Content uploaded by Kan Wang
Author content
All content in this area was uploaded by Kan Wang on Mar 12, 2016
Content may be subject to copyright.
A QoS-based Hybrid Centralized/Distributed
Resource Allocation Algorithm in Downlink
Femtocell Networks
Kan Wang, Yinghong Ma, Member, IEEE, Hongyan Li, Member, IEEE, Peng Liu, Hongguang Sun and Hao Zhang
State Key Laboratory of Integrated Service Networks
Institute of Information Science, Xidian University, Xi’an, Shaanxi, 710071, China
Email: {yhma, hyli, hgsun}@xidian.edu.cn, wangkan@stu.xidian.edu.cn, liupeng0218@gmail.com, zhanghaoxdisn@126.com
Abstract—Femtocells have emerged as an effective solution
to enhance indoor coverage and improve system performance
in cellular networks. However, the inter-cell interference (ICI)
caused by the unplanned nature of femtocells considerably leads
to the degradation in throughput of users with guaranteed
performance (GP). Meanwhile, the existing resource allocation
algorithms bring about high complexity and large overhead.
By tracking channel variation as well as arrival and departure
of users, we propose a hybrid centralized/distributed resource
allocation algorithm to maximize the number of GP users, with
lower complexity and smaller overhead. Simulation results show
that the proposed algorithm can significantly improve the system
performance compared to the existing algorithms such as Q-
FCRA.
I. INTRODUCTION
Femtocells, served as small range indoor access point and
installed by customers, are a cost-effective means of providing
improved indoor coverage in macro cellular networks [1]. Due
to the much smaller range of femtocells than that of macrocell-
s, femtocells can achieve superior signal-to-interference-plus-
noise ratio (SINR). Meanwhile, the spatial reuse of femtocells
permits multiple femtocell users to transmit data on the same
licensed channel, thereby increasing the spectrum efficiency in
the networks. Moreover, 3GPP is now focusing on Long Term
Evolution (LTE) and LTE-Advanced (LTE-A) technologies,
which are based on orthogonal frequency division multiple
access (OFDMA) [1], [2]. LTE is likely to be the dominant
technology for the future, thus the standardization of femto-
cells in LTE is particularly important. Resource allocation in
LTE femtocells networks is the focus of many researches [6]-
[11] and also the subject of this paper.
However, the unplanned nature of femtocells makes the
inter-cell interference (ICI) a challenging task. The femtocells
may be exposed to severe co-tier interference. What’s more, to
perform global optimization, femto base stations (FBSs) must
have knowledge of all users’ channel state information (CSI)
including channel gain and interference suffered from other
FBSs, and then send CSI to a centralized unit, operation and
management (OAM) [4]. However, large backhaul round-trip
delay, ranging from several milliseconds (ms) to dozens of ms
[3], makes that it is impossible to find the optimal solution
instantaneously [1]. Meanwhile, in the dense femtocell net-
works, computational complexity must be taken into account.
Moreover, overhead issues also arise due to the large amount
of CSI in the centralized approach [1], [2].
A large amount of prior work such as [6]-[9] has resorted to
solving resource allocation problem, aiming at maximizing the
system throughput or total utility in the heterogeneous network
(HetNet). However, in the quality of service (QoS)-aware
networks [5], resource allocation must be executed taking into
account QoS requirements of guaranteed performance (GP)
users. The authors in [10], [11] propose a distributed resource
allocation algorithm to satisfy all GP users’ QoS requirements
by utilizing minimal transmit power, assuming that sufficient
resources are available in the femtocell networks. The authors
in [12] consider maximizing the number of GP users in
macrocell networks, provided that resources wouldn’t suffice
to satisfy all users’ QoS requirements. Similar to [12], in
[13], [17], a cluster-based algorithm is proposed to maximize
the num of users in femtocell networks without considering
channel variation (i.e., fast fading and frequency-selective
fading) .
Similar to [12], [13], in this paper, we propose a hybrid
centralized/distributed resource allocation algorithm to maxi-
mize the number of GP users in femtocell networks. Compared
to the scheme where FBSs report all users’ CSI to OAM
through backhaul once new users arrive, our proposed algorith-
m tradeoffs the signaling overhead and system performance,
significantly decreasing the amount of CSI and maintaining
higher throughput as well. Furthermore, compared to [13] we
consider a more practical system model, taking into account
channel variation as well as arrival and departure of GP users
in the long term.
II. SYSTEM MODEL
We consider a LTE HetNet consisting of multiple femtocells
that are randomly distributed in macro cellular networks, as
shown in Fig. 1. As in [10], [11], it is assumed that the
spectrum used by femtocells is orthogonal to that utilized by
macrocells. Therefore, we only need to focus on the co-layer
interference mitigation and resource allocation. The cross-
layer interference case is beyond the scope of this paper.
978-1-4673-6187-3/13/$31.00 ©2013 IEEE
0DFURFHOO
Fig. 1. Network model
In this paper, we consider QoS-aware networks that provide
required QoS of GP users. For GP users, QoS requirements
must be fully met, otherwise the utility of resource allocation
is equal to zero. It is reasonable to assume that the femtocell
network is QoS-aware since various GP users have different
QoS requirements. However, traditional resource allocation
strategy in femtocell networks didn’t take into account QoS
requirements of various users, thereby increasing the block
probability of the system [5], [12].
Note that, resource allocation for Best-Effort (BE) users
has been studied in [12], [13]. Therefore, to highlight the
performance of our algorithm, we only consider GP users in
the system.
In the following, we first describe the network model, then
state and formulate the resource allocation problem.
A. Network Model
We denote by L={1,...,L}and M={1,...,M}the
set of FBS and femto users, respectively. Each femto user is
assumed to be attached to only one FBS, and the set of all
femto users attached to BS L∈Lis denoted by M(l).In
addition, we define as N={1,...,N}the set of Resource
Block (RB) to be allocated in the downlink. RBs could be
reused among cells. However, one RB could only be utilized
by one user in each femtocell.
We also denote by YM×N=[ymn ]the binary allocation
matrices: ymn =1indicates that RB nis assigned to use
mand 0 otherwise. Similar to YM×N,XM×N=[xmn]
represent the binary allocation matrices among femtocells with
xln =1indicating that RB nis utilized by FBS land 0
otherwise. We can infer that xln =m∈M(l)ymn since RB
could only be utilized by one user in each femtocell.
In this paper, we focus on the allocation of RBs; therefore,
the power allocated on each RB is considered as constant.
Given constant power p, the received SINR of femto user m
on subchannel nis calculated as:
SINRmn =p·vB(m)
mn ·ymn
L
l=1,l=B(m)p·vl
mn ·m∈M(l)ymn+σmn
,
(1)
where, vB(m)
mn represent the channel gain on RB nbetween
user mand its FBS B(m).Therefore, the corresponding
achievable rate for user mon subchannel nis determined
by:
Rmn = log 1+ 1
ΓSINRmn (Y),(2)
where, Γ=−ln(5BER)
1.5is the required SINR gap to capacity
for the target bit error ratio (BER) [7].
B. Problem Formulation
Our objective is to maximize the number of GP users whose
QoS requirements is sufficiently satisfied. Introducing metrics
dmand zmrepresenting the QoS requirement of user mand
whether mis satisfied or not, respectively, we formulate the
resource allocation problem as follows:
max
m∈M
zm
s.t. ymn ∈{0,1},∀m, n,
m∈M(l)
ymn ∈{0,1},∀l, n, (3)
ϕm=
N
n=1
Rmn,∀m,
zm=1,ϕ
m−dm≥0,
0,ϕ
m−dm<0,∀m,
where, Rmn is given by (1), (2), and ϕm=N
n=1 Rmn,∀m
denotes the total throughput achieved by user mon all RBs.
This optimization problem is a nonlinear integer programming
problem; therefore, we find it difficult to solve this NP-hard
problem in a centralized approach on small time scales [16].
III. PROPOSAL:HYBRID CENTRALIZED/DISTRIBUTED
RESOURCE ALLOCATION
In this section, we present a hybrid centralized/distributed
resource allocation algorithm in LTE femtocell networks. In
our proposed algorithm, OAM collect and update all users’
CSI periodically in the order of 100ms (much larger than the
femtocell backhaul delay), then performs resource allocation
algorithm and delivers the decision to all FBSs via backhaul.
This process is executed in the centralized approach by OAM
according to various QoS requirements of GP users and mutual
interference between femtocells. Once receiving the decision
form OAM, each FBS will reallocate these RBs assigned to it
to its associated users. Since the channel is fading in nature and
the set of users in the network is time-variant, the reassignment
of RBs must be implemented in each subframe within the
current period. Note that, in each subframe, FBSs reallocate
RBs to its users in a distributed approach, without considering
other femtocells’ RB allocation solution. We decompose the
resource allocation problem as two parts: centralized resource
allocation problem and distributed resource allocation prob-
lem.
A. Centralized resource allocation
We decompose the centralized resource allocation problem
into two sub-problems: RBs Allocation among femtoCells
sub-Problem (RACP) and RBs Allocation among Users sub-
Problem (RAUP). RACP determine how to allocate RBs to
FBSs while the algorithm to RAUP is performed provided
that the RB set for each FBS is predetermined.
Taking FBS lfor example, and assuming that X∈XL×N
has been fixed a prior, i.e., M(l),N(l)has been assigned to
FBS l, RAUP can be formulated as follows:
max
m∈M(l)
zm
s.t. ymn ∈{0,1},∀m∈M(l),n∈N(l),
m∈M(l)
ymn ∈{0,1},∀n∈N(l),(4)
ϕm=
n∈N(l)
Rmn,∀m∈M(l),
zm=1,ϕ
m−dm≥0,
0,ϕ
m−dm<0,∀m∈M(l).
By solving RACP, f(X), representing the number of GP
users in all femtocells, is utilized to evaluate the performance
of allocation strategy X∈XL×N. Note that, for each FBS,
once the set of RBs for other FBSs are predetermined, the
interference to its associated users in the downlink is known
in advance. Therefore, the solution to RAUP can be obtained
with a small cost in OAM through carrying out FBSs loop.
By transforming constraint
zm=1,ϕ
m−dm≥0,
0,ϕ
m−dm<0,∀m∈M(l)(5)
into two constraints:
−⎛
⎝
n∈N(l)
ymnRmn −dm⎞
⎠≤C(1 −zm),∀m∈M(l)(6)
n∈N(l)
ymnRmn −dm<Cz
m,∀m∈M(l)
RAUP can be reformulated as a binary integer programming
(BIP) problem [16]:
max
m∈M(l)
zm
s.t. ymn ∈{0,1},∀m∈M(l),n∈N(l),
m∈M(l)
ymn ∈{0,1},∀n∈N(l),
ϕm=
n∈N(l)
Rmn,∀m∈M(l),(7)
−⎛
⎝
n∈N(l)
ymnRmn −dm⎞
⎠≤C(1 −zm),∀m∈M(l)
n∈N(l)
ymnRmn −dm<Cz
m,∀m∈M(l),
where, Crepresents a constant whose value is usually much
larger than the rate achieved by femto users [16]. Due to the
linear integer property of (7), cutting plane and branch-and-
bound algorithms can be used to deal with this class of linear
BIP problems [16]. Additionally, existing solving package for
BIP can be utilized to efficiently solve this problem within
acceptable time.
As for RACP, Tabu Search (TS) metaheuristic is utilized to
search over the solution space XL×Nand gives the suboptimal
solution to RACP [11], as depicted in Algorithm 1. XL×N
is the RACP solution, indicating the RB set selected for each
femtocell. The quality of RACP solution Xis evaluated based
on the value f(X)through solving RAUP, i.e., (7). Repeating
this process, different RACP solution Xcan be evaluated until
a best solution Xbes t maximizing the number of GP users is
found.
B. Distributed resource allocation
Receiving their RB assignment Xbest from the centralized
resource allocation, FBSs will reassign RBs to its users,
making instantaneous decisions in each subframe within the
current period. It is noteworthy that, similar to RAUP in
the centralized resource allocation, once OAM has made
decisions on which RB is utilized by which FBS, the instan-
taneous interference can be known by FBSs based on the CSI
measurement report. Given Xbest within the current period,
through updating CSI measurement report in each subframe,
the dynamic resource allocation can be made locally by each
FBS in a distributed approach independent of other FBSs’
decisions.
Denote by Intm
n(t)and vB(m)
mn (t)the interference to GP
user mand the channel gain on RB nbetween user m
and its FBS B(m)in subframe t, respectively. Therefore,
in each subframe within the current period, resource alloca-
tion should be made based on the instantaneous CSI, i.e.,
Intm
n(t)and vB(m)
mn (t)by each FBS. The differences between
distributed resource allocation and RAUP are as follows. In
the distributed resource allocation, CSI need to be updated
by FBS instantaneously in every subframe in order to track
channel variation as well as arrival and departure of users. In
addition, distributed resource allocation is carried out by each
FBS instantaneously while RAUP must be solved by OAM
periodically in the order of 100ms.
Similar to RAUP, distributed resource allocation problem
can be efficiently solved by cutting plane or branch-and-bound
algorithm using existing solving package.
As mentioned above, we propose the distributed resource
allocation algorithm in Algorithm 2.
IV. SIMULATION RESULTS
In this section, we verify the performance of our proposed
hybrid centralized/distributed resource allocation algorithm
through extensive simulations, by comparing it with the Q-
FCRA algorithm [13]. Meanwhile, only centralized resource
allocation is used as a benchmark to which the necessity of
distributed resource allocation in our algorithm is verified.
The period of centralized resource allocation is assumed to
be 100ms. In our simulations, we consider a system consisting
Algorithm 1 CENTRALIZED RESOURCE ALLOCATION
ALGORITHM
1. Initialize X,f(X)
2. Initialize best solution Xbest =X, fbes t =f(X)
3. Initialize tabulist=[], count er =0;
4. while count er <counter max (solving RACP)
5. counter =counter +1
6. nbr ter =0,fnbr
best=0
7. while nbr iter <nbr max (neighbor loop)
8. nbr iter =nbr iter +1
9. Select a neighbor Xnbr ∈neighbor (X)
10. Solve RAUP: calculate f(Xnbr)
11. if f(Xnbr )>fbes t
12. Update: Xbes t =Xnbr ;Xnbr
bes t =Xnbr
13. fbes t =f(Xnbr),fnbr
best =f(Xnbr)
14. break
15. endif
16. if Xnbr is in tabulist
17. Calculate next Xnbr ∈neighbor (X)
18. endif
19. if f(Xnbr )>fnbr
bes t
20. Xnbr
bes t =Xnbr ;fnbr
best =f(Xnbr)
21. endif
22. endwhile
23. Enter Xnbr
bes t into a tabulist
24. Update tabulist
25. Update X=Xnbr
bes t ,fbes t =fXnbr
bes t
26. endwhile
Algorithm 2 DISTRIBUTED RESOURCE ALLOCATION
ALGORITHM
1. for each subframe do
2. foreach FBS do
3. Update the CSI of users in subframe t
4. Solve (7) using cutting plane algorithms
5. YM×Nis obtained in subframe t
6. Send RB allocation to its associated user
7. % Allocation strategy is passed on to users
8. endfor
9. endfor
10. % subframe loop is carried out until best Xbes t is updated
of multiple RBs, which is a typical LTE downlink frame
structure. The system bandwidth is assumed to be 5 MHz
[15]. However, due to the higher priority of macrocells and
orthogonality between femtocells and macrocells, a smaller
number of available RBs are assumed in our simulation. The
maximum transmit power of FBS is 20 dBm and standard
deviation of shadow fading is 8 dB. More detailed system
parameters can be found in [15]. Path loss model in suburban
deployment proposed by is utilized.
As shown in Fig. 1, two representative types of femtocell
configuration are considered in our simulations: (i) isolated
case (negligible mutual interference due to the long distance
between femtocells), (ii) symmetric case. The latter case
presents a more sophisticated interference scenario.
00.5 11.5 22.5 3
0
10
20
30
40
50
60
70
80
90
100
Average QoS requirement (Mbps)
Percentage of GP users (%)
hybrid centralized/distributed
algorithm
Q−FCRA
only centralized algorithm
Fig. 2. Number of GP users vs. QoS requirements.
To be fair, similar to [13], it is assumed that all FBSs are
randomly distributed in a 200m×200m region. Meanwhile,
covering area of each FBS is a 10m×10m residence. The
number of femtocells is assumed to be 12, i.e., 6 femtocells in
case (i) and 6 femtocells in case (ii). In our simulation, we only
take into account GP users, whose average QoS requirements
are tunable. All GP users generate real-time service, which is
modeled as a constant rate data flow with 100% activity [14].
A poisson process is used to model the dynamic variation of
GP users, with arrival rate 0.05 per ms and mean service time
100 ms.
Fig. 2 gives the ratio of the number of GP users for the
varying QoS requirements. The corresponding data is averaged
over 300ms, i.e., three periods, and the number of available RB
is 6. As is shown, the hybrid centralized/distributed algorithm
can achieve higher performance compared to Q-FCRA. The
percentage of accepted users in Q-FCRA is only 76% while
our proposed algorithm allows 88% of users to achieve their
QoS requirement (16% improvement), given that the average
QoS requirement is 0.5 Mbps. Note that, the gap between
the hybrid centralized/distributed algorithm and Q-FCRA in
low QoS requirements is much smaller than that in high
requirements.
That is due to the fact that lower QoS requirements result in
the frequency reuse in femtocells and higher QoS requirements
force RBs to be utilized exclusively by single femtocell.
However, in Q-FCRA, RBs are always used exclusively by
only one FBS in case (ii).
Fig 3 shows the ratio of the number of GP users under
various numbers of available RBs, given that the average QoS
requirement is 1 Mbps. The ratio is better with our proposed
algorithm reaching almost 89% compared to 83% in Q-FCRA,
when the number of RBs is 5. Only centralized resource
allocation remains much lower level compared to the hybrid
algorithm (about 50% degradation), verifying that distributed
and dynamic resource allocation in each subframe must be
considered.
Fig. 4 shows the spectrum spatial reuse (SSR) under various
22.5 33.5 44.5 55.5 66.5 7
20
30
40
50
60
70
80
90
100
the number of available RB
Percentage of GP users (%)
hybrid centralized/distributed
algorithm
Q−FCRA
only centralized algorithm
Fig. 3. Number of GP users vs. RB number
22.5 33.5 44.5 55.5 66.5 7
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
the number of available RB
Spectrum spatial reuse (SSR)
hybrid centralized/distributed
algorithm
Q−FCRA
only centralized algorithm
Fig. 4. SSR vs. RB number
numbers of available RBs, provided that the average QoS
requirement is 1Mbps. As is shown, our hybrid algorithm
allows more FBSs to utilize the same RB compared to Q-
FCRA algorithm (about 18% improvement when the number
of RBs is 3). Note that, as the number of RBs increases, GP
users would select RB exclusively utilized by only one FBS in
case (ii) to satisfy their QoS requirements, thereby decreasing
the SSR. In Q-FCRA, SSR remains the same for the varying
number of RBs due to the fact that RB could be utilized by
only one FBS in case (ii). The SSR of the only centralized
resource allocation still remains much lower level.
V. CONCLUSION
This paper proposed a hybrid centralized/distributed re-
source allocation algorithm to maximize the number of GP
users in QoS-aware femtocell networks. Compared to existing
algorithms, we take into account the backhaul delay, channel
variation, signaling overhead as well as the computational
complexity. We have also verified the performance of the
proposed algorithm in terms of percentage of GP users and
SSR through extensive simulations .
ACKNOWLEDGMENT
The author would like to appreciate Prof. Hongyan
Li for her contributions. This work is supported by
the National Science Foundation (61231008), National
S&T Major Project (2011ZX03005-004, 2011ZX03004-
003, 2013ZX03004007-003, 2011ZX03005-003-03), Shaanx-
i 13115 Project (2010ZDKG-26), National Basic Research
Program of China (2009CB320404), Program for Changjiang
Scholars and Innovative Research Team in University
(IRT0852), the 111 Project (B08038) and State Key Labo-
ratory Foundation (ISN1002005, ISN090305).
REFERENCES
[1] J. G. Andrews, H. Claussen, M. Dohler, S. Rangan, M. C. Reed,
“Femtocells: past, present, and future,” IEEE J. Sel. Areas Commun.,
vol. 30, no. 3, pp. 497 - 508, Apr. 2012.
[2] A. Barbieri, A. Damnjanovic, T. Ji, J. Montojo, Y. Wei, D. P. Malladi, O.
Song, and G. Horn, “LTE Femtocells: System Design and Performance
Analysis,” IEEE J. Sel. Areas Commun., vol. 30, no. 3, pp. 586-594,
Apr. 2012.
[3] H. Hariyanto, R. Wulansari, Adit Kurniawan and Hendrawan, “Femtocell
Performance Over Non-SLA xDSL Access Network,” Mobile Networks,
May 9, 2012.
[4] K. Lee, H. Lee and D.-H. Cho, “Collaborative Resource Allocation for
Self-Healing in Self-Organizing Networks,” in Proc. IEEE ICC, 2011,
pp. 1-5.
[5] M. Pischella and J.-C. Belfiore, “Resource allocation for QoS-aware
OFDMA using distributed network coordination,” IEEE Trans. on Ve-
hicular Technology, vol. 58, no. 4, pp. 1766-1775, May 2009.
[6] R. Madan, J. Borran, A. Sampath, N. Bhushan, A. Khandekar, and T.
Ji, “Cell Association and Interference Coordination in Heterogeneous
LTE-A Cellular Networks,” IEEE J. Sel. Areas Commun., vol. 28, no.
9, Dec. 2010.
[7] K. Son, S. Lee, Y. Yi, and S. Chong, “REFIM: A Practical Interference
Management in Heterogeneous Wireless Access Networks,” IEEE J. Sel.
Areas Commun., vol. 29, no. 6, pp. 1260–1272, Jun. 2011.
[8] T. Lan, K. Sinkar, L. Kant, K. Kerpez, “Resource Allocation and
Performance Study for LTE Networks Integrated with Femtocells,” in
Proc. IEEE GLOBECOM, 2011, pp. 1-6
[9] J. Kim and D. Cho, “A joint power and subchannel allocation scheme
maximizing system capacity in dense femtocell downlink systems,” in
Proc. IEEE PIMRC, 2009, pp. 1381-1385.
[10] A. Ladanyi, D. Lopez-Perez, A. Juttner, Xiaoli Chu, Jie Zhang, “Dis-
tributed resource allocation for femtocell interference coordination via
power minimisation,” in Proc. IEEE GLOBECOM Workshops (GC
Wkshps), 2011, pp. 744 – 749.
[11] D. Lopez-Perez, A. Ladanyi, A. Juttner, H. Rivano, Jie Zhang, “Opti-
mization method for the joint allocation of modulation schemes, coding
rates, resource blocks and power in self-organizing LTE networks,” in
Proc. IEEE INFOCOM, 2011, pp.111-115.
[12] M. S. Al Bashar, Z. Ding, and Y. Li, “QoS aware resource allocation
for heterogeneous multiuser OFDM wireless networks,” in Proc. IEEE
SPAWC, 2008, pp.535-539.
[13] A. Hatoum, Rami Langar, N. Aitsaadi, and G. Pujolle, “Q-FCRA: QoS-
based OFDMA Femtocell Resource Allocation Algorithm,” in Proc.
IEEE ICC, 2012, pp. 5151-5156.
[14] G. Li, H. Liu, “Downlink Radio Resource Allocation for Multi-cell
OFDMA System,” IEEE Trans. On Wireless Commun., vol. 5, no. 12,
pp. 3451-3459, December 2006.
[15] Alcatel-Lucent, picoChip Designs, Vodafone, “R4-092042: Simulation
assumptions and parameters for FDD HeNB RF requirements,” in 3GPP
RAN 4, May 2009.
[16] A. Schrijver, “Theory of linear and integer programming,” Wiley, 1986.
[17] P. Liu, J. Li, H. Li, and K. Wang, “An Iteration Resource Allocation
Method to Maximize Number of Users with QoS Demand in Femtocell
Networks,” IEEE ICCC 2013-WCS.