You are on page 1of 7

1858

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 4, MAY 2010

An Efficient Power-Loading Scheme for


OFDM-Based Cognitive Radio Systems
Yonghong Zhang, Member, IEEE, and Cyril Leung, Member, IEEE

AbstractWe study the bit and power-allocation problem for an


orthogonal frequency-division multiplexing (OFDM)-based cognitive radio (CR) system in which the spectrum is licensed to
a primary user (PU) pair consisting of one transmitter and one
receiver. CR pairs (CRPs) may use both active and nonactive PU
bands as long as the generated interference powers are within the
interference temperature limits of the PUs. The optimal solution
and a low-complexity suboptimal solution are proposed. It is found
that significant improvements can be achieved compared with
systems in which CRPs only use the nonactive PU bands.
Index TermsCognitive radio (CR), orthogonal frequencydivision multiplexing (OFDM), power allocation, power loading,
resource allocation and interference management.

I. I NTRODUCTION

OGNITIVE radio (CR) [1], [2] is a concept that can


potentially alleviate the problem of limited unlicensed
frequency bands. As discussed in [3], orthogonal frequencydivision multiplexing (OFDM) is an attractive modulation candidate for CR systems. In [4], the FCC Spectrum Policy Task
Force recommended the use of the interference temperature for
assessing the level of interference. Due to a lack of information
on a reasonable means for implementation, the interference
temperature approach was later shelved [5]. However, it is
noted in [5] that the approach could be reconsidered at some
future time.
The specification of a primary user (PU) pair (PUP) interference temperature limit corresponds to a maximum allowable interference power and makes it possible for CR transceiver pairs
(CRPs) to use active PUP bands in a controlled fashion. Two
types of interference need to be considered, namely, cochannel
interference (CCI) and cross-channel interference (XCI). CCI
arises when a CRP uses an active PUP band. Since a PUP may
not use OFDM, XCI [6] could be generated by a CRP using
an OFDM subchannel close (adjacent) to the PUP band and
from the PUP transmitter to the CRPs subchannels. Only XCI
Manuscript received May 22, 2009; revised September 15, 2009. First
published December 28, 2009; current version published May 14, 2010. This
work was supported in part by the Natural Sciences and Engineering Research
Council of Canada (NSERC) under Grant OGP0001731, by an NSERC postgraduate scholarship, and by the University of British Columbia PMC-Sierra
Professorship in Networking and Communications. The review of this paper
was coordinated by Dr. O. Holland.
Y. Zhang was with the Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, BC V6T 1Z4, Canada.
She is now with Google Inc., Mountain View, CA 94043 USA (e-mail:
joyz@ece.ubc.ca).
C. Leung is with the Department of Electrical and Computer Engineering,
The University of British Columbia, Vancouver, BC V6T 1Z4, Canada (e-mail:
cleung@ece.ubc.ca).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2009.2039154

is treated in [7] and [8], since it is assumed that CRPs may


only use nonactive PUP bands. In [9] and [10], only CCI is
considered since PUPs are assumed to use OFDM. Failure to
consider both XCI and CCI may result in excessive interference
to the PUP operation or unacceptably poor quality of service
for the CRPs. In [11], a recursive algorithm is proposed to find
the subcarrier power allocation that maximizes the capacity,
subject to a total (over all subcarriers) secondary user (SU)
transmit power and individual subchannel SU transmit power
constraints.
In this paper, we propose a low-complexity suboptimal
power allocation algorithm for an OFDM-based CR system in
which CRPs may use both nonactive and active PUP bands as
long as the total XCI and CCI powers do not exceed prescribed
limits. Most existing studies, e.g., [9] and [10], in which CRPs
may use both nonactive and active PUP bands, only consider
CCI. Our results show that, even when both XCI and CCI are
accounted for in allocating resources, it is possible for CRPs
to use both nonactive and active PUP bands to substantially
improve spectrum efficiency.

II. S YSTEM M ODEL


Since our focus is on power loading, we assume that the
assignment of subchannels to CRPs has been made. Thus,
without much loss of generality, we consider one CRP in a
CR system with access to a band of width W Hz. Nonoverlapping portions of this band are licensed to L PUPs, with
the nominal bandwidth of PUP l ranging from fc + FlPU to
fc + FlPU + Wl , l = 1, 2, . . . , L. The maximum interference
power that PUP l can tolerate is Il = Tl Wl [4], where Tl is the
interference temperature limit for PUP l.
A CRP can use nonactive and active PU bands provided that
the total interference in PUP ls band does not exceed Il if PUP
l is active. The CR system accommodates M equally spaced
OFDM subcarriers (subchannels). The nominal bandwidth of
subchannel m, m = {1, 2, . . . , M }, ranges from fc + (m
1)f to fc + mf . Let Ml be the set of subchannels in PUP
ls band and L be the set of active PUPs.
The subchannels are modeled in discrete time, with the timevarying power gain for subchannel m from the CRP transmitter
to its receiver denoted by gm . The power gains for subchannel
m from the CRP transmitter to PUP ls receiver and from PUP
ls transmitter to the CRP receiver are denoted by hl,m and dl,m ,
respectively. It is assumed that there is no intercarrier interference. For notational clarity, we denote the random variables
corresponding to the sample power gains gm , hl,m , and dl,m
by Gm , Hl,m , and Dl,m , respectively.

0018-9545/$26.00 2010 IEEE


Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

ZHANG AND LEUNG: EFFICIENT POWER-LOADING SCHEME FOR OFDM-BASED COGNITIVE RADIO SYSTEMS

CR
Let fl,m
be the interference power experienced by PUP ls
receiver due to the CRP signal in subchannel m when sm = 1,
where sm is the transmit power. Then

 FlPU (m 12 )f +wl
CR (f )df, l L
hl,m F PU
CR
(m 12 )f
(1)
fl,m =
l
0,
l L

where CR (f ) is the baseband power spectral density (PSD)


of the CRP OFDM signal in subchannel m when sm = 1. The
interference power generated by the CRP signal in subchannel
CR
, which represents the XCI
m to PUP ls receiver is sm fl,m
power when subchannel m is outside PUP ls band and the CCI
power when subchannel m is within PUP ls band.
We consider the problem of maximizing the overall rate
achievable by the CRP while keeping the interference powers experienced by the PUPs below the specified interference
power thresholds {Il }, l L. The optimization problem is
formulated as
OP1 : max
sm



sm
log2 1 +
Nm
m=1
M


(2)

subject to
M


sm S

(3)

CR
sm fl,m
Il

(4)

m=1
M

m=1

sm 0,

m = 1, 2, . . . , M.

(5)

In (2), log2 (1 + sm /Nm ) is the number of bits per OFDM


symbol, 
which can be supported by subchannel m [12], Nm =
PU
)/gm is the equivalent noise power, is a
(02 + lL fl,m
signal-to-noise ratio gap parameter that indicates how far the
system is operating from capacity, 02 is the noise power, and
PU
is the interference power generated in subchannel m at the
fl,m
CRP receiver by PUP l. In (3), S is the CRP power limit.
III. O PTIMAL S OLUTION FOR OP1
To find the optimal solution
 for OP1, we rewrite the objective
function in (2) as minsm M
m=1 log2 (1 + (sm /Nm )), which

is convex since the function f (sm ) = log2 (1 + (sm /Nm )) is


convex. Furthermore, since the functions on the left-hand side
of (3)(5) are affine, OP1 is a convex optimization problem.
The Lagrangian [13] can be written as

 M


M


sm
log2 1 +
sm S
F =
+
Nm
m=1
m=1
 M

L
M



CR
l
sm fl,m Il
m sm (6)
+
l=1

m=1

(KKT) conditions [13], we obtain the optimal power


allocation as
+
s(1)
m = [m Nm ]

(7)

CR

where [x]+ = max(0, x), and m = 1/( L


l=1 l fl,m + )
with 0 and l 0.
Solving for the L + 1 Lagrange multipliers { , l , l =
1, 2, . . . , L} is computationally complex. The interior-point
method can also be used to solve OP1 with a complexity
O(M 3 ) [13]. The optimal solution proposed in [11] has a
similar computational complexity to solving (7). We next propose an approximate lower complexity method for solving the
power-allocation problem.
IV. L OW-C OMPLEXITY A PPROXIMATE S OLUTION FOR OP1
Rather than directly solving OP1, we will solve a simpler
problem, motivated by the observation that XCI to a PUP
band mostly comes from the immediately adjacent subchannels
[6]. Assuming that the bandwidth of a PUP is much larger
than that of a subchannel in an OFDM-based CR system and
that there is usually a guard band between two adjacent PUP
bands, XCI from any CRP subchannel mostly impacts one PUP
band, instead of several PUP bands, as in optimization problem
OP1. We thus view each subchannel as belonging to its closest
PUP band and assume that it generates interference only to a
single PUP band, namely, its own. In a practical system, a PUP
band is typically several megahertz wide, whereas an OFDM
subchannel is usually only a few kilohertz wide. The PUP bands
might be separated by a guard band that is several tens of
kilohertz wide.
We formulate an optimization problem, namely, OP2, based
on the approximation described in the previous paragraph and
discuss the optimal solution in Section IV-A. A low-complexity
suboptimal solution for OP2 is proposed in Section IV-B. In
Section IV-C, a suboptimal algorithm for an attendant subproblem OP3 is described. In Section IV-D, the computational complexity and effectiveness of the proposed suboptimal solution
for OP2 are discussed.
A. Problem Formulation
Using the aforementioned approximation, we formulate the
optimization problem OP2 as follows:
OP2 : max
sm

M

m=1


log2

sm
1+
Nm


(8)

subject to
M


sm S

(9)

m=1

m=1

where , {l , l = 1, 2, . . . , L}, and {m , m = 1, 2, . . . , M }


are Lagrange multipliers. Applying the KarushKuhnTucker

1859

CR
sm fl,m
Il ,

lL

(10)

m = 1, 2, . . . , M.

(11)

mMl

sm 0,

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

1860

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 4, MAY 2010

OP2 is the same as OP1, except that constraint (4) is replaced


by (10). The simulation results presented in Section V show that
the proposed suboptimal solution to OP2 is a near-optimal solution to OP1 in most cases and that the simplifying assumption
is reasonable.
Using a similar approach leading to (7), we obtain
+
s(2)
m = [m Nm ]

(12)

where
m =

1
CR
l fl,m

0;

l 0.

(13)

CR
=
For the subchannels in nonactive PUP bands, because fl,m
0, the power allocation has a water-filling interpretation with a
water level 1/ . Note that, for a subchannel m in an active
CR
are nonnegative.
PUP band, m 1/ , since both l and fm
From (12) and (13), we have

s(2)
m

1
Nm

+
.

(14)

CR
Notice that, when fl,m
= fl m, OP2 reduces to a problem
that is essentially the same as that in [9], in which an optimal
iterative water-filling algorithm has been proposed. In OP2,
CR
, m Ml are generally different for a given active
because fl,m
PUP band, the design of a power allocation algorithm for OP2 is
more difficult. Since low-complexity algorithms are preferable,
we propose a fast suboptimal power-allocation algorithm for
OP2 in the next section.

B. Suboptimal Algorithm for Problem OP2


As already noted, if was known, then the power allocation
for subchannels in nonactive PUP bands can simply be obtained
by using water filling with the water level 1/ . For subchannels in active PUP bands, however, we need to find l for each
PUP l, which is not an easy task as implied by (12) and (13). To
design a low-complexity algorithm, we formulate a subproblem
OP3 for each active PUP band to simplify the problem. In the
formulated OP3, apart from the total interference power limit,
each subchannel has a power limit obtained from the optimal
solution for OP2, as suggested by (14). In the rest of this
section, a method for obtaining a suboptimal solution to OP2
is described. The solution for subproblem OP3 is detailed in
Section IV-C.
If no PUP is active, then constraint (10) is automatically
satisfied, and the solution (12) has the standard water-filling
interpretation [14]. To develop a fast suboptimal solution, we
start by assuming that constraint (10) is satisfied and find
the resulting water level and the optimal solution {sm =
all sub[ Nm ]+ , m = 1, 2, . . . , M } using water
 filling for CR
Il
channels with a CRP power limit S. If mMl sm fl,m
for every l, then the optimal power
allocation
has
been
found.

CR
Otherwise, the PUP l for which mMl sm fl,m
/Il is highest
is determined, and the power allocation for the PUP l sub-

channels is adjusted to ensure that the total interference power


is below Il . This power adjustment for PUP l is formulated
as subproblem OP3, and a suboptimal solution is provided in
Algorithm 2 in Section IV-C. According to the optimal solution
for OP2, we know that the power allocated to the subchannels
in Ml should be no higher than [1/ Nm ]+ based on (14),
which shows that the power allocated for PUP l is lowered
after the adjustment. The reduction in CRP transmit power in
subchannels m Ml is then redistributed to the subchannels
with a water level . A new water level is then calculated. If,
with the new (higher) water level, there still exists a PUP l with
an interference power higher than Il , OP3 with l = l is again
solved to lower the interference level for PUP l. This process
continues until the interference power level for each active PUP
l is below Il . An implementation of the procedure is given in
Algorithm 1.
Algorithm 1 Power-Allocation Algorithm.
1) Initialize M = N = {1, 2, . . . , M } and p = S. Sort
{Nm , m M} in decreasing order, with K being the set of ordered subchannel indexes. Find the water level for subchannels
in M using the algorithm proposed in [17] with power limit p
as follows:

1.1) Nsum = mN Nm , = (Nsum + p)/|N |, n = 1.
1.2) While NKn > , let Nsum = Nsum NKn , N =
n = n + 1.
N /{Kn }, = (Nsum + p)/|N |,
1.3) Set sm = [ Nm ]+ , m M.
2) If (10) is satisfied for every active PUP l, stop; otherwise,
go to Step 3).

CR
/Il ), let
3) Determine l = arg maxlL ( mMl sm fl,m
+
Ml = Ml \ {Nm , m Ml }, and let sm = 0, m
(3)
+
Ml \ M+
l ; then, execute Algorithm 2 to get sm , m Ml .

(3)
(3)
Set sm = sm , m M+
sm , Nsum = Nsum
l , p = p
mM+
l

Nm , M = M \ M+ , N = N \ M+ .
+
mMl

4) Find the water level for subchannels in M with power


limit p as follows: Repeat Steps 4.1) and 4.2) until NKn > or
n 0.
4.1) n = n 1, = (p + Nsum + NKn )/(|N | + 1).
4.2) if NKn and Kn M, then Nsum = Nsum +
NKn , N = {N , Kn }.
5) = (Nsum + p)/|N |, sm = [ Nm ]+ , m M, go to
Step 2).
To explain how Algorithm 1 operates, we show an example
in Fig. 1 for a four-PUP OFDM-based CR system. At the
time of resource allocation, PUP 3 is nonactive, and the other
PUPs are active. In the initialization step, i.e., Step 1), we
assume that no PUP is active, and the initial water level is
calculated for all four PUP bands. Since not all PUP bands
interference limits are satisfied, we proceed to Step 3) and
start the first iteration, i.e., IT 1. In IT 1, PUP 2 is selected to
perform interference power reduction because it has the highest
interference-power-to-limit-power ratio. After the power adjustment for PUP 2, a new water level is calculated in Steps 4)
and 5) for PUP bands 1, 3, and 4. Because there still exists
a PUP band l that has interference power higher than Il , we
move to the next iteration IT 2, in which PUP 1 is selected to

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

ZHANG AND LEUNG: EFFICIENT POWER-LOADING SCHEME FOR OFDM-BASED COGNITIVE RADIO SYSTEMS

Fig. 1.

1861

Resource-allocation example for a four-PUP OFDM-based CR system using Algorithm 1.

perform power reduction, and a new water level is found for


PUP bands 3 and 4. At this point, Algorithm 1 stops since
PUP 4s interference power limit is satisfied, and PUP 3 is
nonactive.

(3)

Substituting sm = sm
obtain
=

Il

mS2 (

into


mMl

Nm )flCR
 ,m +
|S1 |

sm flCR
 ,m = Il , we


mS1

Nm flCR
 ,m

. (19)

C. Suboptimal Algorithm for Subproblem OP3


As mentioned in Section IV-B, subproblem OP3 is formulated for PUP l to ensure that the total interference power it
experiences is below Il . Based on the optimal solution for
OP2, the power allocated to the subchannels in Ml should
be no higher than [1/ Nm ]+ from (14). Although we do
not know the exact value of 1/ , it can be approximated
by , i.e., sm [ Nm ]+ = sm , m Ml . The following
optimization problem is thus formulated for PUP l :
OP3 : max

sm



sm
log2 1 +
Nm
+

(15)

mMl

subject to


sm flCR
 ,m Il

(16)

mM+
l

sm 0,

sm Nm ;

m M+
l

(17)

where is the water level for subchannels belonging to PUPs


whose subchannels have yet to undergo power adjustment,

and M+
l Ml contains the subchannels with Nm < . For
subchannels in Ml \ M+
l , sm is set to 0.
Applying the KKT conditions in OP3, we obtain the following optimal power allocation:
s(3)
m

0,
CR

m S0
Nm , m S1

fl ,m
Nm ,

(18)

m S2

where is the Lagrange multiplier with < Nm flCR


 ,m for
CR
CR
m S0 , Nm flCR
 ,m fl ,m for m S1 , and > fl ,m
for m S2 .

If the mutually exclusive sets S0 , S1 , and S2 are known, the


optimal power allocation can be found using (18). Algorithm 2
is a method for obtaining suboptimal solutions for these sets,
i.e., a suboptimal solution for OP3.
Algorithm 2 Suboptimal Algorithm for Solving OP3.
CR
1) Sort A = {Nm flCR
 ,m , m = 1, 2, . . . , M } and B = {fl ,m ,
m = 1, 2, . . . , M } in descending order. Let I and J denote the
sets of ordered subchannel indexes for A and B, respectively.
Let S0 = , S1 = M+
l , and S2 = , and calculate using (19).
> , do S0 = {S0 , In1 },
2) Let n1 = 1. While NIn1 fICR
n1
S1 = S1 \ {In1 }, = [(|S1 | + 1) NIn1 fICR
]/|S1 |, n1 =
n1
n1 + 1.
CR
3) Let n2 = |M+
l |. While fJn2 < , do the following:
3.1) if not Jn2 S0 , then set S2 = {S2 , Jn2 },

S1 = S1 \ {Jn2 }, and = [(|S1 | + 1) ( NJn2 )fJCR


n2
NJn2 fJCR
]/|S
|.
1
n2
3.2) n2 = n2 1.
4) If S1
= or S2 = , then do Step 6); otherwise, go to
Step 5).

CR
. While I > Il , do m =
5) Let I = mS2 ( Nm )fm
CR
), I =
arg minmS2 log2 (1 + ( Nm )/Nm )/(( Nm )fm
CR

I ( Nm )fm , and S2 = S2 \{m }.
(3)
6) Calculate sm , m M+
l , using (18).
In the first stage of Algorithm 2, we determine S0 . We first
set S0 = , S2 = , S1 = M+
l , and n1 = 1; calculate the initial
value of using (19); and sort Nm flCR
 ,m in decreasing order,
with i representing the sorted subchannel index, i.e., i(1) =
CR
. Note that, in the last expression,
arg maxm{1,2,...,M } Nm fm
CR
we have used fm instead of flCR
 ,m to reduce the notational
burden. This should cause no ambiguity since OP3 applies
CR
, all subchannels satisfy
to a given PUP. If Ni(1) fi(1)
CR
CR
,
Nm fl ,m and S0 = . On the other hand, if < Ni(1) fi(1)

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

1862

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 4, MAY 2010

subchannel i(1) is added to S0 since, from (19), removing i(1)


from S1 results in a lower value of , thus ensuring that <
CR
. We then increase n1 by 1 and compare the value
Ni(1) fi(1)
CR
with the value of , updated using (19), until
of Ni(n1 ) fi(n
1)
CR
Ni(n1 ) fi(n1 ) . Then, S0 = {i(1), i(2), . . . , i(n1 1)}.
In the second stage, we determine S2 . We first sort flCR
 ,m
in decreasing order, with j representing the sorted subchannel
CR
index, and start with S2 = and n2 = |M+
l |. If fj(|M+ |) ,
l

all subchannels satisfy flCR


 ,m and S2 = . On the other
+
CR
<

and
j(|M
hand, if fj(|M
+
l |) is not in S0 , subchannel
|)
l

j(|M+
l |) is added to S2 ; from (19), removing subchannel
j(|M+
l |) from S1 results in a higher value of , thus ensuring
CR
< . We then decrease n2 by 1 and compare
that fj(|M
+
|)
l

CR
the value of fj(n
with the value of , which is updated us2)
+
CR
ing (19), until fj(n
. Then, S2 = {j(|M+
l |), j(|Ml |
2)

+
1), . . . , j(n2 + 1)}, and S1 = Ml \ (S0 S2 ).
Finally, at the end of the second stage, if S1
= or S2 = ,
then S0 , S1 , and S2 are determined; otherwise, the following
adjustment is needed to ensure that (10) is satisfied. In this
(3)
case, sm is either 0 or Nm , and one subchannel at a
time is transferred from S2 to S0 until (10) holds. At each
time, the subchannel with the lowest value of log2 (1 + (
CR
) is selected, because it has the
Nm )/Nm )/(( Nm )fm
lowest reward/cost ratio, where reward is the number of bits
allocated to the subchannel, and cost is the interference power
generated on this subchannel.

Fig. 2. ANB for each PUP band as a function of E{H1,m } with S =


0.024 W, I1 = I2 = 8 1015 W, and E{H2,m } = 1012 .

D. Complexity, Convergence, and Suboptimality


The computational complexity of Algorithm 2 is
+
O(|M+
l | log(|Ml |)), which results from the sorting performed
in Step 1) and the procedures in Step 5). The complexity is
O(1) in Step 4) and O(|M+
l |) in Steps 2), 3), and 6). In
Algorithm 1, the computational complexities in Steps 1)
and 2) are 
O(M log M ) and O(L), respectively. For each
PUP l with mMl sm flCR
 ,m > Il , Steps 3) to 5) are performed
once. 
Thus, altogether, the complexity of Step 3) is at
L
most
l=1 O(|Ml | log(|Ml |)), which is no higher than
O(M log(M )). The complexities of Steps 4) and 5) are O(M )
and O(LM ), respectively. Thus, the overall complexity of
Algorithm 1 is O(M log M ) + O(LM ), which is much lower
than that for the optimal algorithm in Section III and/or that for
the optimal recursive algorithm in [11].
The proposed power allocation algorithm, i.e., Algorithm 1,
converges within L + 1 iterations, because after the initial water
filling in Step 1), the other steps are performed at most L
times. There are two reasons why the proposed Algorithm 1 is
suboptimal: First, 1/ is estimated by a slightly lower value,
i.e., , which is the water level of the subchannels in set N .
Although this estimation results in a lower performance within
active PUP bands, the loss is generally small because the noise
powers for the CRP within these bands are normally higher than
those in the nonactive PUP bands. Second, in the formulation of
subproblem OP3, the power constraint for the CRP is converted
to a power limitation on each subchannel, based on the optimal

Fig. 3. ANB for all PUP bands as a function of S with I1 = I2 = 8


1015 W and E{Hl,m } = 1012 , l = 1, 2.

solution for OP2. Comparing the solution (18) for OP3 with
the optimal solution (12) for OP2, we note that (18) is actually
CR
, m
an approximation of (12). In (12), when l fl,m
CR
can be approximated by 1/(l fl,m ), which is essentially the
CR
, m can
same as the case for m S1 . When l fl,m

be approximated by 1/ , which is the same as the case for


m S2 .
V. S IMULATION R ESULTS
A simulation study was performed for the downlink of a
multiuser OFDM-based CR system covering a 3 km 3 km
area with the CR base station located at the center. There are
50 CR users uniformly distributed within the cell. The propagation path loss is calculated using P L = A + 10 log10 (d/d0 )
[15], where A = 100 dB is the path loss at the reference point,
= 4 is the pathloss exponent, d is the distance between the
CR base station and the CR user, and d0 = 100 m is the
distance between the CR base station and the reference point.
The multipath fading is assumed to be Rayleigh. The parameter
is set to 5, which is obtained using ln(5BER)/(1.5) [12]

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

ZHANG AND LEUNG: EFFICIENT POWER-LOADING SCHEME FOR OFDM-BASED COGNITIVE RADIO SYSTEMS

1863

TABLE I
S TATISTICS FOR ACTUAL I NTERFERENCE P OWER E XCEEDING Il , l = 1, 2 BY U SING SUBOPT-APPROX
W ITH I1 = I2 = 8 1015 W AND E{Hl,m } = 1012 , l = 1, 2

with a bit error rate (BER) of 104 . Two subchannel assignment


strategies were considered: In Case A, each subchannel is assigned to the CR user with the highest gain for that subchannel;
in Case B, each subchannel is assigned to the CR user with
the best channel gain relative to its own mean gain for that
subchannel.
In the simulated system, there are three PUPs, M = 24
subchannels with 02 = 1016 W, and |Ml | = 8, l = 1, 2, 3.
Subchannels 1 and 8, 9 and 16, and 17 and 24 serve as guard
bands for PUP 1, 2, and 3, respectively. All links are assumed
to undergo independent Rayleigh fading. PUP 3 is nonactive,
and the total transmit power for PUP 1 and PUP 2 is 0.8 W
each. Following [6], the PUP signal PSDs are assumed to be
those of elliptically filtered white noise processes [16], i.e.,
lPU (f ) = [1 + 2l Rn2 (l , f /f0,l )]1 , where n, l , l , and f0,l
are the filter parameters, and Rn (., .) is the nth-order elliptic
rational function; the PSD of the CRP signal is CR (f ) =
Ts (sin f Ts /(f Ts ))2 , where Ts = 40 s is the symbol duration. Other parameters are E{Dl,m } = 1014 , and the OFDM
symbol guard interval is 8 s.
To illustrate the benefit of allowing the CRP to share active
PUP frequency bands, we compare the average number of bits
per OFDM symbol (ANB) with that of a system in which active
PUP bands may not be used by the CRP. In this case, only
PUP 3s band is available to the CRP. The optimal solution for
such a system is referred to as nonactive band only (NABO).
The optimal solution for OP1, OP2, and the suboptimal solution provided by Algorithm 1 are referred to as OPT, OPTAPPROX, and SUBOPT-APPROX, respectively. A commercial
optimization software package was used to solve OP1. To
assess if the approach in Section IV is reasonable, the actual
interference generated by the CRP tothe active PUPs, i.e.,
CR
PUP 1 and PUP 2, is calculated using M
m=1 sm fl,m , l = 1, 2
for SUBOPT-APPROX.
The ANB values (obtained by averaging over 10 000 realizations of the fading gains) for each PUP band for Case A
for SUBOPT-APPROX and OPT-APPROX, as a function of
E{H1,m }, are shown in Fig. 2, with S = 0.024 W, I1 = I2 =
8 1015 W, and E{H2,m } = 1012 . As E{H1,m } decreases,
the ANB on PUP 1s band increases, since the CRPs interference power to PUP 1s receiver is reduced. The ANBs on
PUP 2s and PUP 3s bands decrease as E{H1,m } decreases
because more CRP power is being diverted to PUP 1s band.
The ANB difference between OPT-APPROX and SUBOPTAPPROX is less than 2%. Although not shown in Fig. 2, the
corresponding ANB values for NABO for PUP bands 1, 2,
and 3 were found to be {0, 0, 72} bits, and the percentage
improvements of SUBOPT-APPROX over NABO are more

Fig. 4. ANB for all PUP bands as a function of I1 with S = 0.024 W, I1 =


I2 , and E{Hl,m } = 1012 , l = 1, 2.

than 50%, 70%, and 85% for E{H1,m } of 109 , 1011 , and
1013 , respectively.
The sum ANB for all subchannels as a function of the
total CRP power, i.e., S, is shown in Fig. 3, with I1 = I2 =
8 1015 W and E{Hl,m } = 1012 , l = 1, 2 for both Case A
and Case B. As expected, the ANBs for OPT, OPT-APPROX,
SUBOPT-APPROX, and NABO increase with S. The ANB
difference between OPT-APPROX and OPT is less than 0.15%,
and the ANB difference between SUBOPT-APPROX and OPTAPPROX is less than 3% for both cases. The ANB for
SUBOPT-APPROX is over 50% (20%) higher than that of
NABO at S = 0.24 mW for Case A (Case B) and over 55%
(70%) at S = 0.024 W.
To show that with SUBOPT-APPROX, the actual generated
interference powers to the active PUPs, i.e., PUP 1 and PUP 2,
are acceptable, we summarize the following in Table I: 1) the
fraction of realizations for which either PUP 1 or PUP 2
experience interference powers in excess of their thresholds Il ;
2) the average percentage by which the interference threshold
limit is exceeded considering only those realizations in which
the threshold is exceeded; and 3) the percentage by which the
maximum interference power (among the 10 000 realizations)
exceeds the thresholds as a function of S, with I1 = I2 =
8 1015 W and E{Hl,m } = 1012 , l = 1, 2. It can be seen
that the actual maximum interference power generated by the
CRP is less than 1.02Il .
The sum ANB for all subchannels as a function of the
interference threshold, i.e., I1 , is shown in Fig. 4, with I1 = I2 ,
S = 0.024 W, and E{Hl,m } = 1012 , l = 1, 2. The ANBs for

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

1864

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 4, MAY 2010

OPT, SUBOPT-APPROX, and OPT-APPROX increase with I1


as more interference can be tolerated by the active PUPs, i.e.,
PUP 1 and PUP 2. Because NABO only makes use of PUP 3s
band, the ANB for NABO does not change with I1 . The ANB
difference between OPT-APPROX and OPT is less than 0.2%,
and the ANB difference between SUBOPT-APPROX and OPTAPPROX is less than 4%. The ANB for SUBOPT-APPROX
is over 40% (30%) higher than that for NABO at I1th = 8
1016 W and over 85% (65%) when I1th 8 1014 W for
Case A (Case B).
VI. C ONCLUSION
A low-complexity suboptimal solution for the powerallocation problem in an OFDM-based CR system, in which
the CRP uses both nonactive and active PUP bands, has been
proposed. The proposed algorithm has a lower complexity
than those of optimal algorithms. The complexity reduction is
achieved by making a validated approximation based on the
following: 1) the fact that XCI from CRUs to PUPs is mainly
limited to a few subchannels adjacent to the PUP bands and
2) the assumption that the bandwidth for a PUP is typically
much larger than that of an OFDM subchannel and that there
is usually a guard band between two adjacent PUP bands.
Simulation results show that the proposed solution is very close
to optimal and provides significant improvement over systems
that use only nonactive PUP bands.
R EFERENCES
[1] S. Haykin, Cognitive radio: Brain-empowered wireless communications, IEEE J. Sel. Areas Commun., vol. 23, no. 2, pp. 201220,
Feb. 2005.
[2] J. Mitola, III and G. Q. Maguire, Jr., Cognitive radio: Making software
radios more personal, IEEE Pers. Commun., vol. 6, no. 4, pp. 1318,
Aug. 1999.
[3] T. A. Weiss and F. K. Jondral, Spectrum pooling: An innovative strategy for the enhancement of spectrum efficiency, IEEE Commun. Mag.,
vol. 42, no. 3, pp. S8S14, Mar. 2004.
[4] Fed. Commun. Comm., Spectrum Policy Task Force Rep., 2002. FCC
02-135.
[5] Fed. Commun. Comm., In the Matter of ET Docket No. 03-237, May
2007. Order FCC 07-78.
[6] T. Weiss, J. Hillenbrand, A. Krohn, and F. K. Jondral, Mutual interference in OFDM-based spectrum pooling systems, in Proc. IEEE 59th
VTCSpring, Milan, Italy, May 2004, vol. 4, pp. 18731877.
[7] G. Bansal, M. J. Hossain, and V. K. Bhargava, Adaptive power loading
for OFDM-based cognitive radio systems, in Proc. IEEE ICC, Glasgow,
U.K., Jun. 2007, pp. 51375142.
[8] T. Qin and C. Leung, Fair adaptive resource allocation for multiuser
OFDM cognitive radio systems, in Proc. 2nd Int. Conf. CHINACOM,
Shanghai, China, Aug. 2007, pp. 115119.
[9] P. Wang, M. Zhao, L. Xiao, S. Zhou, and J. Wang, Power allocation
in OFDM-based cognitive radio systems, in Proc. IEEE GLOBECOM,
Washington, DC, Nov. 2007, pp. 40614065.

[10] P. Cheng, Z. Zhang, H.-H. Chen, and P. Qiu, Optimal distributed joint
frequency, rate and power allocation in cognitive OFDMA systems, IET
Commun., vol. 2, no. 6, pp. 815826, Jul. 2008.
[11] P. Wang, X. Zhong, L. Xiao, S. Zhou, and J. Wang, A general power
allocation algorithm for OFDM-based cognitive radio systems, in Proc.
IEEE ICC Workshops, Dresden, Germany, Jun. 2009, pp. 15.
[12] A. J. Goldsmith and S.-G. Chua, Variable-rate variable-power MQAM
for fading channels, IEEE Trans. Commun., vol. 45, no. 10, pp. 1218
1230, Oct. 1997.
[13] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[14] R. Gallager, Information Theory and Reliable Communication. New
York: Wiley, 1968.
[15] V. Erceg, L. J. Greenstein, S. Y. Tjandra, S. R. Parkoff, A. Gupta,
B. Kulic, A. A. Julius, and R. Bianchi, An empirically based path loss
model for wireless channels in suburban environments, IEEE J. Sel.
Areas Commun., vol. 7, no. 7, pp. 12051211, Jul. 1999.
[16] R. W. Daniels, Approximation Methods for Electronic Filter Design.
New York: McGraw-Hill, 1974.
[17] A. Leke and J. M. Cioffi, A maximum rate loading algorithm for discrete
multitone modulation systems, in Proc. IEEE GLOBECOM, Phoenix,
AZ, Nov. 1997, vol. 3, pp. 15141518.

Yonghong Zhang (M09) received the B.Eng. degree (with honors) in computer science and engineering from Xian Jiaotong University, Xian, China, in
1994 and the M.A.Sc. and Ph.D. degrees in electrical
and computer engineering from The University of
British Columbia, Vancouver, BC, Canada, in 2006
and 2008, respectively.
She has worked in industry in both Canada and
China as a Software Engineer. She is currently with
Google, Inc., Mountain View, CA. Her research
interests include cross-layer optimization for orthogonal frequency-division multiplexing and cognitive-radio wireless communication systems, resource allocation for wireless communication systems, and
software engineering.

Cyril Leung (M76) received the B.Sc. degree (with


honors) from the Imperial College, University of
London, London, U.K., in 1973 and the M.S. and
Ph.D. degrees in electrical engineering from Stanford University, Stanford, CA, in 1974 and 1976,
respectively.
From 1976 to 1979, he was an Assistant Professor
with the Department of Electrical Engineering and
Computer Science, Massachusetts Institute of Technology, Cambridge. From 1979 to 1980, he was with
the Department of Systems Engineering and Computing Science, Carleton University, Ottawa, ON, Canada. Since July 1980, he
has been with the Department of Electrical and Computer Engineering, The
University of British Columbia, Vancouver, BC, Canada, where he is currently
a Professor and is the holder of the PMC-Sierra Professorship in Networking
and Communications. His research interests include wireless communications
systems.
Dr. Leung is a member of the Association of Professional Engineers and
Geoscientists of British Columbia.

Authorized licensed use limited to: Soongsil University. Downloaded on July 03,2010 at 07:33:14 UTC from IEEE Xplore. Restrictions apply.

You might also like