Computational Intelligence - February 2017 - 32

Every BS and MME can only process a limited number
of paging requests per second, which means their
capacities must not be exceeded in TA planning
(so are constraints).
multi-objective genetic algorithm proposed by Deb et al.
[26]) to manage the watershed water quality [27]. Feng et al.
studied MOEA/D and its application in control-structure
integrated multi-objective design for flexible spacecrafts [29].
Liu et al. proposed an iterative power control scheme to plan
WCDMA networks by an EMO algorithm [31]. Ishibuchi et
al. investigated the performance of three EMO algorithms in
optimizing many-objective knapsack problems [32]. More references about EMO algorithms and their applications can be
found in [35].

Cell 2

ve
r

III. The TA Planning Problem
A. Problem Statement

The geographical coverage area of an LTE
network is partitioned into cells, where each
cell is served by a single BS and managed by a
single MME. The MME is used to record the
TA where the UE is registered. The essential
task of TA planning is to group cells to form a
set of TAs that can give the network relatively low location
update cost and paging cost under the condition of guaranteeing
the service quality. Every BS and MME can only process a limited number of paging requests per second, which means their
capacities must not be exceeded in TA planning (so are constraints). According to the analysis, a location update occurs only
when a mobile user crosses the TA boundary. So it is reasonable
to take the number of TA boundary crossings as the measure of
location update cost. The user movements among TAs weigh
heavily in the process of TA planning, while the geographic
environment can directly affect user movements. For instance,
regions with many roads tend to have more user mobility than
regions cut off by high mountains or similar obstacles. Therefore,
the influence of geographic environment is considered when we
undertake TA planning. On the one hand, we try to avoid situations in which groups of cells with no roads connecting them
are assigned to the same TA. On the other hand, the TA boundary should avoid crossing roads or paralleling roads in order to
reduce the so-called Ping-Pong effect.
B. Multi-objective TA Planning Model

Cell 3

Cell1

Figure 1 An example of location update between cells.

Telephone Traffic

Paging Prediction

TA Planning

Geographic Information

Number of Users

Figure 2 The main flowchart of TA planning.

IEEE ComputatIonal IntEllIgEnCE magazInE | FEbruary 2017

In general, most mobile users move on roads and thus the
number of mobile users on the roads can give a good approximation to that of all mobile users in the network. Considering
this, we assume that user movements are confined to the roads
and only consider the traffic movement on the roads. It means
that the TA boundary causes no location updates if there is no
road connecting the adjacent cells separated by the boundary.
Fig. 1 illustrates an instance of this situation in part of an LTE
network. There is a river between cell 1 and cell 2, which hinders user movements between them. The result is that no location update occurs at the boundary between cell 1 and cell 2 if
it is also set as a boundary between two TAs, according to our
hypothesis. Geographic information can be directly reflected
by the traffic flow on each grade (traffic carrying capacity) of
road of the network, and the exact traffic flow data on each
road can be obtained by using statistical method. Roads in an
LTE network can be classified into different types according
to their traffic flow in a realistic TA planning scenario. The
straightforward process of modeling TA planning problem in
this paper is illustrated in Fig. 2.
In our study, roads are roughly divided into three different
types: main roads, streets, and alleys. The main notations used in
the following equations are listed as follows:
❏ " 1, 2, ..., N , : the corresponding cell index set, where N is
the total number of cells in the network.

Table of Contents for the Digital Edition of Computational Intelligence - February 2017