This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access
Secrecy Rate Maximizing Beamforming Schemes
for the DF Relay Wiretap Channels
Duckdong Hwang1, Member, IEEE, Janghoon Yang2, Member, IEEE, Eunchul Yoon3, Member, IEEE,
Hyoung-Kyu Song1, Member, IEEE, and Sung Sik Nam4 , Member, IEEE,
Abstract—We consider beamforming schemes for the multi-
antenna decode-and-forward (DF) relays that maximize the
secrecy rate (SR) in the relay wiretap channel. Both full-duplex
(FD) and half-duplex (HD) relays are considered, where the
FD relay fights the loop-back self-interference while maximizing
the SR. The HD relay beamforming is derived from the FD
case as a byproduct. The facts that the weaker link among the
two hops in the DF relay network dominates the throughput
and that the first hop is shared among the destination and the
eavesdroppers make us divide the beamformer solution space
into three separate regions. Since only two of the three regions
effectively produce non-negative SR, separate beamformer de-
signs for these two effective regions are provided. Numerical
results reveal performance trade-off of FD and HD modes in
various channel configurations, where the two challenges of self-
interference suppression and the protection from eavesdropping
ask different requirements for the beamforming.
Index Terms—Secrecy rate, relay wiretap channel, decode and
forward, beamformer design.
I. I NTRODUCTION
In future networks, it is expected that much more devices
than those of today are connected to the networks to serve
various functions. Thus, the security of these new information
links puts new challenges for such network environment.
Conventional high layer security approaches may require
high cost and complexity for those numerous devices. To
protect such wireless communications from security attacks,
the physical layer security (PLS) issues draw much attention
recently [1]–[3] as a substitute for the expensive higher layer
approaches. Mostly, the PLS relies on signal processing at
the physical layer and hence it asks for new designs and
additional complexity as well. In particular, the security of the
relay wiretap channel (RWC) draws much attention recently
[4]–[11] including the security in the two-way relay channels
[12]–[16]. The beamforming for the multiple antenna relay
is helpful in improving the secrecy rate (SR) either through
protecting the message signals or through confounding the
eavesdroppers with jamming noise [5]. The SR maximization
beamforming in the multiple antenna amplify-and-forward
This work was supported by the Basic Science Research Program through
the National Research Foundation of Korea Funded by the Ministry of Science
and ICT under Grant NRF-2017R1A2B4007398.
Janghoon Yang is the corresponding author (E-mail: jhyang00@smit.ac.kr).
1 Authors are with the Department of Electronics and information, com-
munication Engineering, Sejong University.
2 Author is with Seoul Media Institute of Technology.
3 Author is with Konkuk University.
4 Author is with the Department of Electronic Engineering, Korea Univer-
sity.
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
(AF) RWC is addressed in [4]–[6], [17], [18] and it boils
down to the maximization of products of generalized Rayleigh
quotients.
For the decode-and-forward (DF) relay based wiretap chan-
nel, the outage performance of the SR with single antenna
nodes [7] and the average SR of buffer aided relaying [8]
are studied. Also, the power allocation in the multi-hop DF
relay network is considered in [19]. Recently, there have been
many research works [20]–[27] focusing on the full-duplex
(FD) relaying since it can double the spectral efficiency of the
half-duplex (HD) relaying if the strong self-interference (SI)
of FD mode can be curved sufficiently. However, the relay
beamforming problem in the DF RWC has not been addressed
enough both for the FD mode and for the HD mode. Thus,
we propose beamforming schemes for the DF RWC with these
two relay operating modes, where the FD mode relay protects
the message delivery from the security attacks and improves
the spectral efficiency in the due process through suppressing
the loop-back SI. In HD mode of DF RWC, the role of
relay beamformer becomes less demanding since it only needs
to handle the information protection from the eavesdropping
and thus the key beamforming techniques derived for the FD
mode can be applied to the HD beamforming with minor
modifications.
The RWC is composed of the first hop toward the relay
and a set of second hops toward the legitimate receiver and
eavesdroppers. With the property of DF relaying, where the
weaker link of the two hops of DF relay channel determines
the end-to-end throughput, we can divide the beamformer
solution space into three separate regions according to link
conditions. Among these three regions, only two meet the
conditions for non-negative SR and hence we provide separate
beamforming schemes for those two effective regions of
solution space. For the first region, where the strength of
the first hop link in the relay channel dominates over that
of the second hop link, we propose an optimization scheme
that maximizes the SR over the information leakage level
at which all the beamformed eavesdropper channel gains are
kept. For the other region, we show that the zero forcing of the
self-interference is the necessary condition for the optimality
and provide a beamforming scheme based on the projection
onto the orthogonal space. Numerical results are provided to
compare the performances of FD and HD schemes and identify
the conditions, where the use of computationally complex FD
operation is advocated over the HD mode.
This paper is organized as followings. In Section II, the sys-
tem model of the proposed multi-antenna RWC is presented.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

as hRD . For the k-th eavesdropper, the 1 × MT channel

vector from the relay is denoted as hkRE . Also, there exists
a MR × MT loop-back channel from the transmit antennas
to the receive antennas of the FD relay and it is denoted as
HRR (In HD relaying, the channel through HRR disappears).
The elements of these channel vectors are independent and
identically distributed (i.i.d.) CN (0, I)2 .
The source transmits the message signal x(t) (E[|x(t)|2 ] =
1) at time t with the source power PS . The FD operation at
the relay induces the self-interference through the loop-back
channel HRR so that the received signal at the MR receive
antennas of the FD DF relay is modeled as
p p
yR = PS hSR x(t) + αR PR HRR wT x(t − τ ) + nR , (1)

where wT is MT × 1 relay transmit beamformer; αR is the

path-loss in the loop-back channel; τ is the relay processing
Fig. 1. Multi-antenna full-duplex relay wiretap channel (RWC). In HD
mode, the loop-back channel HRR disappears and the relay has only a set delay; PR is the relay transmit power and the MR × 1 additive
of antennas used for both the reception and the transmission. noise vector nR is CN (0, IMR ). We assume that the relay
first applies MR × 1 receive beamformer wR to yR to make
the relay decision of the message signal x(t) and then, if
The beamforming schemes for the two effective regions of it succeeds in decoding x(t), applies wT to the re-encoded
the FD RWC are provided in subsection III-A and subsec- x(t) before the signal is sent to the antennas at time t + τ .
tion III-B then the schemes are modified for the HD case in The relay receive and transmit beamformers have unit norm
subsection III-C. After the presentation of the corroborating such that kwR k2 = 1, kwT k2 = 1. Here, we assume that
numerical results in Section IV, Section V concludes this pa- the links from source to destination and eavesdroppers are
per. The notation: AT , AH , A∗ and T r[A] are the transpose, negligible. Then, the received signals at the destination and
the Hermitian transpose, the conjugation and the trace of a the k-th eavesdropper at time (t + τ ) are
matrix A, respectively. The function eig(B, C) returns the p
generalized eigen vector of the matrix set B and C with the yD (t + τ ) = αD PR hRD wT x(t) + nD (t + τ ),
largest generalized eigen value given as λ(B, C). P⊥
Π a denotes
p
k
yE (t + τ ) = αkE PR hkRE wT x(t) + nkE (t + τ ), (2)
the projection of vector a onto the orthogonal space of the
space spanned by the columns of Π. Ik denotes the identity where nD (t + τ ) and nkE (t + τ ) are the additive noise
matrix with k × k dimensions and 0k denotes the all zero with CN (0, 1); αD and αkE are path loss terms toward the
matrix with k × k dimensions. CN (0, C) denotes the complex destination and the k-th eavesdropper, respectively.
white Gaussian random vector with zero mean vector 0 and In essence, we are given K + 1 DF relay channels, one
the covariance matrix C. The function Re{x} returns the real toward the legitimate destination and the others toward the
part of a complex number x. eavesdroppers so that the channel capacity expressions of these
K + 1 DF relay channels are given, respectively, as
II. S YSTEM M ODEL
!
Hh
|wR 2
SR | PS
CD = log 1 + min{ , α2D |hRD wT |2 PR } ,
In Fig. 1, the RWC with K eavesdroppers is shown with a α2R |wR
HH 2
RR wT | PR + 1
!
FD DF relay having MR receive antennas and MT transmit Hh
|wR 2
SR | PS
k
CE = log 1 + min{ 2
, α2E |hkRE wT |2 PR } ,
antennas (In HD relaying, the relay has a single set of antennas H
αR |wR HRR wT |2 PR +1
used for the reception and transmission. Also, the transmission k = 1, . . . , K. (3)
period is divided into two phases to separate the reception
and the transmission of the relay, which removes the loop- Here, the term α2R |wR H
HRR wT |2 PR represents the SI through
back SI at the expense of additional spectral resource.). It can the loop-back channel. When the source signal reaches the
be seen that K + 1 FD relay channels co-exist; one toward destination and the eavesdroppers, the second hop link SNRs
the destination and the others toward the eavesdroppers. The should be increased by the amount source signal contributes
source has a single transmit antenna while the eavesdropper in the first phase of the HD relaying. However in FD relaying,
and the destination are equipped with a single receive antenna, the signals from the source and the relay add up and the signal
repectively1. The MR × 1 vector channel between the source models in (2) should be two tap delay spread channels. The
and the relay is denoted as hSR and the 1 × MT vector rate analysis of such a channel is hard since it relies on the
channel between the relay and the destination is denoted Fourier transform of the channel response. From the fact that
1 Here, we assume single antenna source and destination nodes other than 2 We assume that an analogue SI cancellation is applied before the digital
the relay to focus on the relay beamforming problem. As long as rank one beamforming considered in this work so that the powers of SI and of the
transmission is considered, multi-antenna node cases can be easily handled signal from the source are within the dynamic range of signal processing
by maximum ratio transmission/reception. applied.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

TABLE I
D EFINITIONS optimization problem in (6).

PS Source power. A(wR , wT ) SINR of the first hop.

{wR , wT } = arg max CS in (5)
PR Relay power. HRE Stacked eavesdropper s.t. α2E |hkRE wT |2 PR < A(wR , wT ), k = 1, . . . , K.
channels. (6)
wR Relay receive vector. Q̃ = [Q̄, Q] MT × MT unitary
Depending on the comparisons of α2D |hRD wT |2 PR with
matrix composed of Q, A(wR , wT ) in (5), we need to search through the two regions
the orthogonal basis of in the space composed of wT and wR with two non-convex
HRE , and its expressions of CS in (5) so that we can select the best vector
orthogonal complement set among the two sets found from the two region searches.
matrix Q̄.
wT Relay transmit vector. wT = Q̃x wT is parameterized
A. The first region when α2D |hRD wT |2 PR < A(wR , wT )
by x through Q̃
αD R − D path-loss. k
CD , CE Rates at D and Ek
In this subsection, we present the first candidate set of
the relay beamformer vectors wR and wT from the first
αkE R − Ek path-loss. αR Loop-back channel
search region. In (5), the first SR expression is related only
path-loss.
to wT . When K = 1, the second hop channels of the
RWC in Fig. 1 become the multi input single output multi-
eavesdropper (MISOME) channel treated in [28] and the
the worst case eavesdropper channel dominates the SR, the optimal wT is trivially found. First, let us define ΠD = IMT +
SR of the FD DF relay channel can be defined as α2D PR hH 1 2 1 H 1
RD hRD and ΠE = IMT + αE PR (hRE ) hRE . Then,
the maximum of the first SR expression in (5) is achieved with
k
CS = min max{0, CD − CE }. (4) wT = eig(ΠD , Π1E ). However when K > 1, the second hop
k
channels of Fig. 1 is different from the MISOME channel in
Note that the beamforming presented in Section III is per- [28] and we need to find the transmit beamformer such that
formed at the relay and hence the relay is assumed to have 
1 + α2D |hRD wT |2 PR

all the required channel state information through associated max min , (7)
wT k 1 + (αkE )2 |hkRE wT |2 PR
training and learning procedures. In table I, we summarize
definitions used in the paper for the reader’s information. which neither renders a closed form solution nor is easy to
apply any optimization technique. Then, we may consider
the following equivalent optimization using a non negative
III. B EAMFORMING SCHEMES parameter γ.
1 + α2D |hRD wT |2 PR
 
In this work, we try to design wT and wR such that the max ,
SR in (4) is maximized. Let the signal to interference plus wT 1+γ
noise power ratio (SINR) of the first hop be A(wR , wT ) = (αkE )2 |hkRE wT |2 PR ≤ γ, k = 1, . . . , K. (8)
H
|wR hSR |2 PS
α2R |wRHH 2
RR wT | PR +1
. Note that the zero forcing condition
H
For a fixed γ, (8) is a quadratically constrained quadratic
(ZFC) given as wR HRR wT = 0 perfectly suppresses the SI problem (QCQP), which can be efficiently solved by stan-
so that the denominators of A(wR , wT ) in (3) becomes simply dard convex optimization tools. Unfortunately however, the
one. Due to the structure of RWC shown in Fig. 1, positive convexity of the cost function of (8) over γ is not guaranteed
SR requires the channel toward the destination to be better so that we can not be sure if any one dimensional search
off than those toward the eavesdroppers. Considering the fact over γ guarantees the optimal solution. Therefore, instead of
that the first hop is shared between the K + 1 relay channels a standard optimization routine, we take an alternative route
and the minimum operators in (3), all the second hop channels as described below.
toward the eavesdroppers should be worse than or equal to the First, suppose γ is fixed and define a matrix HRE =
first hop ((αkE )2 |hkRE wT |2 PR ≤ A(wR , wT ), k = 1, . . . , K) [(h1RE )T , . . . , (hK T
RE ) ] by stacking the channel vectors toward
to avoid negative SR. Then, we only need to care for the two the eavesdroppers. With the QR decomposition of HRE =
SR expressions from two link conditions shown in (5), which QR, where we obtain the MT × K ortho-normal basis matrix
result in positive SR values. Note that this property holds in Q of HRE . Suppose K < MT and form the MT ×MT unitary
the RWC regardless of duplexing mode. basis matrix Q̃ = [Q̄, Q] by concatenating two basis matrices,
 
1+α2 |hRD wT |2 PR
 where Q̄ is MT × (MT − K) ortho-normal basis complement
mink log 1+(αkD)2 |h k w |2 P
to the space spanned by Q. We can form the relay transmit

T R

 E RE
if α2D |hRD wT |2 PR < A(wR , wT ), 

beamformer wT = Q̃x with the unit norm weight vector x,

CS = 
α |wH H
2
wT |2 PR +1+|wRH
hSR |2 PS
 mink log (1+(αkR)2 |hRk RR | 2 P )(α2 |wH H 2
which is the concatenation of two vectors as x = [xT0 , xT1 ]T ,
w RR wT | PR +1)

RE T R
 E R R
each has size (MT −K)×1 and K ×1, respectively. Lemma 1

if α2D |hRD wT |2 PR ≥ A(wR , wT ).

(5) helps us find the closed form solution of wT if the inequality
For the optimal beamformer design, we need the following in (8) is set to the equality and γ is fixed.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

t H
Lemma 1: The x that builds wT = Q̃x maximizing the cost maximal M and the setting (wR ) = hSR /khSR k achieves
the equality constraint is found from x1 =
function in (8) with p the maximum point.
√ T
γã and x0 = 1 − γkãk2 Q̄H hH H H
RD /kQ̄ hq RD k, where With this, we are given the SR expression for the second region
the row vector a = [1/α1E , . . . , 1/αK 1 −1 in (5) as
E ] and ã = PR aR .
The γ that optimizes (8) can be found from the roots of the 1 + khSR k2 PS
 
equation CS = min log . (10)
k 1 + (αkE )2 |hkRE wT |2 PR
C(γ 2 kãk2 + 2γkãk2 − 1) = (A − B) γ(1 − γkãk2 ), (9)
p
1) When MT ≥ K + 2 and MR ≥ 3: The relay
transmit beamformer wT that maximizes the SR expression
where A = 1 + α2D PR kQ̄H hH RD k ,
2
B = of (10) is the one minimizing the denominator to one by
2 2 H H 2 T 2
αD PR [kãk kQ̄ hRD k + khRD Qã k ], and projecting hH k
RD onto the orthogonal space of hRE , k =
C = 2α2D PR kQ̄H hH RD kRe{hRD Qã }.
T
1, . . . , K so that the leakage signals to the eavesdroppers
Proof: The proof appears in Appendix A. are perfectly suppressed. With the ZFC condition, we find
It is not hard to see that equation (11) is a concave a temporary relay transmit beamformer such as wTt =
function of γ and p thus (9) possesses a single zero within P⊥ H ⊥ H
Π hRD /kPΠ hRD k, where the MT × (K + 1) matrix is given
the interval [0, 1/ 1 − γkãk2 ], where the γ ∗ maximizing the as Π = [(h1RE )T , . . . , (hK T H t
RE ) , HRR wR ]. We still need to
cost function is found. Since we do not need x in solving 2
consider the condition αD |hRD wT | PR ≥ A(wR t 2 t
, wTt ) =
equation (9), we can solve it first before the relay transmit t H 2
|(wR ) hSR | PS to be satisfied since the other condition
beamformer wT is decided with Lemma 1. Once, wT is found α2E |hkRE wTt |2 PR < A(wR t
, wTt ) is obvious. If the condition
as above, we set it as wT1 and find the receive beamformer is met, we have a second candidate set of beamformer vectors
wR maximizing A(wR , wT ), which can be realized with denoted as wT2 = wTt and wR 2
= wR t
.
1 2 1 1 H H
wR = eig(PS hSR hH SR , IMT + PR αR HRR wT (wT ) HRR ). Otherwise, we need to adjust wR such that the con-
1 1
With the found wT and wR , we need to see if the two straint is met with equality such as α2D |hRD wT |2 PR =
link conditions α2D |hRD wT1 |2 PR < A(wR 1
, wT1 ) and γ ∗ = A(wR , wT ) = |wR H
hSR |2 PS instead of pursuing maxi-
k 2 k 1 2 1 1
(αE ) |hRE wT | PR ≤ A(wR , wT ) are met. If those two mization of the numerator of the second q SR expression
t |2 P
conditions are met, we have one candidate set of beamformer in (5). For this, we can define ϑ =
α2D |hRD wT R
.
|(wRt )H h
SR | PS
2
vectors denoted as wT1 and wR 1
. Otherwise, we need to reduce
∗ 1 Then, we form the second receive beamformer vector as
γ from γ so that the second condition γ ≤ A(wR , wT1 ) t a t
∗ wR = ϑwR + ηwR , where wR is found as above and the
is met. Note that reducing γ from γ found above reduces a
auxiliary vector wR from the orthogonal space of the two
|hRD wT1 |2 as well since the cost function is increasing on the
column matrix ∆ = [hSR , HRR wTt ] is found from wR a
=
interval [0, γ ∗ ). For every reduced γ, we find new wT1 with ⊥ ⊥ 3
1 1 P∆ hRD /kP∆ hRDp k . To make kwR k = 1, the parameter η is
Lemma 1 and wR again, which leads us to a new A(wR , wT1 ).
set such that η = ϑ2 cos2 (Θ) + 1 − ϑ2 − ϑ cos(Θ), where
This process is repeated until the two link conditions are met.
Θ = cos−1 (|(wR t H a
) wR |). It is easy to see that wR H
HRR wTt =
2 t 2 H 2
0 and αD |hRD wT | PR = |(wR ) hSR | PS . We, then, form
B. The second region when α2D |hRD wT |2 PR ≥ A(wR , wT ) the second candidate beamformer set as wT2 = wTt and
2
wR = wR .
Now, we present the second candidate set of the relay 2) When MT ≤ (K + 1): If MT ≤ (K + 1), the MT ×
beamformer vectors from the second search region. The ar- (K + 1) matrix Π is a full column rank matrix and hence the
gument of logarithm in the second SR expressions of (5) projection onto orthogonal space of Π will null out wTt . To
α2 |wH H wT |2 PR +1+|wR H
hSR |2 PS
is (1+(αkR)2 |hRk RR
wT | 2 P )(α2 |wH H
R RR wT |2 PR +1)
. Since the term bypass such a situation, we have to allow a portion of leakage
E RE R R
k 2 k 2 signal toward K − MT + 2 of the eavesdroppers (k, k ∈
1+(αE ) |hRE wT | PR related to the k-th eavesdropper chan-
nel from the denominator depends only on wT , we ignore this {1, . . . , K}) by striking out the corresponding columns from
H
|wR hSR |2 PS the matrix Π and making Π a MT × (MT − 1) matrix. In this
term. Then, we are left with M = 1 + α2 |wH H RR wT |2 PR +1
=
R R
1 + A(wR , wT ), which reveals the relation between wR and case, the selection of the K − MT + 2 columns hkRE , k ∈
wT as in Lemma 2. {1, . . . , K} should be made so that the CS in (10) should
Lemma 2: The M is maximized when the ZFC for the SI be maximized. Such objective is achieved if we select the
(wRH
HRR wT = 0) is met with a temporary wR as (wR t H
) = K − MT + 2 eavesdroppers minimizing (αkE )2 |hkRE hH RD | .
2

hSR /khSR k so that we have the maximal M as M = 1 + The adjustment of wR can be carried out in the same way as
khSR k2 PS . above if MR ≥ 3.
Proof: Given wT , the optimal wR 3) When MR = 2: The MR × 2 matrix ∆ will null
a a
maximizing M is found from wR H
= out wR if MR = 2. Instead, we have to form wR =
⊥ H ⊥ H
eig(hSR hH 2 H H PhSR hRD /kPhSR hRD k with a violation of ZFC and hence
SR PS , αR HRR wT wT HRR PR + IMR ), which
results in M = 1 + λ(hSR hH 2 H H with reduction of the first hop SINR A(wR , wT ). The beam-
SR PS , αR HRR wT wT HRR PR +
IMR ). Therefore, a necessary condition for the former wT2 can be found in the same way as above if MT ≥ 3.
optimal wT for the maximal M is to maximize 3 For the auxiliary vector wa , only the fact that it lies in the orthogonal
2
λ(hSR hH H H
SR PS , αR HRR wT wT HRR PR + IMR ). Now, it
R
space of the columns of Π is important while the direction of it does not
is trivial to see the ZFC is a necessary condition for the matter. Hence, hRD is used for wR a here.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

From the two candidate beamformer sets found from the 3.5
two constraint regions, we choose the set with the better SR FD, PS=10, PR=100

value. 3
FD PS=100, PR=10
FD P =100, P =100
S R
HD PS=10, PR=100
C. Half-duplex relay beamforming 2.5
HD P =100, P =10
S R
HD P =100, P =100
In this subsection, we modify the beamforming methods S R

SR(bps/Hz)
2
provided for the FD DF RWC in the previous subsections so
that they work for the HD DF RWC. Here, the relay has only 1.5
one antenna set, which receives the signal from the source
in the first phase and transmits the decoded and remodulated 1

message signal toward the destination in the second phase.

Given that all the other parameters involved with the channel 0.5

models are the same as in Section II, the channel capacity

expressions for the destination and the eavesdropper in this 0
−10 −5 0 5 10 15 20
H
channel model are 1/2 of those in (3) with wR HRR wT = 0 2
10log10(αD/αE)
2

and the first hop SNR expression A(wR ) = |wR H

hSR |2 PS .
To avoid the non-positive SR values, we need all the second Fig. 2. The SR comparison of the FD mode and the HD mode in the DF RWC
against α2D /α2E when α2E = 0.1. Here, α2R = 1, MT = MR = 4, K = 2
hop channels toward the eavesdroppers to be worse than the for the FD mode and the number of antennas is 4 for the HD mode.
fist hop ((αkE )2 |hkRE wT |2 PR < A(wR ), k = 1, . . . , K) . For
such cases, we have the SR expressions as 1/2 of those in (7).
The beamformer optimization procedures in subsection III-A 2.5 FD α2 /α2 =0.1
D E
can be applied since the self-interference does not matter in FD α2 /α2 =1
D E

HD relaying. 2 2
FD αD/αE=10

The first SR expression is maximized with wT1 in sub- 2

HD α2 /α2 =0.1
D E

section III-A and wR = hSR /khSR k, which maximizes 2 2

HD αD/αE=1

A(wR ). If these two found beamformers satisfy the conditions 1.5

HD α2 /α2 =10
D E
SR(bps/Hz)

|hkRE wT |2 PR < A(wR ) and |hRD wT |2 PR < A(wR ), we

set them as wT1 and wR 1
. Otherwise, we apply the γ reducing
process in the last paragraph of subsection III-A though it 1

does not affect wR and hence A(wR ) either. The second SR

expression becomes as (10) with a multiplication by 1/2 and
0.5
hence the same methods in subsection III-B ca be applied. We,
again, form the second candidate beamformer set as wT2 = wTt
2
and wR = wR . We make a similar decision process among 0
−20 −15 −10 −5 0 5 10
1
the two sets (wR , wT1 ) and (wR 2
, wT2 ) as in subsection III-B. 10log10(PR/PS)

IV. N UMERICAL R ESULTS
In this section, we compare the performance of the proposed
beamforming schemes for the FD RWC and the HD RWC. In
Fig. 2, the SR of the two modes for three different pairs of
PS and PR value are plotted against the ratio α2D /α2E , where
the channel becomes a strong wiretap channel as the ratio
α2D /α2E decreases and the other way around. Note that the SR
gains of the FD mode over the HD mode widen as α2D /α2E
increases, which tells us that the FD mode suffers more in the
throughput as the requirement for secrecy becomes stronger.
Regardless of the pair values PS and PR , the gaps narrow
down fast in the other direction of α2D /α2E so that the FD
mode provides not much benefit over the HD mode in strong
eavesdropper situations. Therefore, protecting the legitimate
transmission from the eavesdroppers and acquiring the FD gain
at the same time is a hard task.
Fig. 3 compares the SR of the two modes against the ratio
PR /PS for different pairs of α2D and α2E . Here, PS is set
to 20 (dBW ) while the value of PR varies. As PR increases,
the second hops (toward the destination and the eavesdroppers)
dominate over the first hop so that the second SR expression
Fig. 3. The SR comparison of the FD mode and the HD mode in the DF
RWC against PR /PS when PS = 20 (dBW ). Here, α2R = 1, MT =
MR = 4, K = 2 for the FD mode and the number of antennas is 4 for the
HD mode.
in (5) becomes prevalent and the beamforming scheme in
subsection III-B is applied most of the time. In Fig. 3, it
is shown that the SR of FD mode degrades in this regime
of PR /PS and even performs worse than the HD counterpart
for all the three pairs of α2D and α2E . This failure of the FD
mode (with respect to the HD mode) mainly comes from the
fact that the ZFC for the SI suppression in the beamforming
in subsection III-B incurs painful performance loss that can
not be overcome even by the spectral gain of two coming
from the FD operation. This loss becomes severer when the
eavesdropper channels get stronger than the legitimate channel
toward the destination. From the above numerical results,
we conclude that the benefit of FD operation over the HD
mode can be limited both by the relative intensity of the
eavesdropper channels and by the relative strength of the two
hops.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

1.8
three pairs of destination/eavesdropper channel strength ratio
FD PS=10,PR=100 α2D /α2E compared to the proposed scheme. The degradation of
1.6 FD PS=100,PR=10 SR is severe when PR /PS is low in all pairs of α2D /α2E while
FD PS=100,PR=100
1.4 the difference becomes negligible as PR increases, where the
HD PS=10,PR=100
HD PS=100,PR=10
second SR expression in (5) becomes prevalent as is observed
1.2
HD PS=100,PR=100 in the results of Fig. 3. Also, the SR degradation reduces as
SR(bps/Hz)

1 the eavesdropper channel gets weaker. From the above results,

we conclude that the proposed FD scheme has advantage over
0.8
the ZF based FD scheme when the relay is not allowed to
0.6 use enough power and the eavesdropper channels are stronger
0.4
than the legitimate user channel.

0.2 V. C ONCLUSION
0 Beamforming schemes for the multi-antenna DF RWC that
1 2 3 4 5 6
K maximize the secrecy rate are provided for the FD mode and
the HD mode. The solution space for the beamformer vectors
Fig. 4. The SR comparison of the FD mode and the HD mode in the DF is divided into three regions and the beamformer sets for the
RWC against K, the number of eavesdroppers when α2D = α2E = 0.1. Here, two effective regions are found. From these two sets, the one
α2R = 1, MT = MR = 8 for the FD mode and the number of antennas is providing the best performance is selected. Numerical results
8 for the HD mode.
compare the performances of FD mode and HD mode with the
proposed designs and identify the conditions under which the
2
2 2
FD mode provides the SR benefit over the HD mode. Also, a
FD αD/αE=0.1

1.8 FD α2 /α2 =1
comparison with a ZF based FD scheme shows the condition
D E
2 2
FD αD/αE=10 under which the proposed FD scheme provides advantage.
1.6 2 2
ZF αD/αE=0.1
2 2
Extension of current work to multi-rank transmission case or
ZF αD/αE=1
1.4
ZF α2 /α2 =10
D E
multiple relay case can be considered in future. Also, the case
1.2
with imperfect channel knowledge for the eavesdroppers could
address more practical assumption.
SR(bps/Hz)

0.8 A PPENDIX A
P ROOF OF L EMMA 1
0.6

To meet the condition (αkE )2 |hkRE wT |2 PR ≤ γ, k =

0.4
1, . . . , K, the k-th element of q the vector wTT HRE has the
0.2
magnitude less than or equal to (αk γ)2 PR . The constraint in
E q
0
−20 −15 −10 −5
10log10(PR/PS)
0 5 10
(8) can be met with the equality if we have xT1 R = PγR a.
q
γ −1
Then, we are given xT1 = PR aR or the elements of
Fig. 5. The SR comparison of the proposed FD scheme and the ZF based x1 can be found sequentially using the q triangular structure
FD scheme in the DF RWC against PR /PS when PS = 20 (dBW ). Here, 1 −1
α2R = 1, MT = MR = 4, K = 2 for both schemes. of K × K matrix R. Note that ã = PR aR can be
T √
determined independent of γ and thus x1 = γã scales
the found vector ã accordingly with the magnitude relation
Fig. 4 compares the SRs of the FD mode and the HD kx1 k2 = γkãk2 . To maximize the cost function of (8), we
mode in the DF RWC when the number of eavesdroppers find the direction of x0 from the projection of hRD onto
varies. As the number of eavesdroppers increases and asks the the space spanned by Q̄ so that x0 is a scaled version of
beamformer more burden of hiding the information leakage x̃0 if x̃0 = Q̄H hH RD . Once, x̃0 is found it should be re-
toward them, more loss of the SR is inevitable. However, the scaled to meet kx0 k2 + p kx1 k2 = 1 ⇔ kx0 k2 = 1 −
2
simulation shows the SR degradation is modest in both FD γkãk . Therefore, x0 = 1 − γkãk2 Q̄H hH H H
RD /kQ̄ hRD k.
2
and HD modes. Moreover, the trend of SRs in the two modes Note that we need p γ ≤ 1/kãk to make the argument of
is preserved so that the SR of FD mode outperforms that of the square root 1 − γkãk2 be non negative. From the x0
HD mode consistently in this case. found above, we know hRD wT = hRD [Q̄, Q][x0 , x1 ]T =
p √
Finally, Fig. 5 compares the secrecy rates of the proposed 1 − γkãk2 kQ̄H hH T
RD k + γhRD Qã so that the cost func-
FD scheme and the ZF based FD scheme against the ratio tion in (8) can be written as
PR /PS with the same setting of Fig. 3. Note that the ZF p √
1 + α2D k 1 − γkãk2 kQ̄H hH RD k + γhRD QãT k2 PR
based FD scheme applies the scheme in subsection III-B
1+γ
regardless of the link conditions and hence it does not divide p
the solution space like the proposed FD scheme. In Fig. 5, it A + γB + γ(1 − γkãk2 )C
= . (11)
is shown that the SR of ZF based FD scheme suffers in all the 1+γ

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2884202, IEEE
Access

Now, the extreme points of (11) can be attained by differentia- [22] D. Nguyen, L. Tran, P. Pirinen, and M. Latva-aho, “On the spectral
tion with respect to γ (11) and setting it to zero, which results efficiency of full-duplex small cell wireless systems,” IEEE Trans.
Wireless Comm., vol. 13, no. 9, pp. 4896–4909, Sep. 2014.
in (9). [23] M. Heino, D. Korpi, T. Huusari, E. Antonio-Rodriiguez, S. Venkatasub-
ramanian, T. Riihonen, L. Anttila, C. Icheln, K. Haneda, R. Wichman,
and M. Valkama, “Recent advances in antenna design and interference
R EFERENCES cancellation algorithms for in-band full duplex relays,” IEEE Commun.
Magazine, vol. 53, no. 5, pp. 91–101, May 2015.
[1] H. V. Poor, “Information and inference in the wireless physical layer,” [24] G. Liu, F. R. Yu, H. Ji, V. C. M. Leung, and X. Li, “In-band full-duplex
IEEE Wireless Commun., vol. 19, no. 7, pp. 40–47, Feb. 2012. relaying: A survey, research issues and challenges,” IEEE Commun.
[2] Y.-W. P. Hong, P. Lan, and C.-C. J. Kuo, “Enhancing physical-layer Surveys and Tutorials, vol. 17, no. 2, pp. 500–524, 2015.
securecy in multiantenna wireless systems,” IEEE Signal Processing [25] S. Han, C. Yang, and P. Chen, “Full duplex-assisted intercell interference
Magazine, vol. 30, no. 5, pp. 29–40, Sep. 2013. cancellation in heterogeneous networks,” IEEE Trans. Commun., vol. 63,
[3] Z. Chu, H. Xing, M. Johnston, and S. L. Goff, “Secrecy rate op- no. 12, pp. 5218–5234, Dec. 2015.
timizations for a MISO secrecy channel with multiple multiantenna [26] D. Hwang, J. Yang, and S. S. Nam, “SINR maximizing beamforming
eavesdroppers,” IEEE Trans. Wireless Comm., vol. 15, no. 1, pp. 283– schemes for the full duplex amplify-and-forward relay channel,” IEEE
297, Jan. 2016. Access, vol. 5, no. 1, pp. 18 987–18 998, Dec. 2017.
[4] L. Dong, Z. Han, A. P. Petropulu, and H. V. Poor, “Improving wireless [27] D. Hwang, S. S. Nam, and J. Yang, “Multi-antenna beamforming tech-
physical layer security via cooperating relays,” IEEE Trans. on Signal niques in full-duplex and self-energy recycling systems: Opportunities
Processing, vol. 58, no. 3, pp. 1875–1888, Mar. 2010. and challenges,” IEEE Commun. Magazine, vol. 10, pp. 160–167, Oct.
[5] M. Lin, J. Ge, and Y. Yang, “An effective secure transmission scheme for 2017.
AF relay networks with two-hop information leakage,” IEEE Commun. [28] A. Khisti and G. W. Wornell, “Secure transmission with multiple
Lett., vol. 17, no. 8, pp. 1676–1679, Aug. 2013. antennas I: The MISOME wiretap channel,” IEEE Trans. Inform. Theory,
[6] Y. Yang, Q. Li, W. Ma, J. Ge, and P. Ching, “Cooperative secure vol. 56, no. 7, pp. 3088–3104, Jul. 2010.
beamforming for AF relay netowrks with multiple eavesdroppers,” IEEE
SignalProcessing Lett., vol. 20, no. 1, pp. 35–38, Jan. 2013.
[7] G. Chen, Y. Gong, P. Xiao, and J. A. Chambers, “Physical layer network
security in the full-duplex relay system,” IEEE Trans. Information
Forensics ans security, vol. 10, no. 3, pp. 574–583, Mar. 2015.
[8] A. E. Shafie, A. Sultan, and N. Al-Dhahir, “Physical-layer security of a
buffer-aided full-duplex relaying system,” IEEE Commun. Lett., vol. 20,
no. 9, pp. 1856–1859, Sep. 2016.
[9] B. Fang, Z. Quan, W. Zhong, and W. Shao, “Precoder design for secrecy
maximization in the DF-based MIMO relay networks,” in International
Conference on Wireless Communications & Signal Processing (WCSP),
Oct. 2015.
[10] X. Chen, L. Lei, H. Zhang, and C. Yuen, “Large-scale MIMO relaying
techniques for physical layer security:AF or DF?” IEEE Trans. Wireless
Comm., vol. 14, no. 9, pp. 5135–5146, Sep. 2015.
[11] A. E. Shafie, D. Niyato, and N. Al-Dhahir, “Artificail-noise-aided secure
MIMO full-duplex relay channels with fixed-power transmission,” IEEE
Commun. Lett., vol. 20, no. 8, pp. 1591–1594, Aug. 2016.
[12] H. Wang, Q. Yin, and X. Xia, “Distributed beamforming for physical-
layer security of two-way relay networks,” IEEE Trans. on Signal
Processing, vol. 60, no. 7, pp. 3532–3545, Jul. 2012.
[13] H. Wang, M. Luo, Q. Yin, and X. Xia, “Hybrid cooperative beamforming
and jamming for physical-layer security of two-way relay networks,”
IEEE Trans. Information Forensics ans security, vol. 8, no. 12, pp. 2007–
2020, Dec. 2013.
[14] C. Zhang, H. Gao, T. Lv, Y. Lu, and X. Su, “Beamforming for
secure two-way relay networks with physical layer network coding,”
in Globecom 2014, Dec. 2014, pp. 1734–1739.
[15] Y. Yang, C. Sun, H. Zhao, H. Long, and W. Wang, “Algorithms
for secrecy guarantee with null space beamforming in two-way relay
networks,” IEEE Trans. on Signal Processing, vol. 62, no. 8, pp. 2111–
2126, Apr. 2014.
[16] Q. Li, W.-K. Ma, and D. Han, “Sum secrecy rate maximization for
full-duplex two-way relay networks using Alamouti-based rank-two
beamforming,” IEEE J. Select. Topics in Signal Process., vol. 10, no. 8,
pp. 1359–1373, Dec. 2016.
[17] Y. Gao, Y. Cai, Q. Shi, B. Champagne, and M. Zhao, “Joint transceiver
designs for secure communications over MIMO relay,” in IEEE In-
ternational Conference on Acoustics, Speech and Signal Processing
(ICASSP), Mar. 2016.
[18] X. Wang, Z. Zhang, and K. Long, “Secure beamforming for multiple-
antenna amplify-and-forward relay networks,” IEEE Trans. on Signal
Processing, vol. 64, no. 6, pp. 1477–1492, Mar. 2016.
[19] J. Lee, “Optiaml power alloation for physical layer security in multi-hop
DF relay networks,” IEEE Trans. Wireless Comm., vol. 15, no. 1, pp.
28–38, Jan. 2016.
[20] T. Riihonen, S. Werner, and R. Wichman, “Mitigation of loopback
self-interference in full-duplex MIMO relays,” IEEE Trans. on Signal
Processing, vol. 59, no. 12, pp. 5983–5993, Dec. 2011.
[21] H. A. Suraweera, I. Krikidis, G. Zheng, C. Yuen, and P. J. Smith, “Low-
complexity end-to-end performance optimization in MIMO full-duplex
relay systems,” IEEE Trans. Wireless Comm., vol. 13, no. 2, pp. 913–
9278, Feb. 2014.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.