Professional Documents
Culture Documents
iD1
H
T
2r
i
i
(t
2
)W
i
(t
2
)H
1r
i
, z
2
=
L
iD1
_
H
T
2r
i
i
(t
2
)W
i
(t
2
)n
r
i
(t
1
)
_
n
2
, n
2
is the additive
Gaussian noise vector at T
2
and
i
is a diagonal binary
variable matrix denoting whether the k
th
antenna at the i
th
relay is active or not and its diagonal elements are given by
c
k
i
=
_
1 if the k
th
antenna at the i
th
relay is active,
0 otherwise
(4)
Similarly, during the third time slot t
3
, T
2
transmits its
signal x
2
to the relays with a diagonal matrix power
denoted P
2
(t
3
) with elements
_
P
1
2
(t
3
), . . . , P
M
T
2
2
(t
3
)
_
.
Finally, in the fourth time slot t
4
, the selected relays broad-
cast the amplied signals to T
2
with a diagonal matrix
power P
r
i
(t
4
) with elements
_
P
1
r
i
(t
4
), . . . , P
M
R
i
r
i
(t
4
)
_
. The
received signal at terminal T
1
is given by
y
1
(t
4
) = A
1
(H)x
2
z
1
(5)
where A
1
(H) =
L
iD1
H
T
1r
i
i
(t
4
)W
i
(t
4
)H
2r
i
, z
1
=
L
iD1
_
H
T
1r
i
i
(t
4
)W
i
(t
4
)n
r
i
(t
3
)
_
n
1
and n
1
is the addi-
tive Gaussian noise vector at T
1
.
The noise covariance matrix for OWR transmission at T
m
can be given as
C
z
m
= Ez
m
z
H
m
|
= o
2
n
L
iD1
H
T
mr
i
i
W
i
_
H
T
mr
i
i
W
i
_
H
o
2
n
I
(6)
where
i
and W
i
can be determined from the context.
2.2. Two-way relaying
Figure 2 illustrates a system model of TWR-CR networks.
During the rst time slot, known also as the multiple-access
channel phase, T
1
and T
2
transmit their signals x
1
and x
2
to the relays simultaneously, with a power denoted P
1
, and
P
2
, respectively. In the second time slot, known also as the
broadcast channel phase, the selected relays transmit the
amplied signals to the terminals, with a power denoted
P
r
i
, where i = 1, . . . , L.
In the rst time slot, the complex baseband received
signal at the i
th
relay is given by
y
r
i
(t
1
) = H
1r
i
x
1
H
2r
i
x
2
n
r
i
(7)
During the second time slot, each relay amplies y
r
i
by
multiplying it by W
i
and broadcasts it to the terminals T
1
and T
2
. The received signals in the broadcast channel phase
are given as
Figure 2. Cooperative communication multiple-input multiple-
output system under cognitive radio (CR) scenario for two-way-
relaying-CR networks.
y
1
(t
2
) =
A
1
(H)x
1
Self Interference
A
1
(H)x
2
n
1
(8)
y
2
(t
2
) = A
2
(H)x
1
A
2
(H)x
2
Self Interference
n
2
(9)
respectively, where
A
1
(H) =
L
iD1
H
T
1r
i
i
(t
2
)W
i
(t
2
)H
2r
i
,
A
2
(H) =
L
iD1
H
T
2r
i
i
(t
2
)W
i
(t
2
)H
1r
i
,
A
1
(H) =
L
iD1
H
T
1r
i
i
(t
2
)W
i
(t
2
)H
1r
i
,
A
2
(H) =
L
iD1
H
T
2r
i
i
(t
2
)W
i
(t
2
)H
2r
i
,
n
1
=
L
iD1
_
H
T
1r
i
i
(t
2
)W
i
(t
2
)n
r
i
(t
1
)
_
n
1
and
n
2
=
L
iD1
_
H
T
2r
i
i
(t
2
)W
i
(t
2
)n
r
i
(t
1
)
_
n
2
By using the available knowledge of the CSI that might
be erroneous, the terminals can remove the estimated
self interference. Thus, the received signals at T
1
and T
2
becomes
r
1
= y
1
(t
2
)
A
1
_
H
_
x
1
= A
1
(H)x
2
z
1
(10)
r
2
= y
2
(t
2
)
A
2
_
H
_
x
2
= A
2
(H)x
1
z
2
(11)
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
where z
1
= n
1
_
A
1
(H)
A
1
_
H
__
x
1
and z
2
= n
2
_
A
2
(H)
A
2
_
H
__
x
2
. The noise covariance matrix for
TWR transmission at T
m
can be given as
C
z
m
= E
_
z
m
z
H
m
_
= o
2
n
L
iD1
H
T
mr
i
i
W
i
_
H
T
mr
i
i
W
i
_
H
o
2
n
I
_
A
m
(H)
A
m
_
H
__
C
x
m
_
A
m
(H)
A
m
_
H
__
H
(12)
where
i
and W
i
can be determined from the context.
Note that if perfect CSI is available at the terminals,
A
m
(H) =
A
m
_
H
_
and the noise covariance matrix
becomes dependent only on the noise variance.
3. MULTIPLE ANTENNARELAY
SELECTION AND PROBLEM
FORMULATION
In this section, we formulate the optimization problems
that maximise the secondary sum rate for both OWR-CR
and TWR-CR networks without affecting the QoS of the
PUs. For simplicity, we assume that M
R
i
= M
R
, Vi =
1, . . . , L, i.e. all the relays equipped with the same number
of antennas.
3.1. One-way relaying
The relay power at the k
th
antenna of the i
th
relay can be
expressed as
P
k
r
i
(t
2
) = E
_
w
k
i
(t
2
)y
k
r
i
2
_
=
_
_
M
T
1
zD1
P
z
1
h
kz
1r
i
2
o
2
n
_
_
w
k
i
(t
2
)
2
(13)
From Equation (13), the value of
w
k
i
(t
2
)
can be expressed
as
w
k
i
(t
2
)
P
k
r
i
(t
2
)
M
T
1
zD1
P
z
1
h
kz
1r
i
2
o
2
n
(14)
Similarly, the value of [w
k
i
(t
4
)[ can be expressed as
w
k
i
(t
4
)
P
k
r
i
(t
4
)
M
T
2
zD1
P
z
2
h
kz
2r
i
2
o
2
n
(15)
Thus, the sum rate of the MIMO-OWR can be written as
R
.OWR/
(H) =
1
4
log
2
_
det
_
I
_
A
2
(H)P
1
A
H
2
(H)
_
C
z
1
2
__
1
4
log
2
_
det
_
I
_
A
1
(H)P
2
A
H
1
(H)
_
C
z
1
1
__
(16)
where the factor
1
4
is due to the four time slots that are
needed to accomplish the OWR transmission. Therefore,
the sum rate maximisation problem of OWR-CR multiple
antennarelay selection can be formulated as follows
maximise
P
1
.t
1
/,P
r
.t
2
/,P
2
.t
3
/,P
r
.t
4
/,
V
.t
2
/,
V
.t
4
/
R
.OWR/
_
H
_
(17)
s.t 0 6
M
T
1
vD1
P
v
1
6
P, 0 6
M
T
2
uD1
P
u
2
6
P (18)
0 6
M
R
kD1
P
k
r
i
(t
s
) 6
P
r
, Vi = 1, .., L, s = {2, 4] (19)
M
T
1
vD1
M
PU
jD1
P
v
1
h
jv
1p
2
6 I
th
,
M
T
2
uD1
M
PU
jD1
P
u
2
h
ju
2p
2
6 I
th
(20)
L
iD1
M
PU
jD1
M
R
kD1
c
k
i
(t
s
)P
k
r
i
(t
s
)
h
jk
r
i
p
2
6 I
th
, s = {2, 4] (21)
c
k
i
(t
s
) {0, 1], Vi = 1, .., L, Vk = 1, .., M
R
, s = {2, 4]
(22)
where
V
(t
s
) =
_
c
1
1
(t
s
), .., c
M
R
1
(t
s
), .., c
1
L
(t
s
), .., c
M
R
L
(t
s
)
_
and P
r
(t
s
) =
_
P
1
r
1
(t
s
), .., P
M
R
r
1
(t
s
), .., P
1
r
L
(t
s
), .., P
M
R
r
L
(t
s
)
_
,
are the decision variable vectors of our formulated opti-
mization problem that contain the state and the transmit
power vector of each relay for the second and fourth time
slots, respectively. The constraints (18) and (19) represent
the power budget constraints at the terminals and at the
relays, respectively, while the constraints (20) and (21) rep-
resent the average interference constraints imposed to the
terminals and relays, respectively.
3.2. Two-way relaying
The relay power at the k
th
antenna of the i
th
relay can be
expressed as
P
k
r
i
(t
2
) = E
_
w
k
i
(t
2
)y
k
r
i
2
_
=
_
_
M
T
1
zD1
P
z
1
h
kz
1r
i
M
T
2
zD1
P
z
2
h
kz
2r
i
2
o
2
n
_
_
w
k
i
(t
2
)
2
(23)
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
From Equation (23), the value of
w
k
i
can be expressed as
w
k
i
(t
2
)
P
k
r
i
(t
2
)
M
T
1
zD1
P
z
1
h
kz
1r
i
M
T
2
zD1
P
z
2
h
kz
2r
i
2
o
2
n
(24)
Thus, the sum rate of the MIMO-TWR can be written as
R
.TWR/
(H) =
1
2
log
2
_
det
_
I
_
A
2
(H)P
1
A
H
2
(H)
_
C
z
1
2
__
1
2
log
2
_
det
_
I
_
A
1
(H)P
2
A
H
1
(H)
_
C
z
1
1
__
(25)
where the factor
1
2
is due to the two time slots that are
needed to accomplish the TWR transmission. Thus, the
sumrate maximisation problemof TWR-CRmultiple relay
selection can now be formulated as
maximise
P
1
.t
1
/,P
2
.t
1
/,P
r
.t
2
/,
V
.t
2
/
R
.TWR/
_
H
_
(26)
s.t 0 6
M
T
1
vD1
P
v
1
6
P, 0 6
M
T
2
uD1
P
u
2
6
P (27)
0 6
M
R
kD1
P
k
r
i
(t
2
) 6
P
r
, Vi = 1, .., L (28)
M
T
1
vD1
M
PU
jD1
P
v
1
h
jv
1p
M
T
2
uD1
M
PU
jD1
P
u
2
h
ju
2p
2
6 I
th
(29)
L
iD1
M
PU
jD1
M
R
kD1
c
k
i
P
k
r
i
(t
2
)
h
jk
r
i
p
2
6 I
th
(30)
c
k
i
(t
2
) {0, 1], Vi = 1, .., L, Vk = 1, .., M
R
(31)
where the constraints (29) and (30) represent the average
interference constraint in the rst and second time slots,
respectively.
4. MULTIPLE ANTENNARELAY
SELECTION ALGORITHMS
The optimal solution using continuous power distribu-
tion for our nonlinear optimization problems formulated
in Section 3 are difcult to nd because of the exis-
tence of binary variables c
k
i
, where i = 1, . . . , L and
k = 1, . . . , M
R
[28]. Therefore, we employ heuristic
approaches to nd suboptimal solutions to the problems.
For simplicity, we handle this problem by solving it in
a time slot per time slot fashion for both OWR and
TWR transmissions.
4.1. Quantisation and relay
power distributions
In this section, we propose to use a quantization set
with discrete number of power levels from zero to
the peak relay antenna power (i.e. it is assumed that
the peak power budget allocated at the relays is uni-
formly distributed at each antenna;
P
a
r
=
N
P
r
M
R
). In
fact, each antenna at the relay can transmit the ampli-
ed signal using one of the power level between 0 and
P
a
r
_
P
a
r
i
S =
_
0,
N
P
a
r
N1
,
2
N
P
a
r
N1
, . . . ,
.N2/
N
P
a
r
N1
,
P
a
r
__
, where N
is the number of quantization levels. By this way, cognitive
relays have more exibility to allocate their powers in the
case where continuous power distribution is not available,
which is the case of actual existing systems. This method
is considered as a generalisation of the ONOFF mode
where antennas can either transmit or keep silent. There-
fore, our goal is to nd the optimal power allocation and
antennarelay selection at the relay side. We assume that
the terminal powers at each antenna are equal.
For OWR transmission, the power allocated at the
each antenna of both terminals depends essentially on
two constraints: the peak power constraint (18) and the
interference constraint (20), and their optimal values are
given by
P
a
1
= min
_
_
_
_
_
_
I
th
M
T
1
vD1
M
PU
jD1
h
jv
1p
2
,
P
a
1
_
_
_
_
_
_
(32)
P
a
2
= min
_
_
_
_
_
_
I
th
M
T
2
uD1
M
PU
jD1
h
ju
2p
2
,
P
a
2
_
_
_
_
_
_
(33)
where
P
a
1
=
N
P
M
T
1
and
P
a
2
=
N
P
M
T
2
are the antenna peak
power at antenna a associated with T
1
and T
2
, respectively.
The resulting simplied sum rate maximisation prob-
lem of OWR-CR multiple relay selection can now be
formulated as
maximise
P
r
.t
2
/,P
r
.t
4
/,
V
.t
2
/,
V
.t
4
/
R
.OWR/
_
H
_
(34)
s.t (19), (21), (22) (35)
Similarly, the optimal power allocated at the each
antenna of T
1
and T
2
for TWR transmission can be given
as
P
a
c
= min
_
_
_
_
_
_
I
th
M
T
1
vD1
M
PU
jD1
h
jv
1p
M
T
2
uD1
M
PU
jD1
h
ju
2p
2
,
P
a
c
_
_
_
_
_
_
(36)
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
where c = {1, 2]. Thus, the simplied sum rate maximisa-
tion problem of TWR-CR multiple relay selection can now
be formulated as
maximise
P
r
.t
2
/,
V
.t
2
/
R
.TWR/
_
H
_
(37)
s.t (28), (30), (31) (38)
The objective becomes now to nd the optimal power allo-
cation over relay antennas in order to solve the OWR-CR
and TWR-CR problems expressed in Equations (34) and
(37), respectively. Two approaches are proposed to deal
with these maximisation problems: iterative algorithm and
GA. A comparison between both approaches are given in
Section 5.
4.2. Iteration algorithm
We assume that each antenna has N power levels from zero
to the maximum power, i.e. an antenna cooperates with
one of the quantized power in S without interfering with
the PU. In the proposed algorithm, we aim to maximise
the sum rate by transmitting the signals with the maximum
number of antennas powered with the maximum possible
power without affecting the PUs QoS. At the beginning,
the transmit powers of all antennas at all relays are xed
to
P
a
r
(i.e. the highest power level in the discrete quanti-
zation set S). The algorithm selects the antenna that offers
the highest R and satises the interference constraint at
the same time. Then, it tries to add the maximum num-
ber of antennas that can contribute in maximising the sum
rate. If, during this process, the interference constraint is
not satised, then the new active antennas have to be pow-
ered with the next lower power existing in the discrete
quantized power set
_
P
r
i
S
_
. At the end, the algorithm
converges when P
r
reaches 0 (i.e. no more antenna can
be selected even with the lowest non-zero power). The
proposed algorithm is summarised in Algorithm 1.
4.3. Genetic algorithm
In order to employ the GA, we propose to encode the
power levels into binary words b
.k/
i
, Vi = 1, , L and
Vk = 1, , M
R
such that each power levels is designed
by a binary word. The length of the binary words b
.k/
i
depends on N (i.e. the number of quantization levels) as
follows: length(b
.k/
i
) = {log
2
N where {. denotes the
integer round towards o. For instance, if N = 4, two
bits are sufcient to encode these levels. If N = 11, four
bits are used to encode the code levels. In the last case,
the number of required words is not a power of 2, some
binary words are redundant and they correspond to any
valid word. Several solutions were proposed to solve this
problem by discarding these words as illegal, assigning
them a low utility or mapping them to a valid word with
xed, random or probabilistic remapping [29].
Algorithm 1 Proposed iterative algorithm for OWR-CR
and TWR-CR networks with discrete power levels
Input: N, M
T
1
, M
T
2
, M
R
, I
th
, o
2
n
,
P,
P
r
, L,
H
1r
i
,
H
2r
i
,
H
r
i
p
,
H
1p
and
H
2p
.
Compute P
1
and P
2
using (32) and (33) respectively, for
OWR, or compute P
int
=
V
.
int
(l) = 1.
Compute the sum rate R
.t/
using (16) for OWR or
(25) for TWR.
l = l 1.
end while
Find l
opt
s.t R
opt
= max
l
R
l
.
if R
opt
> R
max
then
(l
opt
) = 1.
R
max
= R
opt
.
L
V
opt
= L
V
opt
L {l
opt
].
else
P
k
r
= P
k
r
N
P
a
r
N1
.
end if
end while
In our GA based approach, we generate randomly T
binary strings to form the initial population set where
T denotes the population length. Each string S
t
, Vt =
1, , T, is built by concatenating LM
R
binary words b
.k/
i
corresponding to a power level of each relay antenna. Thus,
the length of a string is equal to LM
R
log
2
N. Once the
power level of each relay in a string S
t
is known and thus
the values of c
k
i
, Vi = 1, , L, k = 1, , M
R
, (i.e. if
b
.k/
i
refers to a zero power level, then c
k
i
= 0, otherwise,
c
k
i
= 1), the algorithm veries whether the interference
constraint is satised or not. If it is the case, the GA com-
putes the corresponding data rate R
.t/
, which plays the role
of the tness of the string S
t
. Otherwise, R
.t/
= 0. Then,
the algorithm selects t(1 6 t 6 T) strings that provide
the highest data rates and keeps them to the next pop-
ulation while the T t remaining strings are generated
by applying crossovers and mutations to the t survived
parents. Crossovers consist in cutting two selected ran-
dom parent strings at a correspond point that is chosen
randomly between 1 and LM
R
{log
2
(N). The obtained
fragments are then swapped and recombined to produce
two new strings. After that, mutation (i.e. changing a bit
value of the string randomly) is applied with a probability
p. This procedure is repeated until reaching convergence
or reaching the maximum iteration number denoted I.
The proposed GA with discrete power levels is detailed
in Algorithm 2.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
Algorithm 2 Proposed GA for OWR-CR and TWR-CR
networks with discrete power levels
Input: N, M
T
1
, M
T
2
, M
R
, I
th
, o
2
n
,
P,
P
r
, L,
H
1r
i
,
H
2r
i
,
H
r
i
p
,
H
1p
and
H
2p
.
Compute P
1
and P
2
using (32) and (33) respectively, for
OWR, or compute P
iD0
_
LM
R
i
_
(N 1)
i
= O(N
LM
R
) operations
to nd the optimal solution [30] while our proposed itera-
tion algorithm (IA) and GA require (N 1)(LM
R
)
2
and TI
times at most to compute the possible achievable rate until
reaching a suboptimal solution, respectively. However, it is
worth to notice that GA requires more central processing
unit (CPU) time than IA because GA applies crossover and
mutation operations at each step while IA does not require
these operations as it is shown in Table I. Indeed, Table I
shows a comparison between the proposed algorithms and
ES algorithm with average CPU time for 100 channel real-
isations and xed I
th
and
P
r
. It is clear from this table that
the GA requires more processing time than IA even with a
lower rate computations.
Also, it can be seen that the ES algorithm is not a
practical choice because of its high complexity espe-
cially for a large number of relays L, a large number of
relays antenna M
R
and/or a high quantization level N.
Table I. : Central processing unit times for two-way relaying.
ES IA GA
RC, CPU time RC, CPU time (s) RC, CPU time (s)
fM
R
, L, Ng D f1, 4, 64g
2 10
7
, 1 1008, 0.13 1120, 0.45
fM
R
, L, Ng D f2, 4, 64g
3 10
14
, 1 4032, 0.17 1120, 0.6
fM
R
, L, Ng D f4, 4, 64g
8 10
28
, 1 16128, 0.23 1120, 0.76
ES, exhaustive search; RC, rate computation; CPU, central
processing unit; IA, iteration algorithm; GA, genetic algorithm.
Table II. : Complexity comparison for two-way relaying.
M
R
, L, N ES IA GA
M
R
D 1, L D 4, N D 64 2 10
7
1008 1120
M
R
D 1, L D 4, N D 256 4 10
9
4080 1120
M
R
D 1, L D 8, N D 64 3 10
14
4032 1120
M
R
D 1, L D 8, N D 256 2 10
19
16320 1120
M
R
D 2, L D 4, N D 64 3 10
14
4032 1120
M
R
D 2, L D 4, N D 256 2 10
19
16320 1120
M
R
D 2, L D 8, N D 64 8 10
28
16128 1120
M
R
D 2, L D 8, N D 256 3 10
38
65280 1120
M
R
D 4, L D 4, N D 64 8 10
28
16128 1120
M
R
D 4, L D 4, N D 256 3 10
38
65280 1120
M
R
D 4, L D 8, N D 64 6 10
57
64512 1120
M
R
D 4, L D 8, N D 256 1 10
77
261120 1120
ES, exhaustive search; IA, iteration algorithm; GA, genetic
algorithm.
Hence, our proposed algorithms are able to reach a sub-
optimal solution with a considerable saving in terms of
computational complexity as detailed in Table II where
we compute the required number of iterations to achieve
the suboptimal solution for different values of L, M
R
and N. In addition to that, as will be shown in the sequel,
our numerical results show that our proposed algorithms
achieve almost the same performance as the ES method.
Concerning the convergence of the algorithms, by experi-
ments and for a large number of channel realisations, the
proposed algorithms always converge successfully to their
suboptimal solutions.
5. SIMULATION RESULTS
The simulations are executed under the following assump-
tions: (i) all channels are assumed to be independent and
identically distributed (i.i.d) Rayleigh fading channels; (ii)
all cognitive elements have the same peak power, i.e.
P
r
=
P; (iii) all the communication nodes of the sys-
tem are equipped with the same number of antennas, i.e.
M
T
1
= M
T
2
= M
PU
= M
R
= M with o
2
n
= 10
4
; and (iv)
the GA is executed using these parameters: the mutation
probability p is set to 0.5, t = 0.25 T, and the maximum
iteration number is I = 35.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
20 15 10 5 0 5 10 15 20 25 30
0
2
4
6
8
10
12
(a)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
20 15 10 5 0 5 10 15 20 25 30
0
5
10
15
(b)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Figure 3. Achieved sum rate versus the peak power
N
P
r
for the
optimal and iteration algorithm (IA) with I
th
D 10 dBm and
different values of M and N: (a) L D 4 and (b) L D 8.
5.1. Performance of the proposed
algorithms for TWR-CR networks
The merits of MIMO system over single antenna system
are investigated in Figure 3, we plot the TWR secondary
sumrate for different values of M = {1, 4], different values
of N = {256, 64, 16, 2] and different values of L = {4, 8].
It is noticed that we can improve the performance signi-
cantly using the multi-antenna scheme than using the sin-
gle antenna scheme. The benets of using MIMO system
appears clearly with a considerable data rate improvement
when M increases. When N = 16,
P
r
= 10 dBm, L = 8,
and using M = 4 instead of M = 1, our proposed algo-
rithm improves the rate by around 136% because the sum
rate increases from 5.5 to about 13 bits/s/Hz.
In low-SNR region, IA and the optimal solution have
almost the same sum rate, while in the high SNR region, a
gap between both methods is obtained. This gap is increas-
ing with higher
P
r
values. This is justied by the fact that
starting from a certain value of
P
r
the system can not sup-
ply the relays with the whole power budget. Hence, more
relays are deactivated. In fact, at high values of
P
r
, the
20 15 10 5 0 5 10 15 20 25 30
0
1
2
3
4
5
6
7
8
9
P
r
Peak Power [dBm]
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal
GA with N=256
IA with N=256
Best antenna selection
I
th
=10dBm
I
th
=0dBm
Figure 4. Achieved sum rate versus the peak power
N
P
r
for the
optimal and the proposed algorithms with M D 2, N D 256 and
different values of I
th
and L D 4.
20 10 0 10 20 30 40
0
2
4
6
8
10
12
P
r
Peak Power [dBm]
(a)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal
IA with N=512
IA with N=64
IA with N=16
IA with N=2 (ONOFF mode)
20 10 0 10 20 30 40
0
2
4
6
8
10
12
P
r
Peak Power [dBm]
(b)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal
GA with N=512
GA with N=64
GA with N=16
GA with N=2 (ONOFF mode)
Figure 5. Achieved sum rate versus the peak power
N
P
r
for the
optimal and the proposed algorithms with different values of I
th
and N with L D 10, T D 32, and M D 1: (a) iteration algorithm
(IA) and (b) genetic algorithm (GA).
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
interference constraint can be affected. For this reason, we
have introduced the discretization set to get more degrees
of freedom by increasing N as such we enhance the SN
sum rate. It should be noted that with the proposed algo-
rithm, when N o, we achieve the performance of the
optimal solution.
To further improve the performance of the system, we
proposed to employ the GA (with T = 32 random ini-
tial strings) to achieve better sum rate than IA but with
more complexity (CPU time) as discussed in Section 4.4.
In the low-SNR region, we can notice in Figure 4 that
both algorithms and the optimal solution have almost the
same sum rate, while in the high SNR region, the benet of
using GA is clearly observed. Indeed, IA is a deterministic
approach that reaches always the same suboptimal solu-
tion for the same channel realisation while thanks to its
random behaviour, the GA achieves different suboptimal
solutions even for the same channel realisation: it explores
several additional options than IA. In this gure, we com-
pare the performances of the IA and GA with the best
antenna selection. The best antenna selection presented in
[24] attempts to exchange the information between the ter-
minals via the best antenna with the maximum allowed
power that achieves maximum sum rate while respecting
both the interference and peak constraints.
The effect of varying I
th
for different algorithms with
xed M = 2 and N = 256 is also shown in Figure 4
where we plot the TWR secondary sum rate versus
P
r
for
different values of I
th
= {0, 10], dBm.
Figure 5 shows a comparison between the TWR-CR
network performance of the proposed algorithms and the
optimal solution with continuous power distributions for
20 10 0 10 20 30 40
0
2
4
6
8
10
P
r
Peak Power [dBm] P
r
Peak Power [dBm]
P
r
Peak Power [dBm]
P
r
Peak Power [dBm]
(a)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
L=6, N=2
20 10 0 10 20 30 40
0
1
2
3
4
5
6
(b)
A
v
e
r
a
g
e
A
c
t
i
v
e
R
e
l
a
y
s
20 10 0 10 20 30 40
0
2
4
6
8
10
P
r
Peak Power [dBm]
(c)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
L=4, N=8
20 10 0 10 20 30 40
0
1
2
3
4
(d)
A
v
e
r
a
g
e
A
c
t
i
v
e
R
e
l
a
y
s
20 10 0 10 20 30 40
0
2
4
6
8
10
P
r
Peak Power [dBm]
(e)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
L=6, N=8
20 10 0 10 20 30 40
0
1
2
3
4
5
6
(f)
A
v
e
r
a
g
e
A
c
t
i
v
e
R
e
l
a
y
s
ES algorithm
GA
IA
Single Relay
I
th
= 20dBm
I
th
= 20dBm
I
th
= 10dBm
I
th
= 10dBm
I
th
= 20dBm
I
th
= 10dBm
I
th
= 10dBm
I
th
= 20dBm
I
th
= 20dBm I
th
= 10dBm
I
th
= 10dBm
I
th
= 20dBm
Figure 6. The performance of the exhaustive search (ES) algorithm, the iteration algorithm (IA) and the genetic algorithm (GA) with
different values of I
th
, L and N versus
N
P
r
: (a,c,e) achieved sum rate and (b,d,f) average active relays.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
single antenna case. We plot the achieved secondary sum
rate versus
P
r
for different values of I
th
= {10, 20] dBm
and L = 10. For instance, with L = 10, I
th
= 20 dBm
and N = 64, we were able to improve the achievable
data rate by around 16% going from 8.7 bits/s/Hz to more
than 10 bits/s/Hz by using GA instead of IA when
P
r
=
30 dBm. We also notice that with the same quantiza-
tion level, GA is able is able to more maintain the same
slope as the optimal solution with continuous power levels
than IA.
The performances of the ES algorithm, IA, GA (with
T = 32) and the single relay selection with discrete P
r
i
under TWR-CR network scenario with M = 1 are depicted
in Figure 6. It is worth to mention that we can achieve
higher cognitive sum rate by increasing the relay power
budget for a xed interference threshold up to a certain
level. This can be justied by the fact that increasing the
relay power budget will amplify the interference power due
unlicensed users. For instance, Figure 6(a) and (b) plot the
cognitive sum rate and the average number of active relays
versus the peak relay power for L = 6 and N = 2. It
is shown that the proposed algorithms achieve almost the
same secondary sum rate of the ES algorithm by powering
20 10 0 10 20 30 40
0
2
4
6
8
10
12
14
16
18
20
P
r
Peak Power [dBm]
P
r
Peak Power [dBm]
(a)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
OWR
TWR
20 10 0 10 20 30 40
0
10
20
30
40
50
60
(b)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal with peak power constraint only
Optimal with interference constraint only
Optimal with peak power and interference constraints
IA with N=256
IA with N=32
IA with N=8
OWR
TWR
Figure 7. Achieved sum rate of the optimal and iteration algo-
rithm (IA) versus
N
P
r
with L D 6, I
th
D 20 dBm for one-way
relaying (OWR) and tow-way relaying (TWR): (a) M D 1 and (b)
M D 4.
almost the same number of relays. However, by increas-
ing N, we notice a degradation of around 0.5 bits/s/Hz of
IA at the secondary sum rate peak comparing to GA and
ES algorithm while the same performance is reached oth-
erwise as shown in Figure 6(c)(f). However, our proposed
GAmaintains the same performance as ES method even for
high values of L and N. Indeed, thanks to its random evo-
lution process, GA provides more chance to nd a better
combination than IA. In terms of computational complex-
ity, an important saving mainly for large values of N and
L is obtained comparing to the ES algorithm as detailed in
Section 4.4.
In general, by increasing N, M and L, a degradation of
the performance comparing to the ES method at the peak
of the cognitive sum rate is noticed. This can be explained
by the fact that the number of combinations that accommo-
date the interference constraint is very large in that region
and optimal solutions can be reached with ES, which is not
the case with the proposed heuristic approach. In addition
to the performance achieved by the proposed algorithms,
20 10 0 10 20 30 40
0
2
4
6
8
10
12
14
16
18
20
P
r
Peak Power [dBm]
(a)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
OWR
TWR
20 10 0 10 20 30 40
0
10
20
30
40
50
60
P
r
Peak Power [dBm]
(b)
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal with peak power constraint only
Optimal with interference constraint only
Optimal with peak power and interference constraints
GA with N=256
GA with N=32
GA with N=8
OWR
TWR
Figure 8. Achieved sum rate of the optimal and genetic algo-
rithm (GA) versus
N
P
r
with L D 6, I
th
D 20 dBm for one-way
relaying (OWR) and tow-way relaying (TWR): (a) M D 1 and (b)
M D 4.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
an important complexity saving is obtained comparing to
the ES algorithm for TWR transmission as summarised
in Table II.
5.2. OWR transmission versus
TWR transmission
Figures 7 and 8 depict the achieved sum rate of the optimal
and proposed algorithms versus the peak power constraint
P
r
with L = 6, I
th
= 20 dBm and different values of
M = {1, 4] for both OWR and TWR transmissions for IA
and GA, respectively. The sumrate of both OWRand TWR
schemes is compared to the case when only one constraint
is applied (either peak power constraint or interference
constraint). It can be shown that the optimal solution with
interference constraint only is an upper bound for the case
when both constraints are considered. It can be seen that,
we can almost double the secondary sum rate by using
TWR transmission instead of using OWR transmission.
In addition to that, OWR transmission requires more rate
computational analysis than TWR transmission. Indeed, it
requires the double number of operations to solve the opti-
mization problem, because it has to execute the algorithm
twice (i.e. every two time slots).
To investigate the effect of the interference caused by
PUs to the SN for OWR and TWR networks, we plot in
Figure 9 the secondary achievable sum rate as a function
of o
2
n
for xed
P
r
= 10 dBm, I
th
= 20 dBm, L = 4 and
M = 2. It is deduced from this gure that when the value
of o
2
n
increases (i.e. the interference from the PUs to SN
increases), the secondary achievable rate reduces. Also, we
notice that the PU interference has no signicant impact on
the proposed algorithm performance. Indeed, the gap of the
achieved sum rate between the algorithms and the optimal
solution is maintained even for high values of o
2
n
.
Finally, Figure 10 deals with the effect of an erroneous
CSI on the system performance. We vary the variance of
the CSI error o
2
e
between 0 and 1 (i.e. o
2
e
= 0 corresponds
to the perfect CSI scenario) for both algorithms (IA and
GA). We plot the achievable secondary rate versus the error
variance o
2
e
with the following parameters
P
r
= 10 dBm,
1 1.5 2 2.5 3 3.5 4
x 10
4
4
5
6
7
8
9
10
11
12
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
Optimal
GA with N=256
IA with N=256
TWR
OWR
Figure 9. Achieved sum rate using genetic algorithm (GA) and
iteration algorithm (IA) as a function of
2
n
for
N
P
r
D 10 dBm,
I
th
D 20 dBm, L D 4 and M D 2.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
S
u
m
R
a
t
e
(
B
i
t
s
/
s
/
H
z
)
GA with perfect CSI, =0
IA with perfect CSI, =0
GA with imperfect CSI
IA with imperfect CSI
TWR
OWR
Figure 10. Achieved sum rate two-way relaying (TWR) transmis-
sion using genetic algorithm (GA) and iteration algorithm (IA)
as a function of
2
e
under imperfect CSI for
N
P
r
D 10 dBm,
I
th
D 10 dBm, L D 3, N D 256 and M D 2.
I
th
= 10 dBm, L = 3, N = 256 and M = 2. We notice that
the scheme performance is highly affected by the increase
of the CSI error for both algorithms. Indeed, it can be
noticed that for o
2
e
= 0.1, the TWR secondary sum rate
degrades by 27% going from 3.5 Bits/s/Hz to 4.8 Bits/s/Hz
by having imperfect CSI instead of perfect CSI. However,
we can see that TWR network is more affected by the CSI
error than the OWR network. This is because the additional
error observed during the self interference elimination in
Equations (10) and (11).
6. CONCLUSION
In this paper, practical approaches (iterative algorithm and
GA) are designed to maximise the achievable secondary
sum rate by employing a multiple antennarelay selection
scheme for both OWR-CR and TWR-CR networks with
discrete power distributions. We have analysed the perfor-
mance of the proposed algorithms and compared them with
the optimal solution using continuous power distributions
and an ES method for discrete power levels. In many sit-
uations, the proposed algorithms are able to reach a close
solution to both optimal schemes with a considerable sav-
ing in terms of computational complexity. In addition to
that, we have showed that thanks to its random evolution,
the GA provides a better performance than the iterative
one. Furthermore, we showed that comparing to the opti-
mal solution, the performance of our proposed algorithms
follow the same behaviour for high primary interference
and erroneous channel state information. In our ongoing
task, we are working on applying continuous power allo-
cation algorithm based on the particle swarm optimization
technique to our multi-antenna system model.
ACKNOWLEDGEMENT
The work of M.-S. Alouini was made possible by NPRP
grant #52502087 from the Qatar National Research
Fund (a member of Qatar Foundation). The statements
made herein are solely the responsibility of the authors.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
REFERENCES
1. Alsharoa A, Ghazzai H, Alouini M-S. A low complex-
ity algorithm for multiple relay selection in two-way
relaying cognitive radio networks. In Proceedings of
the 5th IEEE Workshop on Cooperative and Cogni-
tive Mobile Networks (COCONET2013) in Conjunction
with IEEE International Conference on Communications
(ICC2013), Budapest, Hungary, June 2013; 327331.
2. Alsharoa A, Ghazzai H, Alouini M-S. A genetic algo-
rithm for multiple relay selection in two-way relaying
cognitive radio networks. In IEEE Vehicular Technol-
ogy Conference (VTC Fall2013), Las Vegas, USA,
September 2013.
3. Mitola J, III, Maguire GQ, Jr. Cognitive radio: making
software radios more personal. IEEE Personal Commu-
nications 1999; 6(4): 1318.
4. Mitola J, III. Cognitive radio: an integrated agent
architecture for software dened radio. Ph.D. Disserta-
tion, Royal Institute of Technology (KTH), Stockholm,
Sweden, 2000.
5. Haykin S. Cognitive radio: brain-empowered wireless
communications. IEEE Journal on Selected Areas in
Communications 2005; 23(2): 201220.
6. Kang X, Liang Y-C, Nallanathan A, Garg HK, Zhang R.
Optimal power allocation for fading channels in cogni-
tive radio networks: Ergodic capacity and outage capac-
ity. IEEE Transactions on Wireless Communications
2009; 8(2): 940950.
7. Zou Y, Yao Y-D, Zheng B. Cooperative relay tech-
niques for cognitive radio systems: Spectrum sensing
and secondary user transmissions. IEEE Communica-
tions Magazine 2012; 50(4): 98103.
8. Zou Y, Zhu J, Zheng B, Yao Y-D. An adaptive coop-
eration diversity scheme with best-relay selection in
cognitive radio networks. IEEE Transactions on Signal
Processing 2010; 58(10): 54385445.
9. Hasna MO, Alouini M-S. Optimal power allocation for
relayed transmissions over Rayleigh-fading channels.
IEEE Transactions on Wireless Communications 2004;
3(6): 19992004.
10. Hasna MO, Alouini M-S. End-to-end performance of
transmission systems with relays over Rayleigh-fading
channels. IEEE Transactions on Wireless Communica-
tions 2003; 2(6): 11261131.
11. Kramer G, Gastpar M, Gupta P. Cooperative strate-
gies and capacity theorems for relay networks. IEEE
Transactions on Information Theory 2005; 51(9):
30373063.
12. Laneman JN, Tse DNC, Wornell GW. Cooperative
diversity in wireless networks: efcient protocols and
outage behavior. IEEE Transactions on Information
Theory 2004; 50(12): 30623080.
13. Rankov B, Wittneben A. Spectral efcient protocols
for half-duplex fading relay channels. IEEE Journal
on Selected Areas in Communications 2007; 25(2):
379389.
14. Jitvanichphaibool K, Zhang R, Liang Y-C. Optimal
resource allocation for two-way relay-assisted OFDMA.
IEEE Transactions on Vehicular Technology 2009;
58(7): 33113321.
15. Li L, Zhou X, Xu H, Li GY, Wang D, Soong A. Simpli-
ed relay selection and power allocation in cooperative
cognitive radio systems. IEEE Transactions on Wireless
Communications 2011; 10(1): 3336.
16. Luo C, Yu FR, Ji H, Leung VCM. Distributed relay
selection and power control in cognitive radio net-
works with cooperative transmission. In Proceedings
of IEEE International Conference on Communications
(ICC2010), Cape Town, South Africa, May 2010; 15.
17. Xu J, Zhang H, Yuan D, Jin Q, Wang C-X. Novel mul-
tiple relay selection schemes in two-hop cognitive relay
networks. In Proceedings of IEEE 3rd International
Conference on Communications and Mobile Computing
(CMC2011), Qingdao, China, April 2011; 307310.
18. Bayat S, Louie RH, Vucetic B, Li Y. Dynamic decen-
tralised algorithms for cognitive radio relay networks
with multiple primary and secondary users utilising
matching theory. Transactions on Emerging Telecommu-
nications Technologies 2013; 24(5): 486502.
19. Choi M, Park J, Choi S. Low complexity multiple
relay selection scheme for cognitive relay networks. In
Proceedings of IEEE Vehicular Technology Conference
(VTC Fall2011), San Francisco, USA, 2011; 15.
20. Naeem M, Lee DC, Pareek U. An efcient multiple
relay selection scheme for cognitive radio systems. In
Proceedings of IEEE International Conference on Com-
munications Workshops (ICC2010), Cape Town, South
Africa, May 2010; 15.
21. Wang B, Zhang J, Host-Madsen A. On the capacity of
MIMO relay channels. IEEE Transactions on Informa-
tion Theory 2005; 51(1): 2943.
22. Fan Y, Thompson J. MIMO congurations for relay
channels: theory and practice. IEEE Transactions on
Wireless Communications 2007; 6(5): 17741786.
23. Cui T, Gao F, Ho T, Nallanathan A. Distributed space-
time coding for two-way wireless relay networks. In
Proceedings IEEE International Conference on Com-
munications (ICC2008), Beijing, China, May 2008;
38883892.
24. Amarasuriya G, Tellambura C, Ardakani M. Two-way
amplify-and-forward multiple-input multiple-output
relay networks with antenna selection. IEEE Journal
on Selected Areas in Communications 2012; 30(8):
15131529.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett
A. Alsharoa et al.
25. Park H, Chun J, Adve R. Computationally efcient relay
antenna selection for af MIMO two-way relay channels.
IEEE Transactions on Signal Processing 2012; 60(11):
60916097.
26. Taki M, Lahouti F. Discrete rate interfering cogni-
tive link adaptation design with primary link spectral
efciency provisioning. IEEE Transactions on Wireless
Communications 2011; 10(9): 29292939.
27. Lee N, Simeone O, Kang J. The effect of imper-
fect channel knowledge on a MIMO system with
interference. IEEE Transactions on Communications
2012; 60(8): 22212229.
28. Boyd S, Vandenberghe L. Convex Optimization.
Cambridge University Press: New York, NY, USA,
2004.
29. Beasley D, Bull DR, Martin RR. An overview of genetic
algorithms: part 2, research topics. University Comput-
ing 1993; 15(4): 170181.
30. Rosen KH. Discrete Mathematics and Its Applications
(6th edn). McGraw-Hill: New York, NY, 2007.
Trans. Emerging Tel. Tech. (2014) 2014 John Wiley & Sons, Ltd.
DOI: 10.1002/ett