Vous êtes sur la page 1sur 6

A Joint Resource Allocation Scheme for OFDMA-

based Wireless Networks with Carrier Aggregation


Fan Wu, Yuming Mao, Xiaoyan Huang, Supeng Leng
School of Communication and Information Engineering
University of Electronic Science and Technology of China
Chengdu, China
Email: {wufan, ymmao, xyhuang, spleng }@uestc.edu.cn

AbstractThe mixture of users with different carrier aggregation
(CA) capabilities presents new challenges to optimize the
performance of the next generation wireless networks. This
paper focuses on the joint resources allocation for OFDMA-based
multi-carrier system. Distinguished from many existing methods,
our approach deploys the joint dynamic spectrum resources
assignment and adaptive power allocation technologies for the
carrier-aggregated systems. A low complexity suboptimal
algorithm, named as the joint CC, RB and power allocation
(JCRPA) algorithm, is proposed in this paper. The algorithm
combines the dynamic component carrier (CC) and resource
block (RB) assignment, as well as adaptive power allocation
iteratively. In contrast to the conventional static CC assignment,
a novel suboptimal dynamic CC and RB assignment algorithm is
designed to maximize network utility. Simulation results
demonstrate that JCRPA is able to improve the system
performance in terms of network utility, average throughput and
fairness.
Keywords-carrier aggregation; network utility; dynamic
spectrum resources assignment; adaptive power allocation
I. INTRODUCTION
The integration and interoperability of mobile
communication technologies, along with new broadband
wireless innovations and intelligent user-oriented services will
lead toward the next-generation wireless networks, which can
provide significantly higher data transmission rate than the
current wireless systems. For instance, one of the targets of
LTE-Advanced is to offer the peak data rates of 1 Gbps for low
mobility and. 100 Mbps for high mobility [1]. With the
aggregation of multiple continuous or non-continuous
component carriers (CCs), carrier aggregation (CA) technology
introduced by 3GPP is one of the efficient methods for data
rate enhancement in wireless networks.
In the multi-carrier system, the base stations are required to
be capable of serving the users on multiple CCs. However, the
users can receive or transmit on one or multiple CCs depending
on their carrier aggregation capabilities. The users in such a
system can be classified to narrowband users and broadband
users. The narrowband users equipped with old or cheap
hardware can work on only one CC, whereas the broadband
users with high performance transceivers can operate on all
CCs.
The mixture of users with different carrier aggregation
capabilities presents new challenges to optimize the network
performance, especially for the OFDMA-based systems. In an
OFDMA-based system, efficient resource allocation becomes a
very important issue. It is difficult to optimize subcarrier
assignment and power allocation simultaneously [2]. This
difficult task can be alleviated by fixing one resource and
controlling the other [3]. In the single-carrier systems, the
optimal solution to each sub-problem can be achieved by some
classic algorithms, for example, the Max-Sum-Capacity (MSC)
rule, the proportional fair (PF) scheduling for subcarrier
assignment problem [4], and the multilevel water-filling
algorithm for power allocation problem [2].
However, in the multi-carrier systems, the resource
allocation problem becomes intractable even if the allocation of
other resources is determined. For instance, given the power
allocation, the subcarrier assignment is still coupled with the
CC assignment, subject to the constraints on the carrier
aggregation capabilities of different users. Resulting from this
case, the resource management of multiple CCs has to be
redesigned to facilitate the implementation of CA.
Recently, there have been a few studies on resource
allocation with CA [5][6][7][8][9]. Most of these schemes
adopt the equal power allocation (EPA) strategy and
decompose the assignment problem into two separate sub-
problems, i.e., CC scheduling and resource block (RB)
assignment on each carrier. First, the base station allocates the
users on proper CCs. The balance of the load across CCs is the
main concern of CC scheduling schemes. The common balance
methods include RR Balancing and Random Carrier. Once the
users are assigned onto certain CC(s), the assignment of RBs in
each CC is carried out. There are two kinds of scheduling
schemes to allocate RBs [5]. One is Independent scheduling
(IC) for each CC, and the other is across CC scheduling (AC).
The IC method is simply derived from the conventional single
carrier scheduling algorithms by performing resource
allocation separately in each CC. The AC approach combines
all CCs assigned to the same user together as one carrier to
allocate RBs [9].
Nevertheless, all of the above approaches are developed on
the 2-step spectrum resource assignment structure, and they
ignore the joint allocation of the power and spectrum resources.
In fact, similar to the single carrier system, the OFDMA-based
multi-carrier systems potentially have high flexibility for radio
resource management [6]. In this paper, we investigate the joint
dynamic CC and RB assignment together with adaptive power
allocation for OFDMA-based multi-carrier system, such that
Corresponding author: Supeng Leng, Email: spleng@uestc.edu.cn
This work is partly supported by the National S&T Major Project of China
under Grant No.2010ZX03005-001, the National Natural Science
Foundation of China under Grant No. 60832007, and the Program for New
Century Excellent Talents in University (NCET-10-0294), China.
2012 IEEE Wireless Communications and Networking Conference: MAC and Cross-Layer Design
978-1-4673-0437-5/12/$31.00 2012 IEEE 1299
improves the overall performance of the system by exploiting
both frequency diversity and multi-user diversity
simultaneously. The following salient features of our work
make it different from the existing schemes.
1) We formulate the design of joint resource management
for the multi-carrier system as a mixed integer non-linear
programming (MINLP) with the objective of the network
utility maximization (NUM), subject to the different carrier
aggregation capabilities of users. By applying the gradient-
based theory, the NUM problem is transformed into the
weighted sum throughput maximization (WSTM) problem in
which the weights are the marginal utility of each user until its
scheduling time.
2) By introducing the concept of System Utility Loss, a
dynamic CC and RB assignment algorithm, i.e., Minimizing
System Utility Loss (MSUL), is proposed for the joint
allocation of CC and RB with the carrier cabability constraints
of different users, so as to approach the optimal system utility.
3) To decrease the high computational complexity in
solving the WSTM problem, we propose a low complexity
suboptimal algorithm, referred to as joint CC, RB and power
allocation (JCRPA) algorithm, which combines dynamic CC
and RB assignment, and adaptive power control. With the
iterative update of marginal utility, both frequency diversity
and multi-user diversity of the system can be fully exploited.
The rest of the paper is organized as follows. Section II
describes the system model. The maximum network utility
model for multi-carrier system is proposed in section III. The
joint resource allocation algorithm is elaborated in section IV.
Section V provides the performance evaluation of the proposed
algorithms via simulations. Finally, Section VI concludes this
paper.
II. SYSTEM MODEL
Consider a multi-carrier system with K users. Among the K
users, K
1
narrowband (NB) users support transmission on only
one CC, whereas the other K
2

broadband (BB) users can be
assigned on all CCs, where K
1
+K
2
=K. Let K
1
and K
2
be the set
of NB users and BB users, respectively.
It is assumed that there are M CCs in the system, denoted
by the set M. The M CCs can be aggregated to form the wide
bandwidth. The set N(m) denotes the resource blocks (RBs) in
the m
th
CC, and the cardinality of each set is N. One RB is the
minimum resource element which can be allocated to a user.
For instance, a RB has 12 consecutive sub-carriers to constitute
an equivalent bandwidth of 180 kHz in LTE-Advanced system.
The time axis is divided into slots. For each time slot, at most
one user can be served over each RB.
Let
,
( )
m
k n
r t denote the instantaneous rate of the k
th
user over
the n
th
RB in the m
th
CC during the t
th
time slot. According to
Shannon capacity,
,
( )
m
k n
r t can be expressed as
, 2 , ,
( ) log (1 ( ) ( ))
m m m
k n k n k n
r t B p t t = + (1)
where B is the bandwidth of each RB, and
,
( )
m
k n
p t denotes the
power allocated to the k
th
user over the n
th
RB in the m
th
CC,
and
,
( )
m
k n
t denotes the received signal-to-noise ratio per unit
power in the t
th
time slot.
,
( )
m
k n
t is defined as follow:
2
, , , 0
( )
m
k n k m n
g t N B (2)
where
, ,
( )
k m n
g t denotes the channel gain, and
0
N B is the
average power of the additive white Gaussian noise on each
RB.
Let ( )
k
r t be the average throughput the k
th
user up to the t
th

time slot. By using an exponentially weighted low-pass time
window [2], ( )
k
r t can be expressed as
( ) (1 1 ) ( 1) 1 ( )
k k k
r t T r t T r t = + (3)
where T is the length of the sliding average window for user
throughput measurement, and ( )
k
r t represents the instantaneous
data rate of the k
th
user at the t
th
time slot. It is shown in [10]
that with sufficient large value of t, ( )
k
r t will weakly converge
to the stationary throughput of user k. In the multi-carrier
system, the instantaneous data rate ( )
k
r t is defined as:
, ,
1 1
( ) ( ) ( ) ( )
M N
m m m
k k n k k n
m n
r t r t t t o |
= =
=

(4)
where ( )
m
k
t o and
,
( )
m
k n
t | are CC and RB allocation indicator,
respectively. ( ) 1
m
k
t o = indicates that the m
th
CC is assigned to
the k
th
user at the t
th
time slot, and ( ) 0
m
k
t o = otherwise. For NB
users, only one CC can be assigned in one slot, but the assigned
CC can be changed over the different slots. For BB users, the
number of the assigned CCs for each BB user must be no more
than the total number of the available CCs in the system.
Consequently, the CC allocation should satisfy the following
conditions.
1
1
( ) 1,
M
m
k
m
t k o
=
s e

K (5)
2
1
( ) ,
M
m
k
m
t M k o
=
s e

K (6)
,
( )
m
k n
t | is the RB allocation indicator such that
,
( ) 1
m
k n
t | = if the
n
th
RB in the m
th
CC is allocated to the k
th
user at the t
th
time
slot and
,
( ) 0
m
k n
t | = otherwise. In order to avoid co-channel
interference, each RB should be assigned to at most one user in
each time slot. Thus,
,
( )
( ) 1, , ( )
m
k n
k m
t m n m |
e
s e e

K
M N (7)
where K(m) is the feasible set of users for the m
th
CC. Due to
the various carrier aggregation capabilities of users, the feasible
sets of users for different CCs are possibly distinct from each
other.
III. PROBLEM FORMULATION
Define the network utility as
1
( )
K
k k
k
U U r
=

where
( )
k
U is the utility function of the k
th
user with respect to the
long-term throughput
k
r . According to [11], optimizing the
NUM problem is equivalent to maximizing
1300
( ) ( )
1 1
( ) ( ) ( 1)
K K
k k k k
k k
U t U r t U r t
= =
A

for each time slot
t. Using one-order Taylor formula, it follows that
( ) ( )
1
( ) ( 1) ( 1) ( ) ( 1)
K
k k k k k
k
U t dU r t d r t r t r t
=
A ~

(8)
Substituting (3) and (4) into (8) , we can get:
( ) ( )
( ) ( )
1
, ,
1 1 1
1
( ) ( 1) ( 1) ( ) ( 1)
1
( 1) ( 1) ( ) ( ) ( ) ( 1)
K
k k k k k
k
K M N
m m m
k k k k n k k n k
k m n
U t dU r t dr t r t r t
T
dU r t dr t r t t t r t
T
o |
=
= = =
A ~
=


(9)
Since ( 1)
k
r t is fixed at a given time slot, the network
utility maximization problem can be formulated as a weighted
sum throughput maximization (WSTM) problem with the
objective expressed by:
,
,
, ,
1 1 1
{ }
{ }{ }
max ( ) ( ) ( ) ( )
m
k n
m m
k k n
K M N
m m m
k k n k k n
k m n
p
w t r t t t
o |
o |
= = =


(10)
where ( )
k
w t =
( )
( 1) ( 1) k k
k
dU r t dr t can be regarded as
the priority of the k
th
. Since scheduling is taking place at the
beginning of each slot, we can omit the time index t for brevity.
Then the joint CC, RB and power allocation problem can
be formulated as a constrained WSTM optimization problem,
denoted by P1.
( )
,
,
, 2 , ,
1 1 1
{ }
{ }{ }
, , max
1 1 1
1
1
2
1
,
P1: max log 1
. .
0, , ( )
1,
,

m
k n
m m
k k n
K M N
m m m m
k k k n k n k n
k m n
p
K M N
m m m
k n k k n
k m n
m
n
M
m
k
m
M
m
k
m
k
w p
s t p P
p m M n N m
k
M k
o |
o |
o |
o
o
|
= = =
= = =
=
=
+
s
> e e
s e
s e

K
K
{ }
{ }
( )
,
1, , ( )
0,1 , ,
0,1 , , , ( )
m
n
k m
m
k
m
k n
m n m
k m
k m n m
o
|
e
s e e
e e e
e e e e

K
M N
K M
K M N
(11)
P1 is a mixed integer non-linear programming (MINLP)
problem. The optimal solution of P1 is practically impossible
due to its complexity. In literatures [3][11], the decomposing
method is used to solve the MINLP problem in a suboptimal
way, where the original optimization problem is decomposed
into tractable sub-problems. The joint subcarrier and power
allocation for the single-carrier network is a MINLP problem
as well, and it can be decomposed into subcarrier assignment
and power allocation sub-problems. Once the power allocation
for each subcarrier is determined, the optimal subcarrier
assignment rule can be used to select the user for achieving the
maximum weighted throughput over the subcarrier. On the
other hand, upon any given subcarrier assignment, the optimal
power allocation can be obtained by the multilevel water-
falling algorithm.
However, it is more challenging in the multi-carrier
systems even if the decomposing method is applied. In P1,
given the power allocation, the dynamic CC and RB
assignment problem is a non-linear integer programming
problem, denoted by P2, which is still difficult to solve when
the value of K, M and N are non-trivial.
( )
{ }
{ }
,
, 2 ,
1 1 1
{ }{ }
1
1
2
1
,
( )
,
P2: max log 1
. . 1,
,
1, , ( )
0,1 , ,
0,1 , ,
m m
k k n
K M N
m m m m
k k k n n k n
k m n
M
m
k
m
M
m
k
m
m
k n
k m
m
k
m
k n
w p
s t k
M k
m n m
k m
k m
o |
o |
o
o
|
o
|
= = =
=
=
e
+
s e
s e
s e e
e e e
e e e

K
K
K
M N
K M
K , ( ) n m e M N
(12)
Consequently, the objective of our work is to propose an
iterative algorithm to obtain the near-optimal solution to P1
efficiently.
IV. JOINT RESOURCE ALLOCATION ALGORITHM WITH CA
Although P2 can be solved by decoupling CC allocation
and RB assignment, it renders a performance degradation of
the system due to lacking of multiuser diversity exploitation.
Therefore, performing CC and RB assignment dynamically is
more constructive to improve the system performance. On the
other hand, most of the existing the works solve the sub-
problems sequentially only once. More improvement is
potentially achieved by refining the resource allocation
iteratively. To address these issues, we propose a suboptimal
iterative carrier aggregation based resource allocation
algorithm, referred to as joint CC, RB and power allocation
(JCRPA) algorithm. In JCRPA, each iteration involves three
steps. First, the Minimizing System Utility Loss (MSUL)
algorithm is devised to allocate CC and RB dynamically with
fixed power allocation. Second, a multilevel water-filling
power allocation algorithm is developed to optimize the power
allocation for the given CC and RB assignment. Finally, update
the marginal network utility.
A. Dynamic CC and RB Assignment Algorithm
To compromise the performance optimality and
computational complexity, we propose a suboptimal algorithm,
i.e., MSUL, for the dynamic assignment of CC and RB
according to the weights of users at each slot. In this algorithm,
a relaxed optimization problem is solved to obtain the upper
bound of the optimum of P2. Then, based on the optimal
solutions of the relaxed optimization problem, an iterative
adjustment algorithm is performed, until all the NB users have
met the constraint of carrier aggregation capability. In each
adjustment iteration, if a NB user is selected, it should adopt
one CC for the maximum utility and releases the RBs on the
other CCs. However, the released RBs will be reassigned
among the other feasible users, which may decrease the utility
of these RBs. The decreasing utility is called as the loss of
system utility. The MSUL algorithm aims at assigning RBs in
accordance with each users carrier aggregation capability and
meanwhile minimizing the loss of system utility caused by the
adjustment process.
It is supposed that the NB users can utilize all the CCs as
well as the BB users. Then, ( ) ,
B
m m = e K K M , with
1301
1 2 B
= K K K . Accordingly, P2 can be reduced to the following
relaxed optimization problem, denoted by P3.
( )
{ }
,
2 , ,
1 1 1
{ }
,
( )
,
P3: max log 1
. . 1, , ( )
0,1 , , , ( )
m
k n
K M N
m m m
k n k n k n
k m n
m
k n
k m
m
k n
w p
s t m n m
k m n m
|
|
|
|
= = =
e
+
s e e
e e e e

K
M N
K M N
(13)
Furthermore, P3 can be divided into M per-CC sub-
problems, that is
( )
{ }
,
2 , ,
1 1
{ }
,
( )
,
P4: max log 1 , 1, 2,...,
. . 1, ( )
0,1 , , ( )
m
k n
K N
m m m
k n k n k n
k n
m
k n
k m
m
k n
w p m M
s t n m
k n m
|
|
|
|
= =
e
+ =
s e
e e e

K
N
K N
(14)
The relaxed optimization problem (P3 and P4) that allows
all users to access all CCs, is the same as optimization in a
conventional single-CC system. The optimal solution of P4 can
be achieved by the following assignment rule. For the m
th
CC,
the n
th
RB is assign to the user k
*
which achieves the maximum
weighted throughput on this resource block, that is
( )
2 ,
( )
( , ) arg max log 1
m m
k n k n
k m
k m n w p
e
= +
K
(15)
It is worth mentioning that the optimum of P4 is the upper
bound of that of P2, since the constraint that each NB user can
be assigned on only one CC is relaxed.
Based on the optimal solutions of the relaxed optimization
problem, iterative adjustment process is performed with the
constraints on the carrier aggregation capabilities of NB users.
The iterative process runs at most K
1
turns. In every iteration,
among the remaining unqualified NB users that cannot satisfy
the constraint of the carrier aggregation capability, the user k
A
with the minimum loss of the system utility
min
k
U A is selected
to adjust its resource from the unqualified users set K
A
. This
procedure will iterate until all the NB users meet the
constraints of the carrier aggregation capability. We can get
min
arg min
A
A k
k
k U
e
= A
K
(16)
where
min
,
( )
min
k k m
m k
U U
e
A A
M
(17)
and
( )
( ) ' '
'
, 2 ,
( )\ ( , )
2
, ( )\ ( , )
( )\
log 1
max log 1
l l
k m k n k n
l k m n k l
l l
n
k k n l k m n k l
k l k
U w p
w p

e e
e e
e
A +
+


M R
M R
K

(18)
In (18), M(k) is the set of CCs assigned to the k
th
user, and
R(k,m) denotes the set of RBs in the m
th
CC allocated to the k
th

user.
The
, k m
U A represents the loss of the system utility, i.e.,
weighted sum throughput, when the k
th
NB user chooses the m
th

CC as its working spectrum whereas reassigns the RBs
in ( , ), ( ) \ k l l k m e R M to other feasible users. The RBs
reallocation decision is made
( )
( ) '
2
,
( )\
( , ) arg max log 1 , ( ) \ , ( , )
A
l l
k n A A A
k n
k l k
k l n w p l k m k n k l
'
'e
' = + e e
K
M R
(19)
On the other hand, among the CCs in M(k
A
), the CC m(k
A
)
with minimum
,
A
k m
U A is selected as the working spectrum of
the
th
A
k NB user, thus
,
( )
( ) arg min
A
A
A k m
m k
m k U
e
= A
M
(20)
In summary, the steps of MSUL are described in TABLE I.
TABLE I. MSUL ALGORITHM
1. Initialization:
1 2 B
= K K K and ( ) ,
B
m m = e K K M;
2. Solve the P4 according to (15) to obtain
*
, k m
o and
*
, , k m n
| ;
* *
( , ), ,
1, 0, ( , )
m m
k m n n k n
k k m n | | = = =
* *
( , )
1, 0, ( , )
m m
k m n k
k k m n o o = = =
3. Update the following sets based on
*
, k m
o and
*
, , k m n
| ;
{ }
*
( ) 1,
m
k
k m m o = = e M M
{ }
*
,
( , ) 1, ( )
m
k n
k m n n m | = = e R N
{ }
*
1 2
1
1,
M
m
B k
m
k k o
=
= > e

K K K
4. Update the set of NB users requiring resource adjustment
{ }
1
|
A B
k ke K K K , and if
A
= u K , then go to Step 5;
otherwise, the algorithm exits;
5. Based on (16), (17) and (18), select a NB user
A
k for
resource adjustment, and choose its working CC ( ),
A
m k
according to (20);
6. According to (19), reallocate the resource released by the
selected NB user
A
k to other users and update
*
, k m
o and
*
, , k m n
| ;
'
'
*
' ,
1 ( , )
( ) \ ( ), ( , )
0 ( , )
l
A A A
k n
k k l n
l k m k n k l
k k l n
|
=
= e e

M R
'
'
*
'
1 ( , )
( ) \ ( ), ( , )
0 ( , )
l
A A A
k
k k l n
l k m k n k l
k k l n
o
=
= e e

M R

7. Update the feasible set of users for CCs, such that
{ } ( ) ( ) \ , ( )
A A
m m k m m k = = K K , and go to Step 3.
B. Adaptive Power Allocation Algorithm
When the CC and RB assignment is given, the adaptive
power allocation is formulated as
( )
( , ) 2 ( , ),
1 1
{ }
max
1 1
P5: max log 1
. .
0, , ( )
m
n
M N
m m
k m n n k m n n
m n
p
M N
m
n
m n
m
n
w p
s t p P
p m M n m

= =
= =
+
s
> e e


N
(21)
where ( , ) k m n denotes the index of the user to which the n
th
RB
of the m
th
CC is assigned.
P5 is a convex programming and has a unique global
optimum which can be obtained in the dual space. Using the
Lagrangian method, the sum power constraint can be relaxed
by introducing a dual price . In the dual space, the primal
problem P5 is separated into MN sub-problems, for each RB.
Then we have
1302
( )
( , ) 2 ( , ),
max log 1
. . 0
m
n
m m m
k m n n k m n n n
p
m
n
w p p
s t p
+
>
(22)
Each problem in (22) is locally convex and has a global
optimum. The optimal power allocation is given by
*
( , ) ( , ),
1 , 0
m m
n k m n k m n n
p w
+
( = >

(23)
where | | ( ) max 0, x x
+
.
According to the subgradient method, the dual price is
updated by
( )
*
max
1 1
( 1) ( ) ( )
M N
m
n
m n
l l s l P p
+
= =
(
+ =
(


(24)
where l is the iteration number, s(l) is a sequence of scalar step
sizes that is chosen to be sufficiently small to guarantee the
convergence. The adaptive power allocation algorithm can be
summarized as TABEL II.
TABLE II. ADAPTIVE POWER ALLOCATION ALGORITHM
1. Initialization:
( )
max
m
n
p P M N =

( , ),
m m
n k m n n
= ,
( , )
, ,
m
n k m n
w w m n =
(0) 0 > , set l
max
and l=0
2. repeat
1) Calculate the optimal power allocation with (23) for
each RB;
2) Update the dual price according to (24)
3) Set 1 l l = +
until dual price converges or
max
l l >
C. JCRPA Algorithm
To decrease the loss of network utility incurred by the
decomposing method, we can iteratively perform the joint CC
and RB assignment, power control, and the update of marginal
utility, until the network utility gain vanishes. The iterative
JCRPA algorithm is given in TABLE III
TABLE III. JCRPA ALGORITHM
1. Initialization:
max
m
n
p P M N =
2. repeat
1) Obtain the RB and CC assignment by performing the
MSUL algorithm;
2) Obtain the power allocation by performing the
adaptive power allocation algorithm;
3) Update the marginal network utility;
until
( )
1 1
( 1) ( )
K K
k k k k
k k
wr l wr l c
= =
+ s


V. PERFORMANCE EVALUATION
Simulation experiments are conducted to verify the
performance of the proposed JCRPA algorithm. We consider a
single cell network with a radius of 1 km, where users are
uniformly distributed. Both NB and BB users are full-buffered.
There are 4 CCs to be aggregated at the BS, and each CC
contains 64 RBs. The bandwidth of each CC is set to be
25MHz. The model of the wireless channel includes the
distance-dependent path loss, shadowing fading and the small
scale fading. The simulation parameters are listed in TABLE
IV. The performance of the JCRPA algorithms is evaluated in
terms of the convergence behavior, the network utility, the
average throughput of users, and the fairness of users.
TABLE IV. SIMULATION PARAMETERS
Parameter Setting / description
CC aggregation pattern 4 CCs with 25MHz per CC
Number of RBs per CC 64
BS maximum transmit power (P
max
) 1 W
Path loss model Okumura-Hata model
Shadowing standard deviation 7 dB
Thermal noise spectral density -174 dB/Hz
Small scale fading distribution Rayleigh distribution
In the simulations, the utility function of the k
th
user is
defined by
( ) ( ) 2
( ) log ( )
k k k
U r t r t =
(25)
It is well-known that if the network utility is defined as the sum
of the logarithmic user throughput, then the PF throughput
allocation achieves the maximum network utility [12].
According to (25), the weight of the k
th
user at the t
th

iteration can be formulated as
( ) ( ( 1)) ( 1) 1 ( 1)
k k k k k
w t dU r t dr t r t = = (26)
By substituting (26) into (15), the RB allocation rule can be
rewritten by
*
,
( )
( , ) arg max ( ) ( 1)
m
k n k
k m
k m n r t r t
e
=
K
(27)
Equation (27) is the classic PF criterion proposed by Kelly [12]
to balance the efficiency and fairness of resource allocation.
The simulations are developed to compare the performance
of four algorithms, i.e., the JCRPA algorithm, the MSUL
algorithm with equal power allocation (MSUL+EPA), the
cross-CC-PF algorithm with equal power allocation (cross-CC-
PF+EPA), and the cross-CC-PF algorithm with adaptive power
allocation (cross-CC-PF+APA). The cross-CC-PF algorithm
[5][9] has been proved as the optimal static CC scheduler to the
NUM with the logarithmic utility function. This algorithm first
evenly distributes the NB users among all CCs, and then
allocates RBs according to the cross PF criterion.
Fig. 1 shows the convergence behavior of the proposed
JCRPA algorithm. It is clear that the proposed JCRPA
algorithm can quickly converge to the final weighted sum
throughput result without much fluctuation.
0 1 2 3 4 5
0
10
20
30
40
50
60
70
80
90
100
The number of iteration
W
e
ig
h
te
d
s
u
m
th
r
o
u
g
h
p
u
t o
f s
y
s
te
m

Figure 1. Typical convergence behavior of the JCRPA algorithm
1303
Fig. 2 illustrates the network utility versus the number of
NB users using the four algorithms. During the simulation,
with 20 users, the ratio of the NB users varies from 0 to 100%.
Its obvious that the proposed algorithms are always more
efficient than the other two algorithms. Especially, the JCRPA
algorithm achieves the highest network utility among the four
algorithms. This is because that the dynamic assignment of
CCs and RBs can exploit the multi-user diversity more
effectively than the separate assignment of them using the
cross-CC-PF algorithm. Moreover, using adaptive power
allocation, more transmission power can be allocated to the
RBs with better conditions so as to improve overall throughput.
Furthermore, Fig.2 indicates that the network utility suffers
from the increasing ratio of NB users. This is due to the fact
that NB users can operate only on one CC, and thus less
multiuser diversity can be exploited.
0 10 20 30 40 50 60 70 80 90 100
390
395
400
405
410
415
420
425
430
Ratio of NB users (%)
N
e
t
w
o
r
k

u
t
ilit
y

(
S
u
m

o
f

lo
g
(
a
v
e
r
a
g
e

t
h
r
o
u
g
h
p
u
t
)

)

JCRPA
MSUL+EPA
across-CC-PF+APA
across-CC-PF+EPA

Figure 2. Network utility with the different ratios of NB users
Fig. 3 shows the average throughput of each user in the four
algorithms, with the equal number of NB and BB users. In this
figure, the first ten users are NB users, and the others are BB
users. Compared with the other three algorithms, it is clear that
the JCRPA algorithm achieves significant throughput gains for
all kinds of users, especially for the NB users. The reason is
that dynamic CC and RB assignment always selects better RBs
from the whole available CCs for each NB user at each
scheduling slot. Although the selected RBs set for each NB
user probably does not coincide with its carrier capability, the
proposed adjustment procedure is performed to guarantee the
carrier capability of each NB user with the least utility loss. In
contrast, the NB user in the cross-CC-PF algorithm suffers
from the predetermined working CC that cannot be changed
even if the channel condition of the user becomes poor.
0 2 4 6 8 10 12 14 16 18 20
0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
User ID (NB user: 1-10, BB user: 11-20)
A
v
e
r
a
g
e
th
r
o
u
g
h
p
u
t (
M
b
it/s
)


JCRPA
MSUL+EPA
across-CC-PF+APA
cross-CC-PF+EPA

Figure 3. Average throughput performance
Furthermore, the network throughput is defined as the total
throughput of simultaneous transmissions in a network. Based
on Fig. 3, we can easily calculate the average network
throughput under the four algorithms, i.e., the average network
throughputs of JCRPA, MSUL+EPA, cross-CC-PF+EPA and
cross-CC-PF+APA are 45.2 Mbps, 40 Mbps, 35.78 Mbps, and
35.34 Mbps, respectively. In particular, the JCRPA algorithm
achieves approximately 26.3% gain than the cross-CC-
PF+EPA algorithm. It is clear that the JCRPA algorithm is
superior to the other three algorithms because of its joint
resource allocation strategy.
VI. CONCLUSION
In this paper, we investigate the joint resource allocation for
OFDMA-based wireless systems with carrier aggregation,
which is formulated as a WSTM problem. Due to combining
the dynamic CC and RB assignment and adaptive power
allocation iteratively, the proposed JCRPA algorithm can fully
exploiting the frequency and multi-user diversity of system.
Numerical results indicate that the JCRPA algorithm can
effectively address the WSTM problem in a multi-carrier
system, so that the system performance can be significantly
improved in terms of network utility, average throughput and
fairness. Practically, the handover from one CC to another may
involve extra signaling and control panel complexities. How to
achieve efficient resource utilization with acceptable
complexity is an interesting topic for future research.
REFERENCES
[1] ITU-R, Framework and overall objectives of the future development of
IMT-2000 and systems beyond IMT-2000, M. 1645, Jun 2003.
[2] Guocong Song, Geoffrey Ye Li, Cross-Layer Optimization for OFDM
Wireless NetworksPart II: Algorithm Development, IEEE
Trans.Wireless Commun., VOL. 4, pp. 625-634, March 2005.
[3] Hyang-Won Lee, Song Chong, Downlink Resource Allocation in
Multi-Carrier Systems: Frequency-Selective vs. Equal Power
Allocation, IEEE Trans.Wireless Commun., VOL. 7, pp. 3738-3747,
Octobor 2008.
[4] Guocong Song, Ye (Geoffrey) Li, and Leonard J. Cimini, Jr., Joint
Channel- and Queue-Aware Scheduling for Multiuser Diversity in
Wireless OFDMA Networks, IEEE Trans. Comm., VOL. 57, pp. 2109-
2121, July 2009.
[5] Y. Wang, K. Pedersen, T. Srensen, and P. Mogensen, Carrier load
balancing and packet scheduling for multi-carrier systems, IEEE
Trans.Wireless Commun., vol. 9, pp. 17801789, May 2010.
[6] L. Zhang, K. Zheng, W. Wang, L. Huang, Performance analysis on
carrier scheduling schemes in the long-term evolution-advanced system
with carrier aggregation, IET Commun., Vol. 5, pp. 612-619, 2011.
[7] Liu Liu, Mingju Li, Juejia Zhou, Xiaoming She, et. al., Component
Carrier Management for Carrier Aggregation in LTE-Advanced System,
Proc. IEEE VTC Spring 2011.
[8] Yao-Liang Chung, Lih-Jong Jang, Zsehong Tsai, An Efficient
Downlink Packet Scheduling Algorithm in LTE-Advanced Systems with
Carrier Aggregation, IEEE Consumer Communications and
Networking Conference (CCNC2011), 2011, pp. 632-636.
[9] Yuanye Wang, Klaus I. Pedersen, Troels B. Srensen, Preben E.
Mogensen, Utility Maximization in LTE-Advanced Systems with
Carrier Aggregation, Proc. IEEE VTC Spring 2011.
[10] A. Jalali, R. Padovani, and R. Pankaj, Data throughput of CDMA-HDR
a high efficiency-high data rate personal communication wireless
system, Proc. IEEE VTC, May 2000, pp. 18541858.
[11] A. L. Stolyar, On the asymptotic optimality of the gradient scheduling
algorithm for multiuser throughput allocation, Operations Research, vol.
53, no. 1, pp. 1225, Jan. 2005.
[12] F. Kelly, Charging and rate control for elastic traffic, European
Transactions on Telecommunications, vol. 8, pp. 3337, 1997.
1304

Vous aimerez peut-être aussi