Published in IET Communications

Received on 20th May 2012

Revised on 12th August 2012
doi: 10.1049/iet-com.2012.0225
ISSN 1751-8628
Sensing matrix setting schemes for cognitive
networks and their performance analysis
H. Shokri-Ghadikolaei M. Nasiri-Kenari
Wireless Research Laboratory, Electrical Engineering Department, Sharif University of Technology, Tehran, Iran
E-mail: mnasiri@sharif.edu
Abstract: Powerful spectrum decision schemes enable cognitive radios (CRs) to nd transmission opportunities in spectral
resources allocated exclusively to the primary users. One of the key effecting factors on the CR network throughput is the
spectrum sensing sequence used by each secondary user (SU). In this study, SUs throughput maximisation through nding an
appropriate sensing matrix (SM) is investigated. First, the average throughput of the CR network is evaluated for a given SM.
Then, an optimisation problem based on the maximisation of the network throughput is formulated in order to nd the optimal
SM. As the optimum solution is very complicated, three novel suboptimal solutions are proposed for various cases including
perfect and non-perfect sensing. Despite having less computational complexities as well as lower consumed energies, the
proposed solutions perform quite well compared to the optimum solution. The structure and performance of the proposed SM
setting schemes are discussed in detail and a set of illustrative numerical results is presented to validate their efciencies.
1 Introduction
Opportunistic spectrum sharing has been developed through
the new promising concept of cognitive radio (CR), in order
to meet ever-growing spectrum demands for new wireless
services. The major issue in designing a CR network
(CRN) is to protect incumbent/primary users (PUs) from
potential interference while providing acceptable quality-of-
service levels for secondary users (SUs) (i.e. unlicensed
users). To this end, sensing capability is exploited in CRs,
which enables them to nd some transmission opportunities
called white spaces. Limited number of possible
observations and dynamic nature of observed signals lead to
imperfect sensing, which is usually described by false alarm
and miss-detection probabilities.
Generally, there exists more than one channel to be sensed
by a CR. As a result, sensing schemes are commonly divided
into two categories, that is, wideband sensing and narrowband
sensing. Sensing is wideband when multiple channels are
sensed simultaneously. When only one channel is sensed at
a time, the sensing is narrowband. Easier implementation,
lower power consumption, and less computational
complexity lead to great interest in narrowband sensing.
When the narrowband sensing is used, an immediate
question arises: which channel should be sensed rst? To
achieve the best possible performance, the channels have to
be sensed in an appropriate order determined by sensing
sequence (SS). In [1, 2], the procedures to determine the
optimal set of candidate channels for sensing are discussed.
In [3, 4], the sequential channel sensing problems are
formulated based on maximising the throughput of the SUs.
Although in these works, the optimum sensing times have
been studied, the effects of the sensing errors have not been
addressed. Channel prioritisation has been considered in [5]
in which an optimal channel sensing framework for a single-
user case including the sensing order and the stopping rule
has been proposed. In [6], the optimal SS has been derived
for channels with homogeneous capacities, and it was shown
that the problem of nding the optimal SS for these channels
is NP-hard. Kim and Shin [7] have suggested an SS which
sorts channels in descending order according to their idle
probabilities. In [8], nding the optimal SS sequence has
been investigated for a single-user case with the aim to
maximise the SUs throughput. The problem of nding
optimal SS for two CR users has been addressed in [9], in
which an exhaustive search has been applied.
To the best of our knowledge, the problem of nding an
optimal set of SSs is investigated in the literature, either for
a single user or two users case. The problem is still open
for a multiple SUs case. In this paper, we address the
problem of setting a proper spectrum sensing matrix (SM)
for a centralised CRN with multiple SUs. Although
distributed CRNs have attracted intensive attentions in the
literatures, the centralised CRNs are also desirable in many
wireless applications especially cellular networks. IEEE
802.22, for instance, is developed based on a centralised
network [10]. Therefore developing some centralised-based
algorithms is essential for those applications. In this paper,
our objective is to maximise the average network
throughput. First, we assume a perfect channel sensing and
formulate an optimisation problem on spectrum SSs of the
SUs. We discuss the exhaustive search solution as well as
its computational complexity. Owing to massive
computational burden of this solution, a novel algorithm,
which nds near-optimal solution, is proposed. The
proposed algorithm, called SM setting (SMS), provides
3026 IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-com.2012.0225
www.ietdl.org
short-term and mid-term fairness among the SUs and offers
near-optimal solution with a tolerable computational
complexity. The consumed energy of the optimal SM as well
as the maximum and minimum consumed energies of the
SMS scheme are analytically derived, and hereby the energy
efciency of the SMS scheme is demonstrated. Then, we
consider the impact of sensing errors on the SMS algorithm,
and propose modied version of the SMS algorithm, called
MSMS algorithm. In addition, for the multiple access
capability among the SUs, we apply the conventional
p-persistent MAC (PMAC) within the MSMS algorithm, and
call the extended algorithm as PMSMS algorithm. Although,
for simplicity of presentation and performance evaluation, we
focus on homogenous channels, the proposed algorithms can
be easily generalised to cover heterogenous channels by
slightly modifying the reward functions used within the
algorithms, as will be discussed later.
In Section 2, we describe the CRN considered. In Section 2,
the throughput of the CRN for a given SM is formulated, and
the conventional approach to nd the optimal SM as well as its
computational complexity are discussed. In Section 4, the
structure, computational complexity and consumed energy of
the novel suboptimal SMS algorithm are described. In
Section 5, the MSMS algorithm is introduced. The PMSMS
algorithm is described in Section 6. Numerical results are
then presented in Section 7, which validate our analysis.
Finally, the paper is concluded in Section 8.
2 System model
We assume a time-slotted CRN with N
s
SUs, which attempt to
opportunistically transmit in the channels dedicated to the N
p
PUs. As in [8, 9, 11], the SUs are time synchronous in time
slots with other SUs and with the PUs. At the beginning of
each time slot, a channel can be established as occupied or
vacant. The sensing process is performed at the beginning
of each time slot. We assume that the SUs are equipped
with simple transceivers; hence they are able to sense only
one channel at a time. The SUs always have packets to
transmit, and as a consequence they will start transmission
when an opportunity is found. Each SU senses the channels
according to its SS sequentially, that is, the SU senses the
rst channel from the top of its SS for a predetermined time
duration (channel sensing time), and then senses the second
channel, if and only if, the rst channel is found busy. This
procedure will continue until a transmission opportunity is
found. The SU might stop its transmission in a channel and
try to choose a new one because of the presence of the PU.
In order to switch to a new channel, which is called
spectrum handover (HO), a secondary device needs a
specic and constant time duration t
ho
to prepare its
sensing circuitry for the next spectrum sensing.
For the SU, each slot contains two phases: (i) sensing phase
and (ii) transmission phase. The sensing phase contains
several mini-slots of duration t. Sensing is carried out by
the SUs in the mini-slots, and once the transmission
opportunity is found, the transmission phase will be started.
The SUs do not have the adaptive modulation and coding
capability; hence they transmit with a constant rate, R,
during the transmission phase. We dene an SM as a
matrix with the dimensions of N
s
N
p
, in which the ith
row contains the SS dedicated to the ith SU. Given the
primary-free probabilities, that is, P
0,j
1 j N
p
, and
predetermined false alarm and detection probabilities, our
objective is to nd the optimal (or near-optimal) SS of each
SU, that is, the optimal SM, in order to maximise the CRN
throughput. Let us dene the slot effectiveness of the kth
element of the SS, denoted by e
k
, as the ratio of the
transmission phase length to the slot length provided that
the SU senses the kth channel free after sensing the rst
(k 21) channels of its SS occupied. Then, we have
e
k
=
transmission time
time-slot duration
= 1
t + (k 1)(t + t
ho
)
T
(1)
and
B
k
= R e
k
= R 1
t + (k 1)(t + t
ho
)
T
_ _
(2)
where B
k
is the average throughput of the SU if the kth
channel of its SS is chosen for the transmission.
3 Optimal SM
The optimal SS for the CRN containing just one SU is derived
in [7]. For that case, all required was to sort the spectrums
based on their primary-free probabilities. However, in the
CRN with multi-users, the impact of collisions among the
SUs transmissions has to be taken into account. Assume
that S denotes the SM which contains N
s
rows and N
p
columns with the element s
i,j
indicating that the ith SU
senses the s
i,j
th spectrum in its jth mini-slots.
For the network throughput evaluation, we note that for
each spectrum s
i,j
two possible cases might occur. First, the
s
i,j
th spectrum has been sensed by some SUs in their
previous mini-slots (the mini-slots before jth mini-slot). In
this case, regardless of the presence or the absence of the
PU, the spectrum is occupied (as we assumed error-free
sensing in this section). That is, the SU that senses this
channel at the rst time will transmit on the channel if the
spectrum is idle. The second case is that the spectrum is
sensed at the jth mini-slot by ith SU at the rst time.
Hence, the occupation probability of this channel only
depends on the PU activity. From the above discussion, the
repetition of a spectrum in the SM when assuming error-
free sensing does not offer any benet to the CRN throughput.
Let matrix Z
j21
demonstrate the spectrums allocated prior
to the jth mini-slot, and array Y contain the spectrums
dedicated at the jth mini-slot to the SUs prior to the user i,
as shown in the SM S,
S =
s
1,1
s
1,2
. . . s
1,j1
s
2,1
s
2,2
. . . s
2,j1
.
.
.
.
.
.
,............................,
z
j1
s
1,j
s
2,j
.
.
.
s
i1,j
_

_
. . .
Y
s
N
s
,1
s
N
s
,2
. . . s
N
s
,j1
s
i,j
.
.
.
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
N
s
N
p
(3)
Assume A
Z
j1
s
i,j
indicate the presence or absence of the
spectrum s
i,j
in Z
j21
,
A
Z
j1
s
i,j
=
0: if s
i,j
[ Z
j1
1: if s
i,j
Z
j1
_
(4)
By the above denition, when not taking into account the
impact of collision caused by multi-SU transmissions at
the same spectrum because of simultaneously nding the
IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035 3027
doi: 10.1049/iet-com.2012.0225 & The Institution of Engineering and Technology 2012
www.ietdl.org
spectrum free at the jth mini-slot, the spectrum s
i,
can be
efciently used by the ith SU with the probability of
P
0,s
i,j
A
Z
j1
s
i,j
, where P
0,s
i,j
is the absence probability of the s
i,j
th
PU. To consider the impact of the collision on the network
throughput as discussed above, for each mini-slot like the jth
mini-slot, we dene the operator as follows (see (5)).
where
1[s
i,j
=s
k,j
] =
1, if s
i,j
=s
k,j
0, if s
i,j
= s
k,j
_
(6)
The operator is used to model the possible collision because
fo multi-SUs nding the same spectrum free at each jth mini-
slot. Each channel in each time slot has a contribution in the
whole throughput, if and only if, it is sensed only by one SU
(i.e. assigned to one SU SS) because of error-free sensing
assumption.
By the above denition of the operator , the average
throughput of the CRN is easily computed as
Q = (P
0,s
1,1
A
Z
0
s
1,1
P
0,s
2,1
A
Z
0
s
2,1
P
0,s
N
s
,1
A
Z
0
s
N
s
,1
)B
1
+ (P
0,s
1,2
A
Z
1
s
1,2
P
0,s
2,2
A
Z
1
s
2,2
P
0,s
N
s
,2
A
Z
1
s
N
s
,2
)B
2
+ + (P
0,s
1,N
p
A
Z
N
p
1
s
1,N
p
P
0,s
2,N
p
A
Z
N
p
1
s
2,N
p
P
0,s
N
s
,N
p
A
Z
N
p
1
s
N
s
,N
p
)B
N
p
(7)
where B
j
is dened in (2) and A
Z
0
s
i,1
W 1, i. Now, the optimal
SM is found by solving the following:
S

= arg max
s
1,1
,s
1,2
,...,s
N
s
,N
p
Q (8)
Based on (8), we can nd the optimal SM by exploiting
exhaustive search. Assume that the computational
complexity of computing (7) for a given SM, S, is in O(1).
Then, the computational complexity of nding the optimal
SM is in O(N
N
p
N
s
p
). There is no much room for solving (8)
through classical optimisation procedures. Further, solving
(8) through the exhaustive search makes no guarantee for
fairness among the SUs. Also, it results in massive
computational burden, which is not scalable regarding to
both N
p
and N
s
. All these facts make a strong motivation in
developing an appropriate suboptimal solution for the
problem in (8), which is considered in next sections.
4 SMS algorithm
4.1 Structure of the SMS algorithm
The proposed algorithm, designed for error-free sensing case,
is composed of N
p
rounds. In the kth round, the coordinator
determines the kth column of the SM, that is, s
i,k
,
1 i N
s
. As mentioned before, repeating a spectrum in
the SM for more than one times either in the same mini-slot
or in different mini-slots does not have any benets on the
network throughput. During each round and for each SU,
the coordinator assigns a reward to each candidate channel
[the channel that has not been assigned to the SS of any
user previously.] to be possibly allocated to the SS of the
SU at that round and then adopts the channel with the
maximum reward. That is, at the round m, for each SU k
and for each unassigned channel i, we dene G
(k)
i
(m) as the
reward of the channel i if selected as the mth component of
the kth SUs SS. This reward is set equal to the contribution
of the kth SU to the network throughput if the ith channel
is selected, as will be described latter. Then, the channel
with the maximum reward is selected.
We denote the set of all assigned channels to the SM by A.
At the beginning, we have A and S , where S is the
SM and denotes the empty matrix. We also denote the set of
all channels by N, where N = {1, 2, . . . , N
p
}. The process is
as follows:
4.1.1 Round-1: For this round, rst the coordinator assigns
a spectrum to the SS of the rst SU at its rst mini-slot. The
coordinator must adopt s
1,1
from the unassigned channels,
that is, A = N\A = N. We have G
(1)
i
(1) = P
0,i
B
1
, where B
1
is dened in (2). The coordinator selects a channel with the
highest reward for the s
1,1
s
1,1
= arg max
i[A
G
(1)
i
(1) (9)
After s
1,1
is determined, A and A are respectively updated to
A = {s
1,1
} and A = N\A. This procedure is repeated for
each SU; so for the l-th user in the rst round, we have
s
,1
= arg max
i[A
G
()
i
(1) (10)
where G
()
i
(1) = P
0,i
B
1
.
4.1.2 Round-m: At the mth round, for each SU, the
coordinator similarly assigns a reward to each left
spectrums and allocates the best spectrum, which has the
maximum reward, to the SU. If the coordinator chooses
the jth channel for the mth sensing mini-slot of the lth SU,
the following reward will be gained
G
()
j
(m) =

m1
i=1
(1 P
0,s
,i
)
_ _
P
0,j
B
m
(11)
Note that G
()
j
(m) is in fact the throughput of the SU if the
spectrum s
,m
is selected by the user for its data transmission.
Hence
s
,m
= arg max
i[A
G
()
i
(m) (12)
At this round, rst it must be determined from which SU the
procedure should be started. For an acceptable level of
fairness among the SUs, the algorithm starts with an SU
that has gained the lowest cumulative rewards during the
:
A B = B A
m N
p
:P
0,s
1,j
A
Z
j1
s
1,j
P
0,s
2,j
A
Z
j1
s
2,j
P
0,s
m,j
A
Z
j1
s
m,j
=

m
i=1

m
k=1
k=i
P
0,s
i,j
A
Z
j1
s
i,j
1[s
i,j
=s
k,j
]
_

_
(5)
3028 IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-com.2012.0225
www.ietdl.org
previous (m21) rounds [previous (m21) mini-slots], which
the cumulative reward for the lth SU is calculated as

m1
k=1
G
()
s
,k
(k) = G
()
s
,1
(1) + G
()
s
,2
(2) + G
()
s
,m1
(m 1) (13)
where G
()
s
,k
(k) is dened in (11).
This process continues until |A|N
p
or equivalently A = .
At the end of the process, the elements of S without any
assigned spectrums are replaced by zero, which indicates
that the sensing is not performed for those elements. Since
each channel is sensed only once in the proposed algorithm,
the energy consumed by the SMS algorithm equals to N
p
E
c
at the worst case, and it does not increase by N
s
.
In order to have mid-term fairness, at the beginning of
second time slot, the process starts with the second SU and
the rst element of the SS of this user is determined and
then the procedure is continued by selecting the rst
element of the third user, and at the last the rst element of
the rst user is selected. The other elements are determined
as described above. This cyclic ordering is continued in the
following time slots. That is, at the beginning of mth run of
the SMS procedure (mth time slot) the process starts with
selecting the rst element of the SS of the kth SU, where
k = mod(m, N
p
) (14)
These procedures are summarised in the owchart presented
in Fig. 1. Note that the algorithm can be easily modied to
cover heterogenous environments. For the heterogenous
channel throughput, only the term B
m
in the reward
function must be replaced by B
()
j,m
= R
()
j
e
m
, where R
()
j
denotes the transmission rate of the lth SU on the jth
channel and e
m
is dened in (1).
4.2 Computational complexity
As we stated before, the computational complexity of nding
the optimal SM is in order of O(N
N
p
N
s
p
), whereas it is in
order of O(1) for our proposed method. In the SMS
algorithm, a channel will be assigned to the SM if it offers
the highest reward, dened in (11), among the left channels.
From (11), it can be easily shown that G
(l)
j
(m) . G
(l)
k
(m) if
P
0,j
. P
0,k
[The reason for dening the award as in (11) is
that it can be easily modied to the non-error-free sensing
case and also for when considering different MAC
schemes.]. Hence, for the error-free case, the information
required to determine the SM is the primary-free
probabilities of the channels.
4.3 Average consumed energy for nding a
transmission opportunity
Let E
c
(t) and E
c
(t
ho
) denote the consumed energies for
sensing of each primary channel and for each HO,
respectively. Hence, the average consumed energy for
nding a transmission opportunity is calculated
(N
s
+ g
1
+ g
2
+ + g
N
s
)E
c
(t)
+ ( g
1
+ g
2
+ + g
N
s
)E
c
(t
ho
)
(15)
where g
i
denotes the average number of HOs required by the
ith SU to nd an idle channel. The processes of channel
sensing and signal transmission consume more energy
compared to the HO [12]. Therefore it is rational to ignore
the second term in (15) compared to the rst one. To
evaluate g
i
, we consider two following cases: (i) the SU
searches among the channels, nds a transmission
opportunity, and then transmits; (ii) The SU searches
among the available channels, but does not nd any free
channel. Then, g
i
is easily calculated as (see (16))
where the term T1 represents the rst case, and the last term
demonstrates the second case. By substituting (16) into
Fig. 1 SMS algorithm
g
i
= P
1,s
i,1
P
0,s
i,2
+ 2P
1,s
i,1
P
1,s
i,2
P
0,s
i,3
+ + (N
p
1)P
0,s
i,N
p

N
p
1
j=1
P
1,s
i,j
.........................................,,.........................................
T1
+N
p

N
p
j=1
P
1,s
i,j
(16)
IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035 3029
doi: 10.1049/iet-com.2012.0225 & The Institution of Engineering and Technology 2012
www.ietdl.org
(15), the total average consumed energy is derived

N
s
i=1
g
i
E
c
= E
c

N
s
i=1
1 +

N
p
k=2
(k 1)

k1
j=1
P
1,s
i,j
_ _
P
0,s
i,k
_ _
+ E
c

N
s
i=1
N
p

N
p
j=1
P
1,s
i,j
(17)
For the SMS algorithm, analytically deriving the average
consumed energy is complicated. Hence, we only focus on
two special extreme cases, that is, the maximum and the
minimum consumed energies. For the worst case, which
consumes the maximum energy, all the channels appeared
in the SM are sensed. In this case, the consumed energy
equals to N
p
E
c
(t). For the best case, which consumes the
minimum energy, each SU nds the rst channel of its SS
free. In this case, the consumed energy is equal to Min(N
p
,
N
s
)E
c
(t). It is worth noting that for P
0j
1, for all j, the
sequential sensing scheme forces the SUs to continue
searching among all channels allocated to their SSs, which
is equivalent to the worst case, and similarly P
0j
0, for all
j is equivalent to the best case with minimum consumed
energy. If we compute the average consumed energy of the
optimum solution given in (17) for theses two cases, the
maximum and the minimum consumed energies will be
equal to E
c
N
s
(1 +N
p
) and E
c
N
s
, respectively, which are
higher than those of the SMS algorithm.
For more elaboration, we study a special case where all
channels have the same primary-free probabilities, that is,
P
0,j
P, P
1,j
1 2P, for all j. Then, we can simplify (17) as
E
c
(t)

N
s
i=1
g
i
=E
c
(t) N
s
+P

N
s
i=1

N
p
k=2
((k 1)(1P)
k1
)
_ _ _
+N
s
N
p
(1P)
N
p
_
=E
c
(t)N
s
1+
1P+( 1+PPN
p
)(1P)
N
p
P
_
+N
p
(1P)
N
p
_
=E
c
(t)N
s
1P(1P)
N
p
+1
P
+1
_ _
(18)
Please note that (18) is a decreasing function of P. Hence, the
minimum and maximum values of consumed energy are
related to the cases P 0 and P 1, as discussed above.
Moreover, as will be shown in the numerical result section
(Fig. 5), the consumed energy associated with the optimal
SM is higher than the consumed energy for the matrix
obtained by the SMS algorithm for all values of P.
In the following, the impacts of the sensing errors are
investigated. In general, the sensing error manifests itself in
two forms: false alarm and miss-detection. In the SMS
algorithm proposed, a channel is allocated to only one SS,
and thus the SUs have no common channels in their SSs.
Although this approach performs well when there is no
sensing error, in the case of non-perfect sensing, this
method is not efcient; since by a false alarm made by an
SU in a sensed channel, a transmission opportunity is lost
for this channel by all SUs. Therefore the coordinator has to
repeat spectrums in the S in order to increase the possibility
of exploiting all opportunities. On the other hand, allocating
a channel to the SSs of multiple SUs increases the average
number of sensed channels and thus raises the average
sensing energy consumption. Moreover, because of miss-
detection, it is possible for an SU to mistakenly transmit on
a channel, which is already used by another SU or PU, and
therefore some collisions might occur. Hence, there is a
trade-off between the average achievable throughput, energy
consumption and the level of collision in the CRN, which
must be addressed in an extension of the SMS algorithm.
5 MSMS algorithm
Since in the SMS algorithm, the SSs of the SUs have no
common channels, the occupancies of channels at the
beginning of each mini-slot only depend on the PUs
activities. However, if the channels are allowed to be
repeated in multiple rows or columns of the SM, the
occupancy of a channel can be because of the presence of
either the PU or an SU. Since all the SUs use the same
sensing schemes with the same sensing time lengths, they
all have the same probabilities of false alarm and miss-
detection. Thus
P
fa,1
= P
fa,2
= = P
fa,N
p
= P
fa
P
d,1
= P
d,2
= = P
d,N
p
= P
d
(19)
To reect the impact of sensing error on the proposed
algorithm, three possible cases must be considered when
the coordinator tends to adopt the lth channel as s
i,j
:
The lth channel has not yet been allocated to any SSs, that
is, l Z
j21
[matrix Z
j21
is dened in (3)].
The lth channel has been adopted at least once for sensing
at the previous mini-slots, that is, l [ Z
j21
.
The lth channel has been allocated simultaneously to
multiple users at the jth mini-slot, that is, l [ Y, [vector Y
is dened in (3)].
Suppose that jth component of the SS of the ith SU, that is,
s
i,j
, is to be selected by the coordinator. Hence, the reward
gained by adopting the kth channel as s
i,j
must be
determined. Considering the denition of the matrix Z
j21
and the vector Y, if n elements of Z
j21
are equal to k, then
the kth channel is sensed by up to n SUs during previous
mini-slots. Also, if two or more elements of Y are equal to
k, then the kth channel will be sensed by two or more SUs
during the jth mini-slot. When channels are allowed to be
sensed by multiple SUs simultaneously, an appropriate
MAC protocol is used to regulate the access of the SUs to
transmission opportunities. As a rst step, we assume that
an SU starts transmitting when it nds a transmission
opportunity. Applying more appropriate MAC protocols to
decrease the collision probability among the SUs will be
considered later. For the transmission policy considered
here, if the kth channel belongs to Y and it is also adopted
as s
i,j
by the coordinator, a collision may occur, and in this
case, the actual throughput contribution will be zero.
Without loss of generality, it is assumed that the
coordinator starts the allocation process for the jth mini-slot
from the top of the jth column of matrix S.
3030 IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-com.2012.0225
www.ietdl.org
Given Z
j21
, we denote the occupation probability of the kth
channel at the beginning of the jth mini-slot as q
(j)
1,k
. Then
q
(j)
1,k
= P
1,k
+ u
1,k,Z
j1
(20)
where u
1,k,Z
j1
represents the probability of transmission on
(or equivalently the occupation probability of) the kth
channel in the rst ( j 21) mini-slots and is easily
computed as
u
1,k,Z
j1
=

|Z
j1
|
n=1
d
k,Z
j1
n
(1 P
0,k
P
n
fa
) (21)
where |Z
j21
| N
s
( j 21) is the number of elements of Z
j21
and d
k,Z
j1
n
is dened as
d
k,Z
j1
n
=
1: if spectrum#k is in Z
j1
for n times
0: if spectrum#k is not in Z
j1
for n times
_
(22)
Similar to the SMS algorithm, the process starts with S
and at the rst step s
1,1
is selected for the SS of the rst SU
by the coordinator. As before, for channel i [ N, G
(k)
i
(1)
denotes the reward (the throughput) contributed by the kth
SU to the overall throughput of the secondary network
when the ith channel is allocated as the rst element of its SS.
5.1 Round-1
At the rst round, G
(1)
i
(1) = q
(1)
0,i
(1 P
fa
)B
1
in which B
1
is
dened in (2). q
(1)
0,i
= 1 q
(1)
1,i
, where q
(1)
1,i
is given in (20).
Therefore
s
1,1
= arg max
i[N
G
(1)
i
(1) (23)
If the ith spectrum is adopted as s
l,1
, the reward added to the
system can be calculated as
G
(l)
i
(1) =

|Y|
n=0
d
i,Y
n
q
(1)
0,i
P
n
fa
(1 P
fa
)B
1
(24)
where
d
i,Y
n
=
1: if spectrum# i is in Y for n times
0: if spectrum# i is not in Y for n times
_
(25)
Finally, the rst channel of the SS of the lth SU is selected as
s
,1
= arg max
i[N
G
()
i
(1) (26)
5.2 Round-m
The lth SU gains a reward by adopting the jth channel as the
mth element of its SS provided that the user has not detected a
transmission opportunity in its previous sensed channels. The
reward gained by the lth SU is
G
()
j
(m) =

m1
i=1
(q
(i)
0,s
,i
P
fa
+ q
(i)
1,s
,i
P
d
)
..............,,..............
C1
_
_
_
_
_
_
_
_
_
_

|Y|
n=0
d
j,Y
n
q
(m)
0,j
P
n
fa
(1 P
fa
)B
m
................,,................
C2
(27)
where C1 indicates the probability of requiring (m21) HO,
and C2 represents the throughput contribution of jth
channel if selected at the mth mini-slots of the lth SU for
the transmission. q
(i)
0,j
= 1 q
(i)
1,j
, where q
(i)
1,j
is given in (20).
Thus, the mth element of the lth SS is determined as
s
,m
= arg max
j[N
G
()
j
(m) (28)
Similar to the SMS scheme, in the MSMS algorithm, at the
round-m (m [ {2, 3, . . . , N
p
}) the coordinator starts with
the SU that has gained less cumulative rewards in its
(m21) previous mini-slots. The cumulative rewards of the
Fig. 2 MSMS algorithm
IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035 3031
doi: 10.1049/iet-com.2012.0225 & The Institution of Engineering and Technology 2012
www.ietdl.org
jth SU is computed
G
(j)
s
j,1
(1) + G
(j)
s
j,2
(2) + G
(j)
s
j,m1
(m 1) (29)
where G
(j)
s
j,i
(i) is calculated as (27). Hereby, a certain level of
fairness is ensured among the SUs.
The stopping rule of the MSMS algorithm is different from
that of the SMS algorithm. For the MSMS algorithm, two
possible rules can be exploited. First, there exist no
constraint on the number of times that each channel can be
used as the elements of the SM. For this case, the process is
stopped when all the elements of the SM have been
allocated. Second, the number of times that each channel is
appeared in the SM is limited. Although the rst rule leads
to the maximum average throughput of the MSMS
algorithm, the second rule is more rational and practical.
The probability of a channel erroneously sensed as busy
exponentially decreases by the number of times that the
channel is sensed. Therefore there is no benet to repeat the
channel too many times, and we use the second stopping
rule. In the numerical result part, we limit the number of
times that each channel is appeared to be three.
To have further mid-term fairness among the SUs, similar
to the SMS algorithm, at the beginning of mth run of the
MSMS procedure, the process starts with the kth SU as
specied in (14). Fig. 2 presents the owchart of the
MSMS scheme.
6 PMSMS algorithm
Regardless of how the SM being created, it is possible for a
channel to be assigned to the several SUs in the same mini-
slot. In this case, various conventional MAC algorithms can
be exploited. In this section, we utilise the well-known
PMAC protocol in the MSMS algorithm and develop
PMSMS algorithm. In this algorithm, in each mini-slot, the
SUs sense the assigned channels with the probability of p.
In order to have a synchronous sensing scheme for all SUs,
the SU will be idle for t seconds (mini-slot time duration)
if its MAC protocol does not allow it to sense the channel.
The stopping rule as well as fairness establishment
techniques are similar to the MSMS algorithm. Considering
PMAC, there are two cases that a free channel is not used
by an SU. First case is due to the false alarm, and the
second is due to the presence of PMAC protocol. In the
latter case, the channel is not sensed with the probability of
(1 2p). Considering these two cases easily leads to the
following modication of u
1,k,Z
j1
[dened in (21)]
u
1,k,Z
j1
=

|Z
j1
|
n=1
d
k,Z
j1
n
1

n
t=0
n
t
_ _
P
0,k
p
t
P
t
fa
(1p)
nt
_ _
0 p ,1

|Z
j1
|
n=1
d
k,Z
j1
n
(1P
0,k
P
n
fa
) p =1
_

_
(30)
Then, the channel occupation probability is obtained by
substituting (30) into (20). The generalised reward of
assigning the jth primary channel to the SS of the lth SU at
the mth mini-slot is simply calculated as
G
()
j
(m) =

m1
i=1
(q
(i)
0,s
,i
P
fa
+ q
(i)
1,s
,i
P
d
)
_ _

|Y|
n=0
d
j,Y
n
q
(m)
0,j
B
m
h

(31)
where
h

=
(1 P
fa
)p

n
k=0
n
k
_ _
p
k
P
k
fa
(1 p)
nk
0 p , 1
(1 P
fa
)P
n
fa
p = 1
_
_
_
(32)
and nally the coordinator adopts a channel with a highest
reward for the mth mini-slot of lth SU as
s
,m
= arg max
i[N
G
()
i
(m) (33)
7 Numerical results
The simulation parameters are given in Table 1. The values of
SNR and sampling frequency which are used by the energy
detector are adopted from [11]. The value of sensing time
of each channel, t, is selected such that the false alarm and
detection probabilities meet the constraints imposed by the
IEEE 802.22 standard [10]. Each SU senses the channels
according to its SS, each for t seconds, until a free channel
is found. Then, the SU transmits on this channel for the rest
of the time slot. For the simulation results, the average
normalised CRN throughput has been evaluated by
simulating the scenario for 100 time slots.
Fig. 3 validates our analysis and depicts the average
throughput (normalised to R) of the SUs against the
normalised sensing time, that is, (t/T), for the optimal SM
and the SM obtained based on SMS algorithm, for a error-
free sensing case. For the optimal SM, both the theoretical
and simulation results are provided. As realised, the
throughput linearly decreases by the sensing time. This is
due to the fact that for the error-free case, there exists no
error in the detection scheme and thus while the increase of
sensing time does not have any positive impact on the
correct detection, it linearly reduces the transmission time.
Fig. 3 also veries the near-optimality of our proposed
algorithm; while it imposes much less complexity burden
than the optimum scheme. The relative difference between
the average throughput obtained by the SMS algorithm and
that obtained by the optimum SM is negligible and about
0.81%.
Table 1 Simulation parameters
Parameter Description Value
P
min
d
minimum allowable detection
probability
0.9
P
max
fa
maximum allowable false alarm
probability
0.1
f
s
receiver sampling frequency 6 MHz
T time-slot duration 200 ms
t
ho
required time for handover 0.1 ms
N
p
number of PUs 5
N
s
number of SUs 3
3032 IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-com.2012.0225
www.ietdl.org
Fig. 4 compares the average throughputs of different SUs
for a CRN with three SUs, again for a error-free sensing
scheme. The number of primary channels, N
p
, is ve. The
maximum relative difference between the SUs throughputs
is 1.84%, which conrms the fairness among the SUs when
using the proposed scheme.
Fig. 5 shows the average consumed energies of the optimal
SM, SMS, MSMS and PMSMS schemes normalised to E
c
[Fig. 5 represents the number of sensed channels. As
demonstrated in (17), multiplying the average number of
sensed channel by E
c
leads to the average energy
consumed.] against the PUs absence probabilities (PAP).
From this gure, the SUs consume less average energies
when sensing the channels based on the SM obtained by
the SMS and MSMS methods compared with the optimal
SM. In the MSMS scheme, each channel can be repeated at
most three times in the SM, and therefore more channels
will be sensed (each channel can be sensed up to three
times) in average in comparison with the SMS scheme, in
which each channel appears only once in the SM. The
PMSMS consumes less energy than the MSMS for the
small values of the PAP, but for the high value of this
probability, it consumes more energy even more than the
optimum scheme. This phenomenon can be explained as
follows. For small values of the PAP, for example,. close to
zero, the channels are probably occupied. In these cases, the
SUs adopted the PMSMS scheme skip some of the
channels assigned to their sensing lists (because of
exploited p-persistent protocol), and consequently less
channels will be sensed compared with the MSMS scheme,
where all channels must be sensed until a free channel is
found. On the other hand, for the large values of the PAP,
the channels are probably free. In the PMSMS scheme,
some SUs do not sense those free channels in their rst
mini-slots (again because of exploited p-persistent
protocol), and once start sensing, the channels may be
occupied by other SUs, and as a result, they must continue
their sensing process to nd a transmission opportunity.
Albeit this event can reduce the collision probability, and
hence can increase the network throughput, the average
consumed energy can be increased. Please note that the
computational complexity imposed by the PMSMS
algorithm to the coordinator is substantially less than that of
the optimum algorithm. Another observation made from
Fig. 5 is that, as expected, in all schemes, the consumed
energies of the SUs increase by the decrease of the PAP.
Fig. 6 compares the average throughputs of various
spectrum allocation schemes proposed in this paper. For the
two MSMS and PMSMS schemes, a practical scenario with
sensing errors has been assumed. In general, as the sensing
time increases, the detector senses the channels more
accurately and nds more transmission opportunities.
Fig. 3 Average throughput against normalised sensing time for various SM selection schemes
Fig. 4 Average throughput against normalised sensing time for different SUs
IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035 3033
doi: 10.1049/iet-com.2012.0225 & The Institution of Engineering and Technology 2012
www.ietdl.org
However, by the increase of the sensing time, less time
remains for the transmission. Hence, there exists a trade-off
between average throughput and detector accuracy. As seen
in Fig. 6, rst the SUs throughput increases by t; then
after an optimum point, where P
md
and P
fa
are in acceptable
levels, the throughput starts decreasing due to the reduction
of the time left for the transmission. For a sensing time
greater than a specic amount (optimum value), the false
alarm and miss-detection probabilities of the detector
become negligible, and the allocation procedure of the
MSMS algorithm as well as its performance will be similar
to those of the SMS algorithm, for which a error-free
sensing has been assumed. In the MSMS scheme, the SUs
sense the channels with the probability of (1 2P
fa
), and
thus some transmission opportunities are lost as a result of
false alarm. This is the reason that the average throughput
of the SUs obtained by the MSMS algorithm is less than
that of the SMS algorithm for which P
fa
is assumed to be
zero. In the PMSMS algorithm, the applied PMAC protocol
leads to loss some transmission opportunities, and thus to
less average throughput compared to the MSMS when the
number of SUs is not too high.
Fig. 7 demonstrates the advantages of the exploited PMAC
for the case that the number of primary channels are less than
the number of SUs. This gure shows the average SUs
throughput against the probability of sensing an assigned
channel (i.e. p in MAC protocol). Note that the performance
for p 1 is the same as that of the MSMS. As can be
realised, for N
p
, N
s
, the exploited PMAC protocol can
signicantly increase the chance of transmission on the
channels by reducing the contention level among the SUs.
Therefore the PMSMS scheme can substantially offer higher
throughput for the CRN than the MSMS scheme provided
that the coordinator appropriately selects the value of p,
which for the example considered it must be larger than
0.165 for the case of {N
p
5, N
s
8}. Interestingly, for
p 0.46 in the case of {N
p
5, N
s
8} and p 0.28 in the
case of {N
p
5, N
s
10}, the improvements in the
throughput when using the PMSMS scheme are, respectively,
about 48.8 and 1019.5% compared with the MSMS scheme.
Fig. 8 represents the SUs average throughput of the
proposed schemes versus detection probability P
d
. As this
gure shows, the increase of P
d
leads to the decrease of the
average throughput. This phenomenon can be explained as
follows. For the energy detection scheme used as spectrum
sensing technique in the numerical results section, by the
increase of P
d
, P
fa
increases as well [13], and hence more
opportunities will be lost during the sensing procedure.
Moreover, in the case of N
p
. N
s
, less contention occurs
among the SU, and consequently more average throughput
can be achieved comparing with the case of N
s
. N
p
.
Another observation is that in the case of N
s
. N
p
, the
Fig. 5 Normalised average consumed energy against the PAP
Fig. 6 Average throughput of the SUs against normalised sensing time for various proposed schemes
3034 IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-com.2012.0225
www.ietdl.org
PMSMS scheme has the best performance; because it reduces
the contention level among the SUs (because of exploited p-
persistent protocol), and consequently a higher average
throughput is achieved.
8 Conclusion
In this paper, the average throughput of a CRN for a given SM
has been derived, and an optimisation problem has been
formulated. In order to mitigate the challenges associated
with the optimal solution, three novel suboptimal
algorithms have been proposed. More specically, the SMS
and MSMS schemes are proposed for error-free and non-
perfect sensing cases, respectively, and then the PMSMS
algorithm is developed by applying the conventional PMAC
protocol within the MSMS algorithm to strengthen the
multiple access capability of the CRN. The performance of
the proposed schemes has been evaluated, and their
efciencies have been demonstrated through theoretical
analysis as well as exhaustive simulation results.
9 References
1 Hamdaoui, B.: Adaptive spectrum assessment for opportunistic access
in cognitive radio networks, IEEE Trans. Wirel. Commun., 2009, 8, (2),
pp. 922930
2 Shohri-Ghadikolaei, H., Fallahi, R.: Intelligent sensing matrix setting in
cognitive radio networks. IEEE Communication Letters, September 2012
3 Sabharwal, A., Khoshnevis, A., Knightly, E.: Opportunistic spectral
usage: bounds and a multi-band CSMA/CA protocol, IEEE/ACM
Trans. Netw., 2007, 15, (3), pp. 533545
4 Jia, J., Zhang, Q., Shen, X.S.: HC-MAC: a hardware-constrained
cognitive MAC for efcient spectrum management, IEEE J. Sel.
Areas Commun., 2008, 26, (1), pp. 106117
5 Chang, N.B., Liu, M.: Optimal channel probing and transmission
scheduling for opportunistic spectrum access, IEEE/ACM Trans.
Netw., 2009, 17, (6), pp. 18051818
6 Kim, H., Shin, K.G.: Fast discovery of spectrum opportunities in cognitive
radio networks. Proc. IEEE DySPAN, October 2008, pp. 112
7 Kim, H., Shin, K.G.: Efcient discovery of spectrum opportunities with
MAC-layer sensing in cognitive radio networks, IEEE Trans. Wirel.
Commun., 2008, 7, (5), pp. 533545
8 Jiang, H., Lai, L., Fan, R., Poor, H.V.: Optimal selection of channel
sensing order in cognitive radio, IEEE Trans. Wirel. Commun., 2009,
8, (1), pp. 297307
9 Fan, R., Jiang, H.: Channel sensing-order setting in cognitive radio
networks: a two-user case, IEEE Trans. Veh. Technol., 2009, 58, (9),
pp. 49975008
10 Stevenson, C.R., Chouinard, G., Lei, W.H.Z., Shellhammer, S.J.: IEEE
802.22: the rst cognitive radio wireless regional area network
standard, IEEE Commun. Mag., 2009, 47, (1), pp. 130138
11 Liang, Y.C., Zeng, Y., Peh, E.C.Y., Hoang, A.T.: Sensing-throughput
tradeoff for cognitive radio networks, IEEE Trans. Wirel. Commun.,
2008, 7, (4), pp. 13261337
12 Monica, M.S., Sharma, A.K.: Comparative study of energy
consumption for wireless sensor networks based on random and grid
deployment strategies, Int. J. Comput. Appl., 2010, 6, (1), pp. 2835
13 Yucek, T., Arslan, H.: A survey of spectrum sensing algorithms for
cognitive radio applications, IEEE Commun. Surv. Tutor., 2009, 11,
(1), pp. 116130
Fig. 7 Advantages of exploiting PMAC
Fig. 8 Average throughput of the SUs against detection probability for the proposed schemes
IET Commun., 2012, Vol. 6, Iss. 17, pp. 30263035 3035
doi: 10.1049/iet-com.2012.0225 & The Institution of Engineering and Technology 2012
www.ietdl.org