Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Wireless Netw

DOI 10.1007/s11276-016-1331-y

A conflict avoidance scheme between mobility load balancing


and mobility robustness optimization in self-organizing networks
Miaona Huang1 Jun Chen2

Springer Science+Business Media New York 2016

Abstract In the self-organizing networks, mobility load


balancing (MLB) and mobility robustness optimization are
two significant functions. There is a close relationship
between them, as they both adjust the handover parameters
to achieve their respective goals. The conflict may happen
when both of them adjust the same handover parameters in
the opposite directions. Conflict avoidance methods have
been proposed in the existing literature. However, all of the
existing methods cannot get the optimum values of handover parameters. Moreover, the load distribution of the
neighbor cells is neglected, which has a great impact on the
network performance. To address these issues, an effective
scheme based on the load level of neighbor cells is presented. Firstly, the objectives for MLB are designed and
the MLB problem is formulated as a linear programming
problem, which can be readily solved by the well-established methods. Furthermore, considering the load distribution of the neighbor cells, the appropriate values of
handover parameters for MLB can be obtained. Finally, we
provide the framework of MLB procedures. The simulation
results verify the performance of the proposed
scheme outperforms the exiting methods.
Keywords Conflict avoidance  Linear programming  Cell
individual offset (CIO)  Self-organizing networks (SON) 
Long term evolution (LTE)

& Miaona Huang


hmn830312@126.com
1

Dongguan University of Technology, Dongguan 523808,


China

Huawei Technologies, Shenzhen 518000, China

1 Introduction
The 3rd Generation Partnership Project (3GPP) has
specified Long Term Evolution (LTE) [1] as a new radio
access technique. Compared to the 3G network system,
LTE can offer the better network performances, such as
the greater capacity, higher throughput and so on. However, with the ever-growing wireless mobile traffic,
deployment and maintenance of cellular mobile networks
by human operation are becoming more and more complex, capital-costing and time-consuming. To simplify
the management of the network and lower down the
maintenance cost, self-organizing network (SON) is
introduced to LTE which is considered as a promising
technique with automate mechanisms instead of manual
operations [2].
Both mobility robustness optimization (MRO) and
mobility load balancing (MLB) are crucial functions of
SON. MRO aims to decrease the radio link failure (RLF)
and the unnecessary handovers [3]. Too early handover or
too late handover can cause the RLF and the unnecessary
handover refers to the ping-pong handover. Meanwhile, the
objective of MLB is to transfer traffic from the edge of
overloaded cells to their under-loaded neighboring cells,
aiming to improve the network performance [4]. Although
plenty of prior literatures research on how to optimize
MLB or MRO [59], few researches focus on the conflict
avoidance between MLB and MRO. Motivated by these
observations, this paper investigates the conflict avoidance
between MLB and MRO.
Although MRO and MLB are two independent modules,
they both adjust the same handover parameters to achieve
their respective objectives. So conflict may happen when
they adjust the same parameters in opposite directions,
which can cause deadlock and deteriorate the network

123

Wireless Netw

performance [10, 11]. In order to address the problem,


3GPP proposed that MLB should have the higher priority
[12]. That is to say, MRO should be paused when MLB is
working. However, in essence, the conflict problem is not
solved. Ref. [13] proposed a coordination mechanism to
mitigate the conflict problem between MLB and MRO,
which could alleviate the disadvantageous impact of conflict between MLB and MRO by means of adjusting the
handover parameters. The authors in Refs. [14, 15] set an
allowable range for MLB in which the conflict problem can
be prevented. However, all of these methods cannot get the
specific and optimum values but just a range for the handover parameters. Thus, MLB has to change the values of
handover parameters step by step over a wide range. This is
unpractical in practice, since it is time consuming and can
lead to poor users performance. Moreover, the distribution
of the neighbor cells load was not considered. The MLB
function in the overloaded cells would do their utmost to
transfer the traffic to their underloaded neighboring cells,
as a result, the original light-loaded cells may become
overloaded which can seriously affect the network
performances.
To solve these problems and obtain the specific and
proper values of handover parameters, a conflict solution
based on load distribution of the neighbor cells is presented, which generalizes the results in [14, 15]. In particular, the main contributions of this work are summarized
as follows.
1.

2.

3.

Based on optimization theory, the objectives for MLB


are designed with the allowed ranges of handover
parameters for MRO as the constraints. The MLB
problem is formulated as a linear programming
problem, which can be readily solved by the wellestablished methods. The optimum values of handover
parameters for MLB can be obtained, which can
mitigate the load of the overloaded cells most
effectively.
In order to achieve more reasonable load distribution
among neighbor cells, we further propose an effective
scheme to obtain the appropriate values of handover
parameters for MLB. The load level of the neighbor
cells is considered, which is neglected in the existing
literature.
In order to bring the proposed scheme into effect, the
framework of the MLB procedures is designed.

The rest of this paper is organized as follows: Sect. 2


describes the system model and the operation principle
of MRO and MLB is introduced in Sect. 3. The proposed scheme is presented in Sect. 4. Section 5 evaluates the performances of the proposed scheme by LTE
system level simulation. The work is concluded in
Sect. 6.

123

2 System model
2.1 Resource assignment
Without loss of generality, the considered LTE multi-cell
network is shown in Fig. 1, where the frequency reuse
factor is 1 and all cells have the same amount of time
frequency resources. Each cell and its corresponding users
is served by an eNodeB. We do not consider that a user
might get Resource Blocks from more than one base station. In LTE network, Physical resource block (PRB) is the
basic unit that can be assigned to users [17]. Let N denote
the set of all cells. L represents the set of PRBs per cell. We
use set K to denote the set of users in the network.
The received signal-to-interference-and-noise-ratio
(SINR) for user k from the serving cell i on PRB l denoted
as SINRi;k;l can be expressed by
SINRi;k;l

N0

P g
Pi i;k;l
xj;l  Pj  gj;k

j2N;j6i

where gi;k;l denotes the channel gain of the lth PRB


between eNodeB i and user k. Pi (or Pj ) is the transmit
power of eNodeB i (or j). N0 represents the power of
Additive White Gaussian Noise. The variable xj;l 2 f0; 1g
is a binary indicator. If PRB l is used in cell j, xj;l equals to
1, otherwise xj;l 0.
Based on the Shannon theorem, we can obtain the
transmitting data rate of user k on PRB l:


2
Ri;k;l B log2 1 SINRi;k;l
where B is the bandwidth of each PRB.

Fig. 1 System model

Wireless Netw

Let Dk denote the required data rate of user k, N k denote


the required number of PRBs to satisfy the data rate of user

M i + ( H i CIOi , j )

Mj

k. Then the number of the required PRBs N k can be


expressed as:

X
_

3
R  Dk g
N k fj Sk j 
 l2S i;k;l

Mi

where Sk denotes the set of PRBs allocated to user k which


is made up by assigning the available PRBs one by one to
user k randomly until the achievable transmitting data rate
of user k is larger than Dk [9]. jSk j indicates the cardinality
of Sk .
In practice, the actual number Nk of PRBs allocated to
user k yields:
(_
_
N k ; Ni;left  N k
Nk
4
_
0 Ni;left \N k
where Ni;left represents the amount of PRBs left for
accessing users.
In accordance with Eq. (4), users will not be allowed to
access the network when there is not enough resources to
meet their required data rate.
2.2 Load measurement
The cell load is a critical parameter which has a great
impact on the network performances when performing
MLB.
The load (or resource utilization) of cell i can be defined
as the utilization of the total PRBs in the cell. Then the load
of cell i, qi can be calculated as follows:
P
Ii;k Nk
qi k2K
5
jLj
where jLj denotes the cardinality of set L and is the total
amount of PRBs in a cell. A variable Ii;k is employed to
indicate the affiliation of users. Ii;k equals to 1 if the user k
is associated to cell i, otherwise Ii;k 0.
Considering the implementation, we will not allocate
resources to new users when there are not adequate PRBs
to satisfy their data rate. Thus, the load of each cell is \1.

3 Description of handover in LTE


3.1 Handover triggering procedure in LTE
The 3GPP A3 event presents the procedures of the handover (HO) [18]. As shown in Fig. 2, it will be triggered if
the following condition (6) is satisfied for a specific time
interval, TTT (time to trigger).

TTT
A
eNB i

B C

UE moves from cell i to cell j

Position

eNB

Fig. 2 Event A3 in LTE

Mj [ Mi Hi  CIOi;j

where Mi and Mj denote the measurement signal strength


of cell i and cell j, respectively. Hi is the hysteresis
parameter for event A3. CIOi;j is the cell individual offset
(CIO) set by cell i for cell j and is the handover parameter
to be adjusted to balance the load among cells in our
proposed scheme.
As shown in Fig. 2, the entering condition of event A3
begins to be met at point A and it will be satisfied for a time
period of TTT while user equipment (UE) moves from A to
B. UE starts to handover at point B and finishes the handover process at point C.
3.2 Mobility load balance
MLB aims to balance load among neighbor cells [12].
MLB can work in the following three modes.
1.
2.
3.

Relax the handover conditions to make UEs handover


from overloaded cell to its neighbor cell in advance.
Tighten the handover conditions to postpone the UEs
handover from neighbor cell to the overloaded cell.
Do the above two modes at the same time.

The MLB chooses adjusting one of the handover


parameters which is the cell individual offset (CIO) to
achieve its optimization objective. In Fig. 2, without loss of
generality, we assume that cell i is overloaded while its
neighbor cell j is under-loaded. Let CIOi;j denote the CIO
set by cell i for cell j and CIOj;i denote the CIO set by cell j
for cell i. The MLB function in cell i will select cell j to
balance the load and increase the parameter CIOi;j to make
UEs handover from cell i to cell j earlier. Meanwhile, in
order to postpone the handover from cell j to cell i, cell i
informs cell j to decrease CIOj;i . Thus, the increment of the
extra load in cell i will be reduced on the whole.

123

Wireless Netw

3.3 Mobility robustness optimization

3.4 Conflict between MLB and MRO

In the 3GPP standards, the main objective of MRO is to


lower down the radio link failure (RLF) rate and cut
down the unnecessary handovers [14]. Specifically, too
early handover and too late handover can give rise to
RLF and here the unnecessary handover means ping-pong
handover.

In this paragraph, we will explain how the conflict happens


between MLB and MRO. As shown in Fig. 4, without loss
of generality, we assume UEs move from cell i to cell j and
cell i and cell j are heavy loaded and light loaded, respectively. Since the load in cell i is heavy, it chooses cell j to
balance its load. Both MLB and MRO adjust CIO to achieve
their respective optimal objectives. Let CIOi;j and CIOj;i are
the cell individual offset which are set by cell i for cell j and
cell j for cell i, respectively. As UEs move from cell i to cell
j, the MLB in cell i will increase the corresponding CIOi;j to
CIOi;j;MLB in order to handover these UEs to cell j as soon as
possible. Unfortunately, the operation of MLB can not
guarantee that it can choose a suitable value for CIOi;j;MLB
and an improper CIOi;j;MLB will lead to too many RLFs
which will trigger MRO. As a results, MRO will decrease
CIOi;j to CIOi;j;MRO \CIOi;j;MLB , and then MLB will
increase CIOi;j immediately. This will leads to deadlock and
this is the conflict between MLB and MRO. This conflict
can greatly degrade the performances of the two functions,
resulting in network resources wasted seriously.

1.

2.

3.

Too early handover: When UE moves from cell i to


cell j, the handover is triggered after the entering
condition for A3 event which is just satisfied for one
TTT (as shown in Fig. 2). However, on account of
improper setting of handover parameters, the
received signal strength of UE from cell j is too
weak for UE to establish a radio link with cell j. So
the RLF between the UE and cell j happens during
the handover process or immediately after UE
successfully handovers to cell j. In that case, UE
will make a connection with cell i.
Too late handover: When UE moves from cell i to cell
j, the handover is triggered at a particular point (as the
point B in Fig. 2). Due to the improper values of
handover parameters, the received signal strength of
UE from cell i is too bad before the UE successfully
handovers to cell j, so that the RLF between UE and
cell i occurs. In that case, the UE will make a
connection to cell j.
Ping-pong handover: As is shown in Fig. 3, when UEs
position is between point X and point Y, the A3 events
for handover from cell i to cell j and handover from
cell j to cell i are both satisfied. Therefore, after UE
handover from cell i to cell j, it will handover back to
cell i immediately. Although ping-pong handover does
not influence users QoS (Quality of Service) too
much, it brings about a serious waste of the network
resources.

3.5 Allowable range for MLB


The main function of MLB is to distribute the uneven
traffic among neighbor cells, by adjusting CIO. The handover problems are may attributed to the inappropriate
operations of MLB. If MLB adjusts the handover parameters in a correct way, handover problems will not happen
and the conflict with the MRO can be avoided. Accordingly, an allowable range of handover parameters for MLB
has been set in the existing scheme [14].
Assume that cell i is overloaded and the MLB function
in cell i chooses its light loaded neighbor cell j to balance
the load, cell i will increase CIOi;j and meanwhile inform

Overlapped area

Mi

MLB

MRO

Mj

Mj

Mi

M j + ( H j CIO j ,i )

M i + ( H i CIOi , j )

M i + ( H i CIOi , j ,MRO )

M i + ( H i CIOi , j ,MLB )

X
eNB i

Position

Position

UE moves from cell i to cell j

Fig. 3 Ping-pong handover

123

eNB

eNB i

UE moves from cell i to cell j

Fig. 4 Conflict between MLB and MRO

eNB

Wireless Netw

cell j to decrease CIOj;i . After the operation of MLB,


assume that CIOi;j and CIOj;i are adjusted to CIO0i;j and
CIO0j;i , respectively. As presented in Ref. [14], the minimum value Hi  CIOi;j;early and maximum value Hi 
CIOi;j;late should be recorded by the MRO function in cell
i, which content inequality (1) that cannot cause RLF.
Therefore, the adjustment of Hi  CIOi;j is confined
between the minimum value and the maximum value as the
shaded part shown in Fig. 5. In other words, if Hi 
CIOi;j is smaller than Hi  CIOi;j;early , too early handover problem would occur. Similarly, if Hi  CIOi;j is
larger than Hi  CIOi;j;late , too late handover problem
would happen. In the same way, the MRO function in cell j
would record the minimum value Hj  CIOj;i;early and
maximum value Hj  CIOj;i;late of entering condition for
A3 event that cannot generate RLF. And the adjustment of
Hj  CIOj;i is also limited between the minimum value
and the maximum value as the shaded part shown in Fig. 6.
Moreover, ping-pong handovers should be prevented. It
means to eliminate the overlapped area as shown in Fig. 3.
To avoid the conflict between MLB and MRO, the
restrictions on the adjustment for CIO0i;j and CIO0j;i can be
summarized as follows [14]:
8
0
0
< CIOi;j CIOj;i  Hi Hj
0
CIOi;j  CIOi;j  CIOi;j;early
7
: CIO
0
j;i;late  CIOj;i  CIOj;i
Ref. [14] can get the allowed range for CIO0i;j and CIO0j;i
as is shown in inequalities (7). But it cannot get the specific
and optimum values for CIO0i;j and CIO0j;i , respectively.
MLB has to change the values of CIO0i;j and CIO0j;i step by
step over a wide range. This is unpractical in practice, since
it is time consuming and can lead to poor users
performance.

M i + ( H i CIOi , j ,late )

Mi

Mj

M j + ( H j CIO j ,i ,late )

Mi

Mj

M j + ( H j CIO j ,i ,early )

Position
eNB i

UE moves from cell j to cell i

4 The proposed scheme


In this section, based on optimization theory, specially the
theory of linear program, we propose an efficient scheme to
avoid the conflict between MLB and MRO. The proposed
scheme can obtain the specific and proper values of handover parameters according to the load distribution of the
neighbor cells, which is not considered in the existing
literature.
4.1 Optimization problem formulation for MLB
MLB aims to balance the load among neighbor cells.
Assume that cell i has heavy load and its neighbor cell j has
light load. The MLB function in cell i will choose cell j to
balance the load. The MLB in cell i will relax the handover
conditions by adjusting CIOi;j larger to CIO0i;j so that UEs
in cell i can handover to cell j in advance. Then the load of
cell i will be decreased quickly. Meanwhile, cell i informs
cell j to adjust CIOj;i lower to CIO0j;i . Therefore the handovers from cell j to cell i will be postponed and the
increase of the additional load of cell i will be slowed
down. Accordingly, one can arrive at the following optimization objectives for MLB:

M i + ( H i CIOi , j ,early )

The larger CIO0i;j , the earlier for UEs to be handed over


from cell i to cell j. Then the optimization objective for
advancing UEs in cell i handed over to cell j as soon as
possible can be expressed as:
max CIO0i;j

2.
eNB i

UE moves from cell i to cell j


eNB

Fig. 5 Allowable range for UEs handover from cell i to cell j

Fig. 6 Allowable range for UEs handover from cell j to cell i

1.

Position

eNB

To postpone the UEs handover from cell j to cell i, the


value of CIO0j;i should be as small as possible. Then the
optimization objective for postponing UEs handed
over from cell j to cell i as late as possible can be
written as:

123

Wireless Netw

min CIO0j;i

We try to maximize CIO0i;j and minimize CIO0j;i simultaneously. Then, the problem is equivalent to the following
multi-objective optimization problem, subject to the constraints in (7). Thus, we have:
P1

max CIO0i;j ; CIO0j;i T


8
0
0
< CIOi;j CIOj;i  Hi Hj
0
s:t: CIOi;j  CIOi;j  CIOi;j;early
: CIO
0
j;i;late  CIOj;i  CIOj;i

10

where T denotes the transpose of a vector. The wellknown linear weighted sum method can be utilized to
construct the two objectives into a single aggregate
objective function. 0\a\1 and 1  a are made as the
weighted coefficients of the two objectives, respectively.
Then the above problem (P1) can be transformed as:
P2

max aCIO0i;j  1  aCIO0j;i


8
0
0
< CIOi;j CIOj;i  Hi Hj
0
s:t: CIOi;j  CIOi;j  CIOi;j;early
: CIO
0
j;i;late  CIOj;i  CIOj;i

11

P3

CIO0j;i

max

8
0
0
CIO

CIO
<
i;j
j;i  Hi Hj
0
CIO

CIO
s:t:
i;j
i;j  CIOi;j;early
: CIO
0

CIO
j;i;late
j;i  CIOj;i

From the above subsection, the allowable range for CIO0i;j


is [CIOi;j , CIO0i;j;max ] and the allowable range for CIO0j;i is
[CIO0j;i;min , CIOj;i ]. That is to say, when UE moves from
cell i to cell j, CIOi;j should be adjusted to CIO0i;j which is
between the value of CIOi;j and CIO0i;j;max to avoid the too
early handover. When UE moves from cell j to cell i,CIOj;i
should be adjusted to CIO0j;i which is between the value of
CIO0j;i;min and CIOj;i to prevent the too late handover.
Considering the load level of the serving cell and the
target cell, the reasonable values of CIO0i;j and CIO0j;i can be
calculated as follows:
qj
13
Di;j CIOi;j;max  CIOi;j  1 
qi
CIO0i;j CIOi;j Di;j

14

Similarly,
Dj;i CIOj;i  CIOj;i;min  1 

qj

qi

12

15

then,
CIO0j;i CIOj;i  Dj;i

The objective function of (P3) is linear and the constraints are linear too. So the optimization problem (P3) is a
linear programming, which can be readily solved by the
well established methods such as primaldual interiorpoint method using freely convex solvers, e.g. SDPT3
[16, 19]. Thus, the optimal value exists, denoted as
CIO0i;j;max  CIO0j;i;min . Let CIO0i;j;max and CIO0j;i;min represent the maximum value of CIO0i;j and the minimum
value of CIO0j;i , respectively.
In terms of MLB, the optimum values of CIO0i;j and
CIO0j;i are CIO0i;j;max and CIO0j;i;min , respectively, which can
mitigate the load of the overloaded cell i most effectively.
However, it doesnt take the load distribution of the serving
cell and the target cell into account. Thus, it can easily lead
to cell i to be underutilized while cell j is overloaded.
Because cell i only considers its load condition at the cost
of suppressing other cells load requirements. In the following subsection, the reasonable values of CIO0i;j and

123

4.2 Obtain the reasonable values of handover


parameters based on the cell load

then,

Note that a only appears in the objective function, not in


the constraints, thus the value of a is independent of the
intervals of CIO. In other words, it has no effect on the
values of CIO0i;j and CIO0j;i when the aggregate objective
function achieves the maximum. Then, (P2) is equivalent
to the following problem:
CIO0i;j

CIO0j;i will be obtained further depending on the load distribution of the serving cell and the target cell.

16

where qi ,qj represent the load (resource utilization) of cell i


and cell j, respectively. The step Di;j and Dj;i are both
determined by the load of cell i and cell j.The smaller the

value of (qj qi ), the larger the values of Di;j and Dj;i .
4.3 MLB procedure in the proposed scheme
The proposed scheme is triggered by MLB. In order to
bring the proposed scheme into effect, procedures for MLB
are designed as shown in Fig. 7. The procedures can be
described as follows [20]:
1.

2.
3.
4.

Cells require their MRO functions to record the


minimum value and maximum value of entering
condition for event A3 that cannot cause RLFs;
Cells detect their respective cell load and neighbor
cells exchange the load information with each other;
If some cell is overloaded (e.g. cell i), the MLB
function in cell i will be triggered;
Cell i detects the eligible target cell (e.g. cell j) and
Cell i inquires about the handover parameters of cell
jHj ; CIOj;i ; CIOj;i;late ;

Wireless Netw

eNB i

eNB j

MRO record (CIOi , j ,early , CIOi , j ,late )

MRO record (CIO j ,i ,early , CIO j ,i ,late )

Load statistics

Load statistics
Load information exchange

MLB is triggered
and select eNB j
to balance the load
Inquire the handover parameters
response to the inquiry ( H j , CIO j ,i , CIO j ,i ,late )

Work out the reasonable


values of (CIOi , j , CIO j ,i )
request eNB j to adjust CIO j ,i
CIO j ,i
accept the request

Table 1 The setting of TTT


Speed of UE (V/km/h)

015

1530

3060

Higher than 60

TTT(ms)

128

100

80

40

shown in Table 1. In the simulation, the value of TTT is set


to be 100 ms. The MRO should be activated when at least
two RLFs occur in 200 s, or more than three ping-pong
handovers in 50 s. When each TTI begins, the network will
detect the load of each cell. The MLB will be activated
when the resource utilization of its cell reaches up to 70 %.
The neighbor cell whose load is lower than 50 % will be
selected to balance the load. Considering the MRO, the
handover parameters of the overloaded cell and its neighbor cell will be readjusted to proper values in accordance
with the proposed scheme. The length of one TTI is 1 ms.
More simulation parameters can be found in Table 2 [23].
5.1 System metrics

Adjust
CIOi , j
CIOi , j

Adjust
CIO j ,i
CIO j ,i

Fig. 7 Procedures for MLB

5.
6.

7.
8.
9.

Cell j responses to the inquiry by sending the handover


parameters setting Hj ; CIOj;i ; CIOj;i;late to cell i;
Based
on
the
handover
parameters
Hi ; CIOi;j ; CIOi;j;early and Hj ; CIOj;i ; CIOj;i;late of
the two cells, Cell i calculates and obtains the proper
values of CIO0i;j ; CIO0j;i according to the proposed
scheme;
Cell i requests cell j to adjust CIOj;i to CIO0j;i ;
Cell j accepts the request and adjusts CIOj;i to CIO0j;i ;
Cell i adjusts CIOi:j to CIO0i;j .

5 Simulation and results


In this section, the proposed scheme is evaluated by the
LTE system level simulator including MRO, MLB,
resource allocation, call admission control and handover
function [21]. In simulation topology, nineteen hexagonal
cells with omindirectinal antenna are considered as shown
in Fig. 1. Call arrival follows the poisson process. UEs
distribute randomly in the simulation topology. The online
time of each UE is 20 s, and after 20 s idle state, the UEs
will try to access the network once again. Queuing Priority
(QP) scheme (which is equivalent to first in first out
scheme) is adopted in the call admission control. In the
simulation, UEs move straightly in random direction with
the speed of 25 km/h. The appropriated setting of TTT in
accordance with the speed was validated in Ref. [22] as

Some related system metrics are introduced in this subsection with the purpose of showing the simulation results
more clearly.
5.1.1 Call blocking ratio (CBR)
The definition of CBR is showed as follows:
CBR

blocked calls
total initiated calls

17

where blocked calls represents the number of calls that are


not accepted by the cells due to the limitation of network
Table 2 Simulation parameters
Parameter

Assumption

Cell layout

Regular hexagonal grid, 19 cell


sites with an omnidirectional
antenna

Inter-site distance

500 m

Carrier frequency
System bandwidth

2 GHz
5 MHz

transmission power of all eNodeBs

43dBm

Distance-dependent path loss

128.1 ? 37.6log10(R), R in km

Shadowing standard deviation

10 dB

Correlation distance of shadowing

10 m

Hysteresis parameter (H)

3 dB

Initial cell specific offset (CIO)

0 dB

Time to Trigger (TTT)

100 ms

UE speed

25 km/h

Traffic model on UE

Constant Bit Rate, 328 kb/s

Simulation time

500 s

123

Wireless Netw

resource; total initiated calls denotes the total number of


initiated calls in the whole simulation time.
5.1.2 Load balancing index
Let qi t denote the load (or resource utilization) of cell i.
To measure the status of load balancing of the entire network, Jains fairness index [24] is used, and represented as
follows:
P
qi t2
P
et
18
jN j q2i t
where jN j is the number of cells. The value of load balance
index et is between 1=jN j; 1. Larger et represents more
reasonable load distribution among cells.

of the neighbor cells into account. The performance of


MRO function is evaluated by the RLF rate and ping-pong
handover rate. The call blocking ratio, load balancing
index, network average throughput and average number of
overloaded cells per TTI are used to evaluated the performance of MLB. These results are obtained by the Monte
Carlo [25] and the simulation times is 1500.
As shown in Figs. 8 and 9, the RLF rate and ping-pong
handover rate in both the proposed scheme and the
improved method are much lower than that of the original
method. Simply because the original method do not avoid
the conflict between MLB and MRO. It demonstrates that
the conflict avoidance method is essential. In Figs. 8 and 9,
we can find that the RLF rate and ping pong rate in the
proposed scheme are also lower than those in the improved
method. The reason is that the MLB in the proposed

5.1.3 Radio link failure (RLF) rate


14

The definition of RLF rate is showed in Eq. (19)


19

where failed HOs denotes the amount of failed handovers


including the too early handovers and too late handovers.
total triggered HOs represents the total number of triggered handovers in the whole simulation.

12

RLF rate (%)

failed HOs
RLF rate
total triggered HOs

20

2
120

The simulation results are presented for performance


comparison among three cases, which are the original
method [2, 12], the improved method [14] and the proposed scheme. In the original method, both MLB and MRO
modify the handover parameters independently to achieve
their respective objectives. In the improved method, the
MLB adjusts the handover parameters in the allowed range
which can avoid conflict with the MRO. In the proposed
scheme, the MLB adjusts the handover parameters to
proper values which can not only avoid the conflict
between MLB and MRO but also take the load distribution

240

300

360

420

480

Fig. 8 RLF rate


40
35

Ping pong handover rate (%)

5.2 Performance evaluation

180

The number of UEs in the simulation topology

where ping pong HOs denotes the amount of ping-pong


handovers. total triggered HOs represents the total number of triggered handovers in the whole simulation.

123

After UE handover from cell i to cell j, it will handover


back to cell i in less than 1 s. In that case, the handover will
be regarded as ping-pong handover. The ping-pong handover rate is defined as:
ping pong HOs
total triggered HOs

10

5.1.4 Ping-pong handover rate

ping pong rate

original
improved
proposed

original
improved
proposed

30
25
20
15
10
5
0
120

180

240

300

360

420

The number of UEs in the simulation topology

Fig. 9 Ping-pong handover rate

480

Wireless Netw

Average throughput of all cells(Mbps)

75

original
improved
proposed

70
65
60
55
50
45
40
35
30
25
120

180

240

300

360

420

480

The number of UEs in the simulation topology

Fig. 11 Average throughput of all cells

5.5

Average number of overloaded cells per tti

scheme adjusts the handover parameters to more proper


values according to both the load level of neighbor cells
and the allowed range for MRO, while the MLB in the
improved method modifies the handover parameters without considering the load distribution of the neighbor cells.
And thus the adjustment may be inappropriate.
Figure 10 shows the call blocking ratio (CBR) varies
with different number of UEs in the network. From Fig. 10,
we can find that the CBR of the proposed scheme is much
better than that of the original method, and also lower than
that in the improved method. As the MLB in the proposed
scheme adjusts the handover parameters to proper values
according to both the load level of neighbor cells and the
allowed range for MRO, so that the adjustment is more
reasonable. Thus, the congestion situation in the overloaded cells can be relieved in a more reasonable way.
While the improved method does not consider the load
level of the neighbor cells when performs MLB. The MLB
function in the overloaded cells would do their utmost to
transfer the traffic to their underloaded neighboring cells.
As a result, the original light-loaded cells may become
overloaded, resulting in more blocking calls. Therefore, the
CBR in the proposed scheme is lower than that in the
improved method.
Figures 11 and 12 show that the proposed scheme has
the largest network average throughput and the least
average number of overloaded cells per TTI. Figure 13
shows the load balancing indexes of the original method,
the improved method and the proposed scheme for 360
UEs in the simulation topology. From Fig. 13, one can
conclude that the load balancing index of the proposed
scheme is higher than that of the improved method and the
original method. Compared with the improved method, the
MLB in the proposed scheme adjusts the handover

original
improved
proposed

5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
120

180

240

300

360

420

480

The number of UEs in the simulation topology

Fig. 12 Average number of overloaded cells per TTI

Call blocking ratio (%)

20

original
improved
proposed

original
improved
proposed

0.9

Load balancing index (%)

25

15

10

0.8
0.7
0.6
0.5

0.4
0
120

180

240

300

360

420

The number of UEs in the simulation topology

Fig. 10 Call Blocking Rate

480

0.3
0

50

100

150

200

250

300

350

400

450

500

Simulation time

Fig. 13 Load balancing index

123

Wireless Netw

parameters with considering the load level of cell load, so


that the adjustment is more reasonable. Thus, the load
distribution among neighbor cells in the proposed
scheme can achieve more rational and more balanced than
that in the improved method. Meanwhile more UEs can get
sufficient resources from their serving cells in the proposed
scheme. While the improved method just give an allowed
range without giving any consideration with the load level
of the neighbor cells. It can easily lead to other neighbor
cells overloaded, but this situation could not happen in the
proposed scheme. Thus, the proposed scheme can achieve
better network performances than the improved method.
From Figs. 8, 9, 10, 11, 12 and Fig. 13, we can see that
the network performances of the original method are the
worst. The reason is that: in the original method, both MLB
and MRO modify the handover parameters independently
to achieve their respective objectives. As the MLB function
in the original method modifies the handover parameters
without considering the allowable range for MRO and the
load distribution of the neighbor cells, the MLB functions
in the overloaded cells would do their utmost to transfer the
traffic to their underloaded neighboring cells. The adjustment may be inappropriate. As a result, the original lightloaded cells may become overloaded and many too early or
too late handovers happens, leading to more blocking calls,
more RLFs and more overloaded cells in the network. Due
to the network resources cannot be effectively used, the
network average throughput is severely deteriorated. So we
can draw a conclusion that the conflict avoidance method
between MLB and MRO is vital and the load distribution
among neighbor cells should also be considered when
performing MLB.

6 Conclusion
In this paper, an effective scheme was proposed to solve
the conflict between MLB and MRO for the SON. Firstly,
based on optimization theory, the objectives for MLB are
designed and the MLB problem is formulated as a linear
programming problem, which can be readily solved by the
well established methods. The optimum values of handover
parameters for MLB can be obtained, which can mitigate
the load of the overloaded cells more effectively. Furthermore, in order to achieve more reasonable the load
distribution among neighbor cells, we further improve the
MLB algorithm considering the adjacent cells load condition and the proper values of handover parameters for
MLB can be obtained. Finally, to apply the proposed
scheme in practice, the framework of MLB procedures has
been designed. The simulation results indicate that our
proposed scheme outperforms the existing methods. Our
future work will generalize the proposed method to the

123

heterogeneous network such as the ones in LTE-A consisting of macro cell and a set of low power base stations
(e.g. microcells, picocells, femtocells) and ultra-dense
network, which are promising technologies for the
upcoming 5G system.
Acknowledgments This work was supported by the National Science
Foundation of China (NSFC) (Grand No. 61340035).

References
1. GPP, TR 25.913 V9.0.0. Technical Specification Group Radio
Access Network, Requirements for Evolved UTRA (E-UTRA) and
Evolved UTRAN (E-UTRAN). http://www.3gpp.org/DynaReport/
25913.htm.
2. GPP, TR 36.902 V9.3.1. Self-configuring and self-optimizing
network (SON) use cases and solutions. http://www.3gpp.org/ftp/
Specs/archive/36_series/36.902/.
3. Schroder, A., Lundqvist, H., & Nunzi, G.(2008). Distributed selfoptimization of handover for the long term evolution. In Proceedings of the 3rd International Workshop on Self-Organizing
Systems, Vienna, Austria, (pp. 281286).
4. Tiwana, M. I., Sayrac, B., & Altman, Z.(2009). Statistical
learning for automated RRM: Application to eUTRAN mobility.
In IEEE International Conference on Communications, 2009.
ICC (pp. 15).
5. Konstantinou, I., Tsoumakos, D., & Koziris, N. (2011). Fast and
cost-effective online load-balancing in distributed range-queriable systems. IEEE Transactions on Parallel and Distributed
Systems, 22(8), 13501364.
6. Tian, W. H., Zhao, Y., Zhong, Y. L., Xu, M. X., & Jing, C.
(2011). Dynamic and integrated load-balancing scheduling
algorithm for cloud data centers. China Communications, 8(6),
117126.
7. Rodoguez, J., De la Bandera, I., Munoz, P., & Barco, R. (2011).
Load balancing in a realistic urban scenario for LTE networks. In
IEEE 73th Vehicular Technology Conference, 2011. VTC 2011Spring (pp. 15).
8. Kim, H., Veciana, G. D., Yang, X. Y., & Venkatachalam, M.
(2012). Distributed-optimal user association and cell load balancing in wireless networks. IEEE/ACM Transactions on Networking, 20(1), 177190.
9. Yang, Y., Dong, W., Liu, W., & Wang, W. (2014). A unified selfoptimization mobility load balancing algorithm for LTE system.
IEICE Transactions on Communications, 97(4), 755764.
10. Li, Z. H., Wang, H., Pan, Z. W., Liu, N., & You, X. H. (2011).
Joint optimization on load balancing and network load in 3GPP
LTE multi-cell networks. In International Conference on Wireless Communications and Signal Processing, Nanjing, China (pp.
15).
11. Rodriguez, J., De la Bandera, I., Munoz, P., & Barco, R. (2011).
Load balancing in a realistic urban scenario for LTE networks. In
IEEE Vehicular Technology Conference, 2011, VTC. 2011Spring, Budapest, Hungary (pp. 15).
12. GPP, TSG RAN WG3 Meeting #64 R3-091032. Dependencies
among SON use cases and CCO priority. http://www.3gpp.org/
13. Yu, J. T., Hu, H. L., Jin, S. Y., & Zheng, X. Y. (2012). Conflict
coordination between mobility load balancing and mobility
robustness optimization. Computer Engineering, 2012(5), 3741.
14. Liu, Z. Q., Hong, P. L., Xue, K. P., & Peng, M. (2010). Conflict
avoidance between mobility robustness optimization and mobility
load balancing. In IEEE Global Telecommunications Conference,
2010 GLOBECOM (pp. 15).

Wireless Netw
15. Li, Y., Li, M., Cao, B., & Liu, W. J. (2012). A conflict avoid
method between load balancing and mobility robustness optimization in LTE. In 1st IEEE International Conference on
Communications in China, 2012. ICCC (pp. 143148).
16. Boyd, S., & Vandenberghe, L. (2004). Convex optimization.
Cambridge: Cambridge University Press.
17. GPP, TS 36.201 V12.2.0. Technical Specification Group Radio
Access Network; LTE physical layer; General description. http://
www.3gpp.org/DynaReport/36201.htm
18. GPP, TS 36.331 V13.0.0. Technical Specification Group Radio
Access Network;Evolved Universal Terrestrial Radio Access (EUTRA);Radio Resource Control (RRC); Protocol specification.
http://www.3gpp.org/DynaReport/36331.htm.
19. Tutuncu, R. H., Toh, K. C., & Todd, M. J. (2003). Solving
semidefinite quadratic-linear programs using SDPT3. Mathematical Programming, 95(2), 14364646.
20. GPP, TSG RAN WG3 Meeting #64 R3-091294. Exchange of
handover parameters directly between eNBs. http://www.3gpp.
org/.
21. Jcolom, M. T. (2010). Vienna LTE simulators system level simulator documentation. Austria: Institute of Telecommunications,
Vennia University of Technology.
22. Lee, Y., Shin, B., Lim, J., & Hong, D. (2010). Effects of time-totrigger parameter on handover performance in SON-based LTE
systems. In Proceedings of the 16th Asia-Pacific Conference on
Communications, Auckland, New Zealand (pp. 492296).
23. GPP, TR 25.814 v7.1.0. Physical layer aspects for evolved
Universal Terrestrial Radio Access (UTRA). http://www.3gpp.
org/ftp/Specs/archive/25_series/25.814/.
24. Chiu, D., & Jain, R. (1989). Analysis of the increase and decrease
algorithms for congestion avoidance in computer networks.
Computer Networks and ISDN Systems, 17(1), 114.
25. Robert, C. P., & Casella, G. (2009). Mente Carlo Statistical
Methods (2nd ed.). Beijing: Beijing Word Publishing
Corporation.

Miaona Huang was born in


Guangdong Province, China, in
1983. She received her B.S.
degree in Electrical Information
Science and Technology from
Hanshan Normal University in
2006 and M.S. degree in Optics
from South China Normal
University in 2009 and the
Ph.D. degree in Communication
Engineering from South China
University of Technology in
2014. She joined the School of
Electronic and Engineering at
Dongguan University of Technology in 2014. Her current research interests include radio resource
management in Self-organizing network, ultra dense network and
application of convex optimization theory in communication.
Jun Chen was born in Hubei
Province, China, in 1981. He
received his B.S. degree in
Electronic and automation from
Civil Aviation University of
China in 2004 and M.S. degree in
Circuit and System from South
China University of Technology
in 2010 and the Ph.D. degree in
Communication
Engineering
from South China University of
Technology in 2014. He joined
the Huawei technologies in 2014.
His research interests include
distributed antenna system, convex optimization and signal processing in wireless communications.

123

You might also like