Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Power Control and Channel Allocation

in Cognitive Radio Networks with


Primary Users Cooperation
Anh Tuan Hoang, Member, IEEE, Ying-Chang Liang, Senior Member, IEEE, and
Md Habibul Islam, Member, IEEE
AbstractWe consider a point-to-multipoint cognitive radio network that shares a set of channels with a primary network. Within the
cognitive radio network, a base station controls and supports a set of fixed-location wireless subscribers. The objective is to maximize
the throughput of the cognitive network while not affecting the performance of primary users. Both downlink and uplink transmission
scenarios in the cognitive network are considered. For both scenarios, we propose two-phase mixed distributed/centralized control
algorithms that require minimal cooperation between cognitive and primary devices. In the first phase, a distributed power updating
process is employed at the cognitive and primary nodes to maximize the coverage of the cognitive network while always maintaining
the constrained signal to interference plus noise ratio of primary transmissions. In the second phase, centralized channel assignment is
carried out within the cognitive network to maximize its throughput. Numerical results are obtained for the behaviors and performance
of our proposed algorithms.
Index TermsWireless communications, dynamic spectrum access, distributed control, joint power control and channel allocation.

1 INTRODUCTION
T
HE traditional approach of fixed spectrum allocation to
licensed networks leads to spectrum underutilization.
In recent studies by the FCC, it is reported that there are
vast temporal and spatial variations in the usage of
allocated spectrum, which can be as low as 15 percent [1].
This motivates the concepts of opportunistic spectrum access
that allows secondary cognitive radio networks to oppor-
tunistically exploit the underutilized spectrum. In fact,
opportunistic spectrum access has been encouraged by both
recent FCC policy initiatives and IEEE standardization
activities [2], [3].
On one hand, opportunistic spectrum access can im-
prove the overall spectrum efficiency. On the other hand,
transmission from cognitive devices can cause harmful
interference to primary users of the spectrum. This
motivates our objective of maximizing the throughput of a
cognitive radio network while maintaining performance of
coexistent primary users.
In this paper, we consider a point-to-multipoint cogni-
tive radio network in which a base station (BS) controls
and supports a set of fixed-location customer premise
equipments (CPEs). The spectrum of interest is divided
into a set of nonoverlapping, independent channels. Some
of these channels are used by a set of point-to-point
primary transmissions. We are interested in the power
control/channel assignment problem for the cognitive
radio network. The objective is to maximize the total
throughput for the cognitive network while maintaining a
required signal to interference plus noise ratio (SINR) for
all primary receivers (PRXs).
We consider both downlink and uplink scenarios in the
cognitive network. For the downlink control scenario, we
propose a two-phase downlink mixed distributed/centra-
lized control algorithm (DL-MDCA) that requires minimal
cooperation between cognitive and primary devices. In the
first phase of DL-MDCA, BS and primary transmitters
(PTXs) participate in a distributed power updating process
that strives to maximize the coverage of the cognitive
network. BS and PTXs need to exchange simple control
signaling to initiate and terminate this power updating
process. Apart from that, no further control information
needs to be exchanged between BS and PTXs. In the second
phase of DL-MDCA, based on the coverage obtained from
the first phase, BS allocates channels to different CPEs in
order to maximize the total downlink transmission rate.
This is achieved by formulating and solving a maximum
weighted bipartite matching problem. For the uplink
control scenario, a similar two-phase control algorithm,
called UL-MDCA, is proposed. In the first phase, distrib-
uted power control is carried out between PTXs and CPEs.
In the second phase, centralized channel assignment is
applied within the cognitive network.
It can be noted that, in our proposed system, secondary
nodes exercise cognition when intelligently adapting
their transmit power such that their SINR constraint is
met while not violating the SINR constraint of the primary
links. This is consistent with the concept of dynamic
spectrum access based on constraint in interference tem-
perature [4]. In traditional priority-based network, second-
ary devices are normally allowed to transmit only when the
348 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
. The authors are with the Institute for Infocomm Research, A-STAR,
01 Fusionopolis Way, #21-01 Connexis, Singapore 138632.
E-mail: {athoang, ycliang, habibul}@i2r.a-star.edu.sg.
Manuscript received 8 Aug. 2008; revised 4 Feb. 2009; accepted 1 May 2009;
published online 29 July 2009.
For information on obtaining reprints of this article, please send e-mail to:
tmc@computer.org, and reference IEEECS Log Number TMC-2008-08-0313.
Digital Object Identifier no. 10.1109/TMC.2009.136.
1536-1233/10/$26.00 2010 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
higher priority devices are not transmitting. To implement
our proposed system, there should be some mechanisms for
cognitive nodes to sense the channels and detect primary
users. This can be done using different spectrum sensing
techniques such as matched filter, energy detector, and
cyclostationary feature detector [5], [6], [7], [8].
The rest of the paper is organized as follows: Work
related to this paper is described in Section 2. In Section 3,
we describe the system model, introduce important nota-
tion, and discuss the operation principles of the cognitive
and primary networks. The problem of power control/
channel assignment to maximize the downlink throughput
of the cognitive radio network is considered in Section 4.
Particularly, the DL-MDCA scheme is proposed and some
important characteristics of this algorithm are discussed. In
Section 5, the uplink control scenario is considered and the
control algorithm UL-MDCA is proposed. In Section 6, we
provide analysis of the complexity of DL-MDCA and
UL-MDCA. To facilitate the study of DL-MDCA and
UL-MDCA, in Section 7, we introduce some simpler control
algorithms. Numerical results showing the behaviors and
performance of our proposed schemes are presented in
Section 8. Finally, we conclude the paper and outline future
research directions in Section 9.
2 RELATED WORK
In our previous work [9], [10], [11], we consider related
problems of centralized power/channel allocation to max-
imize the number of users supported in multichannel
cognitive radio networks. However, [9], [10], [11] only focus
on the downlink scenario and assume that primary users do
not cooperate with the cognitive radio users in any way. This
paper considers both downlink and uplink scenarios and
highlights the significant benefits when primary users
cooperate with cognitive users in a simple, distributed
manner. In [12], the authors study problems of secondary
spectrum access with minimum SINR guarantee and inter-
ference temperature constraint. They consider both scenar-
ios, when all the secondary links can be supported and when
not all the secondary links can be supported with their SINR
requirement. In the first scenario, the control objective is to
maximize the total transmission rate of the secondary users,
while in the second scenario, an access control problem also
needs to be dealt with. A control problem similar to [12] is
also studied in [13]. We note that, unlike this paper, in both
[12] and [13], no cooperation fromprimary users is assumed.
Furthermore, [12], [13] focus on single-channel scenario, and
therefore, channel allocation is not of concern.
Other works that consider channel assignment for
multichannel cognitive radio networks include [14], [15],
[16]. In [14], Wang and Liu consider a problem of
opportunistically allocating multiple licensed channels to
a set of cognitive stations so that the total number of
channel usages is maximized. In [15], Zheng and Peng
consider a similar problem and introduce a reward function
that is proportional to the coverage areas of base stations.
They also allow the interference effect to be channel specific.
The problem in [16] is for a cognitive-radio, multihop
wireless network. It should be noted that transmit power
control is not considered in [14], [15], [16]. Instead,
protection of primary users is based on physical separation
of communication entities. In particular, it is assumed that if
a node is beyond a certain distance from a primary user,
then no harmful interference will be caused. The channel
allocation problems in [14], [15] are then formulated as
graph coloring problems. The problem in [16] is solved via
integer linear programming.
Our paper is also related to [17] and [18], where a
problem of joint channel assignment and transmit power
control for multichannel wireless ad hoc networks is
considered. As the work is not for cognitive radio,
protecting primary users is not of concern. In a broader
context, our paper is related to work on conventional
problem of power control in cellular networks such as [19],
[20], [21]. Compared to this related work, the problem
considered in this paper is different due to the need to
protect primary users and the joint control over multiple
channels. In our system model, different primary users may
transmit and receive on different channels. The channel
usage pattern of primary users, coupled with their loca-
tions, makes the spectrum available for cognitive usage
irregular. This kind of spectrum irregularity, across available
channels and BS-CPE links, makes it necessary to carry out
channel assignment and power control jointly for all
channels in order to achieve a good system throughput.
3 SYSTEM MODEL
We consider an opportunistic spectrum access scenario as
depicted in Fig. 1. The spectrum of interest is divided into
1 channels that are licensed to a primary network of
` PTX-PRX links. Each of the ` PTX-PRX links occupies
one of the 1 channels (this implies ` 1). In the same
area, a secondary cognitive radio network is deployed. This
cognitive network consists of a BS serving a set of ` CPEs
by opportunistically making use of the 1 channels. We
consider both downlink (from BS to CPEs) and uplink (from
CPEs to BS) scenarios in the secondary cognitive network.
We assume that the BS can transmit and receive on up to
1 channels at a time while each CPE can transmit or receive
on only one channel at a time. Furthermore, each channel
can only be used by one CPE (in either downlink or uplink)
at a given time.
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 349
Fig. 1. Deployment of a cognitive radio network. The cognitive network
consists of one BS serving multiple fixed-location CPEs. The primary
network is modeled as a set of point-to-point PTX-PRX links.
3.1 Notation
To facilitate further discussion, let us define the following
notation (see Table 1):
. G
jj
i.,.c
denotes the channel gain from PTX i to PRX ,
on channel c, 1 i. , `, 1 c 1.
. G
::
i.,.c
denotes the channel gain from secondary
node i to secondary node , on channel c. Here, we
denote BS as secondary node 0 and ` CPEs as
secondary nodes 1 to `, 0 i. , `.
. G
:j
i.,.c
denotes the channel gain from secondary
node i to PRX , on channel c, 0 i `, 1 , `.
. G
j:
i.,.c
denotes the channel gain from PTX i to
secondary node , on channel c, 1 i `, 0 , `.
. 1
j
i.c
denotes the transmit power of PTX i on channel c.
If PTX i does not transmit on channel c, then 1
j
i.c
0.
. 1
:
i.c
denotes the transmit power of secondary node i
on channel c. If secondary node i does not transmit
on channel c, then 1
:
i.c
0.
. `
o
denotes the noise spectrum density at BS, CPEs,
and PRXs. Note that we assume the noise figure
being the same for all receivers just for the sake of
brevity. The results of the paper can be applied
directly to the case when different receivers experi-
ence different noise figures.
3.2 Downlink Scenario in Cognitive Network: SINR
at PRXs and CPEs
Consider the downlink scenario in the cognitive radio
network. For each channel c, let
jd
i.c
denote the SINR
experienced by PRX i.
jd
i.c
can be calculated as

jd
i.c

1
j
i.c
G
jj
i.i.c
`
o

P
`
,1.,6i
1
j
,.c
G
jj
,.i.c
1
:
0.c
G
:j
0.i.c
. 1 i `. 1
where it should be noted that 1
:
0.c
is the transmit power of BS
on channel c and G
:j
0.i.c
is the channel gain from BS to PRX i.
Similarly, let
:d
i.c
denote the SINR experienced by CPE i.

:d
i.c
can be calculated as

:d
i.c

1
:
0.c
G
::
0.i.c
`
o

P
`
,1
1
j
,.c
G
j:
,.i.c
. 1 i `. 2
3.3 Uplink Scenario in Cognitive Network: SINR at
PRXs and BS
Consider the uplink scenario in the cognitive radio network.
Assuming that CPE i (1 i `) is assigned to transmit on
channel c and letting
jn
i.i.c
denote the SINR experienced by
PRX i,
jn
i.i.c
can be calculated as

jn
i.i.c

1
j
i.c
G
jj
i.i.c
`
o

P
`
,1.,6i
1
j
,.c
G
jj
,.i.c
1
:
i.c
G
:j
i.i.c
. 1 i `. 3
At the same time, the SINR at BS, denoted by
:n
i.c
, can be
calculated as

:n
i.c

1
:
i.c
G
::
i.0.c
`
o

P
`
,1
1
j
,.c
G
j:
,.0.c
. 4
3.4 Protecting Primary Users
We assume that, for appropriate performance of the
primary network, the received SINR at each PRX must
be above a predefined value of
j
. In particular, let
c
denote the set of all PRXs that receives on channel c, when
the cognitive radio network operates in the downlink
scenario, we must have

jd
i.c
!
j
. 8i 2
c
. 5
When the cognitive radio network operates in the uplink
scenario and CPE i is assigned to transmit on channel c, we
must have

jn
i.i.c
!
j
. 8i 2
c
. 6
350 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
TABLE 1
Important Notation and Acronyms
To support the operation of the cognitive network, the
primary network must be able to tolerate a certain level of
interference from cognitive transmissions. We assume that
for each channel c and PTX i transmitting on c, i.e., i 2
c
,
there exists a transmit power
~
1
j
i.c
such that
~
1
j
i.c
G
jj
i.i.c
`
o

P
`
,1.,6i
~
1
j
,.c
G
jj
,.i.c

!
j
. 8i 2
c
. 7
where is a positive constant. Essentially, this assumption
means that, apart from the interference caused by other
primary transmissions, each primary receiver can tolerate
an extra interference, e.g., from secondary transmission,
equal to .
3.5 Transmission Rates for Secondary Connections
Using adaptive coding and/or modulation, the transmis-
sion rate between BS and each CPE (in either downlink or
uplink direction) can be varied according to the received
SINR. For a given channel, we assume that the transmission
rate between BS and CPE with SINR of can be written as
n
0. if <
:
.
). if !
:
.
&
8
where
:
is the SINR threshold below which no transmis-
sion is possible between BS and CPEs. When the SINR is
!
:
, the corresponding downlink transmission can be
carried out at rate ). Function ) depends on various
factors such as available coding/modulation schemes and
bit error rate requirement. However, in all cases, ) is
monotonically nondecreasing.
4 DOWNLINK THROUGHPUT MAXIMIZATION
In this section, let us consider the problem of power control
and channel assignment to maximize the downlink
throughput of the cognitive radio network while protecting
1 PTX-PRX links. The problem of maximizing the uplink
throughput is considered in Section 5.
Let o
i.c
be a binary variable that indicates whether
channel c is assigned to the downlink transmission from BS
toward CPE i. In particular, o
i.c
is set to 1 if channel c is
assigned to the downlink transmission toward CPE i.
Otherwise, o
i.c
is set to 0. The problem of power control/
channel assignment to maximize the total downlink
transmission rate of the cognitive network can be stated as
arg max
1
:
0.c
.1
j
i.c
.oi.c
X
1
c1
X
`
i1
n

:d
i.c

o
i.c
. 9
subject to:
0 1
:
0.c
1
:
. 8c 2 f1. . . . . 1g. 10
0 < 1
j
i.c
1
j
. 8i 2
c
. 8c 2 f1. . . . . 1g. 11

jd
i.c
!
j
. 8i 2
c
. 8c 2 f1. . . . . 1g. 12
X
1
c1
o
i.c
1. 8i 2 f1. . . . . `g. 13
X
`
i1
o
i.c
1. 8c 2 f1. . . . . 1g. 14
In (10), 1
:
is the maximum transmit power for BS on each
channel, and in (11), 1
j
is the maximum transmit power for
each PTX. These maximum powers can be regarded as the
intrinsic limits for the type of secondary/primary devices,
which are normally set out by spectrum regulators. Inequal-
ity (12) is due to the SINR constraint for primary links.
Inequality (13) is due to the fact that each CPE can only
receive on one channel at a time and (14) is because two CPEs
cannot simultaneously share a single channel.
4.1 General Approach
As the 1 channels are independent of each other and the
rate function ). is monotonically nondecreasing, without
loss of optimality, the downlink throughput maximization
problem can be separated into the following two phases:
. Phase 1Power Control: For each channel, find the
maximum transmit power of BS, together with the
transmit powers of PTXs, so that the SINR constraints
of all PRXs receiving on that channel are met.
. Phase 2Channel Assignment: After Phase 1, given
the maximum transmit powers of BS on 1 channels,
carry out channel assignment for ` CPEs so that the
total downlink transmission rate is maximized.
For Phase 1, one control option is to carry out centralized
power calculation for BS and all PTXs. In our previous
works in [9], [10], [11], we follow this approach and apply
the Perron-Frobenious theorem ([22]) to obtain Pareto-
optimal transmit power vectors. However, the centralized
power control approach requires knowledge of all channel
gains from BS to PRXs and from PTXs to CPEs. This means
a great deal of cooperation between primary and cognitive
networks. In this paper, we follow another power control
approach, which is carried out in distributed manner and
requires minimal cooperation between the two networks.
On the other hand, as Phase 2 only involves BS and CPEs, a
centralized channel assignment scheme is appropriate. We
term our control approach the Downlink Mixed Distrib-
uted/Centralized Algorithm (DL-MDCA). Each phase of
DL-MDCA is discussed below.
4.2 Phase 1 of DL-MDCA: Distributed Power Control
For Phase 1, we propose a synchronous downlink dis-
tributed power updating (DL-DPU) process that strives to
maximize the coverage of BS while always guaranteeing the
SINR constraint of all PRXs. The DL-DPU process is applied
to one channel at a time. For channel c, 1 c 1, the
following actions are carried out:
. Initialization: Either BS or one of the PTXs operating
on channel c initiates the power updating process by
broadcasting some special tone. All PTXs that
transmit on channel c, together with BS, must
participate in the power updating process. PTX i,
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 351
8i 2
c
, and BS set their transmit powers to the
initial values of 1
j
i.c
0 and 1
:
0.c
0, respectively.
. Power updating: At step /, the following activities are
carried out: 1) BS and PTXs transmit pilot signals at
the set power levels of 1
:
0.c
/ and1
j
i.c
/, respectively.
2) PRXs estimate their SINR and feed back to their
corresponding PTXs. 3) PTX i and BS update their
transmit powers according to:
1
j
i.c
/ 1 1
j
i.c
/
c
j

jd
i.c
/
. i 2
c
.
1
:
0.c
/ 1 1
:
0.c
/c.
15
Here,
jd
i.c
/ is the SINR at PRX i after step / and can
be calculated using (1). In addition, c is a power
scaling factor, which is slightly greater than one.
. Termination: The DL-DPU process will be terminated
if at least one of the following conditions is true:
- The SINR experienced by at least one of the
PRXs goes below the threshold
j
.
- At least one of the transmitters (BS and PTXs)
has its transmit power that approaches the
maximum transmit power constraint.
A node can terminate the DL-DPU process by
broadcasting some special tone (control message).
We assume that the special tone to initiate Phase 1 is
broadcasted on the same channel c. This can be achieved by
periodically reserving small time slots so that primary or
cognitive nodes can broadcast the tone. During these time
slots, all nodes will not carry out normal data transmission.
Nodes that wish to broadcast the special initiating tone in a
particular reserved time slot can do so in a random access
manner, e.g., using j-persistent random access or scheme
similar to contention-based ranging in IEEE 802.16 [23]. This
helps avoid collisions of multiple concurrent tones. It is
important to note that due to the simple nature of the tone,
the overhead of reserving time slots to broadcast the tone
should be negligible.
When one node transmits the tone to initiate the power
updating process, it forces other nodes to cease normal
communication to participate inthe power updating process.
This can affect the primary and secondary performances,
especially in mobile environment when frequent power
updates need to be carried out. To keep this negative effect
under control, one or a combination of the following
approaches can be employed: 1) a restriction can be set so
that only primary users can transmit the tone to initiate the
power updating process; 2) we can also set the limit on how
frequent the power updating process can be carried out.
As can be seen from (15), PTX i updates its transmit
power to aim at the SINR value of c
j
. On the other hand,
BS keeps on increasing its transmit power in order to
increase the cell coverage. A pseudocode for the DL-DPU
process is given in Fig. 12. Next, let us prove some
important properties of the proposed DL-DPU process
Proposition 1. The DL-DPU process will be terminated after a
finite number of iterations.
Proof. The transmit power of BS starts from an initial value
of 1
:
0.c
0 and is increased by factor c 1 after each
iteration. So the power updating process will stop after at
most dlog
c

1
:
1
:
0.c
0
e iterations. tu
Proposition 2. If the initial transmit powers 1
j
i.c
0 and 1
:
0.c
0
are selected such that
jd
i.c
0 !
j
. 8i 2
c
, then
jd
i.c
/ !

j
. 8/ ! 1.
Proof. The proof is by induction. Suppose that for some
/ ! 0,
jd
i.c
/ !
j
. We have

jd
i.c
/ 1

1
j
i.c
/ 1G
jj
i.i.c
`
o

P
`
,1.,6i
1
j
,.c
/ 1G
jj
,.i.c
1
:
0.c
/ 1G
:j
0.i.c
!
1
j
i.c
/c
j
,
jd
i.c
/G
jj
i.i.c
`
o

P
`
,1.,6i
1
j
,.c
/cG
jj
,.i.c
1
:
0.c
/cG
:j
0.i.c

jd
i.c
/
c1
j
i.c
/G
jj
i.i.c
`
o
c
P
`
,1.,6i
1
j
,.c
/G
jj
,.i.c
1
:
0.c
/G
:j
0.i.c

!

j

jd
i.c
/
1
j
i.c
/G
jj
i.i.c
`
o

P
`
,1.,6i
1
j
,.c
/G
jj
,.i.c
1
:
0.c
/G
:j
0.i.c

jd
i.c
/

jd
i.c
/
j
.
16
Note that the first inequality in (16) follows from
1
j
,.c
/ 1 c1
j
,.c
/

j

jd
,.c
/
c1
j
,.c
/. 81 , `. 17
and the last inequality in (16) follows from c 1. tu
Proposition 2 states that if we start with the initial
transmit powers for BS and all PTXs so that the SINR
constraints of all PRXs are met, then during the power
updating process, the SINR constraints of all PRXs are
always maintained. Based on (7), we can set 1
j
i.c
0
~
1
j
i.c
.
Then, if 1
:
0.c
is set to a sufficiently small value such that
1
:
0.c
G
:j
0.i
, from (7), we will have
jd
i.c
0 !
j
. 8i 2
c
. In
case the SINR constraint of a PRX is violated at initialization
(e.g., due to the change in channel condition), the power
updating process will be immediately terminated. The BS
then can further reduce its initial power level in subsequent
power updating process.
Proposition 3. The SINR experienced by each CPE increases
after each power updating step.
Proof. Using (2), for 1 i `, we have

:d
i.c
/ 1
1
:
0.c
/ 1G
::
0.i.c
`
o

P
`
i1
1
j
i.c
/ 1G
j:
i.i.c
!
1
:
0.c
/cG
::
0.i.c
`
o

P
`
i1
1
j
i.c
/cG
j:
i.i.c

1
:
0.c
/G
::
0.i.c
`
o

P
`
i1
1
j
i.c
/G
j:
i.i.c

:d
i.c
/.
18
where the first inequality is due to (17) and the last
inequality follows from c 1. tu
352 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
Proposition 3 states that the coverage, i.e., number of CPEs
that BS can cover, is nondecreasing after each iteration.
4.3 Phase 2 of DL-MDCA: Centralized Channel
Assignment
After the power updatingprocess, for eachchannel c, BS has a
maximumtransmit power 1
:
0.c
. Associated with this transmit
power are the SINRs experienced by ` CPEs. The problemis
how to assign 1 channels to different CPEs so that the total
downlink throughput is maximized. This is achieved by first
transforming the probleminto a weightedbipartite matching
and then finding a maximal weighted match.
The weighted bipartite graph is formed as follows: First,
represent the ` CPEs by a set of vertices, which is
connected to another set of vertices representing the 1
channels. An edge exists between the vertex representing
CPE i and the vertex representing channel c if and only if,
for channel c, the SINR at CPE i is not less than
:
. For each
edge, assign an weight that is equal to the corresponding
transmission rate calculated in (8). An example of such an
weighted bipartite graph is given in Fig. 2.
The problem of channel assignment to maximize the total
downlink transmission rate is equivalent to the problem of
finding a maximal matching for the corresponding
weighted bipartite graph. In this paper, we obtain maximal
weighted matching of a bipartite graph by the following
procedure [24]:
Maximal Weighted Bipartite Matching Procedure
. Step 1: Start with an empty match, i.e., without any
edge selected.
. Step 2: Find a maximum augmenting path for the
current match. An augmenting path is a path with
edges alternate between matched and unmatched.
The score of an augmenting path is equal to the sum
of weights of unmatched edges subtracted by the
sum of weights of matched edges. A maximum
augmenting path is the one with the maximum
score. If this score of the maximum augmenting path
is not positive, then finish as the current match is
maximal. Otherwise, go to Step 3.
. Step 3: Flip the maximum augmenting path obtained
in Step 2, i.e., change unmatched edges of the path to
matched and matched edges of the path to un-
matched. Go back to Step 2 to find another
maximum augmenting path and continue.
In our DL-MDCAalgorithm, Phase 1 of distributed power
control does not require the assumption that each CPE only
operates on one channel at a time. This assumption is used in
Phase 2 to simplify the channel assignment process. How-
ever, Phase 2 can also be readily extended to cover multiple-
channel operation case. In particular, if each CPE can operate
on up to 1 channels, then when forming the bipartite graph,
we just need to represent each CPE by 1 vertices and the rest
of the maximal matching operation is unchanged. It should
also be noted that, when channels change, the SINR
constraints for primary users can be violated. In that case,
the primary users should send out warning tone and
probably restart the power updating process.
4.4 Fairness Considerations
It can be noted that by maximizing the transmit power of
the BS, Phase 1 maximizes the set of CPEs that can be
potentially assigned channels. Given that, how fair the
channels are assigned to CPEs depends on specific
algorithm in Phase 2. In this paper, we focus on throughput
maximization and achieve that with bipartite matching. On
the other hand, if the focus is fairness, we need to frequently
reassign the channels. One way to do that is to divide the set
of CPEs into multiple subsets, and consider these subsets in
a round-robin manner. Given a particular subset of CPEs,
we can then apply bipartite matching to maximize the
achievable throughput for such subset.
5 UPLINK THROUGHPUT MAXIMIZATION
Now, let us consider the problem of power control/channel
assignment for maximizing the uplink throughput of the
cognitive radio network. Similar to the downlink scenario,
we propose a two-phase Uplink Mixed Distributed/
Centralized Algorithm (UL-MDCA). In the first phase of
UL-MDCA, distributed power control is carried out among
CPEs and PTXs in order to meet their SINR constraints. In
the second phase of UL-MDCA, centralized channel assign-
ment is applied within the cognitive network to maximize
the uplink throughput. It should be noted that, unlike the
downlink control scenario, in the uplink, the joint power
control/channel assignment is coupled, and our two-phase
approach is not optimal. However, this approach makes the
control problem much more manageable.
5.1 Phase 1 of UL-MDCA: Distributed Power Control
for Uplink Scenario
For the system model considered in this paper, uplink
power control is significantly more complicated than
downlink power control. In the downlink case, there is
only one secondary transmitter, i.e., the BS, which interferes
primary links. On the other hand, for the uplink case,
different CPEs act as different transmitters and interfere
with PRXs in different ways.
We assume that, for each uplink connection, as long as the
SINR (at BS) is above the required threshold of
:
, the
transmission rate is fixed at a predetermined value, i.e.,
) i. i 0. In other words, we do not consider rate
adaptation for uplink transmissions in the cognitive net-
work. Therefore, maximizing uplink throughput is equiva-
lent to maximizing the number of uplink connections with
satisfied SINR. We further assume that there is at most one
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 353
Fig. 2. Representing the coverage within one cell as a bipartite graph.
The number on each link is its capacity.
PTX-PRX link operating on each of the 1 channels. With
these assumptions, we propose the following Uplink
Distributed Power Control (UL-DPU) process that is applied
to one channel at a time. For channel c, c 2 f1. . . . 1g, let PTX
i be the primary transmitter that operates on channel c. The
UL-DPU process is carried out in two rounds as follows:
Round 1: The CPEs are considered one at a time, starting
from CPE 1. CPE i and PTX i carry out a distributed power
updating process in an iterative manner as follows:
. Initialization: Set the initial transmit power of CPE i
to 1
:
i.c
0 and initial transmit power of PTX i to 1
j
.
Similar to the case of downlink control, the initial
transmit power of CPE i is chosen small enough so
that 1
:
i.c
0G
:j
i.i.c
< .
. Iterative Power Updating: At step /, PTX i and CPE i
update their transmit powers as
1
j
i.c
/ 1 max 1
min
i
. 1
j
i.c
/
c
j

jn
i.i.c
/
( )
.
1
:
i.c
/ 1 1
:
i.c
/c.
19
where 1
min
i
is the minimum transmit power for PTX
i when it carries out power updating with CPE i.
We set 1
min
1
0 and calculate 1
min
i
(i 1) as
explained next.
. Termination: At step /, the power updating process
between CPE i and PTX i will terminate if at least
one of the following conditions is true:
- The SINR at PRX i is below the required value of

j
. In this case, the uplink connection between
CPE i and BS will not be supported. We then set
1
min
i1
1
min
i
. 20
- The transmit power of either CPE i or PTX i
exceeds the maximum allowable value (of 1
:
and 1
j
, respectively) before CPE i can achieve
the required SINR. In this case, the uplink
connection between CPE i and BS will not be
supported. We then set 1
min
i1
according to (20).
- The SINR at BS (corresponding to the transmis-
sion from CPE i) reaches the required value of

:
. In this case, CPE i records its transmit power
1
:
i.c
1
:
i.c
/, BS also remembers this CPE by
including it into a set
c
. We then update
1
min
i1
1
j
i.c
/. 21
Note that 1
min
i
is an operating parameter of PTX i. So
when the iteration terminated, BS needs to send a one-bit
update to PTX-PRX i to inform them about the outcome
of the power updating process, i.e., whether the uplink
connection from CPE i can be supported. Based on this
one-bit information, PTX i will set 1
min
i
according to
either (20) or (21). After the power updating process of
CPE i is terminated, BS, acting as the master, transmits a
control message to ask CPE i 1 to start its power
updating process.
Round 2: After all ` CPEs have been processed, a final
round of power updating is carried out as follows:
. PTX i keeps transmitting at power 1
min
`1
, i.e., the
transmit power it obtains after carrying out dis-
tributed power update with CPE `.
. For each CPE i that has been able to achieve the
required SINR of
:
, i.e., i 2
c
, BS measures its SINR
when PTX i transmits at power 1
min
`1
and instructs
CPE i to carry out a final power update according to
1
:
i.c

1
:
i.c

:n
i.c
if 1
:
i.c

:n
i.c
1
:
.
0 if 1
:
i.c

:n
i.c
1
:
.
8
>
>
>
<
>
>
>
:
22
Note that in (22),
:
i.c
is the SINR experienced at BS
when CPE i and PTX i transmit at powers 1
:
i.c
and
1
min
`1
, respectively.
A pseudocode for the UL-DPU procedure is given in
Fig. 13.
We state and prove the following important character-
istics of the UL-DPU process:
Proposition 4. Given that PTX i transmits at power 1
min
`1
, if
CPE i 2
c
is assigned channel c and transmits at power
1
:
i.c
0, the corresponding SINR at BS is
:
.
Proof. The poof follows directly from (22) and the fact
that SINR at BS scales linearly with the transmit power
of CPEs. tu
Proposition 5. For each channel c, no matter what CPE in the set

c
is assigned the channel for uplink transmission, the UL-DPU
procedure ensures that the SINRof the PTX-PRXlink operating
on channel c is always above the required threshold of
j
.
Proof. First, as the initial transmit power of CPE i is chosen
so that 1
:
i.c
0G
:j
i.i.c
< , from (7), we have
jn
i.i.c
0 !
j
.
Then, similar to the proof of Proposition 2, it can be
shown that during Round 1 of UL-DPU, the SINR of the
PTX-PRX link is always above the required threshold of

j
. What we need to prove is that Round 2 of UL-DPU
also ensures that the SINR of the PTX-PRX link to be
above the required threshold. If CPE i, i 2
c
, transmits
on channel c at power 1
:
i.c
and PTX i transmits at power
1
min
`1
, the SINR at PRX i is

jn
i.i.c

1
min
`1
G
jj
i.i.c
`
o
1
:
i.c
G
:j
i.i.c

1
min
`1
G
jj
i.i.c
`
o
1
:
i.c

:n
i.c
G
:j
i.i.c

1
iii
`1
1
iii
i
1
min
i
G
jj
i.i.c
`
o
1
:
i.c

:n
i.c
G
:j
i.i.c
!
1
min
i
G
jj
i.i.c
`
o

1
iii
i
1
iii
`1
1
:
i.c

:n
i.c
G
:j
i.i.c
.
23
354 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
where the last inequality follows from 1
min
`1
! 1
min
i
. We
further note that

:n
i.c

1
:
i.c
G
::
i.0.c
`
o
1
iii
`1
G
j:
i.0.c
24
and

1
:
i.c
G
::
i.0.c
`
o
1
iii
i
G
j:
i.0.c
. 25
as it is assumed that CPE i can achieve its SINR
constraint when transmitting at power 1
:
i.c
while PTX i is
transmitting at power 1
min
i
. Therefore,

:n
i.c

`
o
1
iii
`1
G
j:
i.0.c
`
o
1
iii
i
G
j:
i.0.c

1
iii
`1
1
iii
i
. 26
where the last inequality follows from 1
min
`1
! 1
min
i
.
Substituting (26) into (23) gives us
1
min
`1
G
jj
i.i.c
`
o
1
:
i.c
G
:j
i.i.c
!
1
min
i
G
jj
i.i.c
`
o
1
:
i.c
G
:j
i.i.c
!
j
. 27
So, the proof is completed. tu
5.2 Phase 2 of UL-MDCA: Centralized Channel
Assignment for Uplink Scenario
In this phase, channel assignment is carried out in a
centralized manner in the same way as it is done in Phase 2
of DL-MDCA scheme (Section 4.3). In particular, a bipartite
graph is first constructed, and after that, maximal bipartite
matching is applied to get the optimal channel allocation.
6 COMPLEXITY ANALYSIS FOR DL-MDCA AND
UL-MDCA
Let us look at the complexity of DL-MDCA and UL-MDCA.
For both algorithms, we can separate the overall complexity
into that of Phase 1 and Phase 2.
6.1 Complexity of Phase 1
Consider the DL-DPU process for the downlink scenario.
For each channel, the number of power updating iterations
is bounded and does not depend either on the number of
nodes in the cognitive network or the number of PTX-PRX
links (see Proposition 1). Therefore, the complexity of the
DL-DPU process is O1, where 1 is the number of
channels in the system.
Consider the UL-DPU process for the uplink scenario.
Again, for each channel and each CPE, the number of
power updating iterations is upper bounded by a constant.
Therefore, the complexity of UL-DPU process is O`1.
6.2 Complexity of Phase 2
The complexity of Phase 2 of either DL-MDCA or UL-
MDCA is due to the bipartite matching process. As it is well
known in graph theories [24], the complexity is
Oj\ j
2
j\ k1j, where j\ j is the number of vertices and
j1j is the number of edges in the corresponding bipartite
graph. The number of vertices in our problem is equivalent
to the number of CPEs, i.e., `. The number of edges is upper
bounded by `1. Therefore, the complexity of Phase 2 of
DL-MDCA and UL-MDCA is O`
2
1.
If we combine the complexity of Phase 1 and Phase 2, the
overall complexity of both DL-MDCA and UL-MDCA
scales as O`
2
1.
7 OTHER CONTROL ALGORITHMS
It can be noted that two basic components of the proposed
DL-MDCA and UL-MDCA schemes are the (limited)
cooperation from primary devices during the power updat-
ing process in Phase 1 and the centralized maximal bipartite
matching employed for channel assignment in Phase 2. We
are interested in understanding the impacts of these two
components in the overall performance of DL-MDCA and
UL-MDCA. To study these impacts, we consider the other
control algorithms given below.
7.1 Noncooperative, Simple Matching (NCSM)
Algorithm
The NCSM algorithm can be used in both downlink and
uplink scenarios. By noncooperative, we mean that during
the power control process in Phase 1 (of either downlink or
uplink scenario), PTXs do not adjust their transmit power
together with BS or CPEs. Instead, each PTX fixes its
transmit power at the maximum value of 1
j
. BS or a CPE
then tries to maximize its transmit power on each channel,
subject to the SINR constraints of all cochannel PRXs. By
simple matching, we mean that channel assignment in
Phase 2 (of either downlink or uplink scenario) is carried
out in a simple manner. In particular, CPEs are processed
one by one according to a random order. For each CPE that
has not been assigned any channel, we just randomly pick
one channel that this CPE can operate at (i.e., with SINR
greater than
:
) and assign to it.
By comparing the performance of DL-MDCA or UL-
MDCA to that of NCSM, we can see the combined impact of
having primary cooperation in Phase 1 and maximal
bipartite matching in Phase 2.
7.2 Noncooperative, Maximal Matching (NCMM)
Algorithm
The NCMM algorithm can be used in both downlink and
uplink scenarios. Phase 1 of this NCMM scheme is similar to
that of NCSM scheme described above, i.e., there is no
cooperation from primary transmitters. On the other hand,
in Phase 2, maximal bipartite matching is employed for
optimal channel assignment. By comparing the performance
of DL-MDCAor UL-MDCAto that of NCMM, we can see the
impact of having primary users cooperation in Phase 1.
7.3 Limited-Cooperation, Simple Matching (LCSM)
Algorithm
The LCSM algorithm can be used in both downlink and
uplink scenarios. Phase 1 of this LCSM scheme is similar to
that of DL-MDCA or UL-MDCA, i.e., PTXs participate in
the distributed power updating process described in
Section 4.2. On the other hand, for Phase 2, simple matching
scheme as described for NCSM is employed. By comparing
the performance of DL-MDCA or UL-MDCA to that of
LCSM, we can see the impact of carrying out optimal
bipartite matching in Phase 2.
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 355
7.4 Centralized Optimal Control Algorithm
In this algorithm, primary and secondary users measure
and report all channel conditions to one central node, e.g.,
the BS or a primary node. This central node then calculates
and decides the transmit power levels for all the transmit-
ters. In other words, centralized power control is carried out
in Phase 1. In phase 2, maximal bipartite matching is
employed for optimal channel assignment.
8 NUMERICAL RESULTS AND DISCUSSION
8.1 Simulation Model
We consider a square service area of size 1.000 m1.000 m
in which a cognitive radio network is deployed. A BS is
deployed at the center of the cell to serve a set of CPEs. The
total number of CPEs is ` 10. The number of PTX-PRX
links is set to 5 and 10. All CPEs and PTXs are randomly
deployed across the entire service area with a uniform
distribution. Then each PRX is deployed such that the offsets
of its horizontal and vertical coordinates, relative to the
coordinates of the corresponding PTX, are uniformly
distributed within the range [50 m, 100 m]. This deployment
model represents a practical scenario in the currently being
developed IEEE802.22 WRAN standard, which allows
cognitive BS and CPEs to share spectrum with short-ranged
incumbent wireless microphone devices [3]. A sample
network, with 10 CPEs and5 PTX-PRXlinks is given in Fig. 1.
We model an orthogonal frequency-division multiple
access (OFDMA) system in which the entire bandwidth is
divided into 48 subcarriers. Each subcarrier is regarded as
one channel in our power control/channel allocation
schemes. The fading channel is represented by a six-tap
channel, with exponential decay factor. Although there are
48 channels (subcarriers), we assume that only 10 of them
are considered for sharing between primary and secondary
networks. These 10 subcarriers are equally spaced within the
available bandwidth. The path loss exponent is taken to be 4.
The noise power density at each CPE is `
o
100 dBm. The
required SINR for each CPE is
:
15 dB. The required
SINR for each PRX is varied from 5 to 30 dB. The maximum
transmit power on each channel for BS and PTX is 50 mW.
For the distributed power updating process, we set the
initial transmit power for secondary nodes (BS or CPEs) to
be 0.5 mW and for PTXs to be 50 mW. The scaling factor
used in the power updating process (Section 4.2) is
c 1.259 % 1 dB. In our simulation, we assume that on
each channel, there is at most one PTX-PRX link.
For the transmission rate function discussed in Sec-
tion 3.5, we use the approximation in [25], i.e., )

0.6

1,3
.
We obtained similar results for other approximations of the
rate function.
8.2 Behaviors of DL-MDCADownlink Scenario
Let us fist look at the behaviors of the distributed power
updating process proposed in Section 4.2 (for downlink
scenario). In Fig. 3, we plot the SINRs experienced by a PRX
and different CPEs on a particular channel during a
distributed power updating process. As can be seen, as PTX
starts at its maximum transmit power, the initial SINR of the
PRX is much higher than the constrained value of 15 dB. On
the other hand, the initial SINRs of all CPEs are very low.
Then, when the power updating process is employed, the
SINR at PRX quickly converges to the target value, which is
16 dB in this case. Note that the SINR of PRX converges to
16 dB instead of the constrained value of 15 dB due to the use
of power scaling factor c 1.259 % 1 dB. As can be seen in
Fig. 3, the SINRs of CPEs increase gradually after each
iteration. Some CPEs eventually see their SINRs cross the
cutoff value of 15 dB while others never do. In Fig. 3, it is also
interesting to observe that some CPEs may start at lower
initial SINRs, however, their SINRs growquicker andsurpass
that of other CPEs. This is due to the variations in channel
gains from BS and PTXs to different CPEs. We observed that
with the given simulation parameters, it took 5-20 iterations
for the power updating process to terminate.
8.3 Performance of DL-MDCADownlink
Throughput
In Figs. 4 and 5, we plot the percentage gains in the total
downlink throughput of Optimal, DL-MDCA, NCMM, and
LCSM, relative to that of NCSM. Note that Optimal scheme
is based on centralized control as discussed in Section 7.4.
Fig. 4 is for the case when there are five PTX-PRX links and
Fig. 5 is for 10 PTX-PRX links. As can be seen, the
performance gain of DL-MDCA is very significant and
ranges from 60 to 160 percent. This shows the combined
356 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
Fig. 3. Behavior, in terms of received SINR at CPEs and PRXs, when
the distributed power control scheme is employed.
Fig. 4. Percentage gains in total downlink transmission rate for optimal,
DL-MDCA, LCSM, and NCMM, relative to NCSM. Number of primary
links 5.
impacts of having PTXs participating in the power updating
process in Phase 1 and carrying out maximal matching in
Phase 2. It can also be noted that optimal centralized control
scheme performs slightly better than DL-MDCA. However,
as discussed in Section 7.4, this optimal scheme requires
knowledge of all channel conditions for a central node to
calculate and set the transmit power levels for all secondary
and primary transmitters.
When comparing the performance of DL-MDCA and
NCMM in Figs. 4 and 5, it is evident that having primary
users cooperation in Phase 1 is important when the SINR
constraint of PRXs is low. However, at high SINR constraint
for PRXs, there is not much gain obtained from PTXs
cooperation. This effect is expected, as when the SINR
requirement of PRXs is low, PTXs can cooperate more by
reducing their transmit powers. Also, from Figs. 4 and 5, it
can be observed that the cooperation from PTXs has more
positive impact when the number of primary links increases
from 5 to 10. This is because with more primary links, it is
important that PTXs adjust their power to reduce inter-
ference caused to CPEs.
When comparing the performance of DL-MDCA and
LCSM in Figs. 4 and 5, it can be noted that carrying out
maximal weighted bipartite matching in Phase 2 gives
significant performance gain for DL-MDCA. Moreover, this
performance impact (of using maximal matching) is more
prominent when the SINR constraint of PRXs increases.
This is because when the SINR constraint of PRXs increases,
the channel availability pattern varies more significantly
across CPEs. That makes it important to carry out intelligent
channel assignment, which is achieved with maximal
bipartite matching.
In Figs. 6 and 7, we plot the gains of DL-MDCA,
NCMM, and LCSM, relative to that of NCSM, when the
performance metric is the total number of CPEs served,
instead of the downlink throughput. This is equivalent to
setting the rate function ) 1. 8 !
:
. As can be seen,
the performance trends are similar to that of Figs. 4 and 5.
However, the gains are less significant. This is because by
making ). a constant function, there are less variations in
the system for the control schemes to exploit. For reference,
we plot the absolute number of CPEs served by the baseline
NCSM scheme in Fig. 8.
8.4 Performance of UL-MDCAUplink Throughput
InFigs. 9 and10, we plot the percentage gain, inthe number of
uplink connections being supported, of Optimal, UL-MDCA,
LCSM, andNCMM, relative to that of NCSM. Note that as we
assume that all active uplinks operating at the same SINR
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 357
Fig. 6. Percentage gains in number of CPEs served for DL-MDCA,
LCSM, and NCMM, relative to NCSM. Number of primary links 5.
Fig. 7. Percentage gains in number of CPEs served for DL-MDCA,
LCSM, and NCMM, relative to NCSM. Number of primary links 10.
Fig. 8. Average number of CPEs served in the downlink by the baseline
scheme NCSM.
Fig. 5. Percentage gains in total downlink transmission rate for optimal,
DL-MDCA, LCSM, and NCMM, relative to NCSM. Number of primary
links 10.
(equal to the minimum required SINR of
:
), the percentage
gain in number of active uplink connections is equivalent to
the percentage gain in the total uplink throughput.
As can be observed, the performance trends of
UL-MDCA, LCSM, and NCMM for the uplink scenario
are similar to that for the downlink scenario. When there are
five PTX-PRX links, the gain of UL-MDCA, relative to
NCSM, is from 6 to 13 percent. However, when the number
of PTX-PRX links is increased to 10, the gain becomes much
more significant and ranges from 20 to 120 percent. We also
note that, when there are more primary links, cooperation
from primary nodes becomes much more beneficial to the
cognitive network. This is evident by the performance gap
between LCSM and NCMM in Fig. 10. It can also be
observed that the performance of Optimal scheme is
slightly better than that of UL-MDCA. However, as
discussed in Section 7.4, this optimal scheme requires
knowledge of all channel conditions for a central node to
calculate and set the transmit power levels for all secondary
and primary transmitters. For reference, we plot the
absolute number of CPEs served by the baseline NCSM
scheme in Fig. 11.
9 CONCLUSIONS
In this paper, we consider the problem of maximizing the
throughput of a cognitive radio network while protecting
primary users of the spectrum. For this, we propose two
mixed distributed/centralized control schemes (for down-
link and uplink scenarios) that require minimal cooperation
between cognitive and primary devices. Numerical results
show the desired behaviors of our proposed algorithms and
358 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010
Fig. 11. Average number of CPEs served in the uplink by the baseline
scheme NCSM.
Fig. 12. DL-DPU process for channel c.
Fig. 10. Percentage gains in number of CPEs served for optimal,
UL-MDCA, LCSM, and NCMM, relative to NCSM, in the uplink scenario.
Number of primary links 10.
Fig. 9. Percentage gains in number of CPEs served for optimal,
UL-MDCA, LCSM, and NCMM, relative to NCSM, in the uplink scenario.
Number of primary links 5.
also demonstrate that the algorithms result in significant
performance gain, in terms of the downlink and uplink
throughput of the cognitive network.
For future research, we are considering the problem
under a multicell scenario. In such a scenario, multiple
cognitive cells share the common set of channels. The
problem then becomes more complex, as there is a trade-off
of coverage and throughput among cochannel cells.
ACKNOWLEDGMENTS
Dr. Y.-C. Liang was the corresponding author.
REFERENCES
[1] FCC, SpectrumPolicy Task Force Report, FCC02-155, Nov. 2002.
[2] FCC, Facilitating Opportunities for Flexible, Efficient, and
Reliable Spectrum Use Employing Cognitive Radio Technologies,
Notice of Proposed Rule Making and Order, FCC 03-322, Dec.
2003.
[3] IEEE 802.22 Wireless RAN, Functional Requirements for the
802.22 WRAN Standard, IEEE 802.22-05/0007r46, Oct. 2005.
[4] S. Haykin, Cognitive Radios: Brain-Empowered Wireless Com-
munications, IEEE J. Selected Areas Comm., vol. 23, no. 2, pp. 13-
18, Feb. 2005.
[5] D. Cabric, S. Mishra, and R.W. Brodersen, Implementation Issues
in Spectrum Sensing for Cognitive Radio, Proc. Asilomar Conf.
Signals, Systems and Computers, Oct. 2004.
[6] W.A. Gardner, Exploitation of Spectral Redundancy in Cyclosta-
tionary Signals, IEEE Signal Processing Magazine, vol. 8, no. 2,
pp. 14-36, Apr. 1991.
[7] A. Sahai and D. Cabric, Spectrum Sensing: Fundamental Limits
and Practical Challenges, Proc. IEEE Intl Symp. New Frontiers in
Dynamic Spectrum Access Networks (DySPAN 05), Nov. 2005.
[8] Y.C. Liang, Y.H. Zeng, E. Peh, and A.T. Hoang, Sensing-
Throughput Tradeoff for Cognitive Radio Networks, IEEE Trans.
Wireless Comm., vol. 7, no. 4, pp. 1326-1337, Apr. 2008.
[9] A.T. Hoang and Y.C. Liang, Maximizing Spectrum Utilization of
Cognitive Radio Networks Using Channel Allocation and Power
Control, Proc. 64th IEEE Vehicular Technology Conf. (VTC 06 Fall),
Sept. 2006.
[10] A.T. Hoang and Y.C. Liang, A Two-Phase Channel and Power
Allocation Scheme for Cognitive Radio Networks, Proc. Intl
Symp. Personal, Indoor, and Mobile Radio Comm. (PIMRC 06), Sept.
2006.
[11] A.T. Hoang and Y.C. Liang, Downlink Channel Assignment and
Power Control for Cognitive Radio Networks, IEEE Trans.
Wireless Comm., vol. 7, no. 8, pp. 3106-3117, Aug. 2008.
[12] Y. Xing, C.N. Mathur, M. Haleem, R. Chandramouli, and K.
Subbalakshmi, Dynamic Spectrum Access with Qos and Inter-
ference Temperature Constraints, IEEE Trans. Mobile Computing,
vol. 6, no. 4, pp. 423-433, Apr. 2007.
[13] L. Le and E. Hossain, Qos-Aware Spectrum Sharing in Cognitive
Wireless Networks, Proc. IEEE GLOBECOM, Nov. 2007.
[14] W. Wang and X. Liu, List-Coloring Based Channel Allocation for
Open-Spectrum Wireless Networks, Proc. IEEE 62nd Vehicular
Technology Conf. (VTC 05 Fall), Sept. 2005.
[15] H. Zheng and C. Peng, Collaboration and Fairness in Opportu-
nistic Spectrum Access, Proc. IEEE Intl Conf. Comm. (ICC 05),
May 2005.
[16] M. Thoppian, S. Venkatesan, R. Prakash, and R. Chandrasekaran,
Mac-Layer Scheduling in Cognitive Radio Based Multi-Hop
Wireless Networks, Proc. Intl Symp. World of Wireless, Mobile, and
Multimedia Networks (WoWMoM 06), June 2006.
[17] A. Behzad and I. Rubin, Multiple Access Protocol for Power-
Controlled Wireless Access Nets, IEEE Trans. Mobile Computing,
vol. 3, no. 4, pp. 307-316, Oct.-Dec. 2004.
[18] G. Kulkarni, S. Adlakha, and M. Srivastava, Subcarrier Allocation
and Bit Loading Algorithms for OFDMA-Based Wireless Net-
works, IEEE Trans. Mobile Computing, vol. 4, no. 6, pp. 652-662,
Nov./Dec. 2005.
[19] J. Zander, Performance of Optimum Transmitter Power Control
in Cellular Radio Systems, IEEE Trans. Vehicular Technology,
vol. 41, no. 1, pp. 57-62, Feb. 1992.
[20] G.J. Foschini and Z. Miljanic, A Simple Distributed Autonomous
Power Control Algorithm and Its Convergence, IEEE Trans.
Vehicular Technology, vol. 42, no. 4, pp. 641-646, Apr. 1993.
[21] N. Bambos, S.C. Chen, and G.J. Pottie, Radio Link Admission
Algorithms for Wireless Networks with Power Control and Active
Link Quality Protection, Proc. IEEE INFOCOM, Nov. 1995.
[22] E. Seneta, Non-Negative Matrices. John Wiley & Sons, 1973.
[23] IEEE 802.16, IEEE Standard for Local and Metropolitan Area
NetworksPart 16: Air Interface for Fixed Broadband Wireless Access
Systems, IEEE, Oct. 2004.
[24] K. Mehlhorn and S. Naher, The LEDA Platform of Combinatorial and
Geometric Computing. Cambridge Univ. Press, 1999.
[25] D. Kivanc and H. Liu, Subcarrier Allocation and Power Control
for OFDMA, Proc. 34th Asilomar Conf. Signals, Systems and
Computers, Oct. 2000.
HOANG ET AL.: POWER CONTROL AND CHANNEL ALLOCATION IN COGNITIVE RADIO NETWORKS WITH PRIMARY USERS COOPERATION 359
Fig. 13. UL-DPU process for channel c. Note that PTX i is the only
primary transmitter operating on channel c.
Anh Tuan Hoang received the bachelors
degree (with first class honors) in telecommu-
nications engineering from the University of
Sydney in 2000 and the PhD degree in
electrical engineering from the National Univer-
sity of Singapore in 2005. He is currently a
research fellow at the Department of Network-
ing Protocols, Institute for Infocomm Research,
Singapore. His research focuses on design/
optimization of wireless communication net-
works. His specific areas of interest include cross-layer design,
dynamic spectrum access, and cooperative communications. He is a
member of the IEEE.
Ying-Chang Liang received the PhD degree in
electrical engineering in 1993. He is now a
senior scientist in the Institute for Infocomm
Research (I2R), Singapore, where he has been
leading the research activities in the area of
cognitive radio and cooperative communications
and the standardization activities in IEEE 802.22
wireless regional networks (WRAN) for which his
team has made fundamental contributions in
physical layer, MAC layer, and spectrum sen-
sing solutions. He also holds adjunct associate professorship positions
at Nanyang Technological University (NTU), Singapore, and the
National University of Singapore (NUS), and adjunct professorship
position at the University of Electronic Science and Technology of China
(UESTC). He has been teaching graduate courses at the NUS since
2004. From December 2002 to December 2003, he was a visiting
scholar at the Department of Electrical Engineering, Stanford University.
His research interest includes cognitive radio, dynamic spectrum
access, reconfigurable signal processing for broadband communica-
tions, space-time wireless communications, wireless networking, in-
formation theory, and statistical signal processing. He served as an
associate editor of the IEEE Transactions on Wireless Communications
from 2002 to 2005, lead guest editor of the IEEE Journal on Selected
Areas in Communications special issue on cognitive radio: theory and
applications, and guest editor of the Computer Networks Journal
(Elsevier) special issue on cognitive wireless networks. He received
the Best Paper Awards from the IEEE VTC-Fall 1999 and the IEEE
PIMRC 2005, and the 2007 Institute of Engineers Singapore (IES)
Prestigious Engineering Achievement Award. He has served for various
IEEE conferences as a technical program committee (TPC) member. He
was the publication chair of the 2001 IEEE Workshop on Statistical
Signal Processing, TPC cochair of the 2006 IEEE International
Conference on Communication Systems (ICCS 06), panel cochair of
the 2008 IEEE Vehicular Technology Conference Spring (VTC 08-
Spring), TPC cochair of the 3rd International Conference on Cognitive
Radio-Oriented Wireless Networks and Communications (CrownCom
08), deputy chair of the 2008 IEEE Symposium on New Frontiers in
Dynamic Spectrum Access Networks (DySPAN 08), and cochair of the
Thematic Program on Random Matrix Theory and Its Applications in
Statistics and Wireless Communications, Institute for Mathematical
Sciences, the National University of Singapore, 2006. He holds six
granted patents and more than 15 filed patents. He is a senior member
of the IEEE.
Md Habibul Islam received the MSc degree in
electrical engineering from Tajik Technical Uni-
versity, in 1995, the MBA degree in finance from
the Institute of Business Administration (IBA),
the University of Dhaka, Bangladesh, in 1999,
and the MS degree in electrical engineering and
the PhD degree in telecommunications engi-
neering from the University of Texas at Dallas in
2002 and 2005, respectively. His dissertation
focused on the interference suppression
schemes for the downlink of multiple antenna code-division multiple
access (CDMA) systems. In February 2006, he joined the Institute for
Infocom Research (I2R), Singapore, as a research fellow, where he is
working on the multiple antenna schemes for IEEE 802.22 Wireless
Regional Area Networks (WRAN) standard. His research interests
include cognitive radio, orthogonal frequency-division multiplexing/
orthogonal frequency-division multiple access (OFDM/OFDMA), and
multiple antenna technology, such as space-time coding and beamform-
ing. He is a member of the IEEE.
> For more information on this or any other computing topic,
please visit our Digital Library at www.computer.org/publications/dlib.
360 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 3, MARCH 2010

You might also like