IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO.
10, OCTOBER 2015
[9] J. Wang and P. A. Regalia, Sensor network localization via boundary
projections, in Proc. IEEE Int. CISS, Baltimore, MD, USA, Mar. 2009,
pp. 224229.
[10] C.-L. Wang, Y.-W. Hong, and Y.-S. Dai, A decentralized positioning
method for wireless sensor networks based on weighted interpolation,
in Proc. IEEE ICC, Glasgow, U.K., Jun. 2007, pp. 31673172.
[11] T. S. Rappaport, Wireless Communications: Principles and Practice,
2nd ed. Upper Saddle River, NJ, USA: Prentice-Hall, 2002.
[12] K. Pahlavan and P. Krishnamurthy, Networking Fundamentals: Wide, Lo-
cal and Personal Area Communications, 1st ed. Hoboken, NJ, USA:
Wiley, 2009.
[13] N. Patwari, A. O. Hero, III, M. Perkins, N. S. Correal, and R. J. ODea,
Relative location estimation in wireless sensor networks, IEEE Trans.
Signal Process., vol. 51, no. 8, pp. 21372148, Aug. 2003.
[14] W.-Y. Chiu, B.-S. Chen, and C.-Y. Yang, Robust relative location es-
timation in wireless sensor networks with inexact position problem,
IEEE Trans. Mobile Comput., vol. 11, no. 6, pp. 935946, Jun. 2012.
[15] C.-L. Wang, Y.-S. Chiou, and Y.-S. Dai, An adaptive location estimator
based on alpha-beta filtering for wireless sensor networks, in Proc. IEEE
WCNC, Kowloon, China, Mar. 2007, pp. 32853290.
[16] P. M. Clarkson, Optimal and Adaptive Signal Processing. Boca Raton,
FL, USA: CRC, 1993.
[17] J. M. Martnez and R. J. B. D. Sampaio, Parallel and sequential
Kaczmarz methods for solving underdetermined nonlinear equations,
J. Comput. Appl. Math., vol. 15, no. 3, pp. 311321, Jul. 1986.
[18] A. Nedic and D. P. Bertsekas, Convergence rate of incremental sub-
gradient algorithms, in Stochastic Optimization: Algorithms and Appli-
cations, S. Uryasev and P. Pardalos, Eds. Norwell, MA, USA: Kluwer,
2000, pp. 263304.
[19] N. A. Pantazis, S. A. Nikolidakis, and D. D. Vergados, Energy-efficient
routing protocols in wireless sensor networks: A survey, IEEE Commun.
Surveys Tuts., vol. 15, no. 2, pp. 551591, 2nd Quart., 2013.
[20] R. W. Ouyang, A. K.-S. Wong, and C.-T. Lea, Received signal strength-
based wireless localization via semidefinite programming: Noncoopera-
tive and cooperative schemes, IEEE Trans. Veh. Technol., vol. 59, no. 3,
pp. 13071318, Mar. 2010.
[21] W.-Y. Chiu and B.-S. Chen, A mixed-norm approach using sim-
ulated annealing with changeable neighborhood for mobile location
estimation, IEEE Trans. Mobile Comput., vol. 9, no. 5, pp. 633642,
May 2010.
[22] K. Yu and Y. J. Guo, Statistical NLOS identification based on AOA,
TOA, and signal strength, IEEE Trans. Veh. Technol., vol. 58, no. 1,
pp. 274286, Jan. 2009.
[23] Crossbow Technology Inc., Milpitas, CA, USA, MICAz Datasheet, 2014,
http://www.xbow.com
0018-9545 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
4893
Joint Cooperative Beamforming, Jamming, and
Power Allocation to Secure AF Relay Systems
Hui-Ming Wang, Member, IEEE, Feng Liu, and Mengchen Yang
AbstractThe idea of multiuser (nodes) cooperation is an efficient way
to improve the physical-layer security of a wireless transmission in the
presence of passive eavesdroppers. However, due to the half-duplex con-
straint of the practical transceivers, two phases are required for one round
of data transmission, which grants the eavesdroppers two opportunities
to wiretap the information. Therefore, protecting the data transmissions
in both phases is critical. Toward this end, we propose a joint coopera-
tive beamforming, jamming, and power-allocation scheme to enhance the
security of an amplify-and-forward (AF) cooperative relay network in
this paper. Different from the existing works assuming that the source
node always uses its total power, we show that the secrecy rate is a
quasi-concave function of the power of the source node so that allocating
its total power may not be optimal. The beamformer design and power
optimization problem can be solved by a bisection method together with a
generalized eigenvalue decomposition, which has a semiclosed form and is
computationally very convenient. Simulations show that the joint scheme
greatly improves the security.
Index TermsBeamforming, physical layer security, power allocation,
relay system.
I. I NTRODUCTION
Exploiting multiple-node cooperation to improve the physical layer
security of wireless communications has attracted increasing interest
very recently [1][15]. For a cooperative system where all termi-
nals are only equipped with a single antenna, generally, there are
two efficient ways to take advantage of the multiple nodes in the
system: cooperative beamforming and cooperative jamming. Coop-
erative beamforming [1][5] helps improve the channel quality to
the legitimate destination, whereas cooperative jamming (also called
artificial noise) degrades the channel condition of the eavesdrop-
pers [6][10]. However, the data transmission in relay networks re-
quires two phases, i.e., phase I (broadcasting phase) and phase II
(relaying phase), due to the half-duplex constraint of the transceivers
(let us assume that there is no direct link between the source and
destination). This grants the potential eavesdroppers two opportu-
nities to intercept the information. Therefore, protecting the data
Manuscript received October 18, 2013; revised January 1, 2014 and June 3,
2014; accepted November 11, 2014. Date of publication November 13, 2014;
date of current version October 13, 2015. This work was supported in part by
the National Natural Science Foundation of China under Grant 61102081 and
Grant 61221063; by the Specialized Research Fund for the Doctoral Program
of Higher Education of China under Grant 20110201120013; by the New
Century Excellent Talents Support Fund of China under Grant NCET-13-0458;
by the Industrial Research Fund of Shaanxi Province under Grant 2012GY2-
28; by the Fok Ying Tung Education Foundation under Grant 141063; and
by the Fundamental Research Funds for the Central University under Grant
2013jdgz11. The review of this paper was coordinated by Prof. H.-H. Chen.
The authors are with the Key Laboratory of Intellectual Network and
Network Security of the Ministry of Education, School of Electronic and
Information Engineering, Xian Jiaotong University, Xian 710049, China
(e-mail: xjbswhm@gmail.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2014.2370754
4894 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO. 10, OCTOBER 2015

ence signal to cover the information transmission. The signal vector

received at the relays is
 
(1)
yR = Psf R s + PJ h R z (1) + n R (1)

where y R  [yR,1 , yR,2 , . . . , yR,N1 ]T ; f R  [fR,1 , fR,2 , . . . ,

(1)
fR,N1 ]T and similarly for h R ; Ps and PJ are the transmit power
values of the source and the jammer, respectively; z (1) is the jamming
signal; and n R is the additive noise at the relay nodes. We normalize
E{|s|2 } = 1 and E{|z (1) |2 } = 1. Concurrently, the eavesdropper
will also receive the signal
 
Fig. 1. Joint cooperative beamforming and jamming scheme, where the solid (1) (1) (1)
lines and the dash lines are the transmissions in phases I and II, respectively.
yE = Ps fE s + PJ qE z (1) + nE (2)

(1)
where nE is the additive noise at the eavesdropper.
In phase II, the N 1 relay nodes do a distributed beamforming
transmissions in both phases is critical to guarantee the security of the
to forward the received signal to the destination. The transmitted
data transmission.
signal x R  [xR,1 , xR,2 , . . . , xR,N1 ] is x R = W y R , where W is
However, most previous works only consider taking cooperative
the weight matrix in the form of W = diag([w1 , w2 , . . . , wN1
]),
beamforming or jamming in phase II to protect the transmission
and diag() is a diagonal matrix. Concurrently, the jammer transmits
[1][5]. In phase I, they assume that the source node broadcasts with (2)
its total power and all the relay nodes listen. Some even simply interference signal again as z (2) with power PJ . The received signals
assume that the transmission in phase I is perfectly secured [2], [3], at the destination D and the eavesdropper E are respectively
[10], which is obviously over optimistic. In our previous works [12],  
(1)
[13], we have already shown that the joint cooperative beamforming yD = Psg TRW f R s + PJ g TRW h R z (1) + n
D (3)
and jamming in both phases will greatly improve the secrecy rate.
 
However, power allocation of the source node still has not been (2)
yE = Psc TE W f R s +
(1)
PJ c TE W h R z (1) + n
(2)
E (4)
considered.
In this paper, we aim to enhance the security of an amplify-and- 
(2) (2)
forward (AF) cooperative relay network. Based on the joint coop- where D 
n PJ gJ z (2) + g TRW n R + nD , and E 
n
erative beamforming and jamming scheme in [12], we optimize the 
(2) (2)
power allocation of the source node to improve the secrecy rate further. PJ qE z (2) + c TE W n R + nE , respectively. c E  [cE,1 , cE,2 , . . . ,
(2)
We show that the secrecy rate is a quasi-concave function of the cE,N1 ]T , and nD and nE are additive noise terms at D and E
transmission power of the source node so that allocating its total during phase II, respectively. Equation (3) can be reformulated as
power may even harm the secrecy. This is intuitively reasonable since 
increasing the source transmission power will increase both the rate to  (1)
yD = Psw H a f g s + PJ w H a gh z (1) + n
D (5)
the legitimate destination and that to the eavesdropper. Therefore, the
power allocation should balance these two effects. The beamformer
design and power optimization problem can be solved by a bisec- where a f g  [fR,1 gR,1 , fR,2 gR,2 , , fR,N1 gR,N1 ]T and simi-
tion method together with a generalized eigenvalue decomposition, larly for a gh , and w  [w1 , w2 , . . . , wN1 ]T .
which has a semiclosed form and is computationally very conve- For the eavesdropper, each transmission phase grants it an opportu-
nient. Simulations show that the joint scheme greatly improves the nity to get the information. Combining (2) and (4) yields the receiving
security. model of the eavesdropper in the whole procedure as

y E = H E s + nE (6)
II. S YSTEM M ODEL
where
We consider an AF wireless network in which a source S wants to
 
send information to the destination D under the existence of an eaves-
dropper E. There are N intermediate relay nodes Rn , n = 1, 2, . . . , N , HE = PsH
fE
Psw a cf
between S and D. Each node in the whole network is only equipped  
(1)
with a single antenna and is subject to the half-duplex constraint. n
E

We assume that there is no direct connection between S and D. Our nE = (1) (2)
(7)
joint beamforming and jamming scheme is to divide the intermediate PJ c TE W h R z (1) + n
E
nodes into two groups: One node is jammer J, and all the other N 1
are relay nodes, as shown in Fig. 1. The relay nodes will forward with a cf  [cE,1 fR,1 , cE,2 fR,2 , , cE,N1 fR,N1 ]T , acg 
(1)
E,1 gR,1 , cE,2 gR,2 , , cE,N1 gR,N1 ] ,
the received signal using cooperative beamforming, and the jammer T
[c and n
E =
transmits interference signals to confuse the eavesdropper. The quasi- (1) (1) (1)
stationary flat-fading channels between S, R, J, and E are also shown PJ qE z (1) + nE . We assume that all the noise terms nD , nE ,
(2)
in Fig. 1. nE , and n R are zero-mean and time-spatially white independent
During phase I, S broadcasts its data. In conventional schemes complex Gaussian random variables with variance 2 . We also assume
[1][3], all N relay nodes will listen to the signal, whereas in our that the jamming signals z (1) and z (2) are both complex Gaussian
scheme, the N 1 relay nodes listen and the jammer sends interfer- random variables.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO. 10, OCTOBER 2015 4895

III. J OINT S ECRECY S CHEME W ITH O PTIMAL (1)

We can see that (12) is a decreasing function of PJ , and to make
P OWER A LLOCATION (1)
information leakage as small as possible, we should let PJ = PJ
The achievable maximum secrecy rate is the measurement of the
where PJ is the maximum power constraint of the jammer. Substitut-
physical layer security ing (11) and (12) into (8) and after some simple manipulations, we
obtain the optimization problem as
Rs = max [I (yD ; s) I (yy E ; s)]+ (8)
wHR w
2 + Ps 1+w fg
w H R gg w
where [a] = max(0, a), and I(, ) is the mutual information. In our
+ max
w ,Ps a + bPs
considered problem, specifically, the destination and the eavesdrop-
H
per see an equivalent single-inputsingle-output and 1 2 single- s.t. w a cf = 0 w H a gh = 0
inputmultiple-output channel with correlated equivalent noise, w PR
w H T (Ps )w
respectively; hence, we have I(yD ; s) and I(yy E ; s) at the top of
the next page, where R f f  diag(|fR,1 |2 , |fR,2 |2 , . . . , |fR,N1 |2 ) Ps PT (13)
and similarly for R gg and R cc , respectively, and R f g  a f ga Hf g and
similarly for R cf , R gh , and R ch . We hope to achieve the maximum where a  2 + PJ |qE |2 , b  |fE |2 , T (Ps )  PsR f f + PJ R hh +
(1) (2)
secrecy rate by searching the optimal w , PJ , PJ , and Ps . 2I , PT is the source power constraint, and PR is the sum power
Remark: Most works only consider the fixed Ps case [2][13], i.e., constraint of the relay nodes.
the source node broadcasts its data with its total power.1 However, we Selection of the Jammer: From (12), the leakage rate is inverse to
will show that this may harm the secrecy, and the power allocation the jamming level. To improve the security further, we can select the
should be optimized. We will find that this will further improve the node with the largest |qE |2 as the jammer.
secrecy rate of the system greatly, as compared with the case utilizing Let H  [a acf , a gh ], and H is the projection matrix onto the null
all its power. space of H . Then the equation constraints can be transformed into
Substituting (9) and (10) into (8) w = H v where v is any vector. On the other hand, note that at the op-
timum, the power constraint w H T (Ps )w w PR should be active, i.e.,
I(yD ; s) w oH T (Ps )w
w o = PR . This is because (w w H R f gw )/(1 + w H R gg w )
 is an increasing function of the norm of w . However, the objec-
1 Psw H R f gw tive function is not an increasing function of Ps , which implies
= log 1+ (9)
2 w H R gg w )+PJ(1)w H R ghw +PJ(2) |gJ |2
2(1+w that utilizing the total power of the source may not be optimal.
Problem (13) now becomes
1
1

I(yy E ; s) = log det I + H E H H
E QE
2 vHR
v
  2 + Ps 1+vv H R
fg
gg v
(1)
2 + PJ |qE |2
(1)
PJ qE h H H max
R W cE a + bPs
QE  (10) v ,Ps
PJ c E W h Rh R W H c E
(1) H T (1) T H
PJ qE cEW hR
s.t. v H T (Ps )vv = PR , Ps PT (14)
it has been shown in [12] that the problem is a nonconvex problem
even with fixed Ps . When taking Ps as a new optimization argument, where R f g  H H R f gH , R ff +
gg  H H R gg H , T (Ps )  PsR

the problem will be more difficult to solve. Therefore, we propose a
PJ R hh + I , R f f  H R f f H , and R hh  H R hhH . This
2 H H

heuristic scheme to achieve a suboptimal but reasonably good solution. problem can be decoupled into two concatenate subproblems as
Observing (8)(10), we can see that we hope to increase I(yD ; s) follows:
as large as possible while keeping I(yy E ; s) as small as possible. 
Therefore, we can do the following three steps. vHR v
2 + Ps maxv fg
gg v
vH R
1+v
1) Design w in the null space of a cf to completely eliminate the max
Ps a + bPs
information leakage in phase II, i.e., let H a cf = 0 so that the
second row of H E in (7) can be eliminated. s.t. v T (Ps )vv = PR ,
H
Ps PT (15)
2) Design w in the null space of a gh to eliminate the interference to
the destination by the jamming signal in phase I, i.e., H a gh = 0 where inner optimization is performed over v , solely taking Ps as a
(it has been forwarded by the relay nodes in phase II). constant, and the solution of which is a function of Ps (Note that Ps
3) Since no information leakage happens in phase II (by 1)), the is an argument in T (Ps )), and the outer optimization problem is taken
jammer should stop sending interference so that D will not be over Ps . Let us first focus on the inner optimization.
(2)
jammed in phase II, i.e., PJ = 0. From the definition, T (Ps ) is a positive definite matrix, and there
exists an invertible
matrix A (Ps ) satisfying
A (Ps )H A(Ps ) = T (Ps ).
With all these considerations, (9) and (10) can be rewritten as Let v  (1/ PR )A A(Ps )vv , and v = PRA (Ps )1v. Substituting v
 into (15), we can rewrite the inner optimization as
1 Ps w H R f gw
I(yD ; s) = log 1+ (11)
2 2 1 + w H R gg w vH B (Ps )
v
max

v vH D (Ps )v
1 Ps |fE |2
I(yy E ; s) = log 1+ (1)
. (12) s.t. vH v = 1 (16)
2 2 + PJ |qE |2
where B (Ps ) = PRA (Ps )H R f gA (Ps )1 , and D (Ps ) =
1 Although the source power-allocation problem has been investigated in [1], I + PRA (Ps )H R gg A (Ps )1 . Obviously, the optimization problem
the proposed scheme is a hill-climbing method, which has no optimality (16) is a generalized eigenvalue problem. The optimal value is the
guarantee. largest eigenvalue of D (Ps )1B (Ps ) achieved at the eigenvector
4896 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO. 10, OCTOBER 2015

associated with the largest eigenvalue. Mathematically, we have the Note that f (Ps ) is the nonzero eigenvalue of a rank-1 positive
optimal vo as semidefinite matrix, it is obvious that f (Ps ) 0, and then h(Ps ) 0.

 Recalling that a > 0 and b 0, the denominator of (24) is always pos-
vo = E D (Ps )1B (Ps ) (17) itive in the range of Ps . Therefore, the positivitynegativity of g  (Ps )
depends on the numerator only. Let m(Ps ) = (a + bPs )h (Ps )
where is a scalar to normalize vo to satisfy voH vo = 1, and E [X] is
b 2 bh(Ps ). Since h(Ps ) is a real polynomial function of Ps , it
one eigenvector of matrix X associated with the largest eigenvalue.
f g is a rank-1 matrix; thus, B (Ps ) is a rank-1 matrix is continuous and its first derivative is m (Ps ) = bh (Ps ) + (a +
Note that R
bPs )h (Ps ) bh (Ps ) = (a + bPs )h (Ps ) < 0 due to h (Ps ) < 0.
as well. Therefore, matrix D (Ps )1B (Ps ) has only one nonzero
This indicates that the numerator m(Ps ) is a strictly decreasing
eigenvalue, which is the largest one. Since (18) holds, i.e.,
function. The positivitynegativity of g  (Ps ), Ps [0, PT ] has the


 following cases.
D (Ps )1B (Ps ) D (Ps )1A (Ps )H H H
af g
If m(0) 0, then g  (Ps ) is always negative, and the maximum
= D (Ps )1 PRA (Ps )H H H
af g objective value is g(0) = log( 2 /a) < 0, which suggests that the

1
 secrecy rate is 0 and it is impossible to communicate safely.
aH
f g H A (Ps ) D (Ps )1A (Ps )H H H
af g If m(PT ) 0, then g  (Ps ) is always positive so that g(Ps ) is an

1
 increasing function, and the maximum happens to be g(PT ).
= PRa H
f g H A (Ps ) D (Ps )1A (Ps )H H H
af g
Finally, if m(0) > 0 and m(PT ) < 0, which implies that g  (0) > 0,


D (Ps )1 A(Ps )H H H
af g (18) then g  (PT ) < 0. Since m(Ps ) is a strictly decreasing continuous real
function, there must be a unique point, for example, Pc (0, Ps ),
where, in the first equation, we just substitute B (Ps ) in such that m(Pc ) = 0, and for Ps Pc , m(Ps ) > 0 and for Ps Pc ,
and, in the second equation, we change the positions of the m(Ps ) < 0. Then, we conclude that, for Ps Pc , g(Ps ) is strictly
1
productions, the only nonzero eigenvalue is  PRa H f g H A (Ps ) increasing, and for Ps Pc , g(Ps ) is strictly decreasing. According
1 H H
D (Ps ) A (Ps ) H a f g , and the associated eigenvector is to the definition [17], g(Ps ) is a quasi-concave function Ps . In this
D (Ps )1 B (Ps )) = D (Ps )1 A(Ps )H H H
E(D af g . situation, Ps = Pc achieves the maximum g(Pc ) of problem (22).
We now have Since Pc is the unique real root of the equation m(Ps ) = 0, we can
  calculate Pc by using the bisection method.
= D (Ps )1A (Ps )H H H
af g
 (19) From this discussion, we can see that the optimal Ps depends on the
 positivitynegativity of m(0) and m(PT ). Only when m(PT ) 0,
vo = PRA (Ps )1D (Ps )1A (Ps )H H H
af g (20)
allocating the total power of the source is optimal. In the case that
and the maximum objective function of (16) is m(0) > 0 and m(PT ) < 0, only a part of its total power should
be allocated. This can be intuitively explained as follows. Since the
1
f (Ps )  PRa H
f g H A (Ps ) D (Ps )1A (Ps )H H H
af g legitimate channel from the source to the destination is a two-hop
relay channel, both power values Ps and PR will impact the secrecy
= PRa H 1H H a f g .
f g H (T (Ps ) + PRR gg ) (21) rate. From a traditional two-hop AF relay network without security
consideration, we know that if PR is fixed, increasing Ps will first
Substituting (21) into (15), we get the outer optimization
 increase the rate, but it will achieve a limit and will not increase any
2 + h(Ps ) more no matter how large Ps is since then the fixed PR becomes a
max g(Ps )  log
Ps a + bPs bottleneck of the two-hop channel, as shown in [16]. In the security
transmission of the AF two-hop network, however, increasing Ps too
= log( 2 + h(Ps )) log(a + bPs ) much not only will not increase the rate of the legitimate channel but
s.t. 0 Ps PT (22) also will increase the leakage rate to the eavesdropper [from (12)]
and thus harms the secrecy rate. This is the reason that Ps should not
where h(Ps )  Ps f (Ps ) = PR Psh H J (Ps )h h, h  H H always be the maximal available power PT .
a f g , and
J (Ps )  (PsR f f + PJ R hh + I + PRR gg )1 . The objective
2 After the optimal Pso = Pc has been obtained, the optimal cooper-
function is a difference of two logarithm functions, which is neither ative beamformer w o has a closed form w o = H v o where v o is in
convex nor concave, and generally difficult to solve. However, in the the form of (20). Finally, we substitute the so-obtained Pso and w o
following, we will show that it is a quasi-concave function of Ps . into (11) and (12) to get the secrecy rate. We summarize the whole
Toward this end, let us first evaluate the convexity of h(Ps ). The first algorithm in Table I.
and second derivatives of h(Ps ) to PS are

 IV. S IMULATION R ESULTS
h (Ps ) = PR h H J (Ps )h ffh
h Psh H J 2 (Ps )R
In the simulation cases, all the channel coefficients are randomly
h (Ps ) = 2PRh H J 3 (Ps )(PJ R
hh + 2I + PRR
gg )R
f f h (23)
generated in each simulation run as complex zero-mean Gaussian
respectively. The detailed derivations are given in the Appendix. random vectors with unit covariance. Noise power 2 is normalized to
Since J (Ps ) is a positive definite matrix and so as J 3 (Ps ), we have be at 0 dBm. In addition, 5000 Monte Carlo runs were done for each
h (Ps ) < 0, i.e., h(Ps ) is a concave function of Ps . point in the figures. Jammer power PJ is assumed to be PR /(N 1),
The first derivative of g(Ps ) is where N is the number of the intermediate nodes.
 In Fig. 2, we show the secrecy rate of the proposed joint opti-
 1 h (Ps ) b mization scheme and compare it with the scheme where all N nodes
g (Ps ) =
ln 2 2 + h(Ps ) a + bPs do null-space beamforming in phase II without jamming in phase I
(labeled as Relay only in the figure). We show cases with differ-
(a + bPs )h (Ps ) b 2 bh(Ps ) ent N = 8, 12, respectively. The total power of the source is PT =
= . (24)
ln 2 ( 2 + h(Ps )) (a + bPs ) 10 dBm. The x-axis PM is the total power consumed by all the relay
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO. 10, OCTOBER 2015
TABLE I
P ROPOSED A LGORITHM
Fig. 2. Secrecy rate comparison of the proposed joint scheme and the relay-
only scheme.
nodes and the jammer PM = PR + PJ . To make the comparison fair,
for the Relay only scheme, N relay nodes have the same total
power. We can see that, as the total power increases, the secrecy
rate of the proposed joint scheme has significant improvement to
the Relay only scheme. This is because, in the proposed joint
scheme, the jammer plays an important role and greatly reduces the
information leakage in phase I, whereas in the Relay only scheme,
the transmission in phase I is revealed to the eavesdropper without any
protection. In addition, we can see that, although the total power is
fixed, as N increases, the achievable secrecy rate increases as well.
This is obviously due to the power gain provided by the more relay
nodes.
Fig. 3 shows the secrecy rate enhancement of the optimized source
power allocation (labeled as Optimal Ps ) to the total power alloca-
tion (labeled as Full Ps ) of the joint scheme. We show the cases
when PM = 10, 20, and 30 dBm, respectively. It is obvious that when
4897
Fig. 3. Secrecy rate enhancement of the optimized source power allocation
to the total power allocation of the joint scheme, where PM = 10, 20, and
30 dBm.
Fig. 4. Secrecy rate enhancement of the joint scheme versus the sum power
PM , where PT = 10, 20, and 30 dBm.
PT is small, the optimal value of Ps should be the source power con-
straint PT . With PT being larger and larger, transmitting with the full
power will harm the secrecy rate, which consists in the aforementioned
interpretations. In this situation, increasing PT will not increase the
secrecy rate anymore, and the optimal Ps should be a constant, which
leads to a fixed secrecy rate. Fig. 4 shows this enhancement versus
the sum power PM when PT = 10, 20, and 30 dBm, respectively. We
can find that when PM is small, the secrecy rate of the Optimal Ps
is larger, indicating that in this situation, using full source power is
not the optimal choice. However, when PM is large enough, the total
power allocation is optimal.
In Fig. 5, we show the advantage of the jammer selection. The
x-axis is still the total power. We can see that selecting the node with
the largest |qE |2 as the jammer improves the secrecy rate further. As
the total power increases, the improvement gets smaller. This is also
reasonable since with more power, the function of jammer selection
becomes insignificant.
4898 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 64, NO. 10, OCTOBER 2015

Fig. 5. Secrecy rate improvement of the jammer selection.
V. C ONCLUSION
In this paper, we have proposed a joint cooperative beamforming,
jamming, and power-allocation scheme to enhance the security of
an AF cooperative relay network. The scheme takes both phases of
the cooperative transmissions into protection so that the secrecy rate
will be greatly improved. We show that the secrecy rate is uasi-
concave function of the power of the source so that allocating its
total power may not be optimal. The beamformer design and power
optimization problem can be solved by a bisection method, together
with a generalized eigenvalue decomposition, which has a semiclosed
form and is computationally very convenient.
A PPENDIX
Here, we will give the detailed derivation of (23). First, we have
h(Ps ) = Ps f (Ps ) = PR Psh H J (Ps )h
h, and its first derivative is
[4] H.-M. Wang, Q. Yin, and X.-G. Xia, Distributed beamforming for
physical-layer security of two-way relay networks, IEEE Trans. Signal
Process., vol. 60, no. 7, pp. 35323545, Jul. 2012.
[5] Y. Yang, Q. Li, W.-K. Ma, J. Ge, and P. C. Ching, Cooperative secure
beamforming for AF relay networks with multiple eavesdroppers, IEEE
Signal Process. Lett., vol. 20, no. 1, pp. 3539, Jan. 2013.
[6] M. Bloch, J. Barros, J. P. Vilela, and S. W. McLaughlin, Friendly jam-
ming for wireless secrecy, in Proc. IEEE ICC, Cape Town, South Africa,
2010, pp. 16.
[7] I. Krikidis, J. Thompson, and S. Mclaughlin, Relay selection for secure
cooperative networks with jamming, IEEE Trans. Wireless Commun.,
vol. 8, no. 10, pp. 50035011, Oct. 2009
[8] G. Zheng, L.-C. Choo, and K.-K. Wong, Optimal cooperative jamming
to enhance physical layer security using relays, IEEE Trans. Signal
Process., vol. 59, no. 3, pp. 13171322, Mar. 2011
[9] J. Huang and A. L. Swindlehurst, Cooperative jamming for secure com-
munications in MIMO relay networks, IEEE Trans. Signal Process.,
vol. 59, no. 10, pp. 48714884, Oct. 2011.
[10] K.-H. Park, T. Wang, and M.-S. Alouini, On the jamming power alloca-
tion for secure amplify-and-forward relaying via cooperative jamming,
IEEE. J. Sel. Areas Commun., vol, 31, no. 9, pp. 17411750, Sep. 2013.
[11] Z. Ding, M. Peng, and H.-H. Chen, A general relaying transmission
protocol for MIMO secrecy communications, IEEE Trans. Commun.,
vol. 60, no. 11, pp. 34613471, Nov. 2012.
[12] H.-M. Wang, M. Luo, X.-G. Xia, and Q. Yin, Joint cooperative beam-
forming and jamming to secure AF relay systems with individual power
constraint and no eavesdroppers CSI, IEEE Signal Process. Lett.,
vol. 20, no. 1, pp. 3942, Jan. 2013.
[13] H.-M. Wang, M. Luo, Q. Yin, and X.-G. Xia, Hybrid cooperative beam-
forming and jamming for physical-layer security of two-way relay net-
works, IEEE Trans. Inf. Forensics Security, vol. 8, no. 12, pp. 20072020,
Dec. 2013.
[14] H.-M. Wang, F. Liu, and X.-G. Xia, Joint sourcerelay precoding
and power allocation for secure amplify-and-forward MIMO relay net-
works, IEEE Trans. Inf. Forensics Security, vol. 9, no. 8, pp. 12401250,
Aug. 2014.
[15] M. Lin, J. Ge, and Y. Yang, An effective secure transmission scheme for
AF relay networks with two-hop information leakage, IEEE Commun.
Lett., vol. 17, no. 8, pp. 16761679, Aug. 2013.
[16] X. Tang and Y. Hua, Optimal design of non-regenerative MIMO wireless
relays, IEEE Trans. Wireless Commun., vol. 6, no. 4, pp. 13981407,
Apr. 2007.
[17] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.

h (Ps ) = PRh H J (Ps )h

h PR Psh H J  (Ps )h
h


ffh
h Psh H J 2 (Ps )R
= PR h H J (Ps )h

where J 2 (Ps )  J (Ps )J

J (Ps ). The second derivative of h(Ps ) is


h (Ps ) = PR h H J  (Ps )h ffh
h h H J 2 (Ps )R


J 2 (Ps ))R
PR Psh H (J ffh

= PR 2h
hH J 2 (Ps )R 2 h
f f h + 2Psh H J 3 (Ps )R
ff

hh + 2I +PRR
= 2PRh H J 3 (Ps )(PJ R gg )R
ffh

=R 2
ffR
ff .
where J 3 (Ps ) = J (Ps )J
J (Ps )J
J (Ps ), and R ff

R EFERENCES
[1] L. Dong, Z. Han, A. P. Petropulu, and H. V. Poor, Improving wire-
less physical layer security via cooperating relays, IEEE Trans. Signal
Process., vol. 58, no. 3, pp. 18751888, Mar. 2010.
[2] J. Zhang and M. C. Gursoy, Collaborative relay beamforming for secure
broadcasting, in Proc. IEEE WCNC, Princeton, NJ, USA, Apr. 2010,
pp. 16.
[3] J. Zhang and M. C. Gursoy, Relay beamforming strategies for physical-
layer security, in Proc. CISS, Princeton, NJ, USA, Mar. 2010, pp. 16.