You are on page 1of 12

490

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

On the Spectrum- and Energy-Efficiency Tradeoff


in Cognitive Radio Networks
Hang Hu, Student Member, IEEE, Hang Zhang, Member, IEEE, and Ying-Chang Liang, Fellow, IEEE

AbstractIncreasing spectrum-efficiency (SE) as well as


energy-efficiency (EE) has attracted much attention recently due
to the fact that the future wireless networks need to address the
issues of high throughput and low power consumption. However,
the objective for optimizing SE sometimes conflicts with the one for
optimizing EE, and the methods for improving EE may result in a
decrease in SE. In this paper, we consider the SE-EE tradeoff for
cognitive radio (CR) networks with co-operative spectrum sensing (CSS). First, we formulate the general problem, and analyze
two special cases: the SE maximization problem and the EE maximization problem. The SE and EE are optimized separately via
joint optimization of sensing duration and final decision threshold
in CSS. Based on the solutions of the two special cases, the general problem for SE-EE tradeoff is solved. Then, we consider the
tradeoff of SE and EE from two perspectives: (1) maximizing EE
while satisfying SE requirement; and (2) maximizing SE while satisfying EE requirement. Efficient algorithms for sensing strategy
design are proposed for each scenario. Finally, we demonstrate the
effectiveness of the proposed sensing strategies and illustrate the
tradeoff between SE and EE via simulations.
Index TermsCognitive radio, spectrum-efficiency, energyefficiency, cooperative spectrum sensing, spectrum-efficiency and
energy-efficiency tradeoff.

I. I NTRODUCTION

ITH THE explosive growth of wireless services,


achieving high spectrum-efficiency as well as energyefficiency has become the key design trend for future wireless
communication technology [1], [2]. On one hand, many countries are now facing the problem of spectrum scarcity, but,
some of the allocated spectrum are not efficiently utilized. Thus,
enormous works have been devoted to improving the spectrum
efficiency (SE) in the past few years [3]. On the other hand, the
study on energy efficiency (EE) has attracted a lot of attention
because of the device requirements and environment concerns
[4]. However, in general, methods for improving EE will result
in a decrease in SE [5]. Therefore, how to design the SE-EE
tradeoff is an emerging problem which is worth studying in
wireless communications.

Manuscript received June 17, 2015; revised October 5, 2015 and November
26, 2015; accepted November 27, 2015. Date of publication December 3, 2015;
date of current version February 12, 2016. This research is funded by National
Science Foundation of China (NSFC) with Grant No. 61571100. The associate
editor coordinating the review of this paper and approving it for publication was
H. Dai.
H. Hu and H. Zhang are with the College of Communications Engineering,
PLA University of Science and Technology, Nanjing 210007, China (e-mail:
xd_huhang@126.com; hangzh_2002@163.com).
Y.-C. Liang is with the University of Electronic Science and Technology
of China (UESTC), Chengdu 611731, China, and also with the Institute for
Infocomm Research, A*STAR, Singapore 138632 (e-mail: liangyc@ieee.org).
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/TCOMM.2015.2505281

Cognitive radio (CR) technology has been proposed to support higher SE through dynamic spectrum access [6]. The
secondary users (SUs) first sense the status of primary users
(PUs) and then access the frequency band if the PUs are
detected to be absent [7]. Since additional energy consumption
is required for the spectrum sensing operation in SUs, when the
energies of the SUs are constrained, we need to consider the
important issue of increasing the EE of SUs.
Cooperative spectrum sensing (CSS) has attracted much
attention in recent years since it can be used to further enhance
the spectrum utilization efficiency [8]. In fact, in wireless communications, the received signal would be severely degraded
due to the detrimental effects of fading and shadowing. In this
case, the sensing performance of a single SU could be severely
degraded, and CSS using multiple SUs can greatly improve the
sensing performance [9]. However, when CSS is used, more
energy will be consumed for local spectrum sensing and sensing
result reporting process as compared to the single SU sensing
case. The threshold of energy detection, sensing duration and
fusion strategy used in the fusion center (FC) will affect both
SE (measured in bits/s/Hz) and EE (measured in bits/Joule/Hz)
of the CR network. In this paper, our objective is to design the
sensing parameters by considering the SE-EE tradeoff.
In the literature, there are some works which deal with the SE
optimization and EE optimization. For example, the authors in
[10] optimize the sensing duration and the fusion threshold in
the FC to maximize the SUs throughput. In [11], the spectrum
utilization can be greatly improved by using the joint spectrum sensing duration adaption and data transmission scheme.
A novel CR system was proposed in [12] to solve the problem
of sensing-throughput tradeoff. The energy-efficient sequential
channel sensing strategy was investigated in [13]. In [14], the
authors studied energy-efficient CR systems by jointly determining the spectrum sensing duration and the data transmission
duration. The effect of SUs transmission power on EE was discussed in [15]. However, how to balance the SE and EE for CR
networks with CSS has not been considered.
In this paper, we aim to design the sensing parameters in
the CSS (including the sensing duration and the final decision
threshold) that can balance the SE with EE under the condition
of sufficient protection to PU. The general problem (GP) for
SE and EE tradeoff is formulated. First, we analyze two special cases of the GP, i.e., the SE maximization problem (P1)
and the EE maximization problem (P2). It is proved that SE
is a unimodal function of the sensing duration, and EE has the
unimodal property if there are some some constraints on the
power parameters. Algorithm 1 and Algorithm 2 are proposed
to maximize SE and EE respectively. Based on the solutions of

0090-6778 2015 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.

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

491



 
 t f
s s
1
,
pf = Q
2
2

Fig. 1. Frame structure of CSS.

P1 and P2, the SE-EE tradeoff problem is analyzed and the GP


is solved.
Since the maximal SE and the maximal EE can not be
obtained simultaneously, we balance the SE and EE from the
following two perspectives: (1) maximizing EE while satisfying SE requirement; (2) maximizing SE while satisfying EE
requirement. The corresponding algorithms are proposed to
solve the tradeoff problems. Simulation results show that the
joint optimization of sensing duration and final decision threshold will significantly improve the SE and EE. Different SE (or
EE) requirements need different optimal values of sensing duration and final decision threshold. It is also shown that there is a
tradeoff between the SE and EE.
The rest of this paper is organized as follows. In Section II,
the problem formulation for SE-EE tradeoff with CSS is presented. The solutions of the formulation are given in Section III.
In Section IV, we analyze the problem of balancing SE and
EE. Simulation results are provided in Section V, followed by
concluding remarks in Section VI.
II. P ROBLEM F ORMULATION
We consider a CR network with K SUs and a FC. The
coverage of the secondary network is assumed to be small as
compared to the distance between the primary network and
the secondary network. Fig. 1 illustrates the frame structure
designed for the CR network with periodic CSS where each
frame consists of a spectrum sensing block, a reporting block
and a data transmission block [16]. Suppose that the frame
duration is T , the sensing duration is ts , and the reporting duration for each SU is tr . In the spectrum sensing block, the K
cooperating SUs sense the PUs status simultaneously. In the
reporting block, the K sensing results are sent to the FC sequentially via a common control channel (CCC). Then, according to
some fusion rules, the FC makes a final decision to indicate that
the PU is present or absent. If the PU is absent, one of the secondary users is allowed to conduct data transmission. However,
if the PU is present, the secondary users will not be able to
utilize the spectrum.
In this paper, we consider that the primary signal is binary
phase-shifted keying (BPSK) signal, the noise is real-valued
Gaussian variable with zero mean and variance 2 , and energy
detection is employed to detect the PUs status in the local sensing stage. The detection probability and false alarm probability
at each SU can be calculated as [17]



 

ts f s
1
pd = Q
,
(1)
2(2 + 1)
2

(2)

where  is the threshold of energy detection, denotes the SNR


of PUs signal at each SU when the PU is present, f s is the sampling frequency. In a small area cognitive radio network, the
SUs are close to each other such that they experience almost
similar path loss. Here, for the ease of presentation, we consider the case that the PU signal reaches all SUs with the same
power level. However, the design methodology can be extended
to case when the received powers are different. By combining
(1) and (2), it is derived that




ts f s
1
2 + 1Q ( pd ) +
.
(3)
pf = Q
2
In the FC, all one bit decisions are fused together according to the counting rule, the final detection probability and final
false alarm probability can be computed by [18]
QD =

B(i; K , pd ),

(4)

B(i; K , p f ),

(5)

i=M

QF =

K

i=M

where
final decision threshold in the FC, B(l; m, p) =

m l M is the
ml is the probability that there are l success
p
(1

p)
l
results in successive m Bernoullis trials (each trial with a
probability of success p).
For fading channels, the false alarm probability will be the
same as (2) since p f is independent of sensing SNR, the detection probability p d can be derived by averaging pd in AWGN
case over the SNR fading distribution. The closed-form expressions for p d for fading channels can be referred to [19]. In this
case, the SNR values of PUs signal at the SUs will be different. The weighted decision fusion schemes shown in [20] can
be employed in the design methodology.
The SU transmits data in the following two cases: (1) when
the PU is absent, and it is correctly detected to be absent; (2)
when the PU is present, and it is falsely detected to be absent.
In the former case, the SE (measured in bits/s/Hz) of the CR network can be formulated as T tsTK tr log2 (1 + S ) [17], where
S is the SNR for the secondary link. In the latter case,
 the SE of

S
T ts K tr
1
+
log
the CR network can be presented as
2
T
+1
[17]. The average SE of the CR network is given by
S E = S E + S E ,

(6)

where
T ts K tr
log2 (1 + S ),
(7)
T


S
T ts K tr
log2 1 +
.
= (1 )(1 Q D )
T
+1
(8)

S E = (1 Q F )
S E

In the above equations, is the probability that PU is absent,


and 1 is the probability that PU is present.

492

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

Considering the energy consumption of the CR network, let


us define Ps , Pr , Pc and Pt as the sensing power consumed
by each SU, the reporting power consumed by each SU, the
circuit power consumed by electronic devices, and the transmit power by the SU transmitter, respectively. Four scenarios
will be considered. (1) The CR network correctly detects PUs
presence, with probability (1 )Q D . The SU is not allowed
to transmit data, and the energy consumption is K ts Ps +
K tr Pr + K (ts + tr )Pc ; (2) The CR network falsely detects
PUs presence (missed detection), with probability (1 )(1
Q D ). One SU will transmit data, and the energy consumption is K ts Ps + K tr Pr + K (ts + tr )Pc + (T ts K tr )(Pt +
Pc ); (3) The CR network correctly detects PUs absence,
with probability (1 Q F ). One SU will transmit data, and
the energy consumption is K ts Ps + K tr Pr + K (ts + tr )Pc +
(T ts K tr )(Pt + Pc ); (4) The CR network falsely detects
PUs absence (false alarm), with probability Q F . The SU
is not allowed to transmit data, and the energy consumption
is K ts Ps + K tr Pr + K (ts + tr )Pc . Thus, the average energy
consumption is

missed detection of primary signal can be ignored: The probability that the PU is present is assumed to be a small value,
hence it is economically advisable for the secondary users to
utilize the PUs spectrum band; The value of SNR is high,
then the SUs can detect the PU reliably and the false alarm
S
) will be
probability Q F is small, and the value of log2 (1 + +1
small since the PU signal is regarded as an interference to the
SU receiver when missed detection happens; The target detection probability is set greater than or equal to 0.9 to protect the
PU sufficiently. By comparing (7) and (8), we can see that E E
will be much smaller than E E when the above case happens.
Thus, the original optimization problem can be approximated
by maximizing S E + (1 ) E E subject to C1 , C2 and C3 .
It is obviously that the approximated SE S E and EE E E will
be lower than the original SE S E and EE E E .

E = K ts Ps + K tr Pr + K (ts + tr )Pc
+ [(1 Q F ) + (1 )(1 Q D )]
(T ts K tr )(Pt + Pc ).

(9)

The EE (measured in bits/Joule/Hz) of the CR network can


be expressed as
E E =

S E T
= E E + E E ,
E

In this section, we will first solve (P1) and (P2) separately.


Then, we will analyze the tradeoff between SE and EE and
solve the problem (GP).
A. Solutions of (P1)
When = 1, the optimization problem is maximizing S E
subject to C1 , C2 and C3 . For a given M, we take the first partial
derivative of S E versus ts and obtain
S E
d Q F T ts K tr
log2 (1 + S )
=
ts
dts
T

(1 Q F ) log2 (1 + S ),
T

(10)

where E E = S EE T , E E = S EE T .
In IEEE802.22 WRAN, to sufficiently protect the PU, the target detection probability Q th
D = 0.9 needs to be satisfied [14].
For a given Q th
,
and
the
values of ts and M, we can find
D
an energy detection threshold to satisfy Q D = Q th
D , namely,



4 +2
1
2
 = 1 + + Q ( pd (M)) ts fs .
Our goal is to design the sensing parameters (including the
sensing duration and the final decision threshold) that can balance the SE with EE under the condition of sufficient protection
to PU. Mathematically, these constraints can be described as
follows
C1 : Q D = Q th
D

(11)

C2 : 0 < ts < T K tr
C3 : 1 M K

(12)
(13)

and the general problem for SE-EE tradeoff becomes


GP : max S E + (1 ) E E

(14)

s.t. : C1 , C2 , C3

(15)

ts ,M

III. S OLUTIONS OF THE F ORMULATED P ROBLEMS

where is the balancing factor, and 0 1. When = 1,


the above general problem becomes SE maximization problem (P1) ; when = 0, the general problem becomes the EE
maximization problem (P2).
However, it is very difficult to analyze the objective function in the (GP), hence we consider the following case that the

(16)

where



dp f
d QF
d Q F dp f
K 1 M1
=
=K
p f (1 p f ) K M
,
dts
dp f dts
dts
M 1
(17)
and

2



dp f
f s 12 2 +1Q1 ( pd )+ ts2fs

=
e
.
(18)
dts
4 ts

It

can

be

derived

S E
limts T K tr t
s

that

limts 0

S E
ts

and

< 0 for any value of M. Hence, there


must exist an optimal sensing duration that can maximize the
S E
= 0 exists in the
spectrum efficiency S E , and the root of t
s
range 0 < ts < T K tr . In the following, it will be proved
that S E has a unique maximal value for any given M, and
S E
S E
= 0. Setting t
= 0, it is
there is only one root for t
s
s
derived that

2




12
2 +1Q1 ( pd )+ ts2f s
fs
K K 1
(T ts K tr )
e
4 M1
ts
M1
K
piM+1
=
(1 p f ) Mi .
f
i
i=0

(19)

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

Fig. 2. Illustration of Y (ts ) and


(ts ) in case 1.

493

Fig. 3. Illustration of Y (ts ) and


(ts ) in case 2.

Then, we get Y (ts ) =


(ts ), where

2

t
f
1 
s
s
Y (ts ) =
2 + 1Q1 ( pd ) +
,
2
2


ts
4

(ts ) = ln

K 1
fs
K (T ts K tr ) M1

M1
K
iM+1
Mi
pf

(1 p f )
.
i

(20)

(21)

i=0

Proposition 1: For a given M,


(ts ) is decreasing in ts over
the range 0 < ts < T K tr .
Proof: Please refer to Appendix A.

The first order derivative of Y (ts ) versus ts is




dY (ts )
fs 
ts f s

1
=
2 + 1Q ( pd ) +
. (22)
dts
2 2ts
2
s)
And the first order derivative of
(ts ) versus ts is d
(t
dts =
u(ts ) + (ts ), where u(ts ) and (ts ) can be referred to (41) and
(42) in Appendix A respectively. Four cases should be considered when we analyze the relationship between Y (ts ) and

(ts ).
(1) Q1 ( pd ) 0. In this case, dYdt(ts s ) > 0, Y (ts ) is increasing
in ts . Thus, Y (ts ) intersects
(ts ) once, which is shown in
Fig. 2.
 1
2
Q ( pd )
(2) Q1 ( pd ) < 0, T K tr 4f+2
. In this case,

s


2 + 1Q1 ( pd ) + ts2fs < 0 and dYdt(ts s ) < 0, Y (ts ) is

decreasing in ts . Obviously,
4

fs
ts

d
(ts )
dts

< u(ts ) <


2


t f
21
2 +1Q1 ( pd )+ s2 s

dp f
1
dts 1 p f

e
. Since


Q
2 +1Q1 ( pd )+ ts2f s

2 x <

fs
s)
for x 0 [21], we have d
(t
<
dts
4 ts




2 2 + 1Q1 ( pd )+ ts2fs = dYdt(ts s ) . Thus, Y (ts )


1 x2

e 2
Q(x)

intersects
(ts ) once, which is shown in Fig. 3.
 1
2
Q ( pd )
(3) Q1 ( pd ) < 0, T K tr > 4f+2
and Y (ts )

s
 1
2
Q ( pd )
intersects
(ts ) in the range 0 < ts 4f+2
.

s
 1
2
Q ( pd )
When 4f+2
< ts < T K tr , Y (ts ) is

s
increasing in ts , and Y (ts ) is not able to intersect
(ts )

Fig. 4. Illustration of Y (ts ) and


(ts ) in case 3.

Fig. 5. Illustration of Y (ts ) and


(ts ) in case 4.

dY (ts )
s)
in this region. Recall that d
(t
dts < dts , thus, Y (ts )
intersects
(ts ) once, which is shown in Fig. 4.
 1
2
Q ( pd )
(4) Q1 ( pd ) < 0,
T K tr > 4f+2
and

s
Y (ts ) does not intersect
(ts ) in the range
 1
2
Q ( pd )
0 < ts 4f+2
. In this case, Y (ts ) must

s
 1
2
Q ( pd )
intersect
(ts ) for 4f+2
< ts < T K tr .

s
As Y (ts ) is increasing in ts and
(ts ) is decreasing in
 1
2
Q ( pd )
ts in the range 4f+2
< ts < T K tr , Y (ts )

s
intersects
(ts ) once, which is shown in Fig. 5.
Based on the above analysis, it can be concluded that Y (ts )
S E
= 0 is
only intersects
(ts ) once. Therefore, the root of t
s
unique, the spectrum efficiency S E is a unimodal function and
there exists only one optimal value of ts that maximizes S E .
Thus, Bisection method [22] can be used to obtain the optimal
opt
sensing time ts,S E for any given M.
In the following, we analyze the optimization of M for a
given ts . The individual detection probability pd (M) can be

494

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

K
K
i
K i Q th = 0.
obtained by solving i=M
D
i ( pd ) (1 pd )
To facilitate analysis, we rewrite (4) and (5) as
Q D = I pd (M, K M + 1),

(23)

Q F = 1 I1 p f (K M + 1, M),

(24)

where I x (g, h) is the regularized incomplete beta function. For


a given Q th
D , we have
pd (M) = I 1 (Q th
D , M, K M + 1),

(25)

where I 1 (y, g, h) is the inverse regularized incomplete beta


function (if y = I x (g, h), we have x = I 1 (y, g, h)).
Then, it is derived that
p f (M)



=Q
2 + 1Q1 (I 1 (Q th
,
M,
K

M
+
1))
+

t
f
s s ,
D
(26)
Q F (M) = 1 I1 p f (M) (K M + 1, M).

(27)

And then the SE and EE can be obtained. It is very difficult


to mathematically analyze the property of SE (or EE) over the
final decision threshold M due to the complexity of Q1 (x)
and I 1 (y, g, h). Since M is an integer, it is not computationopt
ally complicated to search M from 1 to K to find M S E . Let
(k) denote  at kth iteration, where  may be ts , M or
S E . Algorithm 1 for spectrum-efficient design is presented as
follows.
Algorithm 1. Spectrum-efficient design
tr
Initialization: k = 0; M(0) = 1; ts (0) = T K
2 ; S E (0) =
0; tolerance level
repeat
1) Find M that maximizes (7) with ts (k).
2) M(k + 1) M
3) Calculate ts of (16) by using Bisection method with
M(k + 1).
4) ts (k + 1) ts
5) Compute S E with ts (k + 1) and M(k + 1).
6) S E (k + 1) S E
7) k k + 1
until [ S E (k + 1) S E (k)]
Output: M(k); ts (k); S E (k)

B. Solutions of (P2)
When = 0, the optimization problem is maximizing E E
subject to C1 , C2 and C3 . Taking the first partial derivative of
E E versus ts , we can get


S E
E
E E
T
,
(28)
= 2 E
S E
ts
ts
ts
E
where


E
= K (Ps + Pc ) (1 Q F ) + (1 )(1 Q th
D)
ts
d QF
.
(Pt + Pc ) (T ts K tr )(Pt + Pc )
dts
(29)

Let



S E
E
T
E
S E
W =
log2 (1 + S )
ts
ts

= (T ts K tr )[K ts Ps + K tr Pr + K (ts + tr )Pc ]



2 d QF
+ (1 ) 1 Q th
(P
+
P
)(T

K
t
)
t
c
s
r
D
dts
(1 Q F )[K (T K tr )(Ps + Pc ) + K tr (Pr + Pc )].
(30)
Since limts 0

d QF
dts

limts T K tr W < 0.
EE
limts T K tr t
s

= , we have limts 0 W = +,
Thus,

limts 0

E E
ts

= +

and

< 0. Therefore, the optimal sensing


duration that can maximize the energy efficiency E E exists
E E
= 0, it is derived
in the range 0 < ts < T K tr . Setting t
s
that


fs
K K 1
(T ts K tr )
4 M 1
ts

K ts Ps + K tr Pr + K (ts + tr )Pc

+ (1 )(1 Q th
)(P
+
P
)(T

K
t
)
t
c
s
r
D

12

2 +1Q1 ( pd )+ ts2f s

e
= [K (T K tr )(Ps + Pc ) + K tr (Pr + Pc )]
M1
K
piM+1
(1 p f ) Mi .

f
i

(31)

i=0

Then, it is derived that Y (ts ) = Z (ts ), where Y (ts ) can be


referred to Equation (20), and


ts
4

K (T ts K tr ) fs [(T K tr )
K (Ps + Pc ) + K tr (Pr + Pc )]

 M1
K iM+1
(1 p f ) Mi
i=0 i p f
Z (ts ) = ln
K 1
. (32)
K Pc
M1 [K ts Ps + K tr Pr +

(ts + tr ) + (1 ) 1 Q th
D

(Pt + Pc ) (T ts K tr )]
The property of EE over ts is related to the values of the
sensing power, the reporting power, the circuit power and the
transmit power. In this paper, we are interested in the unimodal property of EE. In Appendix B, it is derived that d Zdt(ts s ) =
u(ts ) + v(ts ), where v(ts ) can be referred to (43). And we show
that if there are some constraints on the power parameters, v(ts )
will be less than 0. Then, based on the analysis of the four cases
in subsection A, we can also derive that Y (ts ) only intersect
EE
= 0 is unique, the energy
Z (ts ) once. Thus, the root of t
s
efficiency E E is a unimodal function and there exists only
one optimal value of ts that maximizes E E . Bisection method
opt
can also be used to obtain the optimal sensing time ts,E E for
any given M. In the following, we will design an algorithm to
optimize M.
We define a function as (M, ) = T S E E, where
= E < 0, (M, )
is a positive number. Since (M,)

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

is decreasing in . For any > 0, we have (M, +


) < (M, ). Let () = max (M, ), we can obtain
M

that (M, + ) < () for any M. Thus, ( + ) =


max (M, + ) < (), () is decreasing in .
M

For any value of M, (M, E E (M)) = 0. Let us denote


opt
opt
M E E as the optimal M, then E E (M E E ) is the maximal EE and
is also the largest value of to make (M, ) = 0. We define
as the root of () = 0 and M E E as the optimal M value
to maximize (M, ). For any > 0, we have ( + ) =
max (M, + ) < ( ) = 0. Thus, for any M, ( +
M

) < 0. Therefore, must be the largest value of to make


opt
opt
(M, ) = 0, and = E E (M E E ), M E E = M E E .
Since () is decreasing in and the root of () = 0 is
the maximal EE, we can use Bisection method to calculate the
optimal value. To find the value of M that maximizes (M, ),
(M + 1, ) (M, ) = 0, it is derived
we set (M,)
M
that
 

,
(33)
M=

where


[log2 (1 + S ) (Pt + Pc )]
= ln
(1 )(Pt + Pc )

1 pf
1 pd

K 
,

495

Algorithm 2. Energy-efficient design


tr
+
Initialization: k = 0; M(0) = 1; ts (0) = T K
2 ; = 0;
++
+
++
= ; X = 0; X
= K ; E E (0) = 0;

tolerance levels 1 and 2


repeat
1) Find M with ts (k) using the following loop:
+
while ( ++ + ) > 1 (X++
 X ) = 0 do

12 ( + + ++ ), X
if () > 0 then
+ , X+ X
else
++ , X++ X
end if
end while
M X
2) M(k + 1) M
3) Calculate ts of (28) by using Bisection method with
M(k + 1).
4) ts (k + 1) ts
5) Compute E E with ts (k + 1) and M(k + 1).
6) E E (k + 1) E E
7) k k + 1
until [ E E (k + 1) E E (k)] 2
Output: M(k); ts (k); E E (k)

(34)
and

= ln



pd
1
.
1
pf
1 pd

(35)

Proposition 2: For p f and pd given in (2) and (1) respectively, > 0.


Proof: Please refer to Appendix C.

Since M is a positive integer, must be positive, then it is
derived that < , where
=

log2 (1 + S )(1 p f ) K
1

.
Pt + Pc (1 p f ) K + (1 )(1 pd ) K

(36)

Let (k) denote at kth iteration, where may be ts , M or


E E . Algorithm 2 for energy-efficient design is presented as
follows.

Fig. 6. Illustration of S E , E E and S E + (1 ) E E .

Since
we

limts t opt

s,E E

have

limts t opt

s,S E

C. Solutions of (GP)
When 0 < < 1, for a given M, it is difficult to analyze the
property of S E + (1 ) E E over ts . However, from subsection A and subsection B, we know that S E is a unimodal
function, and E E has the unimodal property if there are some
constraints on the power parameters. The optimal sensing time
opt
that maximizes E E is denoted as ts,E E , the optimal sensing
opt
time that maximizes S E is denoted as ts,S E . It is assumed that
opt
opt
opt
opt
ts,E E < ts,S E (If ts,E E ts,S E , the problem can be analyzed in
a similar way), which is illustrated in Fig. 6. Then, both E E
opt
and S E are increasing in ts in the range 0 < ts ts,E E , and
opt
are decreasing in ts in the range ts,E E ts < T K tr .

limts t opt

s,S E

E E
ts

limts t opt

=0

s,E E

E E
ts < 0 and
( S E +(1) E E )
ts

and

limts t opt

s,E E

( S E +(1) E E )
> 0.
ts
S E
limts t opt ts = 0,
s,S E

S E
ts

> 0,
Since

we have

< 0. Thus, the optimal sensing

time that maximizes S E + (1 ) E E exists in the range


opt
opt
ts,E E < ts < ts,S E . Generally speaking, it is difficult to find an
efficient algorithm to obtain the optimal value. Since we know
opt
opt
the range of ts (ts,E E < ts < ts,S E ), finding the optimal sensing
time is still solvable, which can be achieved by the exhaustive
search. The algorithm that can maximize S E + (1 ) E E
is very similar to Algorithm 1, and is not stated here.
IV. BALANCING SE AND EE
It has been proved that, for any given M, SE is a unimodal
opt
function of ts and there exists only one optimal value ts,S E

496

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

Fig.
 7. Illustration ofSE and EE for the CR network with high SE requirement
opt
th
S E S E (ts,E E ) .

Fig.
 8. Illustration ofSE and EE for the CR network with low SE requirement
opt
th
S E < S E (ts,E E ) .

Algorithm 3. Maximizing EE while satisfying SE requirement


that maximizes S E , EE is a unimodal function of ts under
some constraints on the power parameters and there exists only
opt
one optimal value ts,E E that maximizes E E . In general, the
maximal SE and the maximal EE can not be obtained simultaneously. Methods for increasing SE may result in a decrease in
EE. Thus, how to balance SE and EE is a problem which should
be considered. In this section, we solve this problem from the
following two perspectives.
 Maximizing EE while satisfying SE requirement. This
criterion can be employed when the energy of the CR network
is constrained, for instance, the SUs are handheld devices which
are battery-powered. In order to satisfy the SE requirement, we
th
set S E th
S E , where S E is the threshold of SE requirement.
th
For a given S E , we suppose that there are two roots for the
equation S E = th
S E (If there is one root, the problem can be
le f t
analyzed in a similar way). And they are assumed to be ts,S E
right

le f t

opt

for each M [1, K ] do


opt
Using Bisection algorithm to compute the root (ts,S E (M))
le f t

right

of (16), the roots ( ts,S E (M) and ts,S E (M)) of S E th


SE =
opt
0, and the root (ts,E E (M)) of (28).
opt

le f t

if ts,S E (M) < ts,S E (M) then


Determine
the optimal sensing time

 by comparing
le f t
right
E E ts,S E (M) and E E ts,S E (M) ;
else
opt
Choose ts,E E (M) as the value of sensing time.
end if
end for
Compare E E (M), choose the maximal one.

right

and ts,S E , where ts,S E < ts,S E < ts,S E . The SE requirement is


le f t
right
converted as the boundaries of the sensing time ts,S E , ts,S E .
For the CR network with high SE requirement, th
S E may be
opt
opt

larger than or equal to S E (ts,E E ), i.e., th


S E (ts,E E ), as
SE
depicted in Fig. 7. In this case, E E is a monotonic function of ts
le f t
right
for ts,S E < ts < ts,S E . To maximize the EE, the optimal value
le f t

of sensing time can be determined by comparing E E (ts,S E )


right

and E E (ts,S E ).
For the CR network with low SE requirement, th
S E may be
opt
opt
th
lower than S E (ts,E E ), i.e., S E < S E (ts,E E ), as depicted in
le f t

opt

Fig. 8. In this case, E E is increasing in ts for ts,S E < ts ts,E E


opt

right

and is decreasing in ts for ts,E E < ts < ts,S E . To maximize the


opt
EE, ts,E E should be chosen as the value of sensing time.
Algorithm 3 that maximizes EE while satisfying SE requirement is stated as follows.
 Maximizing SE while satisfying EE requirement. This
criterion can be employed for some multimedia applications
with abundant energy. In order to satisfy the EE requirement,
th
we set E E th
E E , where E E is the threshold of EE requireth
ment. For a given E E , we assume that there are two roots for
the equation E E = th
E E (If there is one root, the problem can
le f t
be analyzed in a similar way). And they are assumed to be ts,E E

Fig.
 9. Illustration ofSE and EE for the CR network with high EE requirement
opt
th
E E E E (ts,S E ) .
right

le f t

opt

right

and ts,E E , where ts,E E < ts,E E < ts,E E . The EE requirement is


le f t
right
converted as the boundaries of the sensing time ts,E E , ts,E E .
For the CR network with high EE requirement, th
E E may be
opt
opt
th
larger than or equal to E E (ts,S E ), i.e., E E E E (ts,S E ), as
depicted in Fig. 9. In this case, S E is a monotonic function of ts
le f t
right
for ts,E E < ts < ts,E E . To maximize the SE, the optimal value
of sensing time can be determined by comparing comparing
le f t
right
S E (ts,E E ) and S E (ts,E E ).
For the CR network with low EE requirement, th
E E may be
opt
opt
lower than E E (ts,S E ), i.e., th
<

(t
),
as
depicted
in
E E s,S E
EE

Fig. 10. It is assumed that there is only one root ts,E


for
the
E

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

497

Fig. 10. Illustration of SE and EE for the CR network with low EE requirement

opt
th
E E < E E (ts,S E ) .

equation E E = th
E E . In this case, S E is increasing in ts for
opt
opt

0 < ts ts,S E and is decreasing in ts for ts,S E < ts < ts,E


E.
opt
To maximize the SE, ts,S E should be chosen as the value of
sensing time.
Algorithm 4 that maximizes SE while satisfying EE requirement is very similar to Algorithm 3, and is not stated here.

Fig. 11. The SE & EE vs sensing time for different fusion rules.

V. S IMULATION R ESULTS
Simulations are presented in this section to evaluate the performance of our proposed sensing strategies. We suppose that
the PU is a BPSK modulated signal and the noise is real-valued
Gaussian variable unless otherwise stated. It is assumed that
only one of the SUs is allowed to transmit data to its receiver
if the PU is sensed to be absent. Thus, the SE and EE tradeoff is investigated for only one secondary link. The sampling
frequency f s is equal to the bandwidth of PUs signal, with a
value of 10000 Hz. The number of SUs is K = 8; the probability that PU is absent is = 0.7; Q th
D = 0.9 unless otherwise
stated; the frame duration is T = 100 ms; the individual reporting time is much smaller than T and is set as tr = 0.1 ms; the
transmit power Pt and the circuit power Pc are assumed to be
1.8 W and 80 mW respectively; the sensing power is Ps = 40
mW unless otherwise stated; the reporting power is smaller and
is set as Pr = 10 mW; the SNR for the secondary link is S = 5
dB; = 10 dB unless otherwise stated.
Fig. 11 illustrates the effects of the sensing time as well as the
fusion rule on SE and EE. For any fusion rule, there exists one
opt
optimal value ts,S E that maximizes SE and one optimal value
opt
ts,E E that maximizes EE. The maximal SE and the maximal EE
can not be obtained simultaneously. When our proposed algorithms are employed to optimize the final decision threshold M,
the SE and EE can be improved significantly. When ts initially
increases, OR fusion rule is sub-optimal. However, when ts is
increased further and longer time is allocated as spectrum sensing, Majority (MA) fusion rule becomes sub-optimal. AND
fusion rule is worst either for SE or EE. Therefore, in order
to improve SE and EE, the final decision threshold M is an
important parameter that should be optimized.
The performance of SE and EE with respect to final decision
threshold M is illustrated in Fig. 12. The sensing time is optimized in this simulation. When M = 3, the SE is maximized

Fig. 12. The SE & EE vs the final decision threshold M; sensing time is
optimized.

with a value of 0.95 bits/s/Hz, which is consistent with the


maximum value shown in Fig. 11. And the EE is maximized
when M = 2. Compared with the AND (M = K ) fusion rule,
the SE and EE can be improved significantly when the final
decision threshold is optimized.
In the text followed, we compare the energy-efficient design
with the spectrum-efficient design. The goal of the energyefficient design is to design sensing parameters that can maximize EE, while the goal of the spectrum-efficient design is to
design sensing parameters that can maximize SE. The EE and
the SE are shown in Fig. 13 and Fig. 14, respectively. It is seen
that the EE and the SE calculated by our proposed algorithms
are the same as those employing exhaustive search, which indicates that our proposed algorithms can find the optimal sensing
parameters. In Fig. 13, when S N R = 16 dB, the EE of the
energy-efficient design is about 0.3 bits/Joule/Hz higher than
that of the spectrum-efficient design. The gap decreases as
increases. At 0 dB SNR, the two designs have the same energy
efficiency. In general, the methods that can maximize EE are
not able to obtain the maximum SE, and vice versa. This is
because the energy-efficient design and the spectrum-efficient
design have different targets.

498

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

Fig. 13. EE comparison of spectrum-efficient design and energy-efficient


design.

Fig. 15. EE vs SE (ts and M are jointly optimized to maximize S E +


(1 ) E E ).

Fig. 14. SE comparison of spectrum-efficient design and energy-efficient


design.

Fig. 15 shows the EE versus the SE when ts and M are


jointly optimized to maximize S E + (1 ) E E . We search
M from 1 to K to find optimal M and we use Bisection
search over ts , hence the computation complexity is K , where
tr
) is the number of iterations that Bisection
= log2 ( T K

method takes to terminate and x denotes the smallest integer not less than x [23]. It is seen that the SE and EE with
complex-valued PSK signal and CSCG noise is higher than that
with BPSK signal and real-valued Gaussian noise. The value
of varies from 0 to 1. The case of = 0 corresponds to
the energy-efficient design, and the case of = 1 corresponds
to the spectrum-efficient design. It has been shown that the
maximal EE and the maximal SE can not be obtained simultaneously, there is a tradeoff between the SE and EE. The value
of can be selected properly to balance the EE and the SE.
When 0 < < 1, selecting a right operating point (tradeoff) merits discussion in an operational network. In Fig. 16, we
present the SE and EE versus when ts and M are jointly optimized. Note that M is an integer and 1 M K . The curves
are not smooth due to the fact that we will choose different pairs
of ts and M for different values of . When the energy of the

Fig. 16. SE & EE vs when ts and M are jointly optimized to maximize


S E + (1 ) E E .

CR network is constrained, for example, the SUs are batterypowered devices, we want to maximize EE while satisfying SE
requirement. If the SE is required to be not lower than 0.85
bits/s/Hz, the value of should be not less than 0.2. However,
to maximize EE while satisfying SE requirement, we should
choose = 0.2. When the CR network is supplied with abundant energy, the goal may be maximizing SE while satisfying
EE requirement. In this case, the proper value of can also be
determined in a similar way.
Fig. 17 is simulated to show the EE versus the threshold
of SE requirement th
S E for different fusion rules. For the CR
opt
network with low SE requirement, th
S E < S E (ts,E E ), the maxopt
imal EE can be obtained since the optimal sensing time ts,E E


le f t
right
is always located in the region ts,S E , ts,S E . Thus, EE keeps
the maximal value when the threshold of SE requirement initially increases. However, for the CR network with high SE
opt
requirement, th
S E S E (ts,E E ), the EE decreases as the threshopt
old of SE requirement becomes higher. This is because ts,E E is


le f t
right
le f t
not located in the region ts,S E , ts,S E and we choose ts,S E as
le f t

the value of sensing duration, and E E (ts,S E ) decreases as th


SE

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

499

the power parameters. And we have proposed Algorithm 1 and


Algorithm 2 to maximize SE and EE respectively. Then, based
on the solutions of the two special cases, an efficient algorithm
is proposed to solve the general problem for SE-EE tradeoff.
Then, the EE (SE) is maximized via joint optimization of sensing duration and final decision threshold under the constraint
that the SE (EE) requirement is satisfied. Computer simulations
show that different SE (or EE) requirements need different optimal values of sensing duration and final decision threshold, and
there exists a tradeoff between the SE and EE.
A PPENDIX A
P ROOF OF P ROPOSITION 1
We rewrite
(ts ) as
Fig. 17. EE vs threshold of SE requirement for different fusion rules.

(ts ) = ln{
1 (ts )
2 (ts )},

(37)

where

1 (ts ) =

M1

i=0

2 (ts ) =


K iM+1
pf
(1 p f ) Mi ,
i
4

K (T ts K tr )

K 1
M1

ts
.
fs

(38)

(39)

The first order derivative of


(ts ) versus ts is
[d
1 (ts )/dts ] [d
2 (ts )/dts ]
d
(ts )

= u(ts ) + (ts ),
=
dts

1 (ts )

2 (ts )
(40)
where
Fig. 18. SE vs threshold of EE requirement for various values of target
detection probability.

increases. In addition, to improve EE, there is a need to find the


optimal final decision threshold M.
In Fig. 18, the SE is presented versus the threshold of EE
requirement for various values of target detection probability.
The optimal final decision threshold M is employed. For the
opt
CR network with low EE requirement, th
E E < E E (ts,S E ), the
SE keeps the maximal value when the threshold of EE requirement initially increases. However, for the CR network with high
opt
EE requirement, th
E E E E (ts,S E ), the SE decreases as the
threshold of EE requirement becomes higher. It is also observed
that the SE decreases as the target detection probability Q th
D
increases. The value of Q th
D implies the level of protection to
the primary user. Thus, strengthening protection to the primary
user will result in a lower SE.
VI. C ONCLUSION
In CR networks, the spectrum-efficiency sometimes conflicts
with the energy-efficiency. In this paper, we have studied the
SE-EE tradeoff problem with cooperative sensing. For the two
special cases of the general problem, it has been proved that
the SE is a unimodal function of the sensing duration, and
the unimodal property of EE is based on some constraints on


 M1i
 M1
K
1
i=0
i (M i) p f 1
 M1i
 M1
K  1

1
i=0
i
pf

 Mi
 M1
K
1
(M

i)

1
i=0
i
pf
1

+
,
 Mi




pf
M1 K
1

1
i=0
i
pf

dp f 1
u(ts ) =

dts 1 p f

(41)
(ts ) =
Since

dp f
dts < 0,
d
(ts )
dts <

Thus,
proved.

T + ts K tr
.
2ts (T ts K tr )

(42)

we have u(ts ) < 0. And obviously, (ts ) < 0.


0,
(ts ) is decreasing in ts . Proposition 1 is

A PPENDIX B
T HE C ONSTRAINTS T HAT M AKE v(ts ) < 0.
The first order derivative of Z (ts ) versus ts is d Zdt(ts s ) =
u(ts ) + v(ts ), where u(ts ) < 0 can be referred to Equation (41),
and v(ts ) is computed by
v(ts ) =

G(ts )
,
H (ts )

(43)

500

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 2, FEBRUARY 2016

where
H (ts ) = 2ts (T ts K tr )[K ts Ps + K tr Pr + K (ts + tr )Pc ]
+ (1 )(1 Q th
D )(Pt + Pc )(T ts K tr )]
> 0,
(44)
G(ts ) = ats2 + bts + c,



ts f s


1
have
() < 0,
2(2 +1) <
2


ts f s


1
2
2 . Based on the fact that Q(x) is decreas


ing in x, we derive pd > p f . Thus, p1f 1 1pdpd > 1,
> 0. Proposition 2 is proved.
we


(45)
R EFERENCES

and
a = 3(1 )(1 Q th
D )(Pt + Pc ) 3K (Ps + Pc ),
b = K (Ps + Pc )(T K tr ) K (Pr + Pc )tr
2(1 )(1 Q th
D )(Pt + Pc )(T K tr ),

(46)
(47)

c = K (Pr + Pc )(T K tr )tr


2
(1 )(1 Q th
D )(Pt + Pc )(T K tr ) .

(48)

Next, we investigate the value of G(ts ) over the range


0 < ts < T K tr . First, it is derived that
lim G(ts ) = c < 0,

(49)

ts 0

lim

ts T K tr

G(ts )

= 2(T K tr )[K ts Ps + K tr Pr + K (ts + tr )Pc ] < 0.


(50)
According to the property of quadratic function, if a < 0,
b
b > 0, b
2a < T K tr and G( 2a ) > 0, v(ts ) may not be less
than 0. In this case, it is difficult to analyze the relationship
between Y (ts ) and Z (ts ), the EE may not be a unimodal function of ts . The above four inequations can be equivalent to b > 0
and b2 > 4ac.
Therefore, if the sensing power, the reporting power, the circuit power and the transmit power do not satisfy b > 0 and
b2 > 4ac, v(ts ) < 0.
A PPENDIX C
P ROOF OF P ROPOSITION 2
According
to
(1)



ts f s

Q 2 1
2(2 +1) ,
We define
 

() =

1
2
As

and (2),
pd =


ts f s

.
pf = Q
1
2
2

 
 t f
ts f s
s s

.
1
2(2 + 1)
2
2
(51)

1 1 + 2
()
ts f s
=
< 0,
2

2(2 + 1)

and
lim () =

0



2
1 + 2 1 + 2 +

(52)


ts f s
< 0,
2(2 + 1)
(53)

[1] H. Bogucka, P. Kryszkiewicz, and A. Pliks, Dynamic spectrum aggregation for future 5G communications, IEEE Commun. Mag., vol. 53, no. 5,
pp. 3543, May 2015.
[2] X. Huang, T. Han, and N. Ansari, On green energy powered cognitive
radio networks, IEEE Commun. Surv. Tuts., vol. 17, no. 2, pp. 827842,
May 2015.
[3] C.-X. Wang et al., Cellular architecture and key technologies for 5G
wireless communication networks, IEEE Commun. Mag., vol. 52, no. 2,
pp. 122130, Feb. 2014.
[4] C. He, B. Sheng, P. Zhu, X.-H. You, and Y. Li, Energy- and spectralefficiency tradeoff for distributed antenna systems with proportional
fairness, IEEE J. Sel. Areas Commun., vol. 31, no. 5, pp. 894902, May
2013.
[5] R. Amin, J. Martin, J. Deaton, L. A. Dasilva, A. Hussien, and A. Eltawil,
Balancing spectral efficiency, energy consumption, and fairness in future
heterogeneous wireless systems with reconfigurable devices, IEEE J.
Sel. Areas Commun., vol. 31, no. 5, pp. 969980, May 2013.
[6] Y.-C. Liang, K.-C. Chen, G. Y. Li, and P. Mahonen, Cognitive radio networking and communications: an overview, IEEE Trans. Veh. Technol.,
vol. 60, no. 7, pp. 33863407, Sep. 2011.
[7] S. Atapattu, C. Tellambura, and H. Jiang, Energy detection based cooperative spectrum sensing in cognitive radio networks, IEEE Trans.
Wireless Commun., vol. 10, no. 4, pp. 12321241, Apr. 2011.
[8] D. Treeumnuk, S. L. Macdonald, and D. C. Popescu, Optimizing performance of cooperative sensing for increased spectrum utilization in
dynamic cognitive radio networks, in Proc. IEEE Int. Conf. Commun.,
2013, pp. 46564660.
[9] Q. Chen, M. Motani, W.-C. Wong, and A. Nallanathan, Cooperative
spectrum sensing strategies for cognitive radio mesh networks, IEEE
J. Sel. Topics Signal Process., vol. 5, no. 1, pp. 5667, Feb.
2011.
[10] E. Peh, Y.-C. Liang, Y. L. Guan, and Y. Zeng, Optimization of cooperative spectrum sensing in cognitive radio networks: A sensing-throughput
tradeoff view, IEEE Trans. Veh. Technol., vol. 58, no. 9, pp. 52945299,
Nov. 2009.
[11] W. Yin, P. Ren, and C. Zhang, A joint sensing-time adaption and data
transmission scheme in cognitive radio networks, in Proc. IEEE Global
Commun. Conf., 2011, pp. 15.
[12] S. Stotas and A. Nallanathan, On the throughput and spectrum sensing enhancement of opportunistic spectrum access cognitive radio networks, IEEE Trans. Wireless Commun., vol. 11, no. 1, pp. 97107, Jan.
2012.
[13] Y. Pei, Y.-C. Liang, K. C. Teh, and K. H. Li, Energy-efficient design
of sequential channel sensing in cognitive radio networks: Optimal sensing strategy, power allocation, and sensing order, IEEE J. Sel. Areas
Commun., vol. 29, no. 8, pp. 16481659, Sep. 2011.
[14] Z. Shi, K. C. Teh, and K. H. Li, Energy-efficient joint design of sensing
and transmission durations for protection of primary user in cognitive
radio systems, IEEE Commun. Lett., vol. 17, no. 3, pp. 565568, Mar.
2013.
[15] Y. Gao, W. Xu, K. Yang, K. Niu, and J. Lin, Energy-efficient transmission with cooperative spectrum sensing in cognitive radio networks, in
Proc. IEEE Wireless Commun. Netw. Conf., 2013, pp. 712.
[16] E. Peh, Y.-C. Liang, Y. L. Guan, and Y. Pei, Energy-efficient cooperative spectrum sensing in cognitive radio networks, in Proc. IEEE Global
Commun. Conf., 2011, pp. 15.
[17] Y.-C. Liang, Y. Zeng, E. Peh, and A. T. Hoang, Sensing-throughput
tradeoff for cognitive radio networks, IEEE Trans. Wireless Commun.,
vol. 7, no. 4, pp. 13261337, Apr. 2008.
[18] J. Shen, T. Jiang, S. Liu, and Z. Zhang, Maximum channel throughput via cooperative spectrum sensing in cognitive radio networks,
IEEE Trans. Wireless Commun., vol. 8, no. 10, pp. 51665175, Oct.
2009.

HU et al.: ON THE SPECTRUM- AND ENERGY-EFFICIENCY TRADEOFF IN COGNITIVE RADIO NETWORKS

[19] F. F. Digham, M. S. Alouini, and M. K. Simon, On the energy detection of unknown signals over fading channels, in Proc. IEEE Int. Conf.
Commun., 2003, pp. 35753579.
[20] E. Peh, Y.-C. Liang, Y. L. Guan, and Y. Zeng, Cooperative spectrum sensing in cognitive radio networks with weighted decision fusion
schemes, IEEE Trans. Wireless Commun., vol. 9, no. 12, pp. 38383847,
Dec. 2010.
[21] N. Kingsbury. (2005, Jun.). Approximation Formulae for the Gaussian
Error Integral, Q(x), Connexions [Online]. Available: http://cnx.org/
content/m11067/2.4/
[22] S. C. Chapra and R. P. Canale, Numerical Methods for Engineers, 6th ed.
New York, NY, USA: McGraw-Hill, 2010.
[23] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.

Hang Hu (S12) received the B.S. degree in telecommunications engineering from Xidian University,
Xian, China, and the M.S. degree in communications engineering from PLA University of Science
and Technology, Nanjing, China, in 2010 and 2012,
respectively. He is currently pursuing the Ph.D.
degree in information and communications engineering at the College of Communications Engineering,
PLA University of Science and Technology. His
research interests include cognitive radio technology, cooperative communications, and green
communications.

Hang Zhang (M12) received the B.S. degree


from PLA University of Science and Technology,
Nanjing, China, and the M.S. degree from Southeast
University, Nanjing, China, in 1984 and 1989, respectively. She is currently a Professor and also a Ph.D.
Supervisor with PLA University of Science and
Technology. Her research interests include wireless
communication, satellite communication, and signal
processing in communications.

501

Ying-Chang Liang (F11) is a Principal Scientist and


a Technical Advisor with the Institute for Infocomm
Research (I2R), Singapore, and a Professor with
the University of Electronic Science and Technology
of China (UESTC), Chengdu, China. He was a
Visiting Scholar at the Department of Electrical
Engineering, Stanford University, Stanford, CA,
USA, from December 2002 to December 2003,
and was an Adjunct Faculty with the National
University of Singapore, Singapore, and Nanyang
Technological University, Singapore, from 2004 to
2009. His research interests include wireless networking and communications,
with current focus on applying artificial intelligence, big data analytics, and
machine learning techniques to wireless network design and optimization.
He was elected a Fellow of the IEEE in December 2010, and was recognized by Thomson Reuters as a Highly Cited Researcher in 2014 and 2015.
He is now serving as the Chair of IEEE Communications Society Technical
Committee on Cognitive Networks. He is on the Editorial Board of the IEEE
Signal Processing Magazine, and the IEEE T RANSACTIONS ON S IGNAL
AND I NFORMATION P ROCESSING OVER N ETWORKS , and is an Associate
Editor-in-Chief of the World Scientific Journal on Random Matrices: Theory
and Applications. He served as the Founding Editor-in-Chief of the IEEE
J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS (Cognitive Radio
Series), and was the key Founder of the new journal the IEEE T RANSACTIONS
ON C OGNITIVE C OMMUNICATIONS AND N ETWORKING . He was an Editor
of the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS from 2002
to 2005, an Associate Editor of the IEEE T RANSACTIONS ON V EHICULAR
T ECHNOLOGY from 2008 to 2012, and a Leading Guest Editor of five special
issues on emerging topics published in the IEEE, EURASIP, and Elsevier journals. He was a Distinguished Lecturer of the IEEE Communications Society
and the IEEE Vehicular Technology Society, and has been a member of the
Board of Governors of the the IEEE Asia-Pacific Wireless Communications
Symposium since 2009. He served as the Technical Program Committee (TPC)
Chair of CROWN08 and DySPAN10, the Symposium Chair of ICC12 and
GLOBECOM12, and the General Co-Chair of ICCS10 and ICCS14. He
serves as an Executive Co-Chair and the TPC Chair of GLOBECOM17 to be
held in Singapore. He was the recipient of the IEEE Jack Neubauer Memorial
Award in 2014, the first IEEE Communications Society APB Outstanding
Paper Award in 2012, and EURASIP Journal of Wireless Communications and
Networking Best Paper Award in 2010. He was also the recipient of the Institute
of Engineers Singapore (IES)s Prestigious Engineering Achievement Award
in 2007, and the IEEE Standards Associations Outstanding Contribution
Appreciation Award in 2011, for his contributions to the development of IEEE
802.22 standard.

You might also like