You are on page 1of 14

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO.

9, SEPTEMBER 2010 2683

Performance Analysis of
Bidirectional Communication Protocols
Based on Decode-and-Forward Relaying
Peng Liu, Student Member, IEEE, and Il-Min Kim, Senior Member, IEEE

Abstract—We study and compare the performance of three To more efficiently serve the demand of the bidirectional
very typical bidirectional communication protocols based on the information exchange, the so-called bidirectional cooperative
decode-and-forward relaying: time-division broadcast (TDBC), network should be adopted [6]–[8]. To more elaborate on
physical-layer network coding (PNC), and opportunistic source
selection (OSS). We first derive the exact closed-form outage this, let us consider a bidirectional cooperative network where
probabilities for the PNC and OSS protocols and an exact but two end-sources S1 and S2 intend to exchange information
one-integral form outage probability for the TDBC protocol. with the help of a relay R. To accomplish this task, one may
To gain more insight, we also derive closed-form asymptotic straightforwardly apply the unidirectional relaying in a two-
outage probability expressions for all protocols, with which way manner. That is, S1 transmits information to R and S2
we obtain further analytical results on the asymptotic optimal
power allocation, the asymptotic optimal relay location, and the in the first time slot; in the second time slot, R forwards
performance gain of the OSS protocol over the TDBC protocol (either with AF or DF) the incoming signal to S2 ; in the third
in the high signal-to-noise ratio (SNR) regime. Finally, we study time slot, S2 transmits information to R and S1 ; and in the
the diversity-multiplexing tradeoff (DMT) for each protocol both fourth time slot, R forwards the signal to S1 . Obviously, the
in the finite and infinite SNR regimes. straightforward application of the unidirectional relaying to the
Index Terms—Decode-and-forward, diversity-multiplexing bidirectional network suffers from very low spectral efficiency
tradeoff (DMT), opportunistic source selection (OSS), physical- due to the half-duplex constraint. For this reason, more effi-
layer network coding (PNC), time-division broadcast (TDBC). cient bidirectional protocols which can exploit the underlying
nature of the physical wireless channels should be adopted
[5]–[8]. In this paper, we will refer to the straightforward
I. I NTRODUCTION application of the unidirectional relaying as the unidirectional

I N a unidirectional cooperative network, one source trans- approach.


mits information to one destination with the help of one It was shown that the spectral efficiency can be improved
or multiple relays, which may adopt either an amplify-and- by exploiting the shared broadcast channel (BC) nature of
forward (AF) or a decode-and-forward (DF) relaying [1], [2]. the wireless medium through the network coding concept [9],
In this network, the so-called cooperative diversity can be [10], [11]. A very good example is the time-division broadcast
achieved so that the reliability of the communication from (TDBC) protocol, which combines the transmissions of the
the source to the destination is enhanced [1]–[4]. In spite of third and fourth time slots in the unidirectional approach into
many benefits, the unidirectional techniques suffer from very one single transmission in a broadcast manner [11]–[13].1
low spectral efficiency due to the half-duplex constraint [5]. Specifically, S1 and S2 transmit their own signals in the first
Note that the half-duplex constraint is more practical than the and second time slots, respectively. In the third time slot, R
full-duplex constraint. amplify-and-forwards the superposition of the two received
There exist many applications where two end-sources intend signals to S1 and S2 through the BC as depicted in Fig. 1.
to exchange information [5]–[8]. For such applications, the Another possibility is that R decodes the received signals first
classical unidirectional techniques which support information and performs an XOR operation over the two decoded message
transmission in one direction only should not be adopted. bits from S1 and S2 ; then reencodes the XORed message bit
and transmits the new symbol in the third time slot. Through
Paper approved by D. I. Kim, the Editor for Spread Spectrum Transmission the network coding techniques, it is possible for each end-
and Access of the IEEE Communications Society. Manuscript received May source to detect the signal from its counterpart. This way
28, 2009; revised January 11, 2010 and March 23, 2010.
This paper was presented in part at the 25th Biennial Symposium on the TDBC protocol can accomplish one information exchange
Communications, Kingston, Ontario, Canada, May 2010. in three time slots, meaning that the spectral efficiency is
This research was supported in part by the Natural Sciences and En- increased by 33% over the unidirectional approach [11].
gineering Research Council of Canada (NSERC), and by the Ubiquitous
Computing and Network (UCN) Project, Knowledge and Economy Frontier It was also shown that the spectral efficiency can be
R&D Program of the Ministry of Knowledge Economy (MKE) in Korea, as
a result of UCN’s subproject 10C2-C2-12T. 1 Note that the TDBC protocol was also called straightforward network
The authors are with the Department of Electrical and Computer Engineer- coding in [11]. In [12], TDBC-I was used to specify a TDBC protocol without
ing, Queen’s University, Kingston, Ontario, K7L 3N6, Canada (e-mail: {6pl4, using the direct link between the two end-sources, and TDBC-II was used
ilmin.kim}@queensu.ca). to specify a TDBC protocol which utilizes the direct link. In this paper, we
Digital Object Identifier 10.1109/TCOMM.2010.080310.090292 consider the TDBC protocol which utilizes the direct link.
0090-6778/10$25.00 ⃝
c 2010 IEEE
2684 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

R R

S1 S2 S1 S2

If I1 > I2
Time slot 1
Time slot 1
Time slot 2
Time slot 3 Time slot 2
R
Fig. 1. The TDBC protocol in the bidirectional networks; either AF or DF
may be used at the relay.

S1 R S2 S1 S2
Time slot 1 Time slot 2 If I2 > I1

Fig. 2. The PNC (when DF is used at the relay) and ANC (when AF is Fig. 3. The OSS protocol in the bidirectional networks. 𝐼1 is the mutual
used at the relay) protocols in the bidirectional networks. information between the transmitted signal from S1 and the received signal at
S2 when S1 is selected as the transmitting source; 𝐼2 is the mutual information
between the the transmitted signal from S2 and the received signal at S1
when S2 is selected as the transmitting source. In [16], only AF-based OSS
further increased by allowing the two end-sources to transmit protocol was considered. In this paper, however, the DF-based OSS protocol
is considered.
simultaneously over a multiple-access channel (MAC), and
hence, only two time slots are needed for one information
exchange. This protocol is widely known as physical-layer
network coding (PNC) (see [11]–[14]2) if the DF protocol the OSS protocol. Numerical results of [16] also demonstrated
is employed at the relay or analog network coding (ANC) that the AF-based OSS protocol can achieve higher reliability
(see [15]–[17]) if the AF protocol is employed at the relay, in terms of both the average BER and the outage probability
as depicted in Fig. 2. It is easy to see that the PNC and performance than other AF-based bidirectional protocols such
ANC protocols achieve the highest spectral efficiency since as AF-based TDBC and ANC at some low data rates. The
two-traffic flows are concurrently supported at each time slot. outage probability and the the diversity-multiplexing tradeoff
Due to the half-duplex constraint, however, the direct link (DMT) performance of the AF-based bidirectional protocols
between S1 and S2 can not be utilized even if such direct such as the AF-based TDBC, ANC, and the AF-based OSS
link physically exists, and hence, the PNC and ANC protocols have been well studied in the literature [16], [17]. For the DF-
can only achieve diversity order one. On the other hand, the based bidirectional protocols including the DF-based TDBC,
direct link can be utilized in the TDBC protocol since the two PNC and the DF-based OSS, however, there have been no
end-sources transmit in two different time slots. Therefore, the analytical results on the outage probability and DMT analysis
TDBC protocol can achieve diversity order two.3 in the literature, to the best of our knowledge. This has
When multiple users exist, it was demonstrated that the motivated our work.
time-varying channel fluctuations can be exploited to achieve In this paper, we study and compare the performance of
the well-known multiuser diversity by allocating all resources three DF-based bidirectional protocols including the TDBC,
to the best user whose instantaneous channel quality is near PNC, and OSS protocols. We first derive exact closed-form
the peak [18], [19]. In the TDBC, PNC, and ANC proto- outage probabilities for the PNC and OSS protocols and
cols, however, the time-varying channel fluctuations are not an exact but one-integral form outage probability for the
exploited at all. To overcome this problem, an opportunistic TDBC protocol. To facilitate the comparison, we then de-
source selection (OSS) protocol for a bidirectional cooperative rive closed-form asymptotic outage probability expressions
network with a single AF relay was proposed [16], as depicted for all protocols, with which we prove that the TDBC and
in Fig. 3. In the OSS protocol, S1 and S2 transmit their own OSS protocols achieve full diversity order two and the PNC
information in an opportunistic manner such that the network protocol achieves diversity order one. Moreover, we obtain
throughput is maximized. That is to say, at each time slot, only further results on the asymptotic optimal power allocation,
a single end-source with the best channel condition is allowed relay location, and the performance gain of the OSS protocol
to transmit. Therefore, the multiuser diversity is achieved in over the TDBC protocol in the high signal-to-noise ratio
(SNR) regime. Finally, we study the diversity-multiplexing
2 Note that the PNC protocol was also called multiple access broadcast tradeoff (DMT) for each protocol both in the finite and infinite
(MABC) in [12] and [13]. SNR regimes.
3 The fact that the PNC protocol achieves diversity order one and the TDBC
The remainder of this paper is organized as follows. Sec-
protocol achieves diversity order two was first stated by inferring from whether
or not the direct link can be used, and numerically verified by Kim et al. in tion II introduces the system model and basic assumptions.
[12]. However, there is no analytical proof for this. Section III deals with the exact and asymptotic outage prob-
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2685

abilities and some further analyses on the asymptotic optimal reliability in certain cases, as demonstrated in Section V,
power allocation, relay location, and performance gain. Sec- although the spectral efficiency of OSS is lower than that of
tion IV investigates the DMT performance. Section V presents PNC and TDBC. By fixing the total transmission power 𝐸 for
some numerical results and Section VI concludes this paper. each transaction, we use 𝐸𝑟 to denote the average transmission
Notation: We use 𝐴 := 𝐵 to denote that 𝐴, by definition, power consumed by the relay R. For the TDBC and PNC
equals 𝐵 and we use 𝐴 =: 𝐵 to denote that 𝐵, by definition, protocols, it is further assumed that the average transmission
equals 𝐴. For two functions 𝑔(𝑥) and ℎ(𝑥), 𝑔(𝑥) = 𝒪(ℎ(𝑥)) power at 𝑆1 is identical to the average transmission power
means lim 𝑔(𝑥)/ℎ(𝑥) = 𝑐, where 𝑐 is a non-zero constant at 𝑆2 and the total transmission power at 𝑆1 and 𝑆2 in each
𝑥→𝑥0
and 𝑥0 is any given value possibly equal to ∞. For random transaction is denoted by 𝐸𝑠 . For the OSS protocol, we use
variable 𝑋, 𝑓𝑋 (𝑥) denotes its probability density function 𝐸𝑠 to denote the average transmission power at the selected
(PDF). Also, 𝑋 ∼ 𝒞𝒩 (𝜇, Ω) indicates that 𝑋 is a circularly end-source in each transaction. Therefore, irrespective of the
symmetric complex Gaussian random variable with mean 𝜇 protocols, we have 𝐸𝑠 + 𝐸𝑟 = 𝐸 for each transaction.
and variance Ω. Finally, log(⋅) denotes the base-2 logarithm, For ease of exposition, we introduce some useful notation
log10 (⋅) the base-10 logarithm, and ln(⋅) the natural logarithm. which will be used throughout the paper. we define the average
SNR 𝜌 as the ratio of the total transmission power 𝐸 to the
noise power 𝑁0 , i.e., 𝜌 := 𝑁𝐸0 . Also, we use 𝛽 to denote
II. S YSTEM M ODEL
the power ratio of 𝐸𝑠 to 𝐸, i.e., 𝛽 := 𝐸𝑠 /𝐸. We use 𝑅 to
Consider a bidirectional cooperative network consisting of denote a predetermined target rate for each transaction in the
two different end-sources and a single relay, where each whole network. Then the target rate for each traffic flow in the
terminal is equipped with a single antenna and operates in TDBC and PNC protocols is 𝑅/2 and the target rate for each
a half-duplex mode. We use S1 , S2 , and R to denote the two traffic flow in the OSS protocol is 𝑅. Moreover, we introduce
end-sources and the relay, respectively. It is assumed that relay three quantities related with 𝑅: 𝑇1 := 𝐿1 /𝜌, 𝑇2 := 𝐿2 /𝜌, and
R adopts the DF protocol and all channels are reciprocal and 𝑇3 := 𝐿3 /𝜌, where 𝐿1 := 21.5𝑅 − 1, 𝐿2 := 2𝑅 − 1, and
quasi-static over at least three time slots. In the remainder of 𝐿3 := 22𝑅 − 1.
this paper, we will refer to the DF-based TDBC and the DF-
based OSS protocols simply as the TDBC and OSS protocols,
respectively, whenever there is no ambiguity. Let ℎ1 denote III. O UTAGE P ROBABILITY A NALYSIS
the channel coefficient between S1 and R; ℎ2 the channel
coefficient between S2 and R; ℎ0 the channel coefficient for In this section, we study the outage performance of three
the direct link, if any, between S1 and S2 . The channel co- bidirectional protocols: TDBC, PNC and OSS. We also study
efficients are mutually independent with ℎ𝑖 ∼ 𝒞𝒩 (0, Ω𝑖 ) for the asymptotic outage in the high-SNR regime.
𝑖 = 0, 1, 2. Then ∣ℎ𝑖 ∣2 is an exponential random variable with
parameter 𝜆𝑖 , where 𝜆𝑖 = 1/Ω𝑖 . The complex noise associated
with each channel is modeled as a mutually independent A. TDBC Protocol
additive white Gaussian noise (AWGN) with zero mean and In the TDBC protocol, one information exchange is accom-
variance 𝑁0 . In the bidirectional protocols considered in this plished in three time slots, as depicted in Fig. 1. Let 𝑅1 and
paper, the two end-sources intend to exchange information. 𝑅2 denote the data rates of S1 and S2 , respectively. Then an
In the coherent detection, the receiver must know the channel achievable rate region of the TDBC protocol is the closure of
information. Thus, with coherent detection, it is reasonable the convex hull of the set of points (𝑅1 , 𝑅2 ) satisfying the
to adopt the full channel state information (CSI) assumption following inequalities [12, eqs. (8) and (9)]:
for the bidirectional cooperative networks. Specifically, we
assume that relay R knows both ℎ1 and ℎ2 , and both end- 𝑅1 < 𝐼1TDBC , (1)
sources know ℎ1 and ℎ2 in the PNC protocol, and ℎ0 , ℎ1 , and
ℎ2 in the TDBC and OSS protocols. Note that this full CSI 𝑅2 < 𝐼2TDBC , (2)
assumption has been widely adopted in many publications on
the bidirectional networks [5], [16], [17]. where 𝐼1TDBC and 𝐼2TDBC are defined as follows:
In this paper, we compare three bidirectional protocols ⎧ ( )
TDBC, PNC, and OSS by fixing the total power 𝐸 for  1
log 1 + 𝛽𝜌∣ℎ0 ∣2
, if ∣ℎ1 ∣2 < 2𝑇1


 3 2 𝛽 ,
each transaction in the whole bidirectional network. In the ⎨ ( )
PNC protocol, each transaction requires two time slots and 𝐼1TDBC = 1
log 1 + 𝛽𝜌∣ℎ0 ∣2
supports two traffic flows concurrently. In the TDBC proto- 
 3 ( 2 )


⎩ + 1 log 1 + (1 − 𝛽)𝜌∣ℎ2 ∣2 , if ∣ℎ1 ∣2 > 2𝑇1
col, each transaction requires three time slots and supports 3 𝛽 ,
two traffic flows concurrently as well. Unlike the PNC and (3)
⎧ ( )
TDBC protocols, the OSS protocol supports two traffic flows  1
2
log 1 + 𝛽𝜌∣ℎ2 0 ∣ , if ∣ℎ2 ∣2 < 2𝑇1


 3 𝛽 ,
opportunistically based on the channel condition and each ⎨ ( )
transaction in the OSS protocol requires two time slots and 𝐼2TDBC = 1
log 1 + 𝛽𝜌∣ℎ0 ∣2
supports only one traffic flow. This is a fundamental difference 
 3 ( 2 )


⎩ 1
+ 3 log 1 + (1 − 𝛽)𝜌∣ℎ1 ∣2 , if ∣ℎ2 ∣2 > 2𝑇1
between the OSS protocol and the PNC and TDBC protocols. 𝛽 .
This difference makes the OSS protocol have much higher (4)
2686 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

In the above equations, the pre-log factor 1/3 is because each is given by
traffic flow takes three time slots in the TDBC protocol [12].4,5 TDBC
𝑃out (𝑅, 𝜌)
The system is in outage when the target rate exceeds the ⎧ 4𝜆 (𝜆 +𝜆 )𝐿2 ( )
1 1 1
achievable rate. Therefore, the outage probability of the TDBC 

0 1 2
+ 𝒪 𝜌3 , 0 < 𝛽 ≤ 2/3,


𝛽 2 𝜌2

protocol is given by 
 (( )

 2(𝜆0 +𝜆1 +𝜆2 )2 −2𝜆20 2
( ) 
⎨ − (𝜆 1 +𝜆2 )
𝐿21
𝑅 𝑅 𝛽2 ( 2(1−𝛽) 2
) (8)
TDBC
𝑃out (𝑅, 𝜌) := Pr 𝐼1TDBC < or 𝐼2TDBC < . (5) = (𝜆1 +𝜆2 )2 (1+𝐿1 )(1−𝜃) 𝜃(1+𝐿1 )−(1−𝐿1 )
2 2 
 +

 2(1−𝛽)2 ) ( )

 2𝜆 (𝜆 +𝜆 )(1+𝐿 1 )(1−𝜃+ln 𝜃)
 − 1
+ 𝒪 1
𝜌3 ,
0 1 2
In the following theorem, we derive an exact expression of 
 𝛽(1−𝛽) 𝜌2

the outage probability for the TDBC protocol. 2/3 < 𝛽 < 1.
Theorem 1: For the TDBC protocol, an exact outage prob- Proof: See Appendix B
ability expression is given by
TDBC B. PNC Protocol
𝑃out (𝑅, 𝜌)
⎧ 2(𝜆0 +𝜆1 +𝜆2 )𝑇1 2𝜆 𝑇 2(𝜆1 +𝜆2 )𝑇1 The PNC protocol consists of two phases, namely the MAC
 − − 0𝛽 1 −

 1 + 𝑒 𝛽 − 𝑒 − 𝑒 𝛽 , phase and the BC phase, as depicted in Fig. 2. In each phase,

⎨ 0 < 𝛽 ≤ 2/3, (6) two traffic flows are simultaneously supported. We use 𝑅1 and
= 𝑅2 to denote the transmission rates of S1 and S2 , respectively.

 −
2(𝜆0 +𝜆1 +𝜆2 )𝑇1 2𝜆 𝑇
− 0𝛽 1
(𝜆 +𝜆 )𝑇
− 1 1−𝛽2 1

 1+𝑒 𝛽 −𝑒 −𝑒 An achievable rate region of the PNC protocol is the closure

−Ψ(𝑎, 𝑏, 𝑐, 𝜃), 2/3 < 𝛽 < 1. of the convex hull of the set of points (𝑅1 , 𝑅2 ) satisfying the
following inequalities [12], [13]:
The function Ψ(𝑎, 𝑏, 𝑐, 𝜃) is defined as
∫ 1 𝑅1 < 𝐼1PNC , (9)
𝑒−(𝑎𝑥+ 𝑥 ) 𝑑𝑥,
𝑏
Ψ(𝑎, 𝑏, 𝑐, 𝜃) := 𝑐 (7) 𝑅2 < 𝐼2PNC , (10)
𝜃 PNC
𝑅1 + 𝑅2 < 𝐼sum , (11)
𝜆1 +𝜆2 2𝜆
+ 𝛽𝜌0
where 𝑎 = (𝜆1 +𝜆 2 )(1+𝐿1 )
(1−𝛽)𝜌 , 𝑏= 2𝜆0
𝛽𝜌 , 𝑐 = 𝑎𝑒 (1−𝛽)𝜌 , and where 7
𝛽+2(1−𝛽)𝐿1 { ( )
𝜃 = 𝛽+𝛽𝐿1 . 1 𝛽𝜌∣ℎ1 ∣2
𝐼1PNC = min log 1 + ,
Proof: See Appendix A. 2 2
For 0 < 𝛽 ≤ 2/3, we have an exact truly closed-form 1 ( )}
2
expression for the outage probability of the TDBC protocol; log 1 + (1 − 𝛽)𝜌∣ℎ2 ∣ , (12)
2
for 2/3 < 𝛽 < 1, we have an exact outage probability { ( )
1 𝛽𝜌∣ℎ2 ∣2
expression in a one-integral form. Note that the integration 𝐼2PNC = min log 1 + ,
2 2
lower limit 𝜃 of (7) is a constant irrespective of 𝜌, and it can ( )}
be shown that 0 < 𝜃 < 1 by using the fact 2/3 < 𝛽 < 1. 1
log 1 + (1 − 𝛽)𝜌∣ℎ1 ∣2 , (13)
Solving Ψ(𝑎, 𝑏, 𝑐, 𝜃) in closed-form is extremely hard, to the 2
( ( ))
best of our knowledge; however, this integration can be easily PNC 1 𝛽𝜌 ∣ℎ1 ∣2 + ∣ℎ2 ∣2
computed with numerical approaches since the integration 𝐼sum = log 1 + . (14)
2 2
region is between 0 and 1.6
To provide more insight into the TDBC protocol, we ana- In eqs. (12)–(14), the pre-log factor 1/2 is because each traffic
lyze the asymptotic high-SNR performance in what follows. flow takes two time slots in the PNC protocol [12], [13].
Corollary 1: The TDBC protocol achieves full diversity Therefore, the outage probability of the PNC protocol is given
order two, and the asymptotic high-SNR outage probability by
( )
PNC PNC 𝑅 PNC 𝑅 PNC
4 The sums of logarithms in (3) and (4) are due to the fact that the codebook 𝑃out (𝑅, 𝜌) := Pr 𝐼1 < or 𝐼2 < or 𝐼sum < 𝑅 .
2 2
used at the relay is independent of the codebooks used at the end-sources [12], (15)
[13]. Note that there is no benefit of using a same (common) codebook for
the relay and the end-sources in the TDBC protocol (In the OSS protocol, 7 Note that (12) represents the minimum of the achievable rates of two
however, a same codebook scheme can significantly reduce the detection single-hop channels: the S1 -R channel and the R-S2 channel. Each single-hop
complexity, see footnote 10). First, using a same codebook would lead to channel has an achievable rate given by a logarithm function, which does not
a smaller achievable rate than using independent codebooks [29]. Moreover, require any independent assumption on the codebooks. For the same reason,
there is no benefit of using a same codebook in terms of the detection (13) does not require any independent assumption on the codebooks either.
complexity because the optimum ML detection at the intended destination Furthermore, (14) is the sum-rate of a Gaussian MAC channel, which requires
(either S1 or S2 ) is generally coupled with two different information-bearing the two transmitted symbols from S1 and S2 be independent. However, this
symbols due to an XOR operation at the relay, no matter the same codebook independence is automatically ensured by the independence between the end-
scheme or the independent codebooks scheme is used. Thus, the detection source massages. Therefore, in terms of achievable rate, there is no benefit
complexity stays the same, irrespective of the codebook schemes. For these of using independent codebooks as in the TDBC protocol (see footnote
reasons or more, to the best of our knowledge, independent codebooks are 4). Furthermore, as the direct link cannot be used, the intended destination
generally adopted in the TDBC protocol [12], [13]. always receives one single information-bearing symbol from the relay, and
5 Note that the achievable rate captured by (3) and (4) in this paper is
hence, the detection at the destination is always in a single-symbol decodable
essentially the same as the achievable rate captured by (8) and (9) of [12]; manner. This means that the detection complexity always stays the same.
they yield exactly the same outage probability. Therefore, for the PNC protocol, there is no difference in the achievable
6 In standard softwares such as Matlab, such integration can be easily rate or detection complexity, whether using a same codebook scheme or an
computed by the built-in function quad. independent codebooks scheme.
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2687

In the following theorem, we derive an exact closed-form this is equivalent to the traditional three-node unidirectional
expression of the outage probability for the PNC protocol. DF cooperative network [23, eq. (19)]. Therefore, the mutual
Theorem 2: For the PNC protocol, an exact closed-form information 𝐼1OSS between the transmitted signal from S1
outage probability expression is given by and the received signal at S2 when S1 is selected as the
transmitting source is given by10
𝑃 PNC (𝑅, 𝜌) ⎧ 1 ( )
⎧ out 2
if ∣ℎ1 ∣2 < 𝑇𝛽3 ,
 𝜆1
2(𝜆2 −𝜆1 )𝑇2 −2𝜆2 𝑇3
+ 𝜆1𝜆−𝜆
2(𝜆1 −𝜆2 )𝑇2 −2𝜆1 𝑇3
⎨ 2 log 1 + 𝛽𝜌∣ℎ0 ∣ ,

 1 + 𝜆2 −𝜆1 𝑒
 𝑒 ,
𝛽 2 𝛽
 ( )


2
for Case A, 𝐼1OSS = 1
log 1 + 𝛽𝜌∣ℎ0 ∣2 + (1 − 𝛽)𝜌∣ℎ2 ∣2 , (18)

 
⎩ 2

 2 𝑇3
if ∣ℎ1 ∣ > 𝛽 ,

 2𝜆2 𝑇3 2𝜆1 𝑇3

 1 − 𝑒− 𝛽 − 2𝜆2 (𝑇3𝛽−2𝑇2 ) 𝑒− 𝛽 , for Case B,

 where the pre-log factor 1/2 is because each traffic flow takes

⎨ (𝜆2 −𝜆1 )𝑇2 2𝜆2 𝑇3 (𝜆1 −𝜆2 )𝑇2 2𝜆1 𝑇3 two time slots. Secondly, let us consider the case when S2 is
= 1 + 𝜆2𝜆−𝜆
1
𝑒 1−𝛽 − 𝛽 + 𝜆1𝜆−𝜆 2
𝑒 1−𝛽 − 𝛽 , selected as the transmitting source. Then, in the same way, the

 1 2

 for Case C, mutual information 𝐼2OSS between the transmitted signal from



 ( ) S2 and the received signal at S1 when S2 is selected as the

 2𝜆2 𝑇3 2𝜆1 𝑇3

 1 − 𝑒− 𝛽 − 2𝜆2 𝑇𝛽3 − 1−𝛽 𝑇2
𝑒− 𝛽 , for Case D, transmitting source is given by



 ⎧ 1 ( )

⎩ 2
if ∣ℎ2 ∣2 < 𝑇𝛽3 ,
⎨ 2 log 1 + 𝛽𝜌∣ℎ0 ∣ ,

(𝜆1 +𝜆2 )𝑇2
1 − 𝑒− 1−𝛽 , for Case E.
( )
(16) 𝐼2OSS = 1
log 1 + 𝛽𝜌∣ℎ0 ∣2 + (1 − 𝛽)𝜌∣ℎ1 ∣2 , (19)

⎩ 2
𝑇3
2
In eq. (16), Case A represents 0 < 𝛽 ≤ 2/3 and 𝜆1 ∕= 𝜆2 ; if ∣ℎ2 ∣ > 𝛽 .
Case B: 0 < 𝛽 ≤ 2/3 and 𝜆1 = 𝜆2 ; Case C: 2/3 < 𝛽 ≤ 𝛽0 In the OSS protocol, the two end-sources are opportunisti-
and 𝜆1 ∕= 𝜆2 ; Case D: 2/3 < 𝛽 ≤ 𝛽0 and 𝜆1 = 𝜆2 ; and Case cally selected as the transmitting source such that the mutual
E: 𝛽0 < 𝛽 < 1, where 𝛽0 = 𝑇3𝑇+𝑇 3
2
.8 information of the whole network is maximized. That is, S1
Proof: See Appendix C. is selected as the transmitting source if 𝐼1OSS > 𝐼2OSS , and the
Note that the outage probability of (16) for the PNC mutual information 𝐼 OSS of the OSS protocol equals 𝐼1OSS , i.e.,
protocol only involves standard elementary functions such 𝐼 OSS = 𝐼1OSS . On the other hand, if 𝐼2OSS > 𝐼1OSS , S2 is selected
as the exponential function which can be easily computed. as the transmitting source and 𝐼 OSS = 𝐼2OSS . If 𝐼1OSS = 𝐼2OSS ,
Therefore, it is an exact and truly closed-form expression. To S1 and S2 are selected as the transmitting source with equal
provide more insight into the PNC protocol, we also study the probability.
asymptotic high-SNR performance in what follows. In the following lemma, a unified expression of the mutual
Corollary 2: The PNC protocol achieves diversity order information 𝐼 OSS combining (18) and (19) is derived.
one, and the asymptotic high-SNR outage probability is given Lemma 1: For the OSS protocol, the mutual information
by 𝐼 OSS between the transmitted signal from the selected source
⎧ ( ) and the received signal at the other end-source is given by
 2(𝜆1 +𝜆2 )𝐿2 1 1 ⎧ ( ) { }
⎨ + 𝒪 𝜌2 , 0 < 𝛽 ≤ 2/3,
PNC
𝛽 𝜌 

1
log 1 + 𝛽𝜌∣ℎ0 ∣2 , if max ∣ℎ1 ∣2 , ∣ℎ2 ∣2 < 𝑇𝛽3 ,
𝑃out (𝑅, 𝜌) = ( ) 

2
 
 ( )
⎩ (𝜆1 +𝜆2 )𝐿2 1 + 𝒪 12 , 2/3 < 𝛽 < 1. 

1
2
log 1 + 𝛽𝜌∣ℎ0 ∣2 + (1 − 𝛽)𝜌∣ℎ1 ∣2 ,
1−𝛽 𝜌 𝜌
𝐼 OSS
= if 𝑇𝛽3 < ∣ℎ2 ∣2 < ∣ℎ1 ∣2 or ∣ℎ1 ∣2 < 𝑇𝛽3 < ∣ℎ2 ∣2 ,
(17) 
 ( )




1
log 1 + 𝛽𝜌∣ℎ0 ∣2 + (1 − 𝛽)𝜌∣ℎ2 ∣2 ,
Proof: Using a similar approach as in Corollary 1 yields 
⎩ 2
if 𝑇𝛽3 < ∣ℎ1 ∣2 < ∣ℎ2 ∣2 or ∣ℎ2 ∣2 < 𝑇𝛽3 < ∣ℎ1 ∣2 .
the result. (20)
Proof: From the opportunistic source selection criterion,
C. OSS Protocol the mutual information 𝐼 OSS of the whole network in the OSS
The fundamental idea of the OSS protocol is that one protocol is given by
{ }
end-source out of two is opportunistically selected based on 𝐼 OSS = max 𝐼1OSS , 𝐼2OSS . (21)
the instantaneous channel conditions such that the multiuser
10 The OSS protocol is essentially a unidirectional DF protocol at any given
diversity of the time-varying channels is exploited. The traffic
time instant, although it can be seen as a bidirectional protocol in a long run.
flow in the OSS protocol is essentially unidirectional in each For this reason, we adopt the same assumption on the codebook in the OSS
time instant because only one of the two end-sources is protocol as in the classic unidirectional DF networks; that is, a same codebook
selected and the selected source transmits information in one is used at the selected source and the relay (see [1, eq. (15)], [19, eq. (19)]
and numerous other publications). Therefore, the achievable rate in the OSS
directional only.9 Let us first consider the case when S1 is protocol involves the sum of two SNRs inside a log-function in both the
selected as the transmitting source. Then the signal transmitted second cases of (18) and (19). Note that the second cases of (18) and (19)
from S1 is received at S2 through two paths: the path from S1 only address the case when the relay can successfully decode, and hence, the
forwarded symbol by the relay is the same as the transmitted symbol by the
directly to S2 and the relay path from S1 to S2 via R. Note that source, due to the same codebook assumption. Therefore, the optimum ML
detection at the intended destination is in a single-symbol decodable manner.
8 Note that 𝑇 > 2𝑇 follows from the definitions of 𝑇 and 𝑇 , and hence
3 2 2 3 We admit that using independent codebooks at the selected source and the
2/3 < 𝛽0 < 1 follows correspondingly. relay can increase the achievable rate (see footnote 4). However, the detection
9 In a long run, however, the traffic flow in the OSS protocol is bidirectional complexity would be significantly higher because the decision metric would
over time-varying channels because the two end-sources are opportunistically be coupled with two different symbols. For these reasons, a same codebook
selected as the transmitting source. scheme is adopted for the OSS protocol.
2688 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

TDBC
Substituting (18) and (19) into (21) eventually yields (20). probability 𝑃out (𝑅, 𝜌) of (8) is monotonically decreasing in
Using the mutual information in Lemma 1, we now study 𝛽 for 0 < 𝛽 ≤ 2/3. Therefore, 2/3 ≤ 𝛽0 ≤ 1.
the outage probability performance of the OSS protocol. The Note that it is extremely hard to determine the exact value
OSS protocol is in outage when the target data rate 𝑅 exceeds 𝛽0 for the TDBC protocol. Although it is easy to show that
the maximum achievable rate 𝐼 OSS , that is, TDBC
𝑃out (𝑅, 𝜌) of (8) is monotonically decreasing in 𝛽 for 0 <
OSS
( ) 𝛽 ≤ 2/3, it is extremely hard to show the monotonicity in
𝑃out (𝑅, 𝜌) := Pr 𝐼 OSS < 𝑅 . (22) 2/3 < 𝛽 < 1 because the outage expression for this range is
In the following theorem, we derive an exact closed-form too involved (note that 𝜃 in the asymptotic outage probability
outage probability expression for the OSS protocol. expression is a function of 𝛽 as well). Hence, we could not
Theorem 3: For the OSS protocol, a closed-form exact find the exact value of 𝛽0 for the TDBC protocol. Instead, we
outage probability expression is given by conclude that the asymptotic optimal power ratio must be in
( 𝜆0 𝑇 3
)( 𝜆1 𝑇 3
)( 𝜆2 𝑇 3
) the range 2/3 ≤ 𝛽0 ≤ 1.
OSS
𝑃out (𝑅, 𝜌) = 1 − 𝑒− 𝛽 1 − 𝑒− 𝛽 1 − 𝑒− 𝛽 Interesting insights into the bidirectional protocols can be
easily obtained from the above lemma: 𝛽0 = 1 in the OSS
+ 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ) + 𝐺(𝛽, 𝜆0 , 𝜆2 , 𝜆1 , 𝑇3 ). (23)
indicates that the relay is not needed when the SNR goes to
The function 𝐺(𝑥, 𝑦0 , 𝑦1 , 𝑦2 , 𝑧), 0 < 𝑥 < 1, 𝑧 > 0, 𝑦𝑖 > 0, infinity; however, 𝛽0 = 2/3 in the PNC protocol indicates
𝑖 = 0, 1, 2, is defined at the top of the next page, where Case the relay is needed even if the SNR goes to infinity; and
𝑦0
1 represents 0 < 𝑥 ≤ 1/2 and 𝑥 ∕= 𝑦 +𝑦 ; Case 2: 1/2 < 2/3 ≤ 𝛽0 ≤ 1 indicates that the relay might be needed in the
0 1
𝑦0 𝑦0 𝑦0 TDBC protocol when SNR goes to infinity. All the insights
𝑥 < 1, 𝑥 ∕= 𝑦 +𝑦 , and 𝑥 ∕= 𝑦 +𝑦 +𝑦 ; Case 3: 𝑥 = 𝑦 +𝑦
0 1 0
𝑦0 1 2 0 1
will be further confirmed by numerical results in Section V.
and 𝑦0 ≤ 𝑦1 ; Case 4: 𝑥 = 𝑦 +𝑦 and 𝑦0 > 𝑦1 ; Case 5:
𝑦 0 1 For example, Fig. 6 indicates that the optimal relay location
𝑥 = 𝑦 +𝑦0 +𝑦 and 𝑦0 > 𝑦1 + 𝑦2 .11 is 0.5. Also, Figs. 7 and 8 indicate that the optimal power
0 1 2
Proof: See Appendix D. ratio is around 2/3 for the PNC protocol, 0.8 for the TDBC
To provide more insight into the OSS protocol, we study protocol, and 1 for the OSS protocol.
the asymptotic high-SNR performance of the OSS protocol in In the following, we give some intuitive explanations on
what follows. the asymptotic optimal power ratios for the three protocols.
Corollary 3: The OSS protocol achieves full diversity order For the OSS protocol, when the SNR is substantially high,
two, and the asymptotic outage probability is given by the direct link alone can accomplish the task of transmitting
⎧ 2
( )


𝜆0 (𝜆1 +𝜆2 ) 𝐿3
+𝒪 1
, 0 < 𝛽 ≤ 12 , information from the selected source to the intended des-
2𝛽(1−𝛽) 𝜌2 𝜌3
OSS
𝑃out (𝑅, 𝜌) = ( ) tination, without any help from the relay. This is why the

⎩ 2
𝜆0 (𝜆1 +𝜆2 )(3𝛽−1) 𝐿3 1 1 relay is not needed in the OSS protocol when SNR goes to
2𝛽 3 𝜌2
+𝒪 𝜌3
, 2
< 𝛽 < 1.
(25) infinity. In the PNC protocol, however, even if the SNR is
substantially high, without the relay’s help, the information
Proof: Using a similar approach as in Corollary 1 yields cannot be exchanged between the two end-sources because the
the result. relay is the only medium that helps to forward the information
to both end-sources. This is why the relay is always needed
D. Asymptotic Optimum Power Allocation in the PNC protocol even if the SNR goes to infinity. For
To gain more insight into the three protocols, we investigate the TDBC protocol, the information can be exchanged via
the optimum power allocation in this subsection. For analytical two links: the direct link and the relay-link. Also, the relay
tractability, we choose the asymptotic outage probability as a can simultaneously serve the two end-sources through the BC
target function, and hence, the optimum power allocation aims channel. Compared to the OSS protocol where the relay only
to minimize the asymptotic outage probability by optimally serves one end-source in each time instant, the role of the
allocating power to each terminal in a bidirectional network. relay in the TDBC protocol is more important than in the
Recall that 𝛽 is the ratio of 𝐸𝑠 to the total power 𝐸. OSS protocol. Therefore, when the SNR is substantially high,
The asymptotic optimal power ratio 𝛽0 is presented in the it should be beneficial for the TDBC protocol to exploit both
following lemma. the direct link and the relay-link so as to accomplish the
Lemma 2: For the PNC protocol, 𝛽0 = 2/3; For the OSS information exchange. Hence, as the SNR goes to infinity,
protocol, 𝛽0 = 1; For the TDBC protocol, 2/3 ≤ 𝛽0 ≤ 1. the TDBC protocol still needs the relay.
Proof: For the PNC protocol, it can be shown that the
PNC
asymptotic outage probability 𝑃out (𝑅, 𝜌) of (17) is strictly E. Asymptotic Optimum Relay Location
decreasing in 𝛽 for 0 < 𝛽 ≤ 2/3 and increasing in 𝛽 for 2/3 < In this subsection, we study the asymptotic optimum relay
𝛽 < 1. Therefore, 𝛽0 = 2/3 yields the optimum performance. location for the bidirectional protocols. We assume that the
For the OSS protocol, we can show that the asymptotic outage two end-sources and the relay are located in a straight line.12
OSS
probability 𝑃out (𝑅, 𝜌) of (25) is monotonically decreasing We normalize the distance between S1 and S2 as 1 and use
in 𝛽 for all 0 < 𝛽 < 1, and hence, 𝛽0 = 1. For the 𝑑 (0 < 𝑑 < 1) to denote the relay location (i.e., 𝑑 is
TDBC protocol, we can only show that the asymptotic outage the distance between R and S1 ). The channel variances are
11 The 𝑦0
case when 𝑥 = 𝑦0 +𝑦1 +𝑦2
and 𝑦0 ≤ 𝑦1 + 𝑦2 is included in Case 12 This assumption has also been widely adopted in numerous publications
1. such as [16], [20], [25], [26], etc.
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2689

𝐺(𝑥, 𝑦0 , 𝑦1 , 𝑦2 , 𝑧)
⎧ 𝑦1 𝑦2
𝑒−( 1−𝑥 + 𝑥 )𝑧 ,
𝑦 𝑧 (𝑦 +𝑦 )𝑧
− 2 𝑥𝑦1 − 0 𝑥2 (1−𝑥)𝑦0

 𝑒 𝑥 + (1−𝑥)𝑦0 −𝑥𝑦1 𝑒 − (1−𝑥)𝑦 −𝑥𝑦 for Case 1,

 0 1



 𝑦 𝑧 (𝑦 +𝑦 )𝑧 𝑦 𝑦 (𝑦 +𝑦 )𝑧
 𝑒−( 1−𝑥 + 𝑥 )𝑧 − 𝑦 +𝑦
2 𝑥𝑦 0 2 (1−𝑥)𝑦0 1 2 𝑦1 1 2

 𝑒− 𝑥 + (1−𝑥)𝑦 1−𝑥𝑦 𝑒− 𝑥 − (1−𝑥)𝑦 −𝑥𝑦 𝑒− 𝑥

 0 1 0 1 1 2


(𝑦 +𝑦 )𝑧
𝑒( 𝑥 𝑦0 −𝑦1 −𝑦2 ) 𝑥 + (𝑦 +𝑦 ) (1−𝑥)𝑦 0−𝑥(𝑦
𝑥𝑦1 1−2𝑥 𝑧 (1−𝑥)𝑦 𝑦1 1 2

 − (1−𝑥)𝑦 −𝑥(𝑦 𝑒− 1−𝑥 , for Case 2,

 0 1
+𝑦 2
)
2 (
+𝑦 2 )
)


1 0 1


 𝑦 𝑧 (𝑦 +𝑦 )𝑧
⎨ 𝑒− 2𝑥 − 𝑦1 𝑧 𝑒− 0 𝑥 2 − 𝑒−( 1−𝑥
𝑦1 𝑦
+ 𝑥2 )𝑧
, for Case 3,
1−𝑥
:=

 𝑦 𝑧
𝑦1 𝑧 − (𝑦0 +𝑦2 )𝑧 𝑦 𝑦 (𝑦 +𝑦 )𝑧
− 𝑒−( 1−𝑥 + 𝑥 )𝑧 − 𝑦 +𝑦 𝑒( 𝑥 𝑦0 −𝑦1 −𝑦2 ) 𝑥
 2 1 2 𝑦1 1 2 𝑥𝑦1 1−2𝑥 𝑧

 𝑒− 𝑥 − 1−𝑥 𝑒 𝑥 𝑒− 𝑥 − (1−𝑥)𝑦 −𝑥(𝑦 +𝑦 )

 1 2 0 1 2

 (1−𝑥)𝑦 𝑦1 (𝑦 +𝑦 )𝑧
1 2

 + (𝑦 +𝑦 ) (1−𝑥)𝑦 0−𝑥(𝑦 𝑒− 1−𝑥 , for Case 4,

 ( +𝑦 ) )


1 2 0 1 2
( (𝑦 +𝑦 )𝑧 )



 𝑦
− 2𝑥
𝑧
(1−𝑥)𝑦0 𝑦
−( 1−𝑥
𝑦
1 + 2 𝑧
) 𝑥𝑦1 (𝑦 +𝑦 )𝑧
− 0 𝑥2 𝑦1 − 11−𝑥2
(𝑦 +𝑦 )𝑧
− 1 𝑥2

 𝑒 − (1−𝑥)𝑦 −𝑥𝑦 𝑒 𝑥 + (1−𝑥)𝑦 −𝑥𝑦 𝑒 + 𝑦 +𝑦 𝑒 −𝑒

 0 1 0 1 1 2

⎩ (2𝑥−1)𝑦1 𝑧 − 𝑦0𝑥𝑧
+ 𝑥(1−𝑥) 𝑒 , for Case 5.
(24)

modeled as Ω0 = 1, Ω1 = 𝑑−4 , and Ω2 = (1 − 𝑑)−4 [27]. Applying the above result, we obtain the performance gain of
The asymptotic optimal relay location 𝑑0 is that value of the OSS protocol over the TDBC protocol:
𝑑 which minimizes the asymptotic outage probability. Note
that although the straight-line model is a simple geometric 𝐺OSS∣TDBC
model for the three-node relay networks, our discussion on ⎧ ( )
8(1−𝛽) 𝐿21
the optimal relay location can be generalized to arbitrary 
 5 log 10 , 0 < 𝛽 ≤ 1/2,


𝛽 𝐿32

geometric models. 
 ( )

 8𝛽 𝐿
2

Substituting 𝜆0 = 1, 𝜆1 = 𝑑4 , and 𝜆2 = (1 − 𝑑)4 , we 


 5 log10 3𝛽−1 𝐿123 , 1/2 < 𝛽 ≤ 2/3,


can show that 𝑃outOSS
(𝑅, 𝜌) is monotonically decreasing in 𝑑 
 (( )
⎨ 2(𝜆0 +𝜆1 +𝜆2 )2 −2𝜆20 (26)
(𝜆1 +𝜆2 )2
for 0 < 𝑑 ≤ 0.5 and increasing in 𝑑 for 0.5 < 𝑑 < 1. = 5 log10 𝛽 2 − 2(1−𝛽) 2 𝐿21
Therefore, the optimal relay location for the OSS protocol 


 − 2𝜆0 (𝜆1 +𝜆2 )(1+𝐿 1 )(1−𝜃+ln 𝜃)

is 𝑑0 = 0.5. Similarly, we can show that the optimal relay 


 𝛽(1−𝛽) ( ))


location for both the TDBC and PNC protocols is 𝑑0 = 0.5 
 +
(𝜆1 +𝜆2 )2 (1+𝐿1 )(1−𝜃) 𝜃(1+𝐿1 )−(1−𝐿1 )

 2(1−𝛽)2
as well. These results will be further confirmed by numerical 
 ( )
⎩ −5 log 𝜆0 (𝜆1 +𝜆2 )(3𝛽−1)𝐿23
, 2/3 < 𝛽 < 1.
results (see Fig. 6). 10 2𝛽 3

Note that the optimum relay location 𝑑0 = 0.5 suggests As an example, we consider the case when 𝛽 = 2/3 and
that the relay should be placed in the middle between the 𝑅 = 1 bps/Hz; then using the second case of (26) we obtain
two end-sources when the SNR is sufficiently high so that it 𝐺OSS∣TDBC ≈ 1.5 dB. This performance gain will be further
can optimally serve the demand of bidirectional information confirmed by the numerical results of Figs. 4 and 5.
exchange.
IV. D IVERSITY-M ULTIPLEXING T RADEOFF
F. Asymptotic Performance Gain of OSS over TDBC In order to compare the reliability as well as the spectral
efficiency of the three bidirectional protocols, we consider the
In this subsection, we study the performance gain of the diversity gain and the multiplexing gain simultaneously. The
OSS protocol over the TDBC protocol in the high SNR diversity-multiplexing tradeoff which is a fundamental tradeoff
regime. Specifically, we compare the required SNRs for the for any communication systems provides a whole view on
two protocols to achieve the same outage, and thus, the the diversity gain and multiplexing gain [21]. Recently, the
performance gain is expressed in terms of the difference in DMT performance of multiple-antenna systems has also been
the required SNRs in dB scale. Consider any two protocols studied in the finite SNR regime [22]. In this section, we
of diversity order two and the asymptotic outage probabilities study both the finite-SNR DMT and the infinite-SNR DMT
are given by 𝑃𝑎 = 𝑎/𝜌2𝑎 and 𝑃𝑏 = 𝑏/𝜌2𝑏 , where 𝜌𝑎 and 𝜌𝑏 performance for the three bidirectional protocols.
are the required SNRs. To study the performance gain, we We first study the finite-SNR DMT. The finite-SNR DMT
evaluate the outage probabilities in a log-log scale: log10 𝑃𝑎 = for a single-antenna system is defined as [22]
log10 𝑎 − 2 log10 𝜌𝑎 and log10 𝑃𝑏 = log10 𝑏 − 2 log10 𝜌𝑏 . Then
∂ ln 𝑃out (𝑟 log(1 + 𝜌), 𝜌)
setting 𝑃𝑎 = 𝑃𝑏 we obtain the performance gain of protocol 𝐷(𝑟, 𝜌) := − , (27)
𝑎 over protocol 𝑏: 𝐺𝑎∣𝑏 = 10 log10 𝜌𝜌𝑎𝑏 = 5 log10 𝑎𝑏 . Note that ∂ ln 𝜌
this performance gain actually captures the difference in the where 𝑟 is the multiplexing gain, and 𝐷(𝑟, 𝜌) is the corre-
coding gains for two protocols of the same diversity gain [28]. sponding diversity gain at a finite SNR 𝜌. With the derived
2690 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

1
10
TDBC protocol for 0 < 𝛽 ≤ 2/3 is given at the top of the
1.5𝑟
Exact outage for PNC, eq. (16)
Asymptotic outage for PNC, eq. (17) next page, where 𝜉 = (1+𝜌)𝜌𝛽 −1 .
0
10
Simulated outage for PNC
Exact outage for TDBC, eq. (6) Proof: Substituting (6) for the case 0 < 𝛽 ≤ 2/3 into
Asymptotic outage for TDBC, eq. (8)
Simulated outage for TDBC (27) and taking derivative yields (28).
−1
Exact outage for OSS, eq. (23)
10
Asymptotic outage for OSS, eq. (25) We now consider the infinite-SNR DMT. By definition, the
Simulated outage for OSS
infinite-SNR DMT can be obtained from the finite-SNR DMT
Outage probability

−2
10
by letting 𝜌 go to infinity [17], [21], [22]. However, as the
−3
10
finite-SNR DMT expressions are very involved, it becomes
difficult to take the limit. As an alternative approach, we
−4
10 use the asymptotic high-SNR outage probability expressions
to derive the infinite-SNR DMT, as given in the following
−5
10 theorem.
Theorem 5: The infinite-SNR DMT 𝐷 ˜ TDBC (𝑟) of the
−6
10
0 5 10 15 20 25 30
˜ PNC
TDBC protocol, 𝐷 (𝑟) of the PNC protocol, and 𝐷 ˜ OSS (𝑟)
Average SNR: ρ (dB)
of the OSS protocol are given as follows:
˜ TDBC (𝑟) = 2 − 3𝑟,
𝐷 (29)
Fig. 4. Outage probability against average SNR 𝜌 for the TDBC, PNC, and
OSS protocols with 𝑑 = 0.3, 𝑅 = 1 bps/Hz, 𝐸𝑠 = 2𝐸/3, and 𝐸𝑟 = 𝐸/3. ˜ PNC (𝑟) = 1 − 𝑟,
𝐷 (30)
𝐷˜ OSS (𝑟) = 2 − 4𝑟. (31)
1
10 Proof: We first derive the infinite-SNR DMT for the
Exact outage for PNC, eq. (16)
Asymptotic outage for PNC, eq. (17) TDBC protocol. We rewrite the asymptotic outage probability
Simulated outage for PNC
of the TDBC protocol irrespective of the values of 𝛽 as
0
10
Exact outage for TDBC, eq. (6)
Asymptotic outage for TDBC, eq. (8)
Simulated outage for TDBC follows:
−1
10 Exact outage for OSS, eq. (23) ( )
Asymptotic outage for OSS, eq. (25)
TDBC (𝑎𝐿21 + 𝑏𝐿1 + 𝑐) 1
Simulated outage for OSS
𝑃out (𝑅, 𝜌) = + 𝒪 , (32)
𝜌2 𝜌3
Outage probability

−2
10

−3
where 𝑎 is a non-zero constant, 𝑏 and 𝑐 are constants which
10
can possibly take zero values. By the definition of the infinite-
−4
10
SNR DMT, we substitute 𝑅 = 𝑟 log 𝜌 into (32) and let 𝜌
TDBC
go to infinity. Then 𝑃out (𝑅, 𝜌) is accurately approximated
−5
10 as 𝑃out (𝑅, 𝜌) ≈ 𝑎𝜌3𝑟−2 by ignoring higher order terms.
TDBC

Therefore, 𝐷˜ TDBC (𝑟) = 2 − 3𝑟. Also, the infinite-SNR DMT


−6
10
0 5 10 15 20 25 30
for the PNC and OSS protocols can be derived in the same
Average SNR: ρ (dB)
manner as the TDBC protocol by using the asymptotic outage
probability expressions we have derived earlier.
Fig. 5. Outage probability against average SNR 𝜌 for the TDBC, PNC, and
OSS protocols with 𝑑 = 0.5, 𝑅 = 1 bps/Hz, 𝐸𝑠 = 2𝐸/3, and 𝐸𝑟 = 𝐸/3.
V. N UMERICAL R ESULTS
This section gives extensive numerical results on the out-
age probability performance of the TDBC, PNC, and OSS
outage probability expressions, it is straightforward to derive protocols. We adopt the straight-line model for the three-
the finite-SNR DMT for the bidirectional protocols by using node cooperative networks as described in Section III-E. Also,
(27). Specifically, for the TDBC with 0 < 𝛽 ≤ 2/3, PNC, the channel parameters are modeled as 𝜆0 = 1, 𝜆1 = 𝑑4 ,
and OSS protocols, exact closed-form expressions of the and 𝜆2 = (1 − 𝑑)4 . We note that the derived exact outage
finite-SNR DMT can be easily derived by substituting their probability expressions of (6), (16), and (23) are invariant
respective outage probability expressions of (6), (16), and (23) under interchanging of 𝜆1 and 𝜆2 . Therefore, it is sufficient
into (27). For the TDBC protocol with 2/3 < 𝛽 < 1, however, and no loss of generality to consider the case 0 < 𝑑 ≤ 0.5
the finite-SNR DMT expression is more complex than other only.
cases because taking derivative of Ψ(𝑎, 𝑏, 𝑐, 𝜃) yields another Firstly, we fix the data rate as 𝑅 = 1 bps/Hz and compare
integral whose integral region is between 𝜃 and 1, as in the reliability of the TDBC, PNC, and OSS protocols. Each
Ψ(𝑎, 𝑏, 𝑐, 𝜃). Unfortunately, it is extremely hard to solve the protocol is compared under the condition that the total average
resulting integral in closed-form. However, the integral can be power of the whole network is 𝐸. As there are several factors
easily computed with similar numerical approaches used for such as the average SNR 𝜌, the relay location 𝑑, and the power
Ψ(𝑎, 𝑏, 𝑐, 𝜃). In this paper, we do not present all these derived ratio 𝛽 that can affect the outage probability, we consider
finite-SNR DMT expressions due to a length limit. Instead, as each of the factors separately by fixing the other two factors.
an example, we present the exact finite-SNR DMT expression Specifically, we consider the following three cases:
of the TDBC protocol for 0 < 𝛽 ≤ 2/3 in the following 1) We vary the average SNR 𝜌 while fixing the relay
theorem. location 𝑑 and setting 𝐸𝑠 = 2𝐸/3 and 𝐸𝑟 = 𝐸/3 for
Theorem 4: The finite-SNR DMT 𝐷TDBC (𝑟, 𝜌) of the each protocol. Figs. 4 and 5 show the outage probability
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2691

( )( )
TDBC (3𝑟𝜌 − 2)𝜉 + 3𝑟/𝛽 (𝜆0 + 𝜆1 + 𝜆2 )𝑒−2(𝜆0 +𝜆1 +𝜆2 )𝜉 − 𝜆0 𝑒−2𝜆0 𝜉 − (𝜆1 + 𝜆2 )𝑒−2(𝜆1 +𝜆2 )𝜉
𝐷 (𝑟, 𝜌) = . (28)
1 + 𝑒−2(𝜆0 +𝜆1 +𝜆2 )𝜉 − 𝑒−2𝜆0 𝜉 − 𝑒−2(𝜆1 +𝜆2 )𝜉

TABLE I
against the average SNR 𝜌 for 𝑑 = 0.3 and 𝑑 = 0.5 C OMPARISON OF THE BIDIRECTIONAL PROTOCOLS WHEN THE OUTAGE
respectively. It is easy to see that the simulated outage PROBABILITY IS EQUAL TO 10−2 , WHERE THE NOTATION 𝐴 ≻ 𝐵 MEANS
probability perfectly matches with the exact closed-form 𝐴 OUTPERFORMS 𝐵 IN TERMS OF OUTAGE PERFORMANCE .
outage probability expressions. Also, asymptotic outage
Low Rate, e.g., 𝑅 = 1 bps/Hz OSS≻TDBC≻PNC
probability approaches the exact outage probability in Medium Rate, e.g., 𝑅 = 3 bps/Hz TDBC≻OSS≻PNC
the high-SNR regime (e.g., 𝜌 ≥ 20 dB). Moreover, High Rate, e.g., 𝑅 = 6 bps/Hz PNC≻TDBC≻OSS
the outage probability curve of the OSS protocol lies
below the outage probability curves of the PNC and
TDBC protocols, which indicates that the OSS protocol
All these observations can be explained by the the concept of
is more reliable. Furthermore, the performance gain of
the diversity-multiplexing tradeoff, which is discussed in the
the OSS protocol over the TDBC protocol at the high-
following.
SNR regime is around 1.5 dB, which is consistent with
The DMT curves in both the finite and infinite SNR regimes
our theoretical analysis in Section III-F.
are presented in Fig. 10. Note that the DMT curves are
2) We vary the relay location 𝑑 while fixing the average
obtained based on the derived analytical expressions (rather
SNR 𝜌 and setting 𝐸𝑠 = 2𝐸/3 and 𝐸𝑟 = 𝐸/3 for each
than numerical simulations) in Section IV. It is obvious to
protocol. Fig. 6 shows the outage probability against
see that the finite-SNR DMT curves converge to the infinite-
the relay location 𝑑 when 𝜌 = 20 dB. It is demonstrated
SNR DMT curves, as expected by definitions. Moreover, the
that the outage probabilities of the three protocols get
OSS and TDBC protocols achieve higher maximal diversity
their minimum when 𝑑 = 0.5. This indicates that the
gain than the PNC protocol. On the other hand, the PNC
optimum relay location is the center between the two
protocol achieves higher maximal multiplexing gain than the
end-sources, which is in accordance with the theoretical
TDBC and OSS protocols. For low data rate, the effect of the
results in Section III-E.
diversity gain dominates the effect of the multiplexing gain,
3) We vary the power ratio 𝛽 while fixing the relay location
and hence, TDBC and OSS outperform the PNC protocol. For
𝑑 and the average SNR 𝜌. For each protocol, the total
high data rate and finite SNR, the effect of the multiplexing
power of the whole network is fixed to 𝐸. We compare
gain dominates the effect of the diversity gain, and therefore,
the outage probabilities by varying 𝛽 for each protocol.
PNC outperforms other protocols. However, for the asymptotic
Figs. 7 and 8 show the outage probability with 𝜌 =
high-SNR regime, the effect of the diversity gain dominates
20 dB against various 𝛽 for 𝑑 = 0.3 and 𝑑 = 0.5,
the effect of the multiplexing gain, irrespective of the data rate.
respectively. We observe that the optimal power ratio
Due to this reason, the TDBC and OSS protocols outperform
is around 0.8 for the TDBC protocol, 2/3 for the PNC
the PNC protocol in the asymptotic high-SNR regime.
protocol, and 1 for the OSS protocol, which is consistent
Remark: For two protocols A and B, even if A always
with the analytical results in Section III-D.
outperforms B in the sense of DMT, it does not necessarily
We have demonstrated that the OSS protocol is more
guarantee that A always has better performance than B in
reliable than other protocols for low data rate, e.g., 𝑅 = 1
terms of outage probability. For instance, even if we suppose
bps/Hz (see Figs. 4–8). In the following, we investigate the
A always has a higher multiplexing gain than B with the same
reliability for higher data rates. In Fig. 9, we consider three
diversity gain, A can be inferior to B in outage probability for
data rates: 𝑅 = 1 bps/Hz, 𝑅 = 3 bps/Hz, and 𝑅 = 6 bps/Hz.
low data rates. In this paper, we demonstrated that OSS has
We fix the relay location as 𝑑 = 0.3 and set 𝐸𝑠 = 2𝐸/3
a lower multiplexing gain than TDBC while they have the
and 𝐸𝑟 = 𝐸/3 for all three protocols as before. We observe
same diversity gain; however, for low data rates, OSS can
that the OSS protocol has the highest reliability in the whole
achieve better outage probability performance than TDBC, as
SNR range when data rate is very low, e.g., 𝑅 = 1 bps/Hz,
demonstrated in Figs. 4–8, and 9, although TDBC is eventually
as demonstrated by Fig. 9. This is because only the OSS
superior to OSS in outage probability for high data rates.13
protocol can exploit the multiuser diversity, as discussed in
Section I. However, as the data rate increases, both TDBC and
VI. C ONCLUSIONS
PNC can outperform the OSS. As an example, we consider
the case when outage probability equals 10−2 . If 𝑅 = 1 In this paper, we have analyzed and compared the perfor-
bps/Hz, the OSS protocol requires the lowest SNR to achieve mance of three DF-based bidirectional protocols including the
the same performance as other protocols; If 𝑅 = 3 bps/Hz, TDBC, PNC, and OSS protocols. The outage probabilities of
the TDBC becomes the best protocol which achieves the 13 A similar observation was made in [24], where the non-orthogonal AF
same performance as others at the lowest SNR; If 𝑅 = 6 (NAF) protocol and the distributed Alamouti coding protocol were compared.
bps/Hz, the best protocol is the PNC protocol. A detailed It was demonstrated that even though the NAF protocol always has a higher
comparison is presented in Table I. We also observe that the multiplexing gain than the distributed Alamouti coding with the same diversity
gain, the NAF protocol has worse outage probability performance at low data
TDBC and OSS protocols outperform the PNC protocol at rates, although the NAF is eventually better than the distributed Alamouti
the asymptotic high-SNR regime, irrespective of the data rate. coding in high data rates.
2692 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

−1 −1
10 10
Exact outage for PNC, eq. (16) Exact outage for PNC, eq. (16)
Simulated outage for PNC Simulated outage for PNC
Exact outage for TDBC, eq. (6) Exact outage for TDBC, eq. (6)
Simulated outage for TDBC Simulated outage for TDBC
Exact outage for OSS, eq. (23) Exact outage for OSS, eq. (23)
Simulated outage for OSS Simulated outage for OSS

−2 −2
10 10
Outage probability

Outage probability
−3 −3
10 10

−4 −4
10 10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Relay location: d β

Fig. 6. Outage probability of the TDBC, PNC, and OSS protocols for various Fig. 8. Outage probability of the TDBC, PNC, and OSS protocols for various
𝑑 while 𝜌 = 20 dB, 𝑅 = 1 bps/Hz, 𝐸𝑠 = 2𝐸/3, and 𝐸𝑟 = 𝐸/3. 𝛽 while 𝜌 = 20 dB, 𝑑 = 0.5, and 𝑅 = 1 bps/Hz.

0
−1 10
10
Exact outage for PNC, eq. (16)
Simulated outage for PNC
Exact outage for TDBC, eq. (6)
Simulated outage for TDBC −1 R = 3 bps/Hz
10
Exact outage for OSS, eq. (23)
Simulated outage for OSS

−2
10 R = 1 bps/Hz R = 6 bps/Hz
−2
Outage Probability

10
Outage probability

−3
10

−3
10

Exact outage for PNC, eq. (16)


−4
10 Simulated outage for PNC
Exact outage for TDBC, eq. (6)
Simulated outage for TDBC
Exact outage for OSS, eq. (23)
Simulated outage for OSS
−5
−4 10
10 0 10 20 30 40 50 60
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Average SNR: ρ (dB)
β

Fig. 7. Outage probability of the TDBC, PNC, and OSS protocols for various Fig. 9. Outage probability of the TDBC, PNC, and OSS protocols for various
𝛽 while 𝜌 = 20 dB, 𝑑 = 0.3, and 𝑅 = 1 bps/Hz. rates with 𝑑 = 0.3, 𝐸𝑠 = 2𝐸/3, and 𝐸𝑟 = 𝐸/3.

that the OSS protocol is the best choice for applications with
all the three protocols are extensively studied and compared. very low data rate (e.g., 𝑅 = 1 bps/Hz) because it can exploit
Specifically, we derived exact closed-form outage probability the multiuser diversity.
expressions for the PNC and OSS protocols and an exact In this paper, we have focused on the OSS protocol with
outage probability in a one-integral form for the TDBC proto- a same (common) codebook at the relay and end-sources. It
col. We also derived asymptotic outage probability expressions is obvious that using independent codebooks at the relay and
for all protocols and demonstrated that the TDBC and OSS end-sources can increase the achievable rate, and hence, the
protocols achieve full diversity order two and the PNC pro- outage performance of the OSS protocol will be enhanced.
tocol achieves diversity order one. Based on the asymptotic On the other hand, the independent codebooks will inevitably
outage probabilities, further analytical results including the increase the decoding complexity at the destination, as ex-
asymptotic optimal power allocation, relay location, and the plained in footnote 10. Therefore, it is an interesting issue to
performance gain of the OSS protocol over the TDBC protocol study an independent codebooks scheme for the OSS protocol
were obtained. Finally, we studied the finite-SNR and infinite- as a future work.
SNR DMT performance and demonstrated that the PNC pro-
tocol is able to achieve the highest multiplexing gain, whereas
the TDBC and OSS protocols are able to achieve the highest A PPENDIX A
diversity gain. Therefore, for applications with high data rate, P ROOF OF T HEOREM 1
2
the PNC protocol is the best choice. On the other hand, the Let 𝑋𝑖 = ∣ℎ𝑖 ∣ , 𝑖 = 0, 1, 2. Then 𝑋𝑖 ’s are independent ex-
OSS and TDBC protocols can be better choices than the PNC ponential random variables with parameters 𝜆𝑖 , 𝑖 = 0, 1, 2. Us-
TDBC
protocol when data rate is low. Moreover, we demonstrated ing Total Probability Theorem, the outage probability 𝑃out
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2693

( ) ( )
TDBC 2𝑇1 2𝑇1 2𝑇1 2𝑇1 2𝑇1 2𝑇1
𝑃out = 1 − Pr 𝑋0 > , 𝑋1 < , 𝑋2 < − Pr 𝑋0 > , 𝑋1 < , 𝑋2 >
𝛽 𝛽 𝛽 𝛽 𝛽 𝛽
( ) ( )
2𝑇1 2𝑇1 2𝑇1 𝛽 𝛽(1 − 𝛽)𝜌 2𝑇1
− Pr 𝑋0 > , 𝑋1 > , 𝑋2 < − Pr 𝑋0 + (1 − 𝛽)𝑋2 + 𝑋0 𝑋2 > 𝑇 1 , 𝑋1 > 𝑋2 >
𝛽 𝛽 𝛽 2 2 𝛽
( )
𝛽 𝛽(1 − 𝛽)𝜌 2𝑇1
− Pr 𝑋0 + (1 − 𝛽)𝑋1 + 𝑋0 𝑋1 > 𝑇 1 , 𝑋2 > 𝑋1 >
2 2 𝛽
=: 1 − 𝐴1 − 𝐴2 − 𝐴3 − 𝐴4 − 𝐴5 . (A.1)

2
Therefore,
TDBC (ρ=10 dB) ⎧ 2(𝜆1 +𝜆2 )𝑇1
1.8 OSS (ρ=10 dB) ⎨ 𝑒− 𝛽 , 0 < 𝛽 ≤ 2/3,
PNC (ρ=10 dB)
TDBC (ρ=30 dB) 𝐴4 + 𝐴5 = (A.7)
1.6 ⎩ (𝜆1 +𝜆2 )𝑇1
𝑒−
OSS (ρ=30 dB)
PNC (ρ=30 dB) 1−𝛽 + 𝜓, 2/3 < 𝛽 < 1,
1.4 TDBC (ρ→∞)

1.2
OSS (ρ→∞)
PNC (ρ→∞)
where

Diveristy Gain

𝑇1
1−𝛽 2𝜆0 𝑇1 −(1−𝛽)𝑥2

1
𝜓 := (𝜆1 + 𝜆2 ) 𝑒−(𝜆1 +𝜆2 )𝑥2 𝑒 𝛽 1+(1−𝛽)𝜌𝑥2
𝑑𝑥2 .
2𝑇1
0.8 𝛽
(A.8)
0.6
By some manipulations such as change of variable and some
0.4
normalization techniques, 𝜓 can be written in a more compact
0.2 form as follows:
∫ 1
𝑏
𝑒−(𝑎𝑥+ 𝑥 ) 𝑑𝑥,
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Multiplexing Gain
0.7 0.8 0.9 1 𝜓=𝑐 (A.9)
𝜃
where 𝑎, 𝑏, 𝑐, and 𝜃 are given in Theorem 1. Finally, substi-
Fig. 10. Finite-SNR and infinite-SNR DMT curves of the TDBC, PNC, and tuting (A.2)–(A.4) and (A.7) into (A.1) yields (6).
OSS protocols for 𝑑 = 0.1, 𝐸𝑠 = 2𝐸/3, and 𝐸𝑟 = 𝐸/3.

A PPENDIX B
P ROOF OF C OROLLARY 1
of (5) can be written as (A.1), where 𝐴1 , 𝐴2 , and 𝐴3 can The asymptotic outage probability captures the limiting
be easily derived by using the fact that 𝑋𝑖 ’s are independent behavior of the outage probability as the SNR 𝜌 → ∞,
exponential random variables, where → means goes to some value. Since 𝑇1 = 𝐿1 /𝜌, we
2𝜆0 𝑇1
( 2𝜆1 𝑇1
)( 2𝜆2 𝑇1
) may equivalently study the limit of the outage probability as
𝐴1 = 𝑒− 𝛽 1 − 𝑒− 𝛽 1 − 𝑒− 𝛽 , (A.2) 𝑇1 → 0. In this proof, we use 𝑃out for notational simplicity to
2𝜆0 𝑇1
( 2𝜆1 𝑇1
) 2𝜆2 𝑇1 denote the outage probability of the TDBC protocol. Let us
𝐴2 = 𝑒− 𝛽 1 − 𝑒− 𝛽 𝑒− 𝛽 , (A.3) first consider the case 0 < 𝛽 ≤ 2/3; then 𝑃out is given by the
( ) ′
2𝜆0 𝑇1
𝐴3 = 𝑒− 𝛽 𝑒− 𝛽
2𝜆1 𝑇1
1 − 𝑒− 𝛽
2𝜆2 𝑇1
. (A.4) first case of (6). We observed that both 𝑃out and 𝑃out approach

0 as 𝑇1 → 0, where 𝑃out is the derivative of 𝑃out with respect
Also, 𝐴4 and 𝐴5 are identical in form except that 𝑋1 and to 𝑇1 . To proceed, we apply the L’Hôpital’s rule as follows:
𝑋2 are interchanged. Therefore, 𝐴5 can be obtained by 𝑃out 1
lim = lim 𝑃 ′′ , (B.1)
interchanging 𝜆1 and 𝜆2 in the expression of 𝐴4 . In the 𝑇1 →0 𝑇12 2 𝑇1 →0 out
following, we first derive 𝐴4 . If 0 < 𝛽 ≤ 2/3, with some ′′
where 𝑃out represents the second derivative of 𝑃out with respect
manipulations, 𝐴4 can be simplified as to 𝑇1 . Then with standard limiting analysis, we can show
( )
2𝑇1 𝜆2 2(𝜆1 +𝜆2 )𝑇1 8𝜆0 (𝜆1 + 𝜆1 )
𝐴4 = Pr 𝑋1 > 𝑋2 > = 𝑒− 𝛽 . ′′
lim 𝑃out = . (B.2)
𝛽 𝜆1 + 𝜆2 𝑇1 →0 𝛽2
(A.5) 4𝜆0 (𝜆1 +𝜆1 ) 2 3
If 2/3 < 𝛽 < 1, 𝐴4 can be obtained as follows: Therefore, we can write 𝑃out = 𝛽 2( )𝑇1 + 𝒪(𝑇1 ) or
4𝜆0 (𝜆1 +𝜆2 )𝐿21 1 1
∫ ∞ ( ) equivalently 𝑃out = 𝛽2 𝜌2 + 𝒪 𝜌3 . This yields the
𝑇1 − (1 − 𝛽)𝑥2 first case of (8). For 2/3 < 𝛽 < 1, we write 𝑃out = 𝜙 − 𝜓,
𝐴4 = Pr 𝑋1 > 𝑥2 , 𝑋0 > 𝛽 𝛽(1−𝛽)𝜌 𝑓𝑋2 (𝑥2 )𝑑𝑥2 2(𝜆0 +𝜆1 +𝜆2 )𝑇1 2𝜆0 𝑇1 (𝜆1 +𝜆2 )𝑇1
2 + 𝑥2 where 𝜙 = 1 + 𝑒− − 𝑒− 𝛽 − 𝑒− 1−𝛽
2𝑇1
𝛽 2 𝛽 and
∫ 1−𝛽
𝑇1
2𝜆 𝑇 −(1−𝛽)𝑥2 𝜓 = Ψ(𝑎, 𝑏, 𝑐, 𝜃). Thus, we can separately evaluate 𝜙 and 𝜓.
− 0 1
= 𝜆2 𝑒−(𝜆1 +𝜆2 )𝑥2 𝑒 𝛽 1+(1−𝛽)𝜌𝑥2 𝑑𝑥2 Similar to the case 0 < 𝛽 ≤ 2/3, we can show
2𝑇1
𝛽

𝜆2 𝜙 2(𝜆0 + 𝜆1 + 𝜆2 )2 2𝜆20 (𝜆1 + 𝜆2 )2


+
(𝜆1 +𝜆2 )𝑇1
𝑒− 1−𝛽 . (A.6) lim 2 = − − . (B.3)
𝜆1 + 𝜆2
𝑇1 →0 𝑇1 𝛽2 𝛽2 2(1 − 𝛽)2
2694 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

Also, we can compute 𝜓 ′′ as follows: A. Case 1: 0 < 𝛽 ≤ 2/3


PNC
∫ ∫ ( ) From (C.1) and (C.2), the outage probability 𝑃out for the
1 1
𝑏′ case 0 < 𝛽 ≤ 2/3 can be written as
𝑒 −( ) 𝑑𝑥 − 2𝑐′ 𝑒−(𝑎𝑥+ 𝑥 ) 𝑑𝑥
𝑏 𝑏
𝑎𝑥+ 𝑥
𝜓 ′′ = 𝑐′′ ′
𝑎𝑥+
𝜃 𝜃 𝑥 ( )
∫ 1( ′′
) PNC 2𝑇2 2𝑇3
𝑏 𝑃out = Pr min(𝑋1 , 𝑋2 ) < or 𝑋1 + 𝑋2 <
𝑒−(𝑎𝑥+ 𝑥 ) 𝑑𝑥
𝑏
−𝑐 𝑎′′ 𝑥 + 𝛽 𝛽
𝜃 𝑥 ( )
∫ 1( )2 2𝑇2 2𝑇2 2𝑇3
𝑏′ = 1 − Pr 𝑋1 > , 𝑋2 > , 𝑋1 + 𝑋2 >
𝑒−(𝑎𝑥+ 𝑥 ) 𝑑𝑥, 𝛽 𝛽 𝛽
𝑏

+𝑐 𝑎𝑥+ (B.4) ( )
𝜃 𝑥 2𝑇2 2𝑇2 2𝑇3
= 1 + Pr 𝑋1 > , 𝑋2 > , 𝑋1 + 𝑋2 <
𝛽 𝛽 𝛽
where all the derivatives are taken with respect to 𝑇1 . We ( )
2𝑇2 2𝑇2
rewrite 𝑎, 𝑏, and 𝑐 in terms of 𝑇1 by replacing 𝜌 = 𝐿1 /𝑇1 − Pr 𝑋1 > , 𝑋2 >
and then it can be shown that 𝑎 → 0, 𝑏 → 0, 𝑐 → 0, 𝑎′ → 𝛽 𝛽
( )
(𝜆1 +𝜆2 )(1+𝐿1 )
, 𝑏′ → 𝛽𝐿2𝜆0
, 𝑐′ → (𝜆1 +𝜆 2 )(1+𝐿1 )
, 𝑎′′ → 0, 𝑏′′ → 2𝑇2 2𝑇2 2𝑇3
(1−𝛽)𝐿1 ( 1 ) (1−𝛽)𝐿1 = 1 + Pr 𝑋1 > , 𝑋2 > , 𝑋1 + 𝑋2 <
𝛽 𝛽 𝛽
0, and 𝑐′′ → 2 𝜆1−𝛽 1 +𝜆2
+ 2𝜆𝛽0 (𝜆1 +𝜆 2 )(1+𝐿1 )
(1−𝛽)𝐿21
as 𝑇1 → 0. By 2(𝜆1 +𝜆2 )𝑇2
interchanging limit and integral, we can show − 𝑒− 𝛽 . (C.3)

𝜓 1 Using the fact 𝑇3 > 2𝑇2 , the last term of (C.3) can be derived
lim = lim 𝜓 ′′ as follows:
𝑇1 =0 𝑇12 2 𝑇1 =0 ( )
1−𝜃 1 − 𝜃2 2𝑇2 2𝑇2 2𝑇3
= lim 𝑐′′ − lim (𝑎′ 𝑐′ ) + ln𝜃 lim (𝑏′ 𝑐′ ) Pr 𝑋1 > , 𝑋2 > , 𝑋1 + 𝑋2 <
2 𝑇1 =0 2 𝑇1 =0 𝑇1 =0
𝛽 𝛽 𝛽
∫ 2(𝑇3𝛽−𝑇2 ) ∫ 2𝑇𝛽3 −𝑥2
2𝜆0 (𝜆1 + 𝜆2 )(1 + 𝐿1 )(1 − 𝜃 + ln 𝜃)
= = 𝜆1 𝑒−𝜆1 𝑥1 𝜆2 𝑒−𝜆2 𝑥2 𝑑𝑥1 𝑑𝑥2 ,
𝛽(1 − 𝛽)𝐿21 2𝑇2 2𝑇2
( ) ⎧
𝑥2 = 𝛽 𝑥1 =
𝛽
(𝜆1 + 𝜆2 )2 (1 + 𝐿1 )(1 − 𝜃) 𝜃(1 + 𝐿1 ) − (1 − 𝐿1 ) 2(𝜆1 +𝜆2 )𝑇2 2(𝜆1 −𝜆1 )𝑇2 −2𝜆2 𝑇3
(C.4)
− . 
 𝑒 − 𝛽 + 𝜆2𝜆−𝜆1
𝑒 𝛽
2(1 − 𝛽)2 𝐿21 
 1

 𝜆
2(𝜆1 −𝜆2 )𝑇2 −2𝜆1 𝑇3
(B.5) ⎨ + 𝜆 −𝜆 𝑒
2 𝛽 , 𝜆1 ∕= 𝜆2 ,
1 2
=

 2(𝜆1 +𝜆2 )𝑇2
2𝜆2 (𝑇3 −2𝑇2 ) − 2𝜆1𝛽𝑇3
Therefore, considering both (B.3) and (B.5), we can obtain 

 𝑒− 𝛽 − 𝛽 𝑒

⎩ 2𝜆 𝑇
− 2𝛽 3
the second case of (8). Finally, (8) indicates that the diversity −𝑒 , 𝜆1 = 𝜆2 .
order of the TDBC protocol is two. This completes the proof.
PNC
Therefore, the the outage probability 𝑃out for 0 < 𝛽 ≤ 2/3
is given as follows:

A PPENDIX C 
 𝜆1
2(𝜆2 −𝜆1 )𝑇2 −2𝜆2 𝑇3

 1 + 𝜆2 −𝜆1 𝑒

𝛽
P ROOF OF T HEOREM 2 
⎨ + 𝜆2 𝑒 2(𝜆1 −𝜆2 )𝑇𝛽 2 −2𝜆1 𝑇3 ,
PNC 𝜆1 −𝜆2 𝜆1 =
∕ 𝜆2 ,
𝑃out =
PNC
We rewrite the outage probability 𝑃out of (15) for the PNC 

 2𝜆 𝑇 2𝜆 𝑇
 1 − 2𝜆2 (𝑇3 −2𝑇2 ) 𝑒− 1𝛽 3 − 𝑒− 2𝛽 3 , 𝜆1 = 𝜆2 .
protocol as follows: 
⎩ 𝛽

( ) (C.5)
PNC 𝑅
𝑃out = Pr min(𝐼1PNC , 𝐼2PNC ) < PNC
or 𝐼sum < 𝑅 . (C.1)
2
B. Case 2: 2/3 < 𝛽 < 1
PNC
2
As in Appendix A, let 𝑋𝑖 = ∣ℎ𝑖 ∣ , 𝑖 = 1, 2. Then 𝑋1 and 𝑋2 From (C.1) and (C.2), the outage probability 𝑃out for the
are independent exponential random variables with parameters case 2/3 < 𝛽 < 1 can be written as
𝜆1 and 𝜆2 , respectively. From (12) and (13), it can be shown (
PNC 𝑇2 𝑇2
that 𝑃out = 1 − Pr 𝑋1 > , 𝑋2 > ,
1−𝛽 1−𝛽
) (C.6)
2𝑇3
min(𝐼1PNC , 𝐼2PNC ) 𝑋1 + 𝑋2 > .
( ) 𝛽
1
= log 1 + min (𝛽𝜌/2, (1 − 𝛽)𝜌) min(𝑋1 , 𝑋2 ) 2𝑇2 2𝑇3 𝑇3
2⎧ ( ) If 1−𝛽 > 𝛽 , i.e., < 𝛽 < 1, (C.6) is simplified as
𝑇3 +𝑇2
⎨ 12 log 1 + 𝛽𝜌 2 min(𝑋 1 , 𝑋 2 ) , 0 < 𝛽 ≤ 2/3, ( )
= PNC 𝑇2 𝑇2
⎩ 1 ( ) 𝑃out = 1 − Pr 𝑋1 > , 𝑋2 >
1−𝛽 1−𝛽
2 log 1 + (1 − 𝛽)𝜌 min(𝑋1 , 𝑋2 ) , 2/3 < 𝛽 < 1. (C.7)
(𝜆1 +𝜆2 )𝑇2
(C.2) = 1 − 𝑒− 1−𝛽 .
2𝑇2
In the following, therefore, we divide the discussion into two If 1−𝛽 < 2𝑇𝛽3 , i.e., 2/3 < 𝛽 < 𝑇3𝑇+𝑇
3
2
, the outage probability
PNC
cases: 0 < 𝛽 ≤ 2/3 and 2/3 < 𝛽 < 1. 𝑃out can be derived in the same manner as in the case 0 <
LIU AND KIM: PERFORMANCE ANALYSIS OF BIDIRECTIONAL COMMUNICATION PROTOCOLS BASED ON DECODE-AND-FORWARD RELAYING 2695

𝛽 ≤ 2/3, and the derived expression for this case is given as By some mathematical manipulations, we get
follows:
Pr (𝐴 ∩ 𝐵)
⎧ (𝜆2 −𝜆1 )𝑇2 2𝜆2 𝑇3 𝜆2 𝑇 3
( 𝜆1 𝑇 3 𝜆1 𝑇 3
)

 1 + 𝜆2𝜆−𝜆
1
𝑒 1−𝛽 − 𝛽 = 𝑒− 𝛽 𝑒− 𝛽 − 𝑒− 1−𝛽

 1

 𝜆2
(𝜆1 −𝜆2 )𝑇2 2𝜆 𝑇
− 1𝛽 3
𝜆1 ( − (𝜆1 +𝜆𝛽 2 )𝑇3 )
⎨ + 𝜆1 −𝜆2 𝑒
 , 𝜆1 ∕= 𝜆2 ,
1−𝛽 (𝜆1 +𝜆2 )𝑇3
− 𝑒 − 𝑒− 1−𝛽
PNC
𝑃out = ( ) 2𝜆1 𝑇3 (C.8) 𝜆1 + 𝜆2

 1 − 2𝜆2 𝑇𝛽3 − 1−𝛽 𝑇2
𝑒− 𝛽 ∫ 1−𝛽
𝑇3 (D.4)


𝑒( 𝛽 𝜆0 −𝜆1 )𝑥1 𝑑𝑥1
(𝜆 +𝜆 )𝑇 1−𝛽

 2𝜆2 𝑇3 − 𝜆1 𝑒 − 1 𝛽2 3

⎩ −𝑒− 𝛽 , 𝜆1 = 𝜆2 . 𝑇3
𝛽
∫ 𝑇3
1−𝛽
𝑒( )𝑥1 𝑑𝑥 ,
𝜆0 𝑇 3 1−𝛽

Finally, a combination of (C.5), (C.7), and (C.8) yields the + 𝜆1 𝑒− 𝛽 𝛽 𝜆0 −𝜆1 −𝜆2
1
𝑇3
PNC
closed-form expression 𝑃out of (16). 𝛽

where the integrals of (D.4) can be easily solved. The expo-


nents of the exponential functions in the above integrals can be
A PPENDIX D possibly equal to zero because 𝛽 can take any values satisfying
P ROOF OF T HEOREM 3 1/2 < 𝛽 < 1. Specifically, if 𝛽 = 𝜆0𝜆+𝜆 0
1
and 𝜆0 > 𝜆1 , the
exponent in the first integral equals zero; If 𝛽 = 𝜆0 +𝜆𝜆01 +𝜆2
2
As in Appendix A, we let 𝑋𝑖 = ∣ℎ𝑖 ∣ , 𝑖 = 0, 1, 2. and 𝜆0 > 𝜆1 + 𝜆2 , the exponent in the second integral equals
Then 𝑋𝑖 ’s are independent exponential random variables with zero.14 Therefore, the solution to the integrals of (D.4) can
parameters 𝜆𝑖 , 𝑖 = 0, 1, 2. Using Total Probability Theorem, be possibly different depending on the values of 𝛽. Because
the outage probability 𝑃outOSS
of (22) can be written as (D.1). the final expression of Pr (𝐴 ∩ 𝐵) is very involved, we do not
Since 𝑋𝑖 ’s are independent exponential random variables, the present the final expression of Pr (𝐴 ∩ 𝐵).
first probability of (D.1) can be derived as follows:
(( ) ( )) B. Pr (𝐴 ∩ 𝐶)
𝑇3 𝑇3
Pr 𝑋0 < ∩ max(𝑋1 , 𝑋2 ) < In order to derive Pr (𝐴 ∩ 𝐶), we proceed as follows:
𝛽 𝛽 (D.2)
( 𝜆 𝑇
) ( 𝜆 𝑇
)( 𝜆 𝑇
)
− 0𝛽 3 − 1𝛽 3 − 2𝛽 3 Pr (𝐴 ∩ 𝐶)
= 1−𝑒 1−𝑒 1−𝑒 .
∫ ∞
= Pr (𝐴∣𝑋1 = 𝑥1 ) Pr (𝐶∣𝑋1 = 𝑥1 ) 𝑓𝑋1 (𝑥1 )𝑑𝑥1 . (D.5)
We note that the second and third probabilities of (D.1) are 0
identical in form except that 𝑋1 and 𝑋2 are interchanged. Comparing (D.5) and (D.3), we notice that the derivation
Therefore, their final expressions are identical in form except of Pr (𝐴 ∩ 𝐶) is similar to the derivation of Pr (𝐴 ∩ 𝐵).
that 𝜆1 and 𝜆2 are interchanged. We denote the second Specifically, we also need to discuss two cases separately:
probability as 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ). Then the third probabil- 0 < 𝛽 ≤ 1/2 and 1/2 < 𝛽 < 1. For each case, there also
ity is denoted as 𝐺(𝛽, 𝜆0 , 𝜆2 , 𝜆1 , 𝑇3 ).( For convenience, we) exists a special subcase where the exponent of a certain term
define sets 𝐴, 𝐵, and 𝐶 as: 𝐴 := 𝑋0 + 1−𝛽 𝛽 )𝑋 1 < 𝑇3
𝛽 , in the integral equals zero. Finally, a piecewise solution to
( ) (
𝑇3
𝐵 := 𝛽 < 𝑋2 < 𝑋1 , 𝐶 := 𝑋1 < 𝛽 < 𝑋2 . Then we𝑇3 Pr (𝐴 ∩ 𝐶) can be derived for different 𝛽 values. Due to a
space limitation, we do not present the final expression of
have 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ) = Pr (𝐴 ∩ (𝐵 ∪ 𝐶)), where ∩ and
Pr (𝐴 ∩ 𝐶).
∪ denote the intersection and union operators, respectively. It
Therefore, we can obtain 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ) by summing
can be further shown that 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ) = Pr (𝐴 ∩ 𝐵)+
Pr (𝐴 ∩ 𝐵) and Pr (𝐴 ∩ 𝐶). Also, the last term of (D.1) can
Pr (𝐴 ∩ 𝐶) since 𝐵 ∩ 𝐶 = ∅, where ∅ denotes the empty set.
be derived by interchanging 𝜆1 and 𝜆2 in the expression
In the following, we now derive Pr (𝐴 ∩ 𝐵) and Pr (𝐴 ∩ 𝐶)
of 𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ). Finally, we obtain the closed-form
separately.
expression for the outage probability of the OSS protocol
by substituting all the derived terms into (D.1). In Theorem
3, we have presented a general expression 𝐺(𝑥, 𝑦0 , 𝑦1 , 𝑦2 , 𝑧)
A. Pr (𝐴 ∩ 𝐵) of (24) instead of presenting explicit expressions for both
𝐺(𝛽, 𝜆0 , 𝜆1 , 𝜆2 , 𝑇3 ) and 𝐺(𝛽, 𝜆0 , 𝜆2 , 𝜆1 , 𝑇3 ).
If 1−𝛽
𝛽 ≥ 1, i.e., 0 < 𝛽 ≤ 1/2, it can be shown that
𝐴 ∩ 𝐵 = ∅, and therefore, Pr (𝐴 ∩ 𝐵) = 0. In the flowing,
we now focus on the case with 1/2 < 𝛽 < 1. We proceed as R EFERENCES
follows: [1] J. N. Laneman, D. Tse, and G. W. Wornell, “Cooperative diversity in
wireless networks: efficient protocols and outage behavior,” IEEE Trans.
Inf. Theory, vol. 50, pp. 3062–3080, Dec. 2004.
Pr (𝐴 ∩ 𝐵) [2] J. N. Laneman and G. W. Wornell, “Distributed space-time-coded
∫ ∞
protocols for exploiting cooperative diversity in wireless networks,”
= Pr (𝐴 ∩ 𝐵∣𝑋1 = 𝑥1 ) 𝑓𝑋1 (𝑥1 )𝑑𝑥1 IEEE Trans. Inf. Theory, vol. 49, pp. 2415–2425, Oct. 2003.
∫0 ∞
14 Note 𝜆0
= Pr (𝐴∣𝑋1 = 𝑥1 ) Pr (𝐵∣𝑋1 = 𝑥1 ) 𝑓𝑋1 (𝑥1 )𝑑𝑥1 . (D.3) that 𝜆0 > 𝜆1 ensures that 𝜆0 +𝜆1
is between 1/2 and 1, and
𝜆0
0 𝜆0 > 𝜆1 + 𝜆1 ensures that 𝜆0 +𝜆1 +𝜆2
is between 1/2 and 1.
2696 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 9, SEPTEMBER 2010

( )
OSS 𝑇3 𝑇3
𝑃out = Pr 𝑋0 < , max(𝑋1 , 𝑋2 ) <
𝛽 𝛽
(( ) (( ) ( )))
1−𝛽 𝑇3 𝑇3 𝑇3
+ Pr 𝑋0 + 𝑋1 < ∩ < 𝑋2 < 𝑋1 ∪ 𝑋1 < < 𝑋2 (D.1)
𝛽 𝛽 𝛽 𝛽
(( ) (( ) ( )))
1−𝛽 𝑇3 𝑇3 𝑇3
+ Pr 𝑋0 + 𝑋2 < ∩ < 𝑋1 < 𝑋2 ∪ 𝑋2 < < 𝑋1 .
𝛽 𝛽 𝛽 𝛽

[3] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diversity, [22] R. Narasimhan, “Finite-SNR diversity-multiplexing tradeoff for corre-
part I: system description,” IEEE Trans. Commun., vol. 51, pp. 1927– lated Rayleigh and Rician MIMO channels," IEEE Trans. Inf. Theory,
1938, Nov. 2003. vol. 52, pp. 3965–3979, Sep. 2006.
[4] ——, “User cooperation diversity, part II: implementation aspects and [23] A. S. Avestimehr and D. N. C. Tse, “Outage capacity of the fading relay
performance analysis,” IEEE Trans. Commun., vol. 51, pp. 1939–1948, channel in the low-SNR regime,” IEEE Trans. Inf. Theory, vol. 53, pp.
Nov. 2003. 1401–1415, Apr. 2007.
[5] B. Rankov and A. Wittneben, “Spectral efficient protocols for half- [24] C. Hucher, G. R.-B. Othman, and J.-C. Belfiore, “AF and DF protocols
duplex fading relay channels,” IEEE J. Sel. Areas Commun., vol. 25, based on Alamouti ST code,” in Proc. IEEE ISIT, June 2007, pp. 1526–
pp. 379–389, Feb. 2007. 1530.
[6] T. J. Oechtering and H. Boche, “Bidirectional regenerative half-duplex [25] D. Chen and J. N. Laneman, “Modulation and demodulation for coop-
relaying using relay selection,” IEEE Trans. Wireless Commun., vol. 7, erative diversity in wireless systems,” IEEE Trans. Wireless Commun.,
pp. 1879–1888, May 2008. vol. 5, pp. 1785–1794, July 2006.
[7] T. J. Oechtering, C. Schnurr, I. Bjelakovic, and H. Boche, “Broadcast [26] A. Ribeiro, X. Cai, and G. B. Giannakis, “Symbol error probabilities
capacity region of two-phase bidirectional relaying,” IEEE Trans. Inf. for general cooperative links,” IEEE Trans. Wireless Commun., vol. 4,
Theory, vol. 54, pp. 454–458, Jan. 2008. pp. 1264–1273, May 2005.
[8] T. J. Oechtering and H. Boche, “Optimal time-division for bidirectional [27] T. S. Rappaport, Wireless Communications: Principles and Practice.
relaying using superposition encoding,” IEEE Commun. Lett., vol. 12, Upper Saddle River, NJ: Prentice Hall, 2002.
pp. 265–267, Apr. 2008. [28] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
[9] S.-Y. R. Li, R. W. Yeung, and N. Cai, “Linear network coding,” IEEE Communications. Cambridge, UK: Cambridge Univ. Press, 2003.
Trans. Inf. Theory, vol. 49, pp. 1204–1216, Feb. 2003. [29] T. M. Cover and J. A. Thomas, Elements of Information Theory, 2nd
[10] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, and J. Crowcroft, edition. Hoboken, NJ: John Wiley & Sons, 2006.
“XORs in the air: practical wireless network coding,” IEEE/ACM Trans.
Networking, vol. 16, pp. 497–510, June 2008.
Peng Liu received the B.Eng. and M. Eng. degrees
[11] S. Zhang, S. Liew, and P. Lam, “Physical layer network coding,” in
in electrical engineering from Xidian University,
Proc. ACM MobiCom, Sep. 2006, pp. 358–365.
Xi’an, China, in July 2005 and July 2007, respec-
[12] S. J. Kim, P. Mitran, C. John, R. Ghanadan, and V. Tarokh, “Coded
tively. In September 2007, he joined the Wireless
bi-directional relaying in combat scenarios,” in Proc. IEEE MILCOM,
Information Transmission Lab (WITL) of the De-
Oct. 2007, pp. 1–7.
partment of Electrical and Computer Engineering,
[13] S. J. Kim, P. Mitra, and V. Tarokh, “Performance bounds for bidirec-
Queen’s University, Kingston, Canada, where he
tional coded cooperation protocols,” IEEE Trans. Inf. Theory, vol. 54,
received his M.Sc. degree in August 2009. He is
pp. 5235–5241, Nov. 2008.
currently working towards the Ph.D. degree, and
[14] S. Zhang, S.-C. Liew, and L. Lu, “Physical layer network coding
is serving as a research assistant in the WITL.
schemes over finite and infinite fields,” in Proc. IEEE GLOBECOM,
His research interests include cooperative diversity
Dec. 2008, pp. 1–6.
networks, bidirectional communications, cognitive radios, and MIMO com-
[15] S. Katti, S. Gollakota, and D. Katabi, “Embracing wireless interference:
munication systems.
analog network coding,” in Proc. ACM SIGCOMM, Aug. 2007, pp. 397–
408.
Il-Min Kim received the B.S. degree in electronics
[16] Z. Yi and I.-M. Kim, “An opportunistic-based protocol for bidirectional
engineering from Yonsei University, Seoul, Korea, in
cooperative networks,” IEEE Trans. Wireless Commun., vol. 8, pp. 4836–
1996, and the M.S. and Ph.D. degrees in electrical
4847, Sep. 2009.
engineering from the Korea Advanced Institute of
[17] Z. Yi and I.-M. Kim, “Finite-SNR diversity-multiplexing tradeoff and
Science and Technology (KAIST), Taejon, Korea,
optimum power allocation in bidirectional cooperative networks,” sub-
in 1998 and 2001, respectively. From October 2001
mitted. [Online]. Available: http://arxiv.org/PS_cache/arxiv/pdf/0810/
to August 2002, he was with the Department of
0810.2746v1.pdf
Electrical Engineering and Computer Science at
[18] R. Knopp and P. Humblet, “Information capacity and power control
MIT, Cambridge, USA, and from September 2002 to
in single-cell multiuser communication,” in Proc. IEEE ICC, 1995, pp.
June 2003, he was with the Depatment of Electrical
331–335.
Engineering at Harvard, Cambridge, USA, as a post-
[19] P. Viswanath, D. N. C. Tse, and R. Laroia, “Opportunistic beamforming
doctoral research fellow. In July 2003, he joined the Department of Electrical
using dumb antennas,” IEEE Trans. Inf. Theory, vol. 48, pp. 1277–1294,
and Computer Engineering at Queen’s University, Kingston, Canada, where he
June 2002.
is currently an associate professor. His research interests include bidirectional
[20] J. N. Laneman and G. W. Wornell, “Energy-efficient antenna-sharing
communications, cooperative diversity networks, network coding, femto cells,
and relaying for wireless networks,” in Proc. IEEE WCNC, Sep. 2000,
CoMP, and cognitive radio. He is currently serving as an editor for the
pp. 7–12.
IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS and the Journal
[21] L. Zheng and D. N. C. Tse, “Diversity and multiplexing: a fundamental
of Communications and Networks (JCN).
tradeoff in multipleantenna channels,” IEEE Trans. Inf. Theory, vol. 49,
pp. 1073–1096, May 2003.

You might also like