Sunteți pe pagina 1din 5

Physical Communication 25 (2017) 293–297

Contents lists available at ScienceDirect

Physical Communication
journal homepage: www.elsevier.com/locate/phycom

Full length article

Joint optimization of component carrier selection and resource


allocation in 5G carrier aggregation system
Weidong Gao ∗ , Lin Ma, Gang Chuai
Beijing University of Posts and Telecommunications, Beijing 100876, China

article info abstract


Article history: In this paper, we consider joint optimization of Component Carrier (CC) selection and resource allocation
Received 2 November 2016 in 5G Carrier Aggregation (CA) system. Firstly, the upper-bound system throughput with determined
Received in revised form number of CCs is derived and it is proved by using graph theory that the throughput optimization
16 December 2016
problem is NP hard. Then we propose a greedy based algorithm to solve this problem and prove that
Accepted 22 December 2016
Available online 24 December 2016
the proposed algorithm can achieve at least 1/2 of the optimal performance in the worst case. At last, we
evaluate the throughput and computational complexity performance through a variety of simulations.
Keywords:
Simulation results show that the proposed algorithm can obtain better performance comparing with
Carrier aggregation existing schemes while keeping the computation complexity at an acceptable level.
Component carrier selection © 2017 Elsevier B.V. All rights reserved.
Resource allocation
Joint optimization

1. Introduction that should be carefully designed, which further increases the


complexity of radio resource management. With respect to CC
With the rapid growth of wireless users and the occurrence selection, most existing schemes are either load-based or channel-
of various high-speed multimedia applications, it is urgent for quality-based [3]. Load-based CC selection is fairly simple, but
the wireless network operators to improve their system capacity it is common understanding that this kind of schemes lack full
as well as to boost single-user peak data rate. To meet such utilization of wireless channel characteristics and only lead to sub-
requirements, 3GPP has launched the research of LTE-Advanced optimal system performance. Though channel-quality-based CC
(LTE-A) and one of the most important features of LTE-A system selection strategy can ensure that all users are assigned to the
is Carrier Aggregation (CA) [1]. With CA, multiple spectrum CCs with high quality channels, there is probability that a large
fragments, called Component Carriers (CCs), can be aggregated number of users are assigned to the same CC simultaneously, which
together into a wider spectrum, which can substantially improve may greatly reduce the system spectrum utilization efficiency. In
single user’s throughput. Though Carrier Aggregation is not a new
view of the above mentioned facts, the authors in [4] introduced
technology, there is probability that it will also play an important
a new metric for evaluating a CC: η = ηRSRP /ηload , where the
role in future 5G networks.
CC with the larger η value will be assigned to the user with
Though CA technology has significantly improved the system
higher priority. Though this scheme takes into account both of
performance, it also has increased the complexity of resource
Quality of Service (QoS) and fairness, it has the shortcoming
scheduling for the network. In addition, due to that multiple
component carriers are used by the UE simultaneously, more that the scheduled users may be assigned to the CC with poor
transmission power will be consumed by the eNodeB as well as by channel quality due to light load reasons, thus reducing the system
the UE [2]. Therefore, energy saving problem cannot be ignored for throughput. The authors in [5] studied CC selection method in
Carrier Aggregation and it is necessary to reduce the extra energy carrier aggregation system, but only sub-optimal results could be
consumed by CA operation, thus reduce the carbon emission and achieved because Physical Resource Block (PRB) allocation was
contribute to green communication. not considered jointly with CC selection. Refs. [3,6] studied PRB
In Carrier Aggregation system, except for traditional Resource allocation schemes, but they would significantly increase mobile
Block (RB) allocation, component carrier selection is a new issue terminal’s energy consumption because they did not consider
continuous component carrier selection.
To sum up, most existing researches only consider separate
∗ Corresponding author.
optimization of either carrier selection or resource allocation,
E-mail addresses: gaoweidong@bupt.edu.cn (W. Gao), 9402562@qq.com which could not attain optimal performance in carrier aggregation
(L. Ma), chuai@bupt.edu.cn (G. Chuai). system. In this paper, we propose a new strategy that jointly
http://dx.doi.org/10.1016/j.phycom.2016.12.002
1874-4907/© 2017 Elsevier B.V. All rights reserved.
294 W. Gao et al. / Physical Communication 25 (2017) 293–297

optimizes carrier selection and resource allocation, which can uses under determined transmission power. Then The throughput
obtain higher system throughput than existing schemes. of the system is the total throughput of all the users. Define Energy
Efficiency (EE) as the ratio of system throughput and the consumed
2. System model and problem formulation power, so maximizing EE is equivalent to maximizing the system
throughput under the condition that the eNodeB’s transmit power
is constant. That is to say, the energy efficiency of the system is
2.1. System model
optimal when maximal system throughput is provided.
Consider a single cell network model with a total of K users and
N CCs. Each CC is composed of M PRBs each with bandwidth B. 2.2. Problem analysis
Define K = {1, 2, . . . , K }, N = {1, 2, . . . , N }, M = {1, 2, . . . , M }
as the user set, CC set and PRB set, respectively. Indicator variable
In this section, we will prove that the proposed constrained
ck,n,m ∈ {0, 1} is used to indicate the relationships among CC,
optimization problem as shown in formula (5) can be considered
user and PRB, where value ‘‘1’’ indicates that PRB #m of CC #n is
allocated to user #k and value ‘‘0’’ indicates the opposite. There is as a graph coloring problem. As we know, graph coloring problem
a rule that each PRB can only be assigned to a user at any moment. is a typical NP-hard problem and by far there is no effective way to
According to the Shannon formula, the throughput that user #k solve this kind of problem.
can obtain from PRB #m of CC #n can be written as: Consider a special case of the optimization problem (5) and we
  formulate it as problem S, where each user can only be assigned
pk,n,m × gk,n,m up to 1 CC. As we all know, Bipartite Graph Edge Coloring (BGEC)
rk,n,m = B log2 1+ (1)
BN0 problem is a typical NP-hard problem, so if we can prove that
problem S can be reduced to a BGEC problem in polynomial-time,
where pk,n,m ≥ 0 is the transmit power of the base station, gk,n,m is
then we can deduce that the optimization problem S is also NP-
the channel gain and N0 is Additive Write Gaussian Noise (AWGN)
power spectral density. hard. Firstly, we will give the definition of BGEC problem: For a
Based on formula (1) and the definition of ck,n,m , we can obtain graph G = (V , E ) with vertex set V and edge set E, if we can find a
the total system throughput by summing up the transmission rate subgraph F ⊆ E and the subgraph F can be colored with only two
of all the users as following: colors, then the sub-graphs composed of F are bipartite graphs.
At first, we will transform aforementioned optimization prob-
K 
 N 
M
lem S with graph theory. Assuming that each user v ∈ V corre-
R= rk,n,m × ck,n,m . (2) sponds to a vertex of the graph and each PRB e ∈ E corresponds
k=1 n=1 m=1
to an edge of the graph. All PRBs on any Component Carrier are
With Carrier Aggregation operation, a user can use multiple CCs numbered independently and the PRBs with the same number are
at the same time. The number of CCs occupied by user #k, Ck , is mapped to the same edge, which means that the mapping rela-
given by the following formula: tionship between an edge and PRBs is one-to-many. For simplic-
  ity, assuming that there are two Component Carriers, c1 and c2 .
N M
  If the vertex v (corresponds to user k) is incident with the edge
Ck = τ ck,n,m (3)
e (corresponds to PRB m), then rk,m,c1 = rk,m,c2 = 1, otherwise
n =1 m=1
rk,m,c1 = rk,m,c2 = 0.
where function τ (x)(x ≥ 0) is defined as: In a regular graph, any edge of the graph is associated with
two vertices. Corresponding to the optimization problem S, we can
1, x>0

τ (x) = (4) consider it as the allocation of a PRB group to the two users, where
0, x = 0.
the PRB group is composed of physical resource blocks with the
By combining formulas (1)–(4), we can obtain the maximum same number of all CCs. Therefore, the original resource allocation
throughput of downlink Carrier Aggregation system, which can be optimization problem S can be transformed into an equivalent
modeled as a constrained optimization problem: problem that partition of the users into separate groups. Due to
the assumption that there are two CCs in the system and each
K 
N 
M
 user can be allocated at most one CC, so the set of users will be
max rk,n,m × ck,n,m (5)
partitioned into two groups and it should be ensure that odd length
k=1 n=1 m=1
rings will not appear in the graph. Obviously, this is consistent with
s.t. the definition of bipartite graph edge coloring problem.
K
Next we will prove that the optimal solution of the original

C1 : ck,n,m ≤ 1, ∀ n, m
k=1
optimization problem S is also the optimal solution of the bipartite
graph edge coloring problem. Let F be the largest subset of graph G
 
N M

C2 : τ ck,n,m ≤ T. and e = (v1 , v2 ) be any edge of it. Allocate PRB #m of c1 and c2 to
n=1 m=1 users k1 and k2 , i.e. rk1 ,m,c1 = rk2 ,m,c2 = 1, where (k1 , k2 ) ≡ (v1 , v2 )
The constraint condition C1 indicates that any resource block represents that k1 and k2 corresponds to v1 and v2 respectively.
can be allocated to at most one user during one scheduling interval. Iterate above operations to all the edges of F and ensure that
The constraint condition C2 limits the maximum number of CCs the resource blocks allocated to each user belong to the same CC.
that each user can aggregate simultaneously. It is specified that by Due to that the subset F can be colored with two colors, i.e. the
3GPP that the number of CCs each user can aggregate is less than sub-graphs composed by subset F is a bipartite graph, so above
or equal to 5 [7]. Therefore, the value of T is less than or equal to 5 operations can absolutely be achieved. The resource blocks that
throughout this paper. have not been allocated yet (i.e. not in subset F ) can be assigned to
Assuming that the eNodeB’s transmit power on each PRB is con- any user that they are associated providing that existing grouping
stant, we are able to obtain the throughput of the user on the CCs it status is not destroyed.
W. Gao et al. / Physical Communication 25 (2017) 293–297 295

data rate of each user on each CC rk,c ,m . Initialize the parameters


including virtual user set K ′ , user-CC pair set φ and the resource
allocation solution set S ;
Step 2: Determine whether the user set K is empty, if so go to
step 7, otherwise go to step 3;
Step 3: Calculate f (S ) with respect to S according to formula
(6);
Step 4: For any user-CC pair (k, c ) in φ , calculate the respective
additive system throughput gain comparing with that in Step 3,
i.e. f (S ∪ (k, c )) − f (S );
Step 5: Find the maximum additive gain obtained in Step 4 and
put the corresponding user-CC pair (k∗ , c ∗ ) into S , remove (k∗ , c ∗ )
from the candidate set φ and remove the virtual user k∗ from K ′ ;
Step 6: Go to Step 2;
Step 7: For each PRB of CC #c, find the user with the maximum
transmission rate on it with the following rule:

k∗c .m = arg max rk,c ,m .


 
(7)
(k,c )∈S

Allocate this PRB to user k∗c .m if rk∗ ,c ,m is larger than the preset
threshold, otherwise ignore it.

3.2. Lower bound performance


Fig. 1. Flow chart of the algorithm.
In this section, we will demonstrate the effectiveness of the
3. Joint optimization of component carrier selection and GCSRAA algorithm. Numerical results show that the performance
resource allocation of the proposed algorithm falls above 1/2 of the theoretical optimal
performance at any case. First of all, we will give the definitions
We learned from Section 2 that the constrained optimization used in the proof process.
problem (5) is NP-hard, and by far there is no effective method Submodular function [8]: Given a function f on S, for any x, A, B,
to solve this kind of problem. In this section, we propose a A ∪ {x} ∈ S and B ⊆ A, if we have:
Greedy based Carrier Selected and Resource Allocation Algorithm f (A ∪ {x}) − f (A) ≤ f (B ∪ {x}) − f (B) (8a)
(GCSRAA), which shows a fairly good performance with high
efficiency. f (A ∪ {x}) − f (A) ≥ 0 (8b)
f (φ) = 0 (8c)
3.1. Algorithm design then we call f a submodular function.
Partition matroid [9]: Let S be a series of sub-set of ψ , if S meets
Let K = {1, 2, . . . , K }, C = {1, 2, . . . , C } and M = the following conditions: (i) φ ∈ S; (ii) If A ∈ S and B ⊆ A, then
{1, 2, . . . , M } be the user set, the CC set in the system and the B ∈ S; (iii) If A, B ∈ S and |A| > |B|, there are elements x ∈ A \ B
PRB set of each CC, respectively. Construct an equivalent virtual that make B ∪ {x} ∈ S, then we call S a partition matroid.
user set K ′ that contains p|K | virtual users and each virtual user We will prove the objective function of the optimization
occupies at most one CC. Assuming that one user-CC pair links one problem in Eq. (5) is a submodular function, and its domain is
virtual user and one CC, meaning that the CC is assigned to the a partition matroid. Firstly, we obtain the user-CC pair set as
corresponding user. All of the user-CC pairs construct the user-CC following:
pair set φ , (k, c ) |k ∈ K ′ , c ∈ C . Define the resource allocation
solution set as S and for all (k, c ) ∈ S , we can get the performance ψ = (k, c ) : ∀k ∈ K ′ , c ∈ [1, N] .
 
(9)
metric about S :
After obtaining the user-CC pair set, we can get partition on set
C 
 M
ψ : ϕk = {(k, c ) : ∀c ∈ [1, N]}. Define S as the set of sub-sets on
f (S ) = rk,c ,m .
 
max (6) ψ , then for all A ∈ S, it is clear that: (i) If B ⊆ A, then B ∈ S; (ii) If
k:(k,c )∈S
c =1 m=1 x ∈ A \ B, then B ∪ {x} ∈ S; (iii) For ∀k, |A ∩ ϕk | ≤ 1. According to
The essence of above metric is: for any PRB of the CC, we above analysis, we can find that S is a standard partition matroid.
will find the maximum data rate that it can bear and regard For element A in S, define the following function:
this value as its contribution to the overall throughput of the
C
system. The introduced metric can reflect the effectiveness of a 
resource allocation scheme. On the basis of the above assumptions
f (A) = µc (A) (10)
c =1
and definitions, we could propose a heuristic resource allocation
algorithm based on greedy strategy: all the PRBs on any CC will where, the definition of function µc (A) is as following:
be assigned to the user with the maximum channel gain. In order
M
to improve the overall system performance, ignore the resource 
µc (A) = rk,m,c .
 
max (11)
block if the corresponding maximum channel gain is lower than u:(u,c )∈A
m=1
a predetermined threshold, i.e. it will not be assigned to any user
at all. The flow chart of the algorithm is shown in figure and the It is shown from above equation that µc (B) ≤ µc (A) holds true
specific processes is as following (see Fig. 1): when B ⊆ A. Besides, what makes it different for set A and set B is
Step 1: Obtain the parameters including user set K , CC set C , that a certain of CCs in set A are assigned to more users. Therefore,
PRB set M , the number of CCs occupied by one user p and the the throughput gain of set A is more than that of set B when adding
296 W. Gao et al. / Physical Communication 25 (2017) 293–297

Table 1
Simulation parameters.
Name Value

Cell layout 1 layer, 3 sectors


Channel model SCME model
Simulation scenario City macro
Carrier frequency 3.5 GHz
CC bandwidth 20 MHz
Maximum transmit power 49 dBm
Minimum eNoeB-UE distance 35 m
eNodeB coverage extension parameter 0 dB
Simulation mode Dynamic simulation
Service type Full buffer
Link direction Downlink
TTI length 1 ms
TTI number 5000

a new user to the CC. That is to say, for any element (k, c ), we have:
f (A ∪ {(k, c )}) − f (A) ≤ f (B ∪ {(k, c )}) − f (B) . (12) Fig. 2. Normalized system throughput.
In addition, the throughput of the component carrier will not
decrease when a new user is added to the component carrier. So
we have:
f (A ∪ {(k, c )}) − f (A) ≥ 0. (13)
It is clear that µc (φ) = 0 for any empty set φ , therefore f (φ) =
0. Based on above analysis, we can conclude that the function f is
a submodular function.
According to Ref. [8], for a submodular function like x =
arg maxx∈ϕi f (A ∪ {x})− f (A) whose domain is a partition matroid,
the performance of the sub optimal solution at each iteration is
at least 1/2 of the optimal solution when greedy algorithm is
used to solve the problem. Therefore, we can conclude that the
performance of GCSRAA algorithm is above 1/2 of the optimal
solution in the worst case.

4. Numerical results

Fig. 3. System throughput performance comparison with exhaustive search


In this section, the performance and computational complex-
method.
ity of the proposed GCSRAA algorithm is evaluated through sim-
ulations. At first, the system throughput is demonstrated under
different conditions, including various CC numbers in the sys-
tem and various CC numbers that can be used by single user.
Then, we compare the performance of the proposed GCSRAA al-
gorithm with that of resource allocation algorithms in references,
including Non-Linear Programming Resource Allocation Algorithm
(NLPRAA) [10] and QoS Constrained based Resource Allocation Al-
gorithm (QCRAA) [11]. The simulation parameters used through-
out this paper is shown in Table 1.
For simplicity, we use normalized system throughput in the
following results analysis. Assume there are 5 users and the
available CCs in the system are 5, 10 and 15, respectively. Fig. 2
shows the downlink system throughput with different number of
CCs (i.e. 1–5) that single user can occupy. The more CCs are used
by the user, the more resource blocks it can use and the more
throughputs it can obtain, so all the three curves show a similar
upward trend towards 100% with increasing CC numbers that
single user can adopt simultaneously. Furthermore, when there are Fig. 4. System throughput performance comparison with classical algorithms with
more available CCs in the system, the users are more likely to be different CCs that each user can use.
allocated to more resource blocks with good channel quality, so the
corresponding curve rises faster than that with few available CCs. That is to say, the proposed algorithm could replace the optimal
Fig. 3 shows the system throughput of GCSRAA algorithm solution in most cases.
comparing with that of exhaustive search method. It is clear Fig. 4 illustrates the system throughput that can be achieved by
that for users working in Carrier Aggregation mode, the more GCSRAA algorithm when there are 15 available CCs in the system.
component carriers assigned to them, the higher data rates they The performance of NLPRAA algorithm and QCRAA algorithm are
can achieve. In addition, we can find from the same figure that the also provided for comparison purpose. As can be seen from the
proposed algorithm is comparable to the exhaustive algorithm and figure, the performance improves with the increases CCs for all of
the performance difference is no more than 5% in the worst cases. the three algorithms. Meanwhile, it is clear that GCSRAA algorithm
W. Gao et al. / Physical Communication 25 (2017) 293–297 297

adopted as the optimization metric. Numerical results show that


the performance of the proposed algorithm is close to that obtained
with exhaustive search method. Furthermore, simulations are
performed and it is found that the proposed algorithm can
significantly improve the system throughput while keeping low
computation complexity. Future work will consider fairness
among component carriers and among different users.

Acknowledgment

This work is supported by Major National Scientific &


Technological Specific Project of China under grant number
2015ZX03001025-002.

References

[1] M. Iwamura, K. Etemad, M.H. Fong, et al., Carrier aggregation framework in


3GPP LTE-advanced, IEEE Commun. Mag. 48 (8) (2010) 60–67.
Fig. 5. System throughput performance comparison with classical algorithms with [2] R. Mahapatra, Y. Nijsure, G. Kaddoum, et al., Energy efficiency trade-off
different user in the system. mechanism towards wireless green communication: A survey, IEEE Commun.
Surv. Tutor. 18 (1) (2016) 686–705.
[3] Zhanyang Ren, Shanzhi Chen, Bo Hu, Weiguo Ma, Energy-Efficient Resource
Allocation in Downlink OFDM Wireless Systems with Proportional Rate
Constraints, Vehicular Technology, 2014.
[4] G. Auer, V. Giannini, I. Godor, et al., Cellular energy efficiency evaluation
framework, in: Vehicular Technology Conference, VTC Spring, 2011 IEEE 73rd,
IEEE, 2011, pp. 1–6.
[5] Huaxia Chen, Yonglei Jiang, Jing Xu, Honglin Hu, Energy-efficient coordinated
scheduling mechanism for cellular communication systems with multiple
component carriers, Sel. Areas Commun. (2013).
[6] Yu Ye, Yongning Zhuo, Shizhong Xu, Energy efficient resource allocation
in LTE-A, in: Communications in China, ICCC, 2013 IEEE/CIC International
Conference, 2013.
[7] Zukang Shen, Aris Papasakellariou, Juan Montojo, Dirk Gerstenberger, Fangli
Xu, Overview of 3GPP LTE-advanced carrier aggregation for 4G wireless
communications, IEEE Commun. Mag. 50 (2) (2012) 122–130.
[8] G. Calinescu, C. Chekuri, M. Pál, et al., Maximizing a monotone submodular
function subject to a matroid constraint, SIAM J. Comput. 40 (6) (2011)
1740–1766.
[9] G.L. Nemhauser, L.A. Wolsey, M.L. Fisher, An analysis of the approximations for
maximizing submodular set functions, Math. Program. 14 (1978) 265–294.
[10] G. Yu, Q. Chen, R. Yin, H. Zhang, G.Y. Li, Joint downlink and uplink resource
allocation for energy-efficient carrier aggregation, IEEE Trans. Wireless
Commun. 14 (6) (2015) 3207–3218.
Fig. 6. Convergence speed for different algorithms. [11] B. Barakat, K. Arshad, Energy efficient carrier aggregation for LTE-advanced,
in: GCC Conference and Exhibition, GCCCE, 2015 IEEE 8th, 1–5 Feb. 2015.
outperforms others in all cases and the largest gain is 12% and 32%
comparing with that of NLPRAA and QCRAA, respectively. So we
Weidong Gao received his bachelor degree in 2004 and
have proved that GCSRAA algorithm can get a better performance doctor degree in 2009 from Beijing University of Posts
than those optimization algorithms which only consider separate and Telecommunications (BUPT) in China. From 2009 to
carrier selection or PRB allocation. 2015, he worked as a senior engineer at Potevio Institute
of Technology Co. Ltd. He is currently a lecturer in BUPT
The system throughput is shown in Fig. 5 considering different
and his research activities are focused mainly on 5G
number of users in the system. When the number of users is small, wireless communications including algorithm design and
there is enough resource blocks available, so increasing users will standardization.
also increase system throughput. But with the increasing of users,
the system throughput will not increase unlimitedly and tends
to be stable due to that the resource blocks become constraint.
Lin Ma received his master degree in 2016 from Beijing
Comparing with NLPRAA algorithm and QCRAA algorithm, there is University of Posts and Telecommunications (BUPT) in
an additional of 10% and 23% gain, respectively. China. He is currently working as an engineer at Tencent
In Fig. 6, we evaluate the computation complexity of the three company in China. His research interests include broad-
band wireless access technologies, radio resource manage-
algorithms. As can be seen from the figure, the proposed GCSRAA ment and mobile internet.
algorithm can achieve 90% of the optimal system throughput at
about 100 iterations, which is about 1/4 of that for NLPRAA al-
gorithm and QCRAA algorithm. Furthermore, the optimal perfor-
mance of the two algorithms is at least 10% worse than that of the
algorithm proposed in this paper. Therefore it is confirmed that the
proposed algorithm can obtain better performance while guaran- Gang Chuai received his BS, MS, and Ph.D. degrees in
electrical engineering from Beijing University of Posts
teeing the low complexity. and Telecommunications (BUPT), in 1983, 1999, and
2010, respectively. He is now a professor at the School
of Information and Communication Engineering, BUPT.
5. Conclusion His research interests include 5G wireless network
technology, automatic wireless network planning and
In this paper, we put forward a greedy based component optimization.
carrier selection and resource block allocation scheme for carrier
aggregation system. System capacity and energy efficiency are

S-ar putea să vă placă și