You are on page 1of 5

Opportunistic Spectrum Access in Cognitive Radio

System Employing Cooperative Spectrum Sensing


1 Institute

Hui Huang1 , Zhaoyang Zhang1 , Peng Cheng2 , and Peiliang Qiu1


of Information and Communication Engineering, Zhejiang University, Hangzhou, China
2 Intel Communications Technology Lab (CTL), Beijing, China
Corresponding author: ning ming@zju.edu.cn

Abstract In cognitive radio systems, as the sensing ability


of single cognitive user is limited, the opportunistic spectrum
access has to allow different cognitive users to cooperatively
search for and exploit instantaneous spectrum availability. In this
paper, we address the design of opportunistic spectrum access for
cognitive radio system employing cooperative spectrum sensing.
The opportunistic spectrum access strategy contains two parts:
an access policy to maximize the transmission throughput in each
channel, and a cooperative spectrum sensing strategy to maximize
the overall transmission throughput of the cognitive network. We
first propose the optimal access policy as an optimal tradeoff
point between transmission collision and overlooked opportunity
in each channel. Then the optimal cooperative spectrum sensing
strategy is studied to solve the following questions: to what extent
each cognitive user should cooperate with others and which part
of spectrum it should choose to sense. Simulation results are
given to demonstrate the promising performance.

I. I NTRODUCTION
Public mobile radio spectrum has become a scarce resource
due to the recent boom in wireless technologies. The unlicensed spectrums has been extremely crowded since most
of the wireless technological innovations happen in these
frequency ranges. On the contrary, actual measurements show
that most of the licensed frequency ranges are vastly underutilized at some specific location and time [1]. Cognitive radio
(CR) [2], [3], is proposed as a key technology to deal with
this increasingly tense situation in spectrum use. It enables
public access to those unused spectral bands by allowing the
unlicensed (cognitive) users to opportunistically exploit the
licensed spectrum, and therefore the overall spectral efficiency
can be economically increased [4].
As the cognitive radio has lower priority of the licensed
spectrum, it should transmit without causing any significant
interference to the licensed user (the primary user). How
to sense the spectrum in a reliable manner becomes one of
the major challenges that confront the cognitive radio. Due
to channel fading and shadowing effects, the received signal
strength at the cognitive user may be severely degraded, and
the sensing performance of the single detector is limited. Thus,
cooperative spectrum sensing, which can enhance the sensing
performance, has attracted extensive attention. It has been
proved that relying on the variability of signal strength at
This work was supported by National Basic Research Program of China
(973 Program) (No. 2009CB320405), National Natural Science Foundation of
China (No. 60802012),National High Technology Research and Development
Program (No. 2007AA01Z257), and Natural Science Foundation of Zhejiang
Province (No. Z104252), China.

various locations, cooperative sensing can make the cognitive


system robust to severe or poorly modelled fading environments without drastic requirements on individual cognitive
radios [5].
In literature, a number of opportunistic spectrum access
strategies have been proposed for cognitive network. Most of
them are based on the assumption that each of the cognitive
users has the full spectrum sensing ability for wide spectrum
band. However, due to hardware limitation, the cost to achieve
wide-band spectrum sensing by one single cognitive user is
quite high. And it is realistic that a cognitive user is able to
sense limited bandwidth of spectrum during a certain amount
of time [6] [7]. This sensing limitation brings new challenges
in opportunistic spectrum access in cognitive radio system. To
our knowledge, [7] is the first work that takes into account the
limited sensing ability of each user in cognitive radio system.
Based on the theory of Partially Observable Markov Decision
Process (POMDP), it presents an opportunistic spectrum access strategy for single cognitive user. Lai et al. [8] propose a
cognitive medium access strategy for the single user scenario,
and extends it to the multi-user and multi-channel cases. But
the cooperation among different cognitive users is not taken
into account. [9] proposes the opportunistic spectrum access
protocols based on two collaborative channel spectrum sensing
policies. However, just like in [8], the detection error and
the consequent performance degradation is left beyond any
consideration, and thereby the proposed protocols are still not
very perfect and practical.
In this paper, we consider the design of opportunistic
spectrum access strategy in cognitive network employing
cooperative spectrum sensing. This opportunistic spectrum
access design integrates two basic components: the cooperative spectrum sensing strategy and the access policy. We
assume that the spectrum sensing by single cognitive user
is only conducted within a small portion of spectrum, and
the sensing results are imperfect. The transmission throughput
performance of this cognitive network is investigated by given
the tolerable collision probability of each channel. The optimal
access policy of each channel is then presented when the
cooperative sensing performances are given. And the resulted
average transmission throughput function of each channel is
analyzed to obtain the optimal cooperative spectrum sensing
strategy. At last, we propose the opportunistic spectrum access
strategy with the aim of maximizing the cognitive networks
average transmission throughput. The remainder of the paper

978-1-4244-2517-4/09/$20.00 2009 IEEE

Frequency

t=1

t=2

t=3

...
...

Channel 1
Channel 2

...

t=T

...

...

...

...

...

...

Channel M

Time
Occupied by the primary network
Spectrum opportunities

Fig. 1.

Channel model.

is organized as follows. In section II we introduce the system


model of a cognitive network employing cooperative spectrum
sensing and formulate the problem. Section III investigates the
optimization problem, and proposed an opportunistic spectrum
access strategy to solve it. Section V provides the simulation
results. Conclusions are drawn in Section VI.
II. S YSTEM M ODEL A ND P ROBLEM F ORMULATION
A. System description
As shown in fig. 1, we consider a spectrum consisting of M
channels in this paper. Each channel has the same bandwidth
B. This spectrum is licensed to a primary network, whose
users are operated in a synchronous time-slotted fashion. Since
the primary network does not use the whole spectrum all
the time, it is assumed that the probability of channel i
being occupied by the primary network in one time slot is
i (0 i 1, i = 1, , M ). The value of i depends
on the channel allocation algorithm and traffic statistics of
the primary network, and does not change in a sufficiently
long time. We use Si {0(idle), 1(occupied)} to denote the
availability of channel i in a time slot.
A cognitive network attempts to exploit the spectrum opportunities in these M channels. We assume there are N
cognitive users in this network. According to the Shannon
theory [10], when channel i is used for cognitive transmission,
the theoretical throughput upper bound Ri is:
Ri = B log(1 + SN Rtrans,i );

(1)

where SN Rtrans,i is the received signal to noise ratio of cognitive transmission using channel i. Therefore if the cognitive
network can exploit more channels and fully utilize them,
significant throughput increase can be achieved.
For the protection of the primary network, a reliable spectrum sensing must be performed before cognitive transmission.
For a given channel i, the goal of spectrum sensing is to decide
between the flowing two hypotheses:

0, H0
(2)
Si =
1, H1
Due to some certain constraints, such as hardware limitation, time consumption limitation, etc., it is difficult for a

single sensor to achieve wide-band spectrum sensing. One cognitive user is able to sense limited bandwidth of spectrum at
one time. This bandwidth varies for different spectrum sensing
approaches and different hardware limitations. Without loss
of generality, in the following discussion we assume that a
cognitive user can sense only one channel once. Moreover,
the sensing performance of single cognitive user may be
degraded due to the channel fading and the shadowing effects.
Therefore, the cognitive network uses cooperative spectrum
sensing to achieve wide-band spectrum sensing and increase
the probability of detection.
In such a cognitive network, each time slot is divided
into three successive stages: sensing stage, reporting stage
and transmission stage.1 At the sensing stage, each cognitive
user performs spectrum sensing independently using the same
detection approach. These local sensing results are then transmitted to the fusing center through a control channel at the
reporting stage. For the bandwidth limitation of the control
channel [5], all the cognitive users only transmit the final
sensing results (H0 or H1 ) instead of the raw data of sensing
observations. According to these local sensing results, the
fusing center makes the cooperative sensing results. To further
limit the interference to the primary user and to simplify the
following discussion, the OR-rule (1-out-of-n rule) [11] is used
as the fusion rule. It can be described as follow: if any of the
individual sensing results is H1 , the cooperative sensing result
is H1 ; otherwise, H0 is decided. Finally, at the transmission
stage, the cognitive network transmits using a set of channels
according to the cooperative sensing results and the access
policy. The access policy specifies the access decision of each
channel based on the cooperative sensing results. As the M
channels are independent of each other, the access policy can
be modelled as the accessing probability of each channel when
its cooperative sensing result is given.
B. Problem formulation
Let n = [n1 , n2 , ..., nM ] be the number of cognitive users
choosing each channel to sense. In practical system, the distance from the primary user to the cognitive network is much
greater than the radius of the cognitive network. Therefore, the
primary signal received by each cognitive user experiences
almost identical path loss, and the average receiving SINRs
are equal. And it is reasonable to assume that all the ni
cognitive users sensing channel i have the same miss-detection
probability Pmiss,i and false-alarm probabilities Pfalse,i . As the
cooperative sensing result is decided using the OR-rule, the
miss-detection and false-alarm probabilities of the cooperative
sensing result for channel i can be obtained as follows:
Qmiss,i = Pmiss,i ni ;

(3)
ni

Qfalse,i = 1 (1 Pfalse,i )

(4)

The access policy for channel i can be modelled as a pair of


transmission probabilities (fi (0) , fi (1)), where fi (k) is the
transmission probability of channel i when the cooperative
1 Notice that this assumes that the time slot is sufficiently long to allow an
appropriate detection time interval for the spectrum sensing.

sensing result is Hk , k = 0, 1. As mentioned above, channel


i is occupied by the primary network with probability i . For
channel i, the cognitive transmission is successful only when
the channel is idle, so the successful transmission probability
for channel i can be expressed as:
Ptrans,i = (Qf alse,i fi (1) + (1 Qf alse,i ) fi (0)) (1 i ) .
(5)
As we all know, when a transmission collision occurs, the
primary transmission will be interfered. Therefore, for the
protection of the primary network, the collision probability of
each channel should be constrained below a pre-determined
threshold i (i = 1, , M ).2 And the collision probability of
channel i should be:
Pcolli,i = (1 Qmiss,i ) fi (1) + Qmiss,i fi (0) i .

(6)

The value of i depends on the channel condition and the


transmission throughput requirement of the primary network.
Based on the previous analysis, the objective of the opportunistic spectrum access strategy is to optimally allocate the
M channels to N cognitive users for cooperative spectrum
sensing and choose an optimal access policy for each channel
under the collision probability constraints, in order to maximize the cognitive networks average transmission throughput.
And the optimization problem can be given as:
M


max

(Qf alse,i fi (1) + (1 Qf alse,i ) fi (0)) (1 i ) Ri

i=1

s.t.
(1 Qmiss,i ) fi (1) + Qmiss,i fi (0) i
ni
Qmiss,i = Pmiss,i
ni

Qf alse,i = 1 (1 Pf alse,i )
M


ni N.

(7)

i=1

The optimization variables of this problem are the number of


cooperative sensing users for each channel (ni ) and the access
policy (fi (0) , fi (1)), where i = 1, 2, ..., M .
III. T HE P ROPOSED O PPORTUNISTIC S PECTRUM ACCESS
S TRATEGY
In this section we will solve the optimization problem shown
in (7). Obviously, this is a non-convex optimization problem,
and it is hard to find the closed-form joint optimal design of
the number of the cooperative sensing users ni and the access
policy (fi (0) , fi (1)). We node that for a given channel i, the
false-alarm probability Qf alse,i and miss-detection probability
Qmiss,i are independent of the access policy (fi (0) , fi (1)).
Thus, according to Proposition 2 in [7], the optimal access
policy (fi (0) , fi (1)) for a chosen channel i can be expressed
as:


1, i Qmiss,i , Qmiss,i < i
 1Qmiss,i
(fi (0) , fi (1)) =
(8)
i

Qmiss,i i
Qmiss,i , 0 ,
2 We assume that the primary network is delay-insensitive, and no constraints are imposed on the average delay in the following discussion.

As shown in (3), the miss-detection probability Qmiss,i


depends on the number of cooperative sensing users ni . The
turning point Qmiss,i = i can then be changed into:

log i
ni =
,
(9)
log Pmiss,i
where x is the greatest integer less than or equal to x.
Applying the optimal access policy (fi (0) , fi (1)) given in
(8) to the optimization problem (7), it is easy to obtain the
average transmission throughput of channel i when there are
ni cognitive users choosing channel i to sense:


n

f alse,i ) i )
1 (1i )(1(1P
(1i )Ri , ni > i
ni
1Pmiss,i
Gi (ni )=
ni
i (1Pf alse,i )

(1 i ) Ri ,
ni i
ni
Pmiss,i
(10)
where i is used to denote the turning point of channel i shown
in (9).
Obviously, Gi (ni ) is non-convex. However, it has a nice
property as stated in the following proposition:
Proposition 1: For a chosen channel i, its average transmission throughput Gi (ni ) is increasing and concave when
the number of cooperative users ni i . And the maximum
average transmission throughput is reached when ni = i or
ni = i + 1.
Proof: See Appendix.
Then we can easily obtain the following corollary:
Corollary 1: Let n = [n1 , ..., nM ] be the optimal number
of cognitive users choosing each channel to sense. There is at
most one channel k, 1 k M , satisfying 0 < nk < k .
And for all the other i = k, ni {0, i , i + 1}.
Proof: Firstly, if there is any channel l satisfying nl >
l + 1, according to Proposition 1, it is obviously that
Gl (nl ) < Gl (l + 1) .

(11)

So the cognitive networks average transmission throughput


can be increased if the number of cognitive users sensing
channel l is set as l + 1.
Then, we consider the case when there are two channels,
k and k  , 0 < nk < k and 0 < nk < k . Without
loss of generation, we assume that Gk (nk ) Gk (nk 1)
Gk (nk ) Gk (nk 1). As Gi (ni ) is increasing and concave when ni i , if k nk + nk , we can prove that:
Gk (nk )+Gk (nk ) (Gk (nk ) Gk (nk 1)) nk + Gk (nk )
(Gk (nk )Gk (nk 1)) nk + Gk (nk )
< Gk (nk + nk ) Gk (nk ) + Gk (nk )
= Gk (nk + nk ) .
(12)
Similarly, when ni > i , it can be proved that:
Gk (nk ) + Gk (nk ) < Gk (nk + nk k ) + Gk (k ).
(13)
The inequations (12) and (13) are in conflict with the assumption that nk and nk are the optimal number of cognitive
users sensing channel k and channel k  . And this discussion

10

+1

8
10
12
14
the number of the cognitive users

16

18

20

1
0.8

f (0)

0.6

f (1)

i
i

0.4

+1

0.2
2

8
10
12
14
the number of the cognitive users

16

18

20

the average throughput of the cognitive network (bits per slot)

the average throughput (bits per slo


the access policy

15

140
the proposed strategy
the exhaustive search
the randomchoosing strategy

120

100

80

60

40

20

0.05

0.1
0.15
0.2
the collision probability constraint

0.25

0.3

Fig. 2. the average transmission throughput of the cognitive network versus


the receiving SINR .

Fig. 3. the average transmission throughput of the cognitive network versus


the occupied probability of channel i.

is easy to extend to the case when there are more than two
channels. Therefore, there is at most one channel k, 1 k
M , and 0 < nk < k .
As a result, the optimal number of cognitive users for each
channel can be obtained by a search over all the possible
combinations of [n1 , n2 , ..., nM ] satisfying Proposition 2. And
the optimization problem shown in (7) can be solved by the
following strategy:

detection for local spectrum sensing in the following simulations, and the miss-detection probability and false-alarm
probability of energy detector in a Rayleigh fading channel
have been given by [12]. Moreover, we assume that all the
channels have the same theoretical throughput upper bond
R = 100 bits/slot.
Fig. 2 is the average transmission throughput Gi and the
optimal access policy (fi (0) , fi (1)) of a single channel versus
the number of cognitive users choosing this channel to sense.
The collision probability constraint is i = 0.1, and the
occupied probability of this channel is i = 0.5. The missdetection probability and the false-alarm probability are set
as: Pmiss,i = 0.70, Pf alse,i = 0.17. As proved by Proposition
1, Gi is increasing and concave when the number of cognitive
users ni i . And when ni > i , Gi decreases as ni
increases. So its maximal value is reached when ni = i
or ni = i + 1. The optimal access policy (fi (0) , fi (1))
also changes with ni : When ni i , the miss-detection
probability of the cooperative sensing result is high, and
the cognitive network adopts a conservative access policy to
protect the primary transmission. On the other hand, when
ni > i , the cooperative sensing result tends to overlook the
idle channel, thus the access policy should be aggressive to
get more transmission opportunity.
Fig. 3 shows the average transmission throughput of the
proposed opportunistic spectrum access strategy versus the
collision probability constraint. We consider four channels
with the same collision probability constraint . The occupied
probabilities of these channels are: 1 = 0.2, 2 = 0.3, 3 =
0.4 and 4 = 0.5. A cognitive system containing 8 cognitive
users attempts to use the idle channels. For comparison,
the performances of other two opportunistic spectrum access
strategies are also given: a) the exhaustive search, in which
the optimal number of cooperative users for each channel is
obtained by an exhaustive search; b) the random-choosing
strategy, which assumes that each cognitive user randomly
chooses a channel to sense. The access policies of these two
strategies are both calculated using (3) and (8). It can be

The proposed opportunistic spectrum access strategy


S = ;
for k = 1 to M do
for all n = [n1 , ..., nM ], such that nk = 0 and ni
{0, i , i + 1}, i = k do
M

if
ni N then
i=1

ni , i + 1};
nk = min{N
i=k

S = S {n}
end if
end for
end for
M

Gi (ni ) ;
n = arg max
nS i=1

calculate (fi (0) , fi (1)) using (3) and (8);


As presented above, in the worst case, there are M 3M 1
combinations of [n1 , ..., nM ] in S. And the proposed strategy
M

has to calculate
Gi (ni ) for each of these combinations.
i=1

As the exhaustive search has to search over all the M N


combinations of [n1 , ..., nM ], the proposed strategy has a much
lower complexity of computation when the total number of
cognitive users N is large.
IV. S IMULATION R ESULTS A ND A NALYSIS
In this section, the conclusions stated in the previous section are corroborated by simulation results. We use energy

observed from the graph that the performances of all the three
strategies increase as increases. And as we expected, the
proposed opportunistic spectrum access strategy has the same
average transmission throughput as the exhaustive search, and
outperforms the random-choosing strategy. Considering its
lower computational complexity, the proposed opportunistic
spectrum access strategy is more suitable for implementation.
V. C ONCLUSION
In this paper, the design of opportunistic spectrum access in cognitive network is considered. We investigate the
average transmission throughput of the cognitive network
when cooperative spectrum sensing is employed. The optimal
opportunistic spectrum access problem is formulated, and a
reduced search algorithm is proposed to solve this optimization
problem. Simulation results are also presented to show the
promising performances of the proposed opportunistic spectrum access strategy. We hope this analysis will shed some
light on the research and application of opportunistic spectrum
access.
A PPENDIX : P ROOF OF P ROPOSITION 1
As shown in (10), when the number of cooperative users
ni > i , the average transmission throughput function of
channel i can be expressed as:


n
(1 i ) (1 (1 Pf alse,i ) i )
1
(1 i ) Ri .
Gi (ni ) = 1
ni
1 Pmiss,i
(14)
The first derivative of G1i (x) is
d 1
(1i ) (1i ) Ri
G (x) = 
2
x
dx i
1 Pmiss,i


x
x
(1Pf alse,i ) ln (1Pf alse,i )
1Pmiss,i

x
x
(1 (1 Pf alse,i ) ) Pmiss,i
ln Pmiss,i .
(15)
It should be noted that no matter what the channel state is,
the sensor can always choose H0 with a probability of Pmiss,i .
And in this case, its miss-detection probability and false-alarm
probability are Pmiss,i and 1Pmiss,i respectively. Therefore,
for practical sensor, there is:
Pmiss,i < 1 Pf alse,i .
Furthermore, when x 0 and 0 < y < 1, we have


y x ln y
y x1 (x ln y + 1 y x )
< 0.
=
2
x
y 1 y
(1 y x )

(16)

(17)

According to (16) and (17), we can obtain:


x
x
Pmiss,i
ln Pmiss,i
(1 Pf alse,i ) ln (1 Pf alse,i )
>
. (18)
x
x
1 Pmiss,i
1 (1 Pf alse,i )
d
And dx
G1i (x) < 0. As a result, when ni > i , the average
transmission throughput function of channel i decreases as
ni increases, and the maximum value is reached when ni =
i + 1.

On the other hand, when ni i , the average transmission


throughput function of channel i can be expressed as:
G2i (ni ) =

i (1 Pf alse,i )
ni
Pmiss,i

ni

(1 i ) Ri .

(19)

The first derivative of G2i (x) is


d 2
i
Gi (x) = x
(ln (1 Pf alse,i )ln Pmiss,i )
dx
Pmiss,i
x

(1Pf alse,i ) (1i ) Ri


>0.

(20)

And the second derivative of G2i (x) can be given as:


d2 2
i (1 i ) R
x x
Gi (x) = 
2 (1 Pf alse,i ) Pmiss,i
2
dx
Px
miss,i

(ln (1 Pf alse,i ) ln Pmiss,i )


>0.

(21)

Thus the average transmission throughput Gi (ni ) of channel


i is an increasing and concave function when the number of
cooperative users ni i . Moreover, the maximum average
transmission throughput is reached when ni = i or ni =
i + 1. So we have Proposition 2.
R EFERENCES
[1] Federal Communications Commission, Cognitive radio technologies
proceeding, Rep. ET Docket, no. 03-108, 2003.
[2] J. Mitola, et al., Cognitive radio: Making software radios more
personal, IEEE Personal Communications, vol. 6, no. 4, pp. 1318,
Aug. 1999.
[3] J. Mitola, Cognitive radio: An integrated agent architecture for software defined radio, Doctoral dissertation, Royal Inst. Technol. (KTH),
Stockholm, Sweden, 2000.
[4] S. Haykin, Cognitive radio: brain-empowered wireless communications, IEEE Journal on Selected Areas in Communications, vol. 23,
no. 2, pp. 201-220, Feb. 2005.
[5] Cabric D., Mishra S.M. and Brodersen R.W., Implementation issues in
spectrum sensing for cognitive radios, Signals, Systems and Computers, 2004. Conference Record of the Thirty-Eighth Asilomar Conference
on, Vol. 1, pp. 772-776, Nov. 2004.
[6] Juncheng Jia, Qian Zhang and Xuemin Shen, HC-MAC: A HardwareConstrained Cognitive MAC for Efficient Spectrum Management,
IEEE Journal on Selected Areas in Communications, Vol 26, Issue
1, pp. 106 - 117, Jan. 2008.
[7] Yunxia Chen; Qing Zhao; Swami, A., Joint Design and Separation
Principle for Opportunistic Spectrum Access in the Presence of Sensing
Errors, IEEE Transactions on Information Theory, Vol 54, Issue 5, pp.
2053 - 2071, May 2008.
[8] Lifeng Lai, Hesham El Gamal, Hai Jiang and H. Vincent Poor,
Cognitive medium access: exploration, exploitation and competition, IEEE/ACM Trans. on Networking, 2007. Submitted, available
at www.princeton.edu/ llai.
[9] Hang Su, Xi Zhang, Cross-Layer Based Opportunistic MAC Protocols
for QoS Provisionings Over Cognitive Radio Wireless Networks, IEEE
Journal on Selected Areas in Communications, Vol 26, Issue 1, pp. 118
- 129, Jan. 2008.
[10] T. S. Rappaport, Wireless Communications: Principles and Practice.
Englewood Cliffs, NJ: Prentice-Hall, 1996.
[11] P. K. Varshney, Distributed detection and data fusion. NewYork:
Springer-Verlag, 1997.
[12] Digham F. F., Alouini M.-S. and Simon M. K., On the Energy Detection of Unknown Signals Over Fading Channels, IEEE Transactions
on Communications, Vol 55, Issue 1, pp. 21 - 24, Jan. 2007.

You might also like