IEEE COMMUNICATIONS LETTERS, VOL. 14, NO.

2, FEBRUARY 2010 145

Power Allocation for Bidirectional AF Relaying over

Rayleigh Fading Channels
Yuanyuan Zhang, Yi Ma, Member, IEEE, and Rahim Tafazolli, Senior Member, IEEE

Abstract—This letter presents two novel power allocation derived, which can’t be optimized at the same time. Accord-
schemes for bidirectional amplify-and-forward (AF) relaying over ingly, a power allocation strategy is proposed to make a trade-
Rayleigh fading channels through the exploitation of channel off between these two terminals. It is noticed in the numerical
mean strength. The first scheme aims to maximize the upper
bound of average sum rate, and the other aims to achieve the results that the proposed strategies outperform the traditional
trade-off of outage probability between two terminals. Numerical equal power allocation in the average sum rate and outage
results show considerable performance improvement in compar- probability.
ison with conventional power allocation approaches.
II. S YSTEM M ODEL
Index Terms—Amilify-and-forward, average sum-rate, bidirec-
tional relaying, power allocation, outage probability. Consider a three-node bidirectional relaying system consist-
ing of two terminals, S1 and S2 , one relay, R. In Phase I, S1
and S2 transmit their signals simultaneously to the relay R
I. I NTRODUCTION and the received signal at R is given by
INCE 1960s Shannon’s ground-breaking work about the �� = ℎ1 �1 + ℎ2 �2 + �� (1)
S two-way channel [1], a lot of research efforts have been
paid to find the fundamental capacity limit of this special com-
where �1 and �2 are the transmitted signals with transmit
power �� from S1 and S2 respectively, ℎ1 and ℎ2 are inde-
munication model. However, the capacity for this seemingly pendent complex Rayleigh fading channel gains of channels
simple channel has not been found to this date. Recently, there from S1 and S2 to R, respectively, and �� is the complex
are increasing research activities towards the combination Additive White Gaussian Noise (AWGN) with noise power
of two-way channel and relay channel, namely, two-way �0 . The channel mean strengths of these two channels are �1
relay channel (TWRC) or bidirectional relaying [2][3]. In and �2 , respectively. The channels are assumed to be invariant
the TWRC system, bidirectional communication between two for the consecutive two phases.
users can enjoy improved spectral efficiency compared with At the relay, with amplification factor �, the received signal
the traditional one-way relay channel [2]. The achievable rate is broadcast to both terminals in Phase II. Then, the received
of the system is investigated for relaying protocols including signal at terminal S�,�=1,2 is,
amplify-and-forward (AF), and decode-and-forward (DF) for
deterministic channels in [3]. For more efficient use of the �� = ℎ� (��� ) + �� (2)
power resource, the power allocation has been exploited. In With the knowledge of its own signal at each terminal, the
[4][5], the power allocation is to maximize the system capacity self-interference part can be subtracted from �� resulting
and achievable rate for the deterministic channel; while a fixed
�˜� = �ℎ1 ℎ2 �� + �ℎ� �� + �� (3)
power allocation ratio independent of channel quality is given
to improve the upper bound of the average sum rate in [6]. where � = 2 for � = 1 and � = 1 for � = 2, � is chosen
In this letter, we add to this area by investigating power subject to the power at relay �� as [6],
√
allocation strategies for AF bidirectional relaying system un- ��
der total power constraint1 . Requiring knowledge of only the �= 2 2 (4)
∣ℎ1 ∣ �� + ∣ℎ2 ∣ �� + �0
channel mean strength, these power allocation strategies are
applicable even under rapidly time-varying channels. First, a Then the received SNR for signals from S1 and S2 are,
power allocation strategy is proposed to maximize the upper 2 2
∣ℎ1 ∣ ∣ℎ2 ∣ �� ��
bound of the average sum rate in high average SNR region. �1 = 2 2 , (5)
∣ℎ1 ∣ �� + ∣ℎ2 ∣ (�� + �� ) + 1
Second, to avoid one of the terminals suffering from severe
2 2
outage, the outage probability of each individual terminal is ∣ℎ1 ∣ ∣ℎ2 ∣ �� ��
�2 = , (6)
∣ℎ2 ∣2 �� + ∣ℎ1 ∣2 (�� + �� ) + 1
Manuscript received November 12, 2009. The associate editor coordinating
the review of this letter and approving it for publication was G. Mazzini. respectively, where �� = �� /�0 and �� = �� /�0 . In
This work has been partially supported by the EU-ICT WHERE project most early publications, it is assumed all the three nodes
and UK MVCE Core 4-Efficiency. transmit with the same power, �� = �� , referred as equal
The authors are with C.C.S.R., University of Surrey, U.K., GU2 7XH (e-
mail: y.ma@surrey.ac.uk). power allocation in this paper. However, the performance
Digital Object Identifier 10.1109/LCOMM.2010.02.092227 of this system can be improved by allocating power among
1 The total power constraint is widely considered in relay networks to
these nodes. Next we propose two different power allocation
provide useful insight into the relay optimization [7][8]. Practically, this is
motivated by the fact that in networks such as sensor network, where the strategies with the total power constraint 2�� + �� = �0 in
long-term power consumption is a major concern. high average SNR range.
c 2010 IEEE
1089-7798/10$25.00 ⃝

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
146 IEEE COMMUNICATIONS LETTERS, VOL. 14, NO. 2, FEBRUARY 2010

III. P OWER A LLOCATION S TRATEGIES B. Outage Probability

A. Average Sum Rate In some cases, the sum rate of the system is able to
satisfy certain requirement, with one direction transmission
Δ
Define �� = log2 (1 + �� ). The average sum rate of the at high data rate and the other direction in outage. With this
bidirectional AF relaying is given by consideration, to improve the outage probability of individual
terminals is important. In this section, we derive the average
�[�] = �[�1 ] + �[�2 ] outage probability of each terminal and propose the power
(7)
= �[log2 (1 + �1 )(1 + �2 )], allocation to make a balance of the two terminals in terms of
the individual outage probability.
where �[⋅] denotes the expectation. The optimization problem The outage probability is the probability for each terminal’s
is to solve the following cost function instantaneous rate �� falls below the threshold rate �0 as
����,� = Pr{�� ≤ �0 }. Take terminal S1 as an example, its
max (�[log2 (1 + �1 )(1 + �2 )]) , s.t. 2�� + �� = �0 . (8)
outage probability is given as,
[ ( ) ]
However, optimization based on �[�] is still an open math- �(1 − �)∣ℎ1 ∣2 ∣ℎ2 ∣2 �0
ematical problem to this date. Alternatively, we can reduce ����,1 = Pr log2 1 + ≤ � 0 .
�∣ℎ1 ∣2 + (1 − �)∣ℎ2 ∣2
the optimization problem by maximizing the following upper (12)
bound Equivalently, we have
[ ]
2
∑ 2�∣ℎ1 ∣2 (1 − �)∣ℎ2 ∣2 2(1 − �)(2�0 − 1)
�[�] ≤ log2 (1 + �[�� ]) ����,1 = Pr ≤ .
�∣ℎ1 ∣2 + (1 − �)∣ℎ2 ∣2 (1 − 2�)�0
( �=1
) (13)
� 1 � 2 �� �� (9) Note that the term at the left side of ‘≤’ is the harmonic
≤ log2 1 +
� � + �2 (�� + �� ) + 1 mean of two random variables, �∣ℎ1 ∣2 and (1−�)∣ℎ2 ∣2 . Using
( 1 � )
� 1 � 2 �� �� Theorem 1 in [10], the outage probability can be expressed
+ log2 1 + as,
�2 �� + �1 (�� + �� ) + 1
( ( ))
� 1 1
where the 1st step follows Jensen’s inequality, and the 2nd ����,1 = 1 − ��1 (�) exp − + (14)
step follows the computation of average SNR in [9]. In the 2 ��1 (1 − �)�2
√
high SNR region (�� �� >> 1, � = �, �), this upper bound can where � = 2(1−�)(2
�0
� = � 1
, �(1−�)�1 �2 , �1 (�) is the
−1)
(1−2�)�0
be approximately written into
1st order Bessel function of the 2nd kind. When � is small,
( )
�12 �22 ��2 ��2 �1 (�) ≈ 1/�, which is true in high average SNR region,
�[�] ≤ log2 . i.e., �0 >> 1. Thus, the outage probability can be further
(�1 �� + �2 (�� + �� )) (�2 �� + �1 (�� + �� ))
(10) simplified into,
Then, our objective reduces to maximize (10) subject to the ( ( ))
�(1 − �) 1 1
total power constraint in (8). Due to the employment of ����,1 = 1 − exp − + (15)
1 − 2� ��1 (1 − �)�2
Jensen’s inequality, the upper bound (10) is not tight enough.
Optimization based on (10) does not offer an optimum solution 2�0 −1
with � = as a constant. Similarly, the outage probabil-
�0 ( ( ))
to (8). On the other hand, our numerical results show that the
ity for S2 is ����,2 = 1 − exp − �(1−�)
1−2�
1
��2 + 1
(1−�)�1 .
proposed optimization scheme outperforms the state-of-the-art
power allocation schemes in terms of average sum rate. Both ����,1 and ����,2 are convex functions √ of �, thus the
2�2 − 2�2 +2�1 �2
Define a parameter �(0 < � < 0.5), as the power ratio optimal � to minimize them are ����,1 = 2
and
√ 2 �2 −�1
allocated to each terminal over the total power �0 . Thus, 2�1 − 2�1 +2�1 �2
����,2 = �1 −�2 , respectively. However, ����,1 and
each terminal transmits with power �� = ��0 , and relay with
����,2 are different, thus the outage probability for the two
power �� = (1 − 2�)�0 . Let �0 = �0 /�0 denote the total
terminals can’t be optimized simultaneously with the same �.
transmit SNR, thus, �� = ��0 , �� = (1 − 2�)�0 . Substituting
This is shown in Fig. 1. The outage probability is convex
� into (10), our power allocation strategy is to determine the
for both terminals, with the minimum point at ����,1 and
optimal � which can maximize the following as,
����,2 , respectively. To keep both terminals’ outage probability
� 2 (1 − 2�)2 �12 �22 �02 at a considerable level, we propose to choose ���� as the
max , s.t.0 < � < 0.5. optimal � for the terminal with larger outage probability as a
� (��1 + (1 − �)�2 )((1 − �)�1 + ��2 )
(11) trade-off solution to balance the outage performance of the
By taking the first order derivation of this function with respect two terminals. For example in Fig. 1, ���� = ����,2 , and
to �, it can be proved via standard analysis of continuous mathematically it is to find ���� to the following,
functions that the optimal solution to (11) exists and is in
min max{����,1 , ����,2 }, s.t. 0 < � < 0.5. (16)
the area � ∈ (0.1910, 0.25]. In practice, the optimal � can
be calculated numerically. Contrastively, � is fixed 1/3 in With this power allocation, both ����,1 and ����,2 are in-
equal power allocation; the power allocation suggested in [6] creasing functions versus � in [���� , 0.5); the terminal with
is equivalent to � = 0.1910, but our analysis suggests better larger outage probability is optimized and the other’s outage
performance can be achieved with � ∈ (0.1910, 0.25]. probability is already at a relatively low level.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
ZHANG et al.: POWER ALLOCATION FOR BIDIRECTIONAL AF RELAYING OVER RAYLEIGH FADING CHANNELS 147

0
10 5.2

4.8
average outage probability

4.6

average sum rate

4.4
1
10
4.2
euqal power allocation
S1
4
S2 proposed strategy
optimal for S1 3.8 strategy in [6]
optimal for S2
3.6

10
2 3.4
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
beta d

Fig. 1. The relationship of individual average outage probability with � Fig. 2. Average sum rate versus normalized distance (�0 =10dB).
(�1 > �2 ).

0.055
IV. N UMERICAL R ESULTS
0.05
In this section, computer simulations are performed to
outage for S1 (proposed)
0.045
present the performance of the proposed power allocation outage for S2 (proposed)
strategies. The total average SNR is set to be �0 =10dB and
average outage probability

0.04 outage for S1 (euqal power)

20dB, respectively, and the relay is assumed to be located on outage for S2 (equal power)
0.035
the line that passes through S1 and S2 . The distance between
S1 and S2 is normalized to 1, and the normalized distance 0.03

from S1 to R is �. Thus the channel mean strengths of the 0.025

two channels are �1 = �−3 , and �2 = (1 − �)−3 , respectively.
0.02
Fig. 2 shows the average sum rate versus �. The proposed
power allocation strategy is compared with the power alloca- 0.015

tion strategy in [6], and equal power allocation. The proposed 0.01
one provides significant improvement over the equal power
0.005
allocation, and is slightly better than in [6], which is in line 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
d
with our analytical results. Fig. 3 compares the two termi-
nals’ average outage probability of the 2nd proposed power
allocation strategy with equal power allocation (threshold rate
Fig. 3. Average outage probability versus normalized distance (�0 =20dB).
�0 = 1). The individual outage probability with the proposed
power allocation strategy for either S1 or S2 is lower than with
[3] P. Popvski and H. Yomo, “Physical network coding in two-way wireless
equal power allocation. The performance gain is larger when relay channels,” in Proc. IEEE Int. Conf. Commun. (ICC’07), pp. 707–
the relay is close to one terminal (� → 0 or � → 1). 712.
[4] A. Agustin, J. Vidal, and O. Munoz, “Protocols and resource allocation
for the two-way relay channel with half-duplex terminals,” in Proc.
V. C ONCLUSION IEEE Int. Conf. Commun. (ICC’09), pp. 14–18.
[5] C. K. Ho, R. Zhang, and Y. Liang, “Two-way relaying over OFDM:
This paper has presented simple power allocation strategies optimized tone permutation and power allocation,” in Proc. IEEE Int.
Conf. Commun. (ICC’08), pp. 3908–3912.
for bidirectional AF-relaying system. With only requirement [6] Y. Han, S. H. Ting, C. K. Ho, and W. H. Chin, “High rate two-way
of the channel mean strength, the proposed strategies aimed amplify-and-forward half-duplex relaying with OSTBC”, in Proc. IEEE
to improve the average sum rate and individual average out- VTC 08’Spring, pp. 2426–2430.
[7] F. Gao, T. Cui, and A. Nallanathan, “Optimal training design for channel
age probability under Rayleigh fading channels, respectively. estimation in decode-and-forward relay networks with individual and
Numerical results have shown visible improvement over the total power constraints”, IEEE Trans. Signal Process., vol. 56, no. 12,
equal power allocation. pp. 5937–5949, Dec. 2008.
[8] L. Lai, K. Liu, and H. E. Gamal, “The three-node wireless network:
achievable rates and cooperation strategies”, IEEE Trans. Inf. Theory,
vol. 52, no. 3, pp. 805–828, Mar. 2006.
R EFERENCES [9] X. Deng and A. M. Haimovich, “Power allocation for cooperative
[1] C. E. Shannon, “Two-way communication channels,” in Proc. 4th relaying in wireless networks”, IEEE Trans. Commun. Lett., vol. 9,
Berkeley Symp. Probability and Statistics, 1961, vol. 1, pp. 611–644. no. 11, pp. 994–996, Nov. 2005.
[2] B. Rankov and A. Wittneben, “Spectral efficient protocols for half- [10] M. O. Hasna and M. S. Alouini, “Performance analysis of two-hop
duplex fading relay channels”, IEEE J. Sel. Areas Commun., vol. 25, relayed transmission over rayleigh fading channels”, in Proc. IEEE
no. 2, pp. 379–389, Feb. 2007. VTC 02’Fall, pp. 1992–1996.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.