + 2,
(6)

R1

S1

Rl

S2

RL
(a) ANC protocol with multiple relays

R1

Consequently, using the approximate in (4), the instantaANC

ANC
neous SNR 1,
of the signal 2,
in (6) is given by

Rl

ANC
1,
=

RL

S1

(b) TDBC protocol with multiple relays

Time slot 2

ANC
2,
=

Time slot 3

Fig. 1. System models for the ANC and TDBC protocols with multiple
relays. In the ANC protocol, the direct-path between S1 and S2 is not plotted,
because it cannot be utilized due to the half-duplex constraint even if such
direct-path physically exists.

where is the transmission power at S1 and S2 , and ANC

is the additive white Gaussian noise (AWGN) at R with

(0, 1). At the second time slot, R amplifies
ANC

ANC with an amplifying coefficient and then broadcasts it

ANC
to S1 and S2 over the broadcast channel, and the signal ,
received by S is given by

ANC
ANC
,
= , ANC + ,

= , 1, 1 + , 2, 2

ANC
+ , ANC
+ ,
,
(2)

ANC
is the
where is the transmission power at R , and ,
ANC
AWGN at S with , (0, 1) for = 1, 2. In order to
ensure that the transmission power at R is , the amplifying
coefficient should be determined as follows:

1
.
=
2
(1, + 2, 2 ) + 1

(3)

With the above , however, the analysis becomes very

difficult. Adopting a widely-used approximation [27][29],
therefore, we approximate of (3) as follows:
1

.
(1, 2 + 2, 2 )

(7)

where 1, = 1, 2 and 2, = 2, 2 for = 1, , .

ANC
ANC
of the signal 1,
in
Similarly, the instantaneous SNR 2,
(5) is given by

S2

Time slot 1

1, 2,
,
1, + ( + )2,

(4)

This approximation was originally proposed for the high SNR

regime; but it was also numerically demonstrated that the
approximation was still very accurate even if the SNR was as
low as 0 dB [27]. Through numerical results, in Section VI,
we will also demonstrate that the approximate in (4) yields
very accurate results over the entire SNR range.

1, 2,
.
( + )1, + 2,

(8)

ANC
in
In Section III.A, based on the instantaneous SNRs 1,
ANC
(7) and 2, in (8), we will study RS for the ANC protocol.
Remark: Comparison of the ANC and physical-layer network coding (PNC): The ANC and PNC are very similar
in many aspects: they need two time slots for bidirectional
communication between two sources with the help of a relay;
and under a half-duplex constraint, they cannot exploit the
direct-path between two sources even if such direct-path
physically exists. Thus, the diversity order of both protocols
is just one. On the other hand, the two protocols have many
different features. The capacity region of the ANC is expressed
by the end-to-end SNRs at both sources, while the capacity
region of the PNC is limited by the MAC capacity region. The
detection complexity of the PNC protocol is much higher than
that of ANC protocol. The bit-error rate (BER) performance
of the PNC protocol is slightly better than that of the ANC
protocol when binary phase shift keying (BPSK) is adopted
and there is one relay, which can be seen in Fig. 2.

B. TDBC Protocol With Multiple Relays

The TDBC protocol has also widely been studied [10][12].
The system model for the TDBC protocol with multiple relays
is depicted in Fig. 1(b). In this protocol, there are two different
signal components: direct-path signals and relay-path signals.
TDBC
We first consider the direct-path signals. The signal 1,0
transmitted from S2 and received by S1 at the second time
TDBC
slot and the signal 2,0
transmitted from S1 and received
by S2 at the first time slot are given by, respectively,

TDBC
TDBC
TDBC
TDBC
= 0 2 +1,0
; 2,0
= 0 1 +2,0
,
1,0
(9)
TDBC
TDBC
where ,0
is the AWGN at S with ,0
(0, 1)
for = 1, 2. Consequently, the instantaneous SNRs of the two
TDBC
TDBC
and 2,0
in (9), are identical
direct-path signals, 1,0
TDBC
, where 0TDBC = 0 2 .
and they are denoted by 0
TDBC
We then consider the relay-path signals. We let 1,
denote the signal transmitted from S1 and received by R at

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

10

PNC
ANC

10

Average BER

10

10

own symbol and all the channel coefficients, it can remove

the self-interference. Consequently, S1 and S2 can obtain new
TDBC
TDBC
interference-free signals 1,
and 2,
, respectively, as
follows:

TDBC
= 2, 1, 2, 2 + 1, 1, TDBC
1,
1,

TDBC
+ 2, 1, TDBC
+

,
(15)
2,
1,

TDBC
2,
= 1, 1, 2, 1 + 1, 2, TDBC
1,

TDBC
TDBC
+ 2, 2, 2,
+ 2,
.
(16)
Then, using the approximate , in (13), the instantaneous
TDBC
TDBC
TDBC
SNRs 1,
of the signal 2,
in (16) and 2,
of the
TDBC
signal 1,
in (15) are given by, respectively,

10

10

3503

10

15

20
10 log10 E

25

30

35

40

Fig. 2. Average BER against 10 log10 of the ANC and PNC with one
relay when BPSK is adopted. = = . 1,1 = 2,1 = 1.

TDBC
the first time slot, and let 2,
denote the signal transmitted
from S2 and received by R at the second time slot. Then
TDBC
,
is given by

TDBC
,
= , + TDBC
,
(10)
,

where TDBC
is the AWGN at R with TDBC
(0, 1)
,
,
for = 1, 2. At the third time slot, R first combines the two
TDBC
TDBC
received signals 1,
and 2,
, and then it broadcasts the
TDBC
combined signal. Specifically, the two received signals 1,
TDBC
and 2,
are combined as follows:
TDBC
TDBC
TDBC
= 1, 1,
+ 2, 2,
.

(11)

TDBC 2
Since [,
] = , + 1, we can make TDBC
in (11)

be unit power by selecting , as follows:

,
,
(12)
, =
, + 1

for = 1, 2. In the above equation, the value , is positive

but smaller than one, i.e. 0 < , < 1, and the values 1,
TDBC
and 2, determine how R combines the two signals 1,
TDBC
TDBC
and 2, , where 1, + 2, = 1. Note that
in (11)
always has unit power irrespective of the values 1, and 2, .
As we approximated in (4), we also approximate , of
(12) in the same way [27][29]:

,
,
.
(13)
,
Through numerical results, in Section VI, we will also demonstrate that the approximate , in (13) yields very accurate
results over the entire SNR range.
After combining the two received signals, R broadcasts the
in (11) to S1 and S2 , and the signal
combined signal TDBC

TDBC
,
received by S is given by

TDBC
TDBC
,
= , TDBC
+ ,
,
(14)

TDBC
TDBC
where ,
is the AWGN at S with ,
(0, 1)
for = 1, 2. Similar to the ANC protocol, since S knows its

1, 1, 2,
,
( + 2, )1, + 1, 2,
2, 1, 2,
=
.
2, 1, + ( + 1, )2,

TDBC
1,
=

(17)

TDBC
2,

(18)

In Section III.B, based on the instantaneous relay-path

TDBC
TDBC
in (17) and 2,
in (18), we will study RS
SNRs 1,
for the TDBC protocol.
III. RS IN ANC AND TDBC P ROTOCOLS
In this section, two RS schemes for the ANC and TDBC
protocols are proposed based on a max-min criterion to
minimize the outage probabilities.
A. RS in ANC Protocol
In this subsection, we propose an RS scheme for the
ANC protocol. We first formulate mutual information of two
opposite traffic flows for the -th relay-path of the ANC
protocol: from S1 via R to S2 and from S2 via R to S1 . For
the traffic flow from S1 via R to S2 , using (7), the mutual
ANC
is given by5
information 1,
1
ANC
log2 (1 + 1,
),
(19)
2
where we use the pre-log factor 1/2 because communication
from S1 to S2 is done during two time slots. For the traffic
flow from S2 via R to S1 , using (8), the mutual information
ANC
is given by
2,
ANC
1,
=

1
ANC
log2 (1 + 2,
).
(20)
2
In a multiuser system, the system is in outage if any user
is in outage [23, eq. (25)], [24]. Since there are two opposite
traffic flows from two different users in the same channel, the
ANC protocol can be considered as a two-user system, which
is a special case of a multiuser system. Therefore, the -th
ANC
in
relay-path of the ANC protocol is in outage if either 1,
ANC
(19) or 2, in (20) is smaller than a target rate. Since the
two end-sources are equivalent terminals, it is fair to set the
target rate of each source as /2, where denotes a target rate
ANC
=
2,

5 Since detection/decoding is not involved at the relay, the mutual information of the ANC is expressed by the end-to-end SNRs at both end-sources.
On the other hand, the achievable rate region of the PNC is limited by the
sum-rate constraint as well as individual rate constraints at the MAC phase
because the relay detects/decodes the incoming signals from two end-sources.

3504

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

in bps/Hz for the whole network. Then the outage probability

ANC
out,
() of the -th relay-path of the ANC protocol is given
by
[
]

ANC
ANC
ANC
out,
or 2,
() = Pr 1,
<
<
2
2
]
[

ANC
ANC
, 2,
]<
.
(21)
= Pr min[1,
2

way, for the traffic flow from S2 to S1 , the mutual information

TDBC
2,
is given by
TDBC
2,
=

1
TDBC
log2 (1 + 0TDBC + 2,
).
3

(25)

TDBC
() of the
Similar to (21), the outage probability out,
-th relay-path with the direct-path of the TDBC protocol is
given by
[
]

TDBC
TDBC
TDBC
out,
() = Pr min[1,
, 2,
]<
.
(26)
2

Based on this outage probability for each relay-path, we

now consider RS in the ANC protocol with multiple relays.
The objective of RS is to choose a single relay which minimizes the outage probability of the system. To this end, a relay
ANC
ANC
, 2,
] is maximized
must be selected such that min[1,

over all the relays. That is, the index of the selected relay
must be determined as follows:6

Taking a step similar to (22), we select a single relay to

TDBC
TDBC
, 2,
] over all the relays.
maximize the value min[1,
That is, the index of the selected relay must be determined
as follows:

= arg max min[ ANC , ANC ].

1,
2,

= arg max min[ TDBC , TDBC ].

1,
2,

=1, ,

(22)

The RS in the ANC protocol is implemented in a system as

follows. At the beginning of every two-time-slot,7 the mutual
ANC
information {,
: = 1, 2, = 1, , } is calculated
at both sources. Then one of the sources broadcasts selection
information to all the relays, and only the selected relay R
helps bidirectional communication between S1 and S2 over
two time slots. Specifically, at the first time slot, S1 and S2
simultaneously broadcast their signals. At the second time
slot, only the selected single relay R among all the relays
multiplies the incoming signal, which is the sum of two signals
transmitted from S1 and S2 , with an amplifying coefficient .
Then R broadcasts it to S1 and S2 .
Using (21) with (22), the achieved system outage probabilANC
() of the RS in the ANC protocol is given by
ity out
[
]

ANC
ANC
ANC
out () = Pr max min[1, , 2, ] <
. (23)
=1, ,
2
Note that the selection rule in (22), which maximizes the
ANC
ANC
and 2,
, also
minimum mutual information between 1,
minimizes the outage probability in (23). In Section IV, using
(23), we will derive the outage probability of the RS in the
ANC protocol.
B. RS in TDBC Protocol
In this subsection, we propose an RS scheme for the TDBC
protocol. Similar to the ANC protocol, there are two opposite
traffic flows between S1 and S2 . For the traffic flow from S1
to S2 , two channels are involved: the relay-path channel from
S1 via R to S2 and the direct-path channel from S1 directly to
TDBC
of the two channels
S2 . Then the mutual information 1,
is given by
1
TDBC
log2 (1 + 0TDBC + 1,
),
(24)
3
where we use the pre-log factor 1/3 because communication
from S1 to S2 is done during three time slots. In the same
TDBC
1,
=

6 This RS criterion can be considered as a generalization of [14, eq. (26)],

which considered RS in AF-based unidirectional relay networks.
7 In practice, RS is done every channel coherence time. Also, we can
calculate the total number of pilot signals, and it is 2 + 2. Consequently,
the signaling overhead of this selection is 2 + 2 per channel coherence
time.

=1, ,

(27)

The RS in the TDBC protocol is implemented in a system

as follows. At the beginning of every three-time-slot, the
TDBC
: = 1, 2, = 1, , } is
mutual information {,
calculated at both sources. Then one of the sources broadcasts
selection information to all the relays, and only the selected
relay R helps bidirectional communication between S1 and
S2 over three time slots. Specifically, at the first time slot, S1
broadcasts its signal; and at the second time slot, S2 broadcasts
its signal. At the third time slot, only the selected single relay
R among all the relays combines the two received signals and
broadcasts the combined signal to S1 and S2 .
Using (26) with (27), the achieved system outage probaTDBC
bility out
() of the RS in the TDBC protocol is given
by
[
]

TDBC
TDBC
TDBC
out () = Pr max min[1,
, 2,
]<
.
=1, ,
2
(28)
Note that the selection rule in (27), which maximizes the
TDBC
TDBC
and 2,
,
minimum mutual information between 1,
also minimizes the outage probability in (28). In Section V,
using (28), we derive the outage probability of the RS in the
TDBC protocol.
IV. O UTAGE P ROBABILITY FOR RS IN ANC P ROTOCOL
In this section, we derive the outage probability of the RS
in the ANC protocol. Using (19) and (20), we first rewrite the
ANC
() in (23) as follows:
outage probability out
ANC
out
()
]
[
[
]
ANC
ANC
), log2 (1 + 2,
) <
= Pr max min log2 (1 + 1,
=1, ,
[
]
)
(
ANC
ANC
= Pr max log2 1 + min[1,
, 2,
] <
(29)
=1, ,
]
[
(
)
ANC
ANC
, 2,
] < .
(30)
= Pr log2 1 + max min[1,
=1, ,

In order to solve the probability in (30), it is essential to

ANC
ANC
, 2,
]. Let us define as the
know the CDF of min[1,
ANC
ANC
minimum value between 1, and 2,
as follows:
ANC
ANC
:= min[1,
, 2,
].

(31)

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

(
)
1 ; 1 , 2 , 3 , 4 , 5 =

(
)
5
4
(4 + 5 )(1 + (2 + 3 ))
+
exp
4 + 5
4 + 5
1 3 1, 2,
(
)
(
)[
)
( +
2 + 3
2 + 3
1
2
3
+
1 2
exp
2
1 3 4
3 5
1 32 1, 2,
1 32 1, 2,
)+1
(
)]
(

(1) (2 + 3 )

+2
!
1 3 5
3 4
=0

In the following theorem, we derive the CDF of .

Theorem 1: The CDF () of is given by
(
)
() = 1 ; , , , 1, , 2,
(
)
+ 1 ; , , , 2, , 1, .
(32)
(
)
In the above equation, 1 ; 1 , 2 , 3 , 4 , 5 is given at the
top of the next page. In the equation (33), = 1 (1 + 3
2 )/(2 (2 + 3 1 )), () is the modified Bessel function
of the second kind with order , and () is the exponential
integral function for = 1, 2, .
Proof: See Appendix A.

Note that 1 (; 1 , 2 , 3 , 4 , 5 ) in (33) involves only standard functions: the exponential function, the modified Bessel
function, and the exponential integral function. Also, one can
calculate recursively the exponential integral function +1 ()
for = 1, 2, , as follows [32]:
)
1(
exp() () ,
for > 0.
+1 () =

Thus, only with 1 (), one can evaluate +1 () for =

1, 2, . Since 1 (; 1 , 2 , 3 , 4 , 5 ) in (33) is given in
infinite series, we test the convergence of the infinite series
expression in the following lemma.
Lemma 1: The infinite series expression in (33) is convergent.
+1
1, = exp(1 ) (2 ) +2 (3 ), where
Proof: Let
!
1 = (2 + 3 )/(1 3 5 ) + 1/(3 5 ), 2 = (2 + 3 )/(1 3 5 ),
and
3 = 1/(3 4 ). Then the series expansion in (33) is given

by
=0 (1) 1, . Since exponential integral function ()
is monotonically decreasing in , it can be shown that
1,+1

2 +3 (3 )
2
< lim
= 0.
lim
= lim
1,
+ 1 +2 (3 )
+ 1
By the ratio test [33],8 therefore, it can be shown that the
series expansion in (33) is convergent.

It follows from Lemma 1 that the CDF () of in

(32) is convergent. Using (30) with the CDF () in (32),
we derive outage probability in the following theorem.
ANC
Theorem 2: The outage probability out
() of the RS in
the AF protocol is given by
ANC
out
()

(),

(34)

=1

= 2 1.
where

a series
=0 , we define := lim +1 / . If < 1,
then the series converges absolutely; if > 1, then the series diverges; and
if = 1, then the series is inclusive [33].
8 For

3505

(33)

Proof: Let 1 = max=1, , . Since the CDF 1 ()

is given by 1 () = =1 (), one can obtain (34).

V. O UTAGE P ROBABILITY FOR RS IN TDBC P ROTOCOL
In this section, taking a step similar to (30), we first rewrite
TDBC
the outage probability expression out
() in (28) for the
RS in the TDBC protocol as follows:
(
[
TDBC
() = Pr log2 1 + 0TDBC
out
]
)
TDBC
TDBC
+ max min[1,
, 2,
] < 1.5 .
=1, ,

(35)

Using (35), we derive the outage probability in a one-integral

form for the RS in the TDBC protocol and its lower bound in
closed-form.
A. Outage Probability in One-Integral Form
Similar to in (31), we define as the minimum value
TDBC
TDBC
and 2,
as follows:
between 1,
TDBC
TDBC
, 2,
].
:= min[1,

(36)

In the following theorem, we derive the CDF of .

Theorem 3: The CDF () of is given by
()

(
)
1 ;

( 2, , 1, , , 1, , 2, )

1,
+(1 ; 1, , 2, , , 2, , 1, ,

2 ; ( + 2, )1, , 1, 2, , 1 + 2, / ,
=
2,

(
)

,
(
+

)
,
1
+

/ ,

2
2,

1,

1,

2,
1,

3,
(37)
where 1: (0 1, 1/2 and (1
21, ) ) or (1/2 < 1, 1 and (21, 1) );
2: 0 1, 1/2 and < (1 21, ) ; and
3: 1/2 < 1, 1 and < (21, 1) . In the above
equation,
(
)
2 ; 1 , 2 , 3
(
)
) (
(1
23
23
1)
=1
exp 3
+
. (38)
1
1
2
1 2
1 2
Proof: See Appendix B.

Using (35) with the CDF () in (37), we derive outage

probability in the following theorem.

3506

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

TDBC
Theorem 4: The outage probability out
() of the RS
in the TDBC protocol is given by
)
(

exp (/( 0 ))
TDBC

out () =
( )
,
0
0
=1
(39)
= 21.5 1.
where
TDBC
() in (35)
Proof: Let 2 = max=1, , . Then out
can be given by
(
)
]
[
TDBC
() = Pr log2 1 + 0TDBC + 2 < 1.5
out

[
]
TDBC ().
Pr 2 <
(40)
=
0
0

Since the CDF 2 () is given by 2 () =

=1 ()
and 0TDBC is an exponentially distributed random variable
with mean 0 , one can obtain (39).

TDBC
Although the obtained outage probability out
() in
(39) is very accurate, the final expression is given in a oneintegral form, which requires numerical integration. Therefore,
it should be also useful to derive a closed-form bound of the
outage probability. In the next subsection, we derive a lower
bound of the outage probability in closed-form, which does
not involve any numerical integration.
B. Lower Bound of Outage Probability in Closed-Form
In the performance analysis of relay networks, the following
inequality has widely been used [13], [27]: 1 2 /(1 + 2 ) <
min[1 , 2 ], where 1 > 0 and 2 > 0. Using this inTDBC
TDBC
and 2,
,
equality, the minimum value between 1,
TDBC
TDBC
min[1,
, 2,
], is upper-bounded by
TDBC
TDBC
, 2,
]
min[1,
[
]
[
1,
2, ,
< min min 1, ,
+ 2,
]]
[
2,
min
1, , 2,
+ 1,
[
]
1,
2,
= min 1, ,
2, ,
1, , 2,
+ 2,
+ 1,
U
(41)
=: .

In the following lemma, we derive the CDF U () of U .

Lemma 3: The CDF U () of U is given by
U () = 1 exp (()) ,
where
() =

2, ( +2, )1, +1, ( +1, )2,

,
1, 2, 1, 2,
1, +2, 1, +1, 2,
,
1, 1, 2,
2, +2, 1, +1, 2,
,
2, 1, 2,

(42)

1,
2,

3.
(43)
Proof: See Appendix C.

Note that the CDF U () in (42) is expressed by the

exponential function. Taking a step similar to (39) and using
the CDF U () in (42), we derive a closed-form lower bound
of outage probability in the following theorem.

TDBC
() of the
Theorem 5: The closed-form lower bound out,LB
outage probability of the RS in the TDBC protocol is given
by

(
TDBC
out,LB
() = 1 exp

+
(1)

0
1 =1
=1
=1


!

1 <<

)
(
)
( ) exp /(
0 )
exp
=1
.

1 0 =1 ( )
(44)
Proof: See Appendix D.

TDBC
Note that since the lower bound out,LB
() of (44) is given
in closed-form, it does not require any numerical integration.
TDBC
() of (44) reduces to [12, eq. (26)]
Also, note that out,LB
when = 1 and 1, = 2, = 1/2 for = 1, , .
Therefore, the analysis in this subsection can be considered
as a generalization of that of [12].
VI. S IMULATION R ESULTS
In this section, we first discuss the convergence speed of
the series expression in (33), which was analytically shown
to be convergent in Section IV. Then, for the proposed RS in
the ANC and TDBC protocols, we check the accuracy of the
obtained outage probabilities by comparing our analysis with
Monte Carlo simulations. Finally, we compare the RS in the
ANC protocol with the RS in the TDBC protocol.
A. Convergence Speed of Series Expression in (33)
In this subsection, we check the convergence speed of the
( ) (; 1 , 2 , 3 , 4 , 5 )
series expression in (33). We let
1
denote a truncated version of 1 (; 1 , 2 , 3 , 4 , 5 ) with
truncation
window size , i.e.,
1 (; 1 , 2 , 3 , 4 , 5 ) of

(33) with
=0 replaced by
=0 . Then we define ( )
as the truncation error normalized by 1 (; 1 , 2 , 3 , 4 , 5 )
as follows:
( ) :=



()

1 (; 1 , 2 , 3 , 4 , 5 )
1 (; 1 , 2 , 3 , 4 , 5 )

1 (; 1 , 2 , 3 , 4 , 5 )
(
)
( +
1 )
2
3
+
= exp
1 3 4
3 5

(
)+1
(
)


(1)
(2 +3 )
+2
 =+1 !

1 3 5

3 4
.

1 (; 1 , 2 , 3 , 4 , 5 )
(45)

Fig. 3 shows the normalized truncation error ( ) against

truncation window size . We set = 2, 1 = 2 = 3 = ,
and 4 = 5 = 1. For various values, = 10, 20, 30
dB, we can see the normalized truncation error is very small
even with a small value . For instance, the normalized
truncation error is smaller than 107 when = 20 dB and
= 1. Therefore, even with a small value , substituting
( ) (; 1 , 2 , 3 , 4 , 5 ) into (32) and (37) gives very ac
1
( )
( )
curate closed-form approximations () and () to
( )
actual () and (), respectively. Using () and
( )
(), in the following, we calculate outage probabilities
of (34) and (39).

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

3507

10

(N) for E = 10 dB
(N) for E = 20 dB
(N) for E = 30 dB

Analysis using eq. (34)

Simulation with exact l in eq. (3)

10

Simulation with approximate in eq. (4)

l

10

10

L=1
2

10
Outage probability

Normalized trucation error

10

10

10

L=2

10

10

L=3
10

10

10

L=4
12

10

3
N

10

0.1

0.2

0.3

0.4

0.5
d

0.6

0.7

0.8

0.9

S ,R
1

Fig. 3. Normalized truncation error ( ) in (45) against truncation window

size . = 2, 1 = 2 = 3 = , and 4 = 5 = 1.
0

10

Fig. 5. Outage probability against 1 , of the RS in the ANC protocol.

= 1, 2, 3, 4. = 1.5 bps/Hz. = = 10 dB. 1, = 4
1 , and
()
2, = (1 , )4 where = 1, , . = 1 in
() for

() of (34).

Analysis using eq. (34)

Simulation with exact in eq. (3)

Simulation with approximate l in eq. (4)

10

C. Accuracy of Outage Probability for RS in TDBC protocol

2

Outage probability

10

L=1
3

10

10

10

L=2

L=3

L=4

10

10

15

20
10 log

10

25
E

30

35

40

45

Fig. 4. Outage probability against 10 log10 of the RS in the ANC protocol.

= 1, 2, 3, 4. = 1 bps/Hz. = = . 1, = 2, = 1 where
()
= 1, , . = 1 in () for () of (34).

In this subsection, we check the accuracy of the obtained

outage probabilities of (39) and (44). Fig. 6 shows the outage
probability against 10 log10 of the RS in the TDBC protocol,
where we set = 1, 2, 3, 4, = 1 bps/Hz, = = ,
0 = 1, = 2, = 1, and 1, = 2, = 1/2 for =
( )
1, , . Also, we set = 2 in () for () of
(39). From Fig. 6, we can see that (39) is very accurate even
with = 2, and that (44) is tight. As in Fig. 5, we investigate
the effect of relay location. Fig. 7 shows the outage probability
against 1 , of the RS in the TDBC protocol, where we set
= 1, 2, 3, 4, = 2 bps/Hz, = = 15 dB, 0 = 1,
4
1, = 4
, and 1, = 2, = 1/2
1 , , 2, = (1 1 , )
( )
for = 1, , . Also, we set = 2 in () for ()
of (39). Irrespective of relay location in Fig. 7, we can see
that (39) is very accurate even with = 2, and (44) is tight.

D. Comparison Between RS With ANC and RS With TDBC

B. Accuracy of Outage Probability for RS in ANC protocol
In this subsection, we check the accuracy of the obtained
outage probability of (34). Fig. 4 shows the outage probability
against 10 log10 of the RS in the ANC protocol, where we
set = 1, 2, 3, 4, = 1 bps/Hz, = = , and 1, =
( )
2, = 1 for = 1, , . Also, we set = 1 in ()
for () of (34). From Fig. 4, we can see that (34) is
very accurate even with = 1. In Fig. 5, we investigate the
effect of relay location. Fig. 5 shows the outage probability
against 1 , of the RS in the ANC protocol, where we set
= 1, 2, 3, 4, = 1.5 bps/Hz, = = 10 dB, 1, =
4
for = 1, , . Also, we
4
1 , , and 2, = (1 1 , )
( )

set = 1 in () for () of (34). Irrespective of

relay location in Fig. 5, we can see that (34) is very accurate
even with = 1.

In this subsection, we compare the RS with the ANC

protocol and the RS with the TDBC protocol. Fig. 8 shows
the outage probabilities against 10 log10 of the RS with the
ANC protocol and the RS with the TDBC protocol, where
we set = 4, = 1, 2, 3, 4 bps/Hz, = = ,
0 = 1, = 2, = 1, and 1, = 2, = 1/2 for the
RS in the TDBC protocol with = 1, , . Also, we set
( )
( )
= 2 in both () for () of (34) and ()
for () of (39). With a simple modification of [12], for
the RS with the ANC protocol and the RS with the TDBC
protocol, it can be shown that the diversity orders are and
+ 1, respectively, and the multiplexing gains are 1 and 2/3,
respectively. Thus, in the high SNR regime, the RS with the
TDBC protocol always outperforms the RS with the ANC
protocol irrespective of the target rate . In the low and
medium SNR regimes, however, one scheme outperforms the

3508

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

10

Analysis using eq. (39)

Lower bound using eq. (44)
Simulation with exact i,l in eq. (12)

RS with ANC: simulation with approximate in eq. (4)

l

Simulation with approximate i,l in eq. (13)

10

RS with ANC: analysis using eq. (34)

RS with ANC: simulation with exact l in eq. (3)

10

RS with TDBC: analysis using eq. (39)

RS with TDBC: simulation with exact i,l in eq. (12)

10

RS with TDBC: simulation with approximate in eq. (13)

i,l

10

Outage probability

Outage probability

10

10

10

10

10

10

R=1
4

L=3

10

R=4
R=3

L=2
5

10

L=4

10

R=2

10

L=1

10

15

20

25

30

10

35

10

15

20
10 log

10 log10 E

10

Fig. 6. Outage probability against 10 log10 of the RS in the TDBC protocol. = 1, 2, 3, 4. = 1 bps/Hz. = = . 0 = 1, = 2, = 1
()
and 1, = 2, = 1/2 where = 1, , . = 2 in () for

() of (39).

25

30

35

Fig. 8. Outage probabilities against 10 log10 of the RS with the ANC

protocol and the RS with the TDBC protocol. = 4. = 1, 2, 3, 4 bps/Hz.
= = . 0 = 1, = 2, = 1, and 1, = 2, = 1/2 for the RS
()
with the TDBC protocol where = 1, , . = 2 in both () for

()
() for () of (39).
() of (34) and

10

10

Analysis using eq. (39)

Lower bound using eq. (44)
Simulation with exact i,l in eq. (12)

10

RS with ANC: simulation with approximate in eq. (4)

l

Simulation with approximate i,l in eq. (13)

RS with ANC: analysis using eq. (34)

RS with ANC: simulation with exact in eq. (3)

10

RS with TDBC: analysis using eq. (39)

RS with TDBC: simulation with exact in eq. (12)
i,l

RS with TDBC: simulation with approximate in eq. (13)

i,l

L=1

10

Outage probability

Outage probability

10

L=2

10

R=3
2

10

R=2

10

10

L=3
R=1

10

10

L=4
5

10
6

10

0.1

0.2

0.3

0.4

0.5
d

0.6

0.7

0.8

0.9

0.1

S ,R

Fig. 7. Outage probability against 1 , of the RS in the TDBC protocol.

= 1, 2, 3, 4. = 2 bps/Hz. = = 15 dB. 0 = 1, 1, = 4
1 , ,
2, = (1 1 , )4 , and 1, = 2, = 1/2 where = 1, , .
()
= 2 in () for () of (39).

0.3

0.4

0.5
d

0.6

0.7

0.8

0.9

Fig. 9. Outage probabilities against 1 , of the RS with the ANC protocol

and the RS with the TDBC protocol. = 2. = 1, 2, 3 bps/Hz. = =
4 , and
15 dB. 0 = 1, 1, = 4
1, = 2, =
1 , , 2, = (1 1 , )
1/2 for the RS with the TDBC protocol where = 1, , . = 2 in both
()
()
() for () of (34) and () for () of (39).

other scheme depending on and SNR value. As in Figs. 5

and 7, we investigate the effect of relay location. Fig. 9 shows
the outage probabilities against 1 , of the RS with the ANC
protocol and the RS with the TDBC protocol, where we set
= 2, = 1, 2, 3 bps/Hz, = = 15 dB, 0 = 1,
4
1, = 4
, and 1, = 2, = 1/2
1 , , 2, = (1 1 , )
for the RS in the TDBC protocol with = 1, , . Also, we
( )
( )
set = 2 in both () for () of (34) and ()
for () of (39). Irrespective of relay location in Fig. 9, as
the target rate increases, we can see the RS with the ANC
protocol outperforms the RS with the TDBC protocol.

0.2

S ,R

VII. C ONCLUSIONS
In this paper, we have studied RS for the ANC and TDBC
protocols in a bidirectional relay network consisting of two
end-sources and multiple relays. A single best relay was
selected depending on channel conditions to help bidirectional
communication between the two end-sources. Specifically, we
have selected a single best relay based on a max-min criterion
to minimize the outage probabilities of the ANC and TDBC
protocols. Then, for the RS in the ANC protocol, we have
derived a closed-form expression of the outage probability.
Also, for the RS in the TDBC protocol, we have derived

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

a one-integral form of the outage probability and its lower

bound in closed-form. For future work, we will investigate
RS with optimal power allocation when each terminal has a
single antenna, and RS with precoding or beamforming when
each terminal has multiple antennas.
A PPENDIX A
P ROOF OF T HEOREM 1
ANC
ANC
From (7) and (8), the difference 1,
2,
is given by
ANC
ANC
2,
1,

(
)
1, 2,
.
=
( 1, + ( + )2, )(( + )1, + 2, )
(A.1)
2 1, 2,

Depending on the values (1, , 2, ), is given by

{ ANC
1, , 1, < 2, ,
(A.2)
=
ANC
2,
, 1, 2, .
Using (A.1), the CDF () of is given by
[ ANC
]
< , 1, < 2,
() = Pr 1,
[ ANC
]
+ Pr 2,
< , 1, 2,
[
]
1, 2,
= Pr
< , 1, < 2,
1, + ( + )2,
]
[
1, 2,
< , 1, 2,
+ Pr
( + )1, + 2,
(A.3)
=: 1 + 2 .

3509

Then 1 in (A.7) can be calculated by

(
)
1+ (+ )

1 =
1, ()2, ()
=0

=0

)
= 1+ (+

2, ()

)
1+ (+

( + )

=0

) 1,

1+

( + )

1, ()2, ()
)

1, ()2, ()
)
( + )
2, ()

=: 11 12 13 ,

(A.8)

where 1, () = 1 1, (). Since 1, and 2, are

exponentially distributed random variables with mean 1, and
2, , respectively, the difference 11 12 in (A.8) can be
solved into
2,
1,
+
11 12 =
1, + 2,
1, + 2,
(
)
(1, + 2, )(2 + )
exp
. (A.9)
1, 2,
Also, 13 in (A.8) can be rewritten as

exp(( + )/( 1, ))
( ( + ) )

2,
1+

(
)
2
( + )

exp

2 1, ( / ) 2,
(
)

( + )2

exp
=
,
We first solve 1 as follows:
2
( + )

2,
1,
2,
]
[

1 = Pr ( 2, )1, < ( + )2, , 1, < 2, .
(A.10)
(
(
))
(A.4)
Since the first condition, ( 2, )1, < ( + where = exp ( + )/( 1, ) + 1/( 2, ) .
)2, , of the probability in the right-hand side of (A.4) Since we have no closed-form solution to (A.10), we rewrite
is always satisfied for 2, / , the probability 1 in (A.10) as follows:
(
)

( + )2
(A.4) can be divided into two parts:

exp
13 =

[
]
2, 0
2 1,
2,

1 = Pr 1, < 2, , 2,
(
)
)
( +

( + )2

[
]

exp

.
]
[ ( + )

2, 0
2 1,
2,

2,
+ Pr 1, < min
, 2, , 2, >
.
(A.11)
2,

(A.5) Using [32, eq. (3.324.1)], the first part of the right

It can be shown that

hand side of (A.11) can be solved into 2 2+
1, 2,
(
)
]
[ ( + )

2,

. For the second part of the right1 2 2+

, 2,
min
1, 2,
2,
hand side of (A.11), since we have no closed-form solution,

(
)
( + )

2, ,
we tackle this problem by adopting
expansion of
1
+
,
<

2,

a series

(
)
=
exponential
function,
exp()
=
.
Substituting
the
(
+
)

(
+
)

2,

=0 !

.
2, , 2, > 1 +

series expansion of exp(/2, ) into the second part of the

(A.6) right-hand side of (A.11) yields
(
)
( + )
Referring to Fig. 10 and substituting (A.6) into (A.5) yield

[
]
(
)

exp

( + )

2, 0
2 1, 2,
1 = Pr 1, < 2, , 2,
1+
1

( )
(1) +1

[
2
1
exp

(A.12)
=
( + )2,
!

+ Pr 1, <
,
0 =0
2,

(1) +1
1
( + ) )
(
=

+2 (2 ),
(A.13)
1+
2, >
.
(A.7)
!

=0
13 =

3510

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 12, DECEMBER 2010

X 2,l = X 1,l

X 2,l
w w( E s + E r )
+
Er
Es Er
w
Er

P ROOF OF L EMMA 3

w( E s + E r ) X 2,l
P1 = Pr X 1,l <

s E r X 2 ,l E s w

Depending on the values ( , , 1, ), U in (41) is given

by

P1 = Pr[ X 1,l < X 2,l ]

[
]
(11, )
1,

min +

, 1,

1,
2,

+(1
)

1,

]1,
[
1,
2,
min 1, , +(1
=
2, ,
1, )

[
]

(1
)

1,
min
3.
+1, 1, , 2, ,
(C.1)

P1 = Pr[ X 1,l < X 2,l ]

X 1,l
Fig. 10.

1 on the (1, , 2, ) plane.

where 1 = ( + )/( 2, ) and 2 = /( 1, ).

Finally, substituting (A.13) into (A.11) yields 13 , which in
turn is substituted into 1 of (A.8) with 11 12 of (A.9).
Taking steps similar to those used from (A.4) to (A.13), one
can obtain 2 . Then, using (A.3) with 1 and 2 , one can
obtain the final expression of the CDF (), which is given
in (32).

A PPENDIX B
P ROOF OF T HEOREM 3
TDBC
TDBC
2,
is
From (17) and (18), the difference 1,
given at the top of the next page. Depending on the values
( , , 1, , 1, , 2, ), in (36) is given by

TDBC
TDBC
, 2,
], 1,
min[1,
TDBC
,
2,
1,
=
TDBC
2,
,
3.

(B.2)

(B.3)

where = (11, )( (21, 1) )/(1, ( +(21,

1) )). Then, taking steps similar to those used from (A.3) to
(A.13), one can obtain the first case of (37). For 2 of
(B.2), taking steps similar to the proof of [23, Lemma 1], one
can obtain the CDF () of as follows:
()
]
[
1, 1, 2,
<
= Pr
( + 2, )1, + 1, 2,
(
)
= 2 ; ( + 2, )1, , 1, 2, , 1 + 2, / ,
(B.4)
where 2 (; , , ) is given in (38). Finally, for 3
of (B.2), taking a step similar to (B.4), one can solve
the( CDF () of , which is given by ) () =
2 ; 2, 1, , ( + 1, )2, , 1 + 1, / .
A PPENDIX C

A PPENDIX D
P ROOF OF T HEOREM 5
Taking a step similar to (39), one can obtain the lower bound
TDBC
out,LB
() as follows:
)
(

exp (/( 0 ))
TDBC

out,LB () =
U ( )
.
0
0
=1
(D.1)
)
(
Using the product identity
1 ()
= 1 +
=1

1 =1, , =1
=1 (1)
=1 ( ), one can rewrite the
1 <<
expression =1 () in (D.1) as follows:

=1

For 1 of (B.2), the CDF () of is given by

]
[ TDBC
() = Pr 2,
< , 1, < 2,
[ TDBC
]
+ Pr 1,
< , 1, 2, ,

For 1 of (C.1), the CDF U () of U is given at the

top of the next page. With a simple integration, one can obtain
(42) with the first case of (43). Taking steps similar to those
used from (C.2) to (C.3), one can obtain (42) with the second
and third cases of (43).

U ()

=1+

=1

(1)

1 =1

(
exp

=1

1 <<

)
( ) .

=1

(D.2)

Then, substituting (D.2) into (D.1) and taking a simple integration, one can obtain the final closed-form result in (44).
R EFERENCES
[1] A. Sendonaris, E. Erkip, and B. Aazhang, User cooperation
diversitypart I, II, IEEE Trans. Commun., vol. 51, pp. 19271948,
Nov. 2003.
[2] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity in wireless networks: efficient protocols and outage behavior,
IEEE Trans. Inf. Theory, vol. 50, pp. 30623080, Dec. 2004.
[3] B. Rankov and A. Wittneben, Spectral efficient protocols for halfduplex fading relay chnnels, IEEE J. Sel. Areas Commun., vol. 25,
pp. 379389, Feb. 2007.
[4] S. Katti, S. Gollakota, and D. Katabi, Embracing wireless interference: analog network coding, in Proc. ACM SIGCOMM, 2007, pp.
397408.
[5] P. Popovski and H. Yomo, Wireless network coding by amplify-andforward for bi-directional traffic flows, IEEE Commun. Lett., vol. 11,
pp. 1618, Jan. 2007.
[6] X. Tang and Y. Hua, Optimal design of non-regenerative MIMO
wireless relays, IEEE Trans. Wireless Commun., vol. 6, pp. 1398
1407, Apr. 2007.
[7] R. Zhang, Y.-C. Liang, C. C. Chai, and S. Cui, Optimal beamforming
for two-way multi-antenna relay channel with analogue network
coding, IEEE J. Sel. Areas Commun., vol. 27, pp. 699712, June
2009.

JU and KIM: RELAY SELECTION WITH ANC AND TDBC PROTOCOLS IN BIDIRECTIONAL RELAY NETWORKS

TDBC
TDBC
1,
2,

(
)
1, 2, 1, ( + (21, 1) )2, (1 1, )( (21, 1) )1,
)(
)
= (
( + (1 1, ) )1, + 1, 2, (1 1, ) 1, + ( + 1, )2,

]
(1 1, )
1, ( + 1, )
2,
U () = Pr
1, < , 1, <
+ 1,
(1 1, )( + (1 1, ) )
]
[
1, ( + 1, )
1,
2,
2, < , 1,
+ Pr
+ (1 1, )
(1 1, )( + (1 1, ) )
+1,
(11, )
=
1, ()2, ()
(1
)( +(1
) )

3511

(B.1)

=0

( +(11, ) )

1,

=0

1,
1,
1, ( +1, )

1, ( +1, )

1, )( +(11, ) )

= (1

[8] Y. Han, S. H. Ting, C. K. Ho, and W. H. Chin, Performance bounds

for two-way amplify-and-forward relaying, IEEE Trans. Wireless
Commun., vol. 8, pp. 432439, Jan. 2009.
[9] Q. Li, S. H. Ting, A. Pandharipande, and Y. Han, Adaptive two-way
relaying and outage analysis IEEE Trans. Wireless Commun., vol.
8, pp. 32883299, June 2009.
[10] S. J. Kim, P. Mitra, and V. Tarokh, Performance bounds for bidirectional coded cooperation protocols, IEEE Trans. Inf. Theory, vol.
54, pp. 52355241, Nov. 2008.
[11] S. J. Kim, N. Devroye, P. Mitra, and V. Tarokh, Achievable rate
regions for bi-directional relaying, IEEE Trans. Inf. Theory, to be
published.
[12] Z. Yi, M. Ju, and I.-M. Kim, Outage probability and optimum
combining for time division broadcast protocol, to be published.
[13] A. Bletsas, A. Khisti, D. P. Reed, and A. Lippman, A simple
cooperative diversity method based on network path selection, IEEE
J. Sel. Areas Commun., vol. 24, pp. 659672, Mar. 2006.
[14] A. Bletsas, H. Shin, and M. Z. Win, Cooperative communications
with outage-optimal opportunistic relaying, IEEE Trans. Wireless
Commun., vol. 6, pp. 34503460, Sep. 2007.
[15] Y. Zhao, R. Adve, and T. J. Lim, Improving amplify-and-forward
relay networks: optimal power allocation versus selection, IEEE
Trans. Wireless Commun., vol. 6, pp. 31143123, Aug. 2007.
[16] E. Beres and R. Adve, Selection cooperation in multi-source cooperative networks, IEEE Trans. Wireless Commun., vol. 7, pp. 118127,
Jan. 2008.
[17] D. S. Michalopoulos and G. K. Karagiannidis, Performance analysis
of single relay selection in Rayleigh fading, IEEE Trans. Wireless
Commun., vol. 7, pp. 37183724, Oct. 2008.
[18] R. Tannious and A. Nosratinia, Spectrally-efficient relay selection
with limited feedback, IEEE J. Sel. Areas Commun., vol. 26, pp.
14191428, Oct. 2008.
[19] Y. Jing and H. Jafarkhani, Single and multiple relay selection
schemes and their achievable diversity order, IEEE Trans. Wireless
Commun., vol. 7, pp. 14141423, Mar. 2009.
[20] I. Krikidis, J. S. Thompson, S. McLaughlin, and N. Goertz, Maxmin relay selection for legacy amplify-and-forward systems with
interference, IEEE Trans. Wireless Commun., vol. 8, pp. 30163027,
June 2009.
[21] A. Bletsas, A. G. Dimitriou, and J. N. Sahalos, Interference-limited
opportunistic relaying with reactive sensing, IEEE Trans. Wireless
Commun., vol. 9, pp. 1420, Jan. 2010.
[22] K. S. Hwang, Y. C. Ko, and M.-S. Alouini, Performance bounds for
two-way amplify-and-forward relaying based on relay path selection,
in Proc. IEEE VTC, Apr. 2009, pp. 15.
[23] D. Tse, P. Viswanath, and L. Zheng, Diversity-multiplexing tradeoff
in multiple access channels, IEEE Trans. Inf. Theory, vol. 50, pp.
18591874, Sep. 2004.
[24] D. Tse and P. Viswanath, Fundamentals of Wireless Communications.
Cambridge, MA: Cambridge University Press, 2005.
[25] J. Wang and J. Chen, Performance of wideband CDMA systems with
complex spreading and imperfect channel estimation, IEEE J. Sel.
Areas Commun., vol. 19, pp. 152163, Jan. 2001.
[26] T. S. Rappaport, Wireless Communications: Principle and Practice.
Upper Saddle River, NJ: Prentice Hall, 2002.

(C.2)

1, ()2, ()

(C.3)

[27] P. A. Anghel and M. Kaveh, Exact symbol error probability of a

cooperative network in a Rayleigh-fading environment, IEEE Trans.
Wireless Commun., vol. 3, pp. 14161421, Sep. 2004.
[28] M. O. Hasna and M.-S. Alouini, End-to-end performance of transmission systems with relays over Rayleigh-fading channels, IEEE
Trans. Wireless Commun., vol. 2, pp. 11261131, Nov. 2003.
[29] A. Ribeiro, C. Xiaodong, and G. B. Giannakis, Symbol error
probabilities for general cooperative links, IEEE Trans. Wireless
Commun., vol. 4, pp. 12641273, May 2005.
[30] I. Hammerstom, M. Kuhn, C. Esli, J. Zhao, A. Wittneben, and G.
Bauch, MIMO two-way relaying with transmit CSI at the relay, in
Proc. SPAWC, June 2007, pp. 15.
[31] N. Lee, H. J. Yang, and J. Chun, Achievable sum-rate maximizing
AF relay beamforming scheme in two-way relay channels, in Proc.
IEEE ICC, May 2008, pp. 300305.
[32] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and
Products, 6th edition. San Diego, CA: Academic Press, 2000.
[33] W. Rudin, Principles of Mathematical Analysis, 3rd edition. New
York: McGraw-Hill, 1976.
MinChul Ju received the B.S. degree in electrical
engineering from Pohang University of Science and
Technology (POSTECH), Pohang, Korea, in 1997,
the M.S. degree in electrical engineering from Korea Advanced Institute of Science and Technology
(KAIST), Taejon, Korea, in 1999, and the Ph.D.
degree in electrical and computer engineering from
Queens University, ON, Canada, in 2010. From
1999 to 2005, he was a Researcher at Korea Electronics Technology Institute (KETI), Korea, where
he was involved in many projects related with
WPAN systems such as Bluetooth, IEEE802.15.3, IEEE802.11, and HomeRF.
In 2010, he returned to the Wireless Network Research Center at KETI, where
he is currently a Senior Researcher. His research interests are in the areas
of opportunistic transmissions, cooperative networks, and synchronization in
communication.
Il-Min Kim received the B.S. degree in electronics
engineering from Yonsei University, Seoul, Korea, in
1996, and the M.S. and Ph.D. degrees in electrical
engineering from the Korea Advanced Institute of
Science and Technology (KAIST), Taejon, Korea, in
1998 and 2001, respectively. From October 2001 to
August 2002 he was with the Department of Electrical Engineering and Computer Science at MIT,
Cambridge, USA, and from September 2002 to June
2003 he was with the Department of Electrical Engineering at Harvard University, Cambridge, USA,
as a Postdoctoral Research Fellow. In July 2003, he joined the Department
of Electrical and Computer Engineering at Queens University, Kingston,
Canada, where he is currently an Associate Professor. His research interests
include cooperative diversity networks, bidirectional communications, CoMP,
femto cells, and green communications. He is currently serving as an Editor
for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS and the
Journal of Communications and Networks (JCN).