Intelligent GSM Cell Coverage Analysis System

Based on GIS

Lina Lan
Beijing University of Posts and Telecommunications/School of Network Education, Beijing, China
Email: lindalan2002@sina.com

Xuerong Gou, Yunhan Xie and Meng Wu
Beijing University of Posts and Telecommunications/School of Network Education, Beijing, China
Email: xrgou@126.com



AbstractIn mobile network, a deviation of cell coverage
area influences many network performance indexes. Cell
coverage analyses are vital to network optimization. The
traditional check method is DT (Drive Test) or FSP (Field
Strength Prediction) by manpower which costs much time
and resources. This paper presents an intelligent multiple
factors analysis method on cell coverage, and designs the
relevant software system based on GIS platform. This
system derives a cell coverage analysis chart and identifies
the cells with cross-boundary coverage or poor coverage
problem by collecting a huge number of mobile phone
measure data in OMC and analyzing multiple factors based
on the measure data and the basic data of cells. The measure
data analysis aims to compute signal level distribution,
sample point distribution, category of interferences. The
basic data of cells includes neighborhood relationship,
azimuth ward, location and distance between two cells. The
base station site level can be computed from the basic data
of cells by the triangulation method. The calculation and
analysis results are presented in the map based on GIS
platform to improve visualization. This method and system
are validated by a large number of actual datasets from an
in-service GSM network. Contrast with the traditional cell
analysis method, this method and system demonstrate
advantages in intelligence, accuracy, timeliness, and
visualization.

I ndex Termsmobile network optimization, cell coverage
analysis, mobile measure data, signal level distribution,
sample point distribution, cell coverage metrics, base station
level, OMC (Operation & Maintenance Center), GIS
(Geographic Information System)

I. INTRODUCTION
Mobile network is a dynamic network. The contents of
mobile network optimization are optimizing the
allocation of the resources, adjusting the networks
parameters reasonably, and making the network run in the
best state [1]. Every cell has its own coverage area. One
of the important factors which influence the quality of the
mobile communication is cell coverage deviation such as
cross-boundary coverage, critical coverage, poor
coverage etc. Therefore, how to determine the cell
coverage area in the mobile network is vital to mobile
network optimization [2].
The mobile communicat ion coverage area commonly
refers to the region: call quality is assured in 90% scope
of the region. The field boundary is the mobile station
receiving the minimum signal level. Among the existing
technologies, there are mainly two ways to determine the
cell coverage area.
1. Drive Test [3]. That is the survey crew go along a
certain route in the serving cell and the target cells by car
or on foot, to measure the field intensity point by point by
a coverage test machine when they march forward. In this
way, they acquire the field strength distribution of the
serving cell. Constrained by factors like ground,
topography and cost, its difficult to get the field intensity
at every point in the measurement area. Only some
discrete values of the field strength distribution are
retrieved. Besides, drive test consumes large amount of
manpower and material resources and is not able to
provide real t ime data.
2. Field Strength Prediction [4]. That is using the field
strength prediction algorithm and the known field
strength distribution to predict the field strength
distribution of the area where it is hard to get the field
strength. There are some classical algorithms like
Okumura Model. In these algorithms, not only the
propagation model must conform to the actual situation
of the current network, but during the prediction process,
a large amount of measured data are also needed to be
analyzed, fitted, counted. So large amounts of data,
matching geography-data are needed if using this method,
let alone that the computational complexity is high but
the effective time is short.
Both the two methods are deficiencies, the former can
not provide a comprehensive real-time analysis, and
spend a lot of human resources and material resources;
the latter model requires high precision, and can not be
well adapted to frequent changes in the cells and various
land types.
The voice quality in digital mobile communication
system is not only determined by signal strength, but also
by the same frequency interference and mult i-path effects.
Somet imes, even if the signal strength is high, the voice
quality is still poor [5, 6]. Therefore, the cell coverage
analysis not only need to analyze the level dis tribution,
JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011 897
2011 ACADEMY PUBLISHER
doi:10.4304/jcp.6.5.897-904
but also to take into account inter-cell interference,
sample point distribution to discover various cell
coverage issues, such as cross-boundary coverage, poor
coverage and so on.
The cross-boundary coverage is prone to cause island
effect, switch failure, dropped calls and other serious
impacts on communication quality. Therefore, the
identification of the cross-boundary coverage is the focus
of coverage problems. The traditional method requires
much human efforts on data check and choosing which
depends on the experiences of the network optimizat ion
engineer to confirm the problem cell. The approach has
low efficiency, consuming a lot of human resources.
This paper presents a method to analyze mult iple
factors to verify the cell coverage area and problem based
on the measure data in OMC system and basic
informat ion of cells. The measure data includes signal
level distribution, sample point distribution, and category
of interferences. The basic cell informat ion includes
neighborhood relationship, antenna azimuth and distance
between two cells etc. The method is implemented in a
cell coverage analysis system based on GIS platform.
GIS platform is used to develop the applications which
are associated with geographic informat ion. The coverage
analysis system based on GIS can improve the visibility
of coverage area [7]. Network optimization engineer can
analyze the cell coverage conveniently and precisely.
This paper describes the system design and validation.
The system is tested by real data of an in-service GSM
network. The analysis result is compared with the
traditional approach. The advantages of new method are
summarized based on validation and comparison result.
II. BASIC PRINCIPLE
In mobile network, MS ( Mobile Station ) upload some
measure data ( For example, the identification of the
serving cell and the interference cells, the downlink
signal level of the serving cell and the interference cells
etc.) to the network-side of the mobile network, and the
data are stored in OMC system. One record uploaded by
MS is called a sample point. The acquired field strength
distribution based on these information statistics is
complete and more accurate than that is acquired through
DT (Drive Test). Fig. 1 is the schematic diagram of the
field strength distribution of the serving cell.
Theoretically, the level value of the serving cell will
decrease when the distance to the serving cell increase [8].
For example, in Fig. 1, the level value measured in Cell A
should smaller than that measured in Cell B. The level
values in all the cells are collected through cell phones. If
the level value in Cell A is bigger than or equal to that in
Cell B, then there must be some problem in cell coverage.
.
Figure 1. The schemat ic diagram of the field strength dist ribut ion
However, the field strength distribution can not reflect
the cell coverage situation thoroughly. The information of
the sample points reflects the contact frequency and the
disturbance intensity between the serving cell and the
interference cell. The more the sample points, the closer
the relationship between the two cells, which means the
contact frequency is higher. The disturbance intensity is
presented through CIR (Carry Interference Rate). On the
network-side, for example, BSC (Base Station Control)
can get all the CIR between the serving cell and the
interference cells based on the measured data of all the
sample points. The formula of the CIR is:
C/I (dB) = downlink level serving cell (dBm) -
downlink level interference cell (dBm)
In theory, the farther away from the serving cell, the
remoter the relationship between the two cells, the sample
points should be less and the disturbance intensity should
be weaker. In this case, in Fig. 1, even though the level
value of the serving cell measured in Cell A isnt unusual,
but there are lots of sample points, which means the
contact frequency between this cell and the serving cell or
the disturbance intensity is high, this phenomenon is also
show that there is cell coverage problem. Similarly, if the
sample points measured in Cell B are few or the
relationship with the serving cell is remote also tells the
same thing.
III. ANALYSIS METHOD OF CELL COVERAGE
A. Method flowchart
This article presents an analysis method of cell
coverage based on the principle mentioned above. The
method flowchart is shown as Fig. 2.
This method consists of several parts: data collection,
data analysis and getting the analysis result. At first,
collect the raw data from OMC system. Secondly,
analyze the measure data. Measure data analysis includes
two parts: the field strength distribution analysis and
sample point number analysis. Thirdly, analyze the cell
basic data including base station level, antenna azimuth
and neighborhood relationship. At the end, draw the cell
coverage chart and identify the problem cells.
898 JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011
2011 ACADEMY PUBLISHER
Data collection from OMC
Drawing the cell coverage chart and
determining the problem cell
Signal level
distribution
Sample points
distribution
Measure data analysis
Base station
level
Antenna
azimuth
Cell basic data analysis
Neighborhood
relationship

Figure 2. Flow of the analysis method of cell coverage
The measure data analysis is the most important part in
the method. The analysis of signal level distribution and
sample point distribution is complex. The following Fig.
3 and Fig. 4 show the flows of field strength analysis and
sample point analysis.
Calculate the probability of all the
level value
Draw the level probability plot
Get the sub-point of the level interval
Classify all the cells according to the
level intervals

Figure 3. Flow of the signal level distribut ion analysis
Field strength distribution analysis include four steps:
calculate the probability of all the level value, draw the
level probability plot, get the sub-point of the level
interval and classify all the cells according to the level
intervals.
The sample points number analysis flow is shown as
Fig. 4.
Calculate the average value of Class1,
Class2, Class3 of all the cell pairs
Compare the sample points number of
each cell pair with the average value,
get the sample points information
Look up the sample points
classification table, get the sample
classification code

Figure 4. Flow of the sample point s dist ribut ion analysis
Sample points number analysis include three steps:
calculate the average value of 3 classes of all the cell
pairs, compare the sample points number of each cell pair
with the average value then get the sample points
informat ion and look up the sample points classification
table then get the sample classification code.
Based on the analysis of the two parts of the measure
data, the cell coverage chart can be shown on GIS map.
The planning coverage can be analyzed through cell basic
informat ion such as site level, azimuth and relationship.
Finally comparing with the planning coverage area, the
problem of cell coverage can be identified.
B. Measure data analysis
1) Signal level distribution analysis
In GSM network, the range of the average downlink
level is -47dBm-110dBm. There are many ways to
classify the average downlink level of the source cell
(serving cell) which is included in the collected raw data.
The average downlink level of the source cell can be
divided into several ranges. In this method, the levels are
divided into ranges based on level value statistical
probability.
The probability of every level value is calculated by
examining the number of t imes every source cell
downlink level value occurrence. With the average
downlink level of the interference cell as the horizontal
axis, the probability as the vertical axis, connect all the
points into a line, and then use the moving average
algorithm to make the line into a smooth curve. Draw the
level probability plot as Fig. 5.

Figure 5. Level probabilit y plot s
The level ranges (A, B, C, D.) are defined by taking
the wave troughs of the probability plot as the sub-points
of the source cell average downlink level. The probability
plot in each range approximates a normal distribution
with this partit ion. That is to say, in each source cell
average downlink level range, the probability of every
source cell downlink level value occurrence approximates
a normal distribution. This conforms to the ideal
distribution of the received downlink level of a cell in the
whole coverage area.
Then, all the interference cells are classified into level
A,B,C,D. according to the average downlink level of
the source cell received in the interference cells acquired
from the OMC. For example, in a record where the
serving cell ID is S and the interference cell ID is T, if the
JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011 899
2011 ACADEMY PUBLISHER
average downlink level of the cell S is level A where
received in cell T, cell T belongs to level A. Like this, all
the interference cells (the analyzing cells) can be
classified into level A, B, C, and Daround cell S. In
other words, the level strength distribution of cell S is
presented.
2) Sample point distribution analysis
I/C which is the reciprocal of C/I, expresses the
disturbance intensity. In engineering, -12dB is used as the
same frequency interference protection ratio. If I/C is
smaller than -12dB, then the interference is weak
interference. If I/ C is bigger than 0dB, then the
interference level is higher than the signal level and the
interference is strong interference. Therefore, (-,-12dB],
(-12dB, 0dB) and [0dB, ) represent weak interference,
critical interference and strong interference respectively.
The number of sampling points of the three ranges in a
certain time from the OMC data is represented as Class1,
Class2, and Class3 respectively. As Class1, Class2,
Class3 represent three levels of interference, each cell is
divided into different interference intensity by the
statistical analysis of the sampling points of the three
ranges.
First of all, work out the average number of all the
recording numbers of the sampling points. Avg1
represents the number of sampling number of Class1,
Avg2 of Class2, and Avg3 of Class3. That is:
Avg1 = sum (the number of sampling points of
Class1)/total number of points
In the formula above, sum (the number of sampling
points of Class1) represents all the recorded numbers of
sampling points of Class1. Avg2 and Avg3 are calculated
by the same function.
Then comparing the sample points number of Class1,
Class2, Class3 of each record with Avg1, Avg2, Avg3, if
the number is bigger than mean value, mark it as Big. On
the contrary situation, mark it as Small. In this way get
the classificat ion combination, and then look up the
TABLE I, get the interference intensity and the sample
points classification code. For example, after classify the
sample points number of Class1, Class2, Class3 of a
record (the serving cell ID is S, the interference cell ID is
T), the classification combination is (Big, Big, Big).
According to TABLE I, the classification code from cell
T to cell S is 1 and the interference intensity is strong
interference.
TABLE I. SAMPLE POINT CLASSIFICATION TABLE
Sample point

Classi fi cati on code
Class1
(weak interference)
Class2
(cri tical interference)
Class3
(strong interference)
Interference intensi ty
1 Big Big Big St rong int erference **
2 Big Big Small Moderat e int erference *
3 Big Small Big St rong int erference **
4 Big Small Small Weak int erference
5 Small Big Big St rong int erference **
6 Small Big Small Moderat e int erference *
7 Small Small Big St rong int erference **
8 Small Small Small Weak int erference

TABLE I shows the sample point classification. In
actual network, because the situation when the
interference intensity is strong interference is rare, the
sample points number where the interference intensity is
weak interference and moderate interference is much
bigger than that where the interference intensity is strong
interference, in the table, Big means the number is
bigger than mean value, Small means the number is
smaller than mean value. As a result, in the table the
priority is weak interference, moderate interference,
strong interference, Big is in front of Small, this
reflect the sample points number decrease from class 1 to
8.
The more the sample points number, it means the
relationship between the serving cell and the interference
cell is closer. So in the table, when the sample points of
strong interference is Big, the interference intensity is
defined as strong interference, among the rest, if the
sample points of critical interference is Big, the
interference intensity is defined as moderate
interference, the rest is defined as weak interference, so
in the table, the interference intensity of class 1,3,5,7 is
strong interference, class 2,6 is moderate interference,
class 4,8 is weak interference.
In this method, the number classificat ion (Class1 to
Class 8) of the sampling points of the interference cells
and the interference level (strong interference moderate
interference weak interference) are deduced according
to the statistical analysis of the number of the sampling
points in the OMC data and TABLE I .
C. Cell basic information analysis
Cell basic data includes base station locate level,
antenna azimuth and neighborhood relat ionship. These
data are valuable for cell coverage analysis.
1) Base station level stratification
There are large amounts of base stations in the large
mobile network. The base station intensity varies much in
different regions. In countryside, the distances among
base stations are much farer than in city. The cell
coverage radius in countryside is much bigger than in city.
The coverage radius cant be defined as a constant value.
Thus, the cell coverage is usually indicated by base
station level and azimuth side.
The forward of the azimuth is the area of 120around
the center line of the azimuth, and the other side means
backward. In general, the coverage area of a cell should
be in the forward site level 3 and the backward site level
900 JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011
2011 ACADEMY PUBLISHER
1. The area of inside of forward level 1 to 3 and backward
level 1 is the right coverage region. The area of outside of
forward level 3 and outside of backward level 1 should
not be covered.
To divide the site level of the base station is the key
problem to confirm the cell coverage area. Commonly the
famous Delaunay triangulation principle is used to
partition the base station nodes intelligently. Delaunnay
triangulation is special. It requires that every circumcircle
of each triangle does not contain any other point in the
triangle network, this ensures a triangle is formed by the
most three adjacent nodes [9-11]. Fig. 6 is the chat of cell
stratification by the triangulation method.

Figure 6. Cells strat ificat ion by triangulat ion method
In Fig. 6, the base stations in mobile network are
partitioned into triangle network with the triangulation
method. In the triangle network, the site level can be
defined as the length of the shortest road between two
cells [12]. For example, cell A is the serving cell. The
shortest road from cell B to cell A is 1, so cell B is site
level 1. The shortest road from cell C to cell A is 2, so
cell C is site level 2, and the cell D is site level 2 too. The
partition result is satisfied with the real engineering
experience.
D. Result showing on GIS
Based on the result of cell coverage analysis, combine
the longitude and latitude information of the serving cell
and interference cells in the OMC data, draw the cell
coverage chart on GIS, and then find the cells which have
problems intuitively, as shown in Fig. 7.
S
A2*
A2*
A1
**
D6*
E4
E6
*
B3
**
C4
C6*
D3
**
E4
E6*
D3
**
E4
A2
*
C1
**
BS1
BS2
BS3
BS6
E4
BS4
BS5
BS8
BS7


Figure 7. Coverage analysis chart
Fig. 7 shows the coverage analysis of serving cell S. A,
B, C, D, E represents the signal level rank, the level
decrease in alphabetical order. Number 1, 2 to 8 represent
the sample classificat ion code, the interference level of
the cells marked by ** is strong interference, the
interference level of the cells marked by * is moderate
interference, the interference level of the cells without
mark is weak interference.
In Fig. 7, in BS7 and BS 8 the received level strength
is high (level C, A), the sample classificat ion code is
small (class 1, 2), that means their contact frequency with
the serving cell is high, the interference intensity to the
two cells are strong interference (**) and moderate
interference (*). But the distance between the two cells
and the serving cell S are far, that means possibly that the
configuration of the two cells has some problems. Cell S
maybe have the cross-boundary coverage problem to BS7
and BS8. The reasons which can cause the problems are
the configuration parameters of the related BS, for
example, there is a deviation of antenna azimuth,
longitude or latitude, or maybe the actual environmental
factors like the weather, for example, a strong wind can
make the antenna height, azimuth and the pitch angle
change.
E. Identification of the problem cells
Considering of the above factors, the results of cell
coverage analysis are shown as following TABLE II.
The coverage analysis result is conducted according to
multiple factors such as the sample point class, the signal
level, the site level, the azimuth, and the relat ionship. In
the table, the deviation coverage problems are listed only,
and the right coverage is not listed because of
unnecessary. The cell site which is cross-boundary
coverage or poor coverage is clearly shown in the table.
For example, No 1 in the table means that if the sample
point class of interference cell T is 1 or 3(it means Strong
inference with much sample points), the site level of cell
T is forward outside level 3(it means the distance is far
away from the coverage) from the serving cell S, and cell
T is not the neighborhood cell of cell S, so cell S might
has the cross-boundary coverage problem in cell T.
No 5 in the table means that if the inference cell T is 8
(it means little sample points), the signal level is E (it
means lowest signal level), and site level is forward
inside level 3(it should be in the coverage area), but
relationship is neighborhood; therefore, cell S should has
poor coverage in cell T.
No 7 in the table means that if the inference cell T is
the neighborhood cell of the serving cell S, but there is no
sample point in cell T, so cell S should has the poor
coverage problem in cell T.









JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011 901
2011 ACADEMY PUBLISHER
TABLE II. RESULTS OF MULTI-FACTORS ANALYSIS OF CELL COVERAGE
No Sample point class Signal level Site level Ward of azimuth Rel ati onshi p Result
1 1,3 Out side level 3 forward No cross-boundary coverage
2 1,3 Out side level 1 backward No cross-boundary coverage
3 2,4,5,6,7 A,B Out side level 3 forward No cross-boundary coverage
4 2,4,5,6,7 A,B Out side level 1 backward No cross-boundary coverage
5 8 E Inside level 3 forward Yes poor coverage
6 8 E Inside level 1 backward Yes poor coverage
7 No sample point Yes poor coverage

IV. CELL COVERAGE SYSTEM DESIGN
A. Function module structure
The relevant system to implement the above multiple
factors analysis of cell coverage method is designed as
three layers structure following as Fig. 8.
User Interfaces Layer
GIS GUI Coverage analysis GUI
Result data Data import
Map basic functions Cell coverage map
Data Processing Layer
Data Import
Data Analysis
Antenna
azimuth
analysis
Base station
level analysis
Sample points
number analysis
Signal level
distribution
analysis
Neighborhood
relation
analysis
Raw Data Layer
Map File
OMC data
Neighborhood.txt RxLevels.txt Dac.txt Cf.txt Sectors.txt

Figure 8. Syst em st ruct ure of mult i-factors analysis of cell coverage
Raw data collection layer and data processing layer are
designed as backend services. Raw data collection layer
is responsible for collection of map data and OMC data.
OMC data can be categorized per file format as shown in
Fig. 8.
Data processing layer can import raw data as
intermediate data objects for application consume. Data
processing modules implement various analysis functions:
sample points number analysis, signal level distribution
analysis, neighborhood relation analysis, Base station
level analysis, Antenna azimuth analysis.
User interfaces layer provides 2 kinds of data
presentation method: coverage display based on GIS and
coverage analysis display by data grid.
B. System development and deployment environment
The development platform is Visual Studio 2005 C#
on Windows XP system.
The implementation of GIS features (e.g. the base
stations location and color) employs MapInfo
MapXtreme components.
The system is deployed on Windows XP system.
V. DISCUSSION OF COVERAGE ANALYSIS RESULT
A. Validation with real datasets from in-service network
This method and system has been tested using the real
data from the GSM network in a Chinese province. The
testing data is the mobile phone measure data collected
from the OMC system in Apr, 2009.
For example, the cell RENMINRD1 is analyzed
overall. The coverage chart is shown as following Fig. 9.

Figure 9. Mult i-fact ors analysis of cell coverage graph
In Fig. 9, the red cell RENMINRD1 is serving cell, and
the other cells are interfere cells. The different colors
mean the interfere class 1 to 8 of sample points.
ABCEDE indicate the signal level range. A:[-47-54], B:
(-54-70], C: (-70-82], D: (-82-98], E: (-98-110]
dbm
From the Fig. 9, the cell coverage status is good. In the
nearer cells with s maller station site level to the serving
cell, the interference intensity is stronger and the signal
level is higher. In the farer cells with bigger station site
level to the serving cell, the interference intensity is
weaker and the signal level is lower.
In Fig. 9, the upper left interference cell TUDIJU2 is
far from the serving cell RENMINRD1 and the sit e level
is outside the forward site level 3, so the TUDIJU2 cell
should be out of the coverage boundary of the
RENMINRD1 cell. But the analysis data show that the
sample point number is high (class 1) and signal level is
high (level C). The TUDIJU2 cell is not the
neighborhood cell of the RENMINRD1 cell. It means that
902 JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011
2011 ACADEMY PUBLISHER
the RENMINRD1 cell probably has cross-boundary
coverage in the TUDIJU2 cell.
The left interference cell WANGBAOSHOUJI is near
to the serving cell RENMINRD1 and the site level is in
forward site level 1. The WANGBAOSHOUJI cell shall
be in the coverage of the RENMINRD1 cell. But the
analysis data show that there are no sample points in
WANGBAOSHOUJI cell. The WANGBAOSHOUJI cell
is the neighborhood cell of the RENMINRD1 cell. It
means that the RENMINRD1 cell probably has poor
coverage in the WANGBAOSHOUJI cell.
B. Comparison with traditional analysis approach
In engineering pract ices, there is a tradit ional analysis
approach to analyze the cross-boundary coverage
problems. The engineering practices prove that this
traditional analysis approach is reliable. Therefore, the
comparison with the traditional approach can validate the
accurateness of multiple factors analysis method of cell
coverage.
The analysis result by mult iple factors analysis method
of cell coverage is shown in TABLE III.
The interference level is the list of sample points class
and signal level. From TABLE III, the serving cell
RENMINRD1 has cross-boundary coverage problem in
the cell TUDIJU2 and YINDU2. The serving cell
RENMINRD1 has poor coverage problem in the cell
TAISHANZL2 and WANBAOSHOUJI.
TABLE III. RESULTS OF MULTI-FACTORS ANALYSIS OF CELL COVERAGE
Serving cell Interference cell Interference level Site level Ward of
azimuth
Rel ati onshi p Result
RENMINRD1 TUDIJU2 1C Outside level 3 forward No cross-boundary
coverage
RENMINRD1 YINDU2 7B Outside level 3 forward No cross-boundary
coverage
RENMINRD1 HONGRIXIQU1 1C Inside level 1 backward Yes
RENMINRD1 LHDANNISI1 8C Inside level 1 forward Yes
RENMINRD1 LHDANNISI2 8C Inside level 1 forward Yes
RENMINRD1 TAISHANLU2 1C inside level 2 forward Yes
RENMINRD1 TAISHANZL2 No sample point inside level 2 forward Yes Poor coverage
RENMINRD1 WANBAOSHOUJI No sample point Inside level 1 forward Yes Poor coverage
RENMINRD1 LHWENHUAGONG2 2C Inside level 3 forward Yes

TABLE IV. RESULTS OF TRADITIONAL METHOD OF CELL CROSS-BOUNDARY COVERAGE ANALYSIS
Serving cell Interference cell Interference factor Distance Is cross-boundary coverage
RENMINRD1 HONGRIXIQU1 19.95 10.09 No
RENMINRD1 LHDANNISI1 44.84 15.68 No
RENMINRD1 LHDANNISI2 48.89 17.1 No
RENMINRD1 LHWENHUAGONG2 41.89 13.59 No
RENMINRD1 TAISHANLU2 35.49 14.03 No
RENMINRD1 TAISHANZL2 27.88 11.07 No
RENMINRD1 TUDIJU2 25.57 20 Yes
RENMINRD1 WANBAOSHOUJI 73.17 20 No
RENMINRD1 YINDU2 30.14 20 Yes
RENMINRD1 XINDAXIN3 28.47 11.32 Yes


In traditional approach, the distance between cells and
neighborhood relation and C/I are considered and the
analysis result is shown in TABLE IV. The interference
factor item value varies inversely with C/I value. The
small interference factor means that the interference from
serving cell is strong to interference cell.
When the cross-boundary coverage occurrence in non-
neighborhood interference cells exceeds the threshold, the
serving cell is deemed as cross-boundary coverage cell.
The threshold is 3 in engineering practices. From TABLE
IV, the serving cell RENMINRD1 has cross-boundary
coverage in the cell TUDIJU2 and YINDU2.
From the comparison of TABLE III and TABLE IV,
the serving cell RENMINRD1 has cross-boundary
coverage in the cell TUDIJU2 and YINDU2. The real on-
site testing result proved this analysis conclusion.
TABLE III identified the poor coverage problems in the
cell TAISHANZL2 and WANBAOSHOUJI.
Validation on large amount of real data shows that the
accurateness of these 2 approaches exceeds 80%. The
traditional approach requires much analysis on redundant
informat ion and manually verification efforts is
significant. The new approach in this paper can save
much efforts on manually verification and has the ability
to identify the poor coverage problems.
JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011 903
2011 ACADEMY PUBLISHER
VI. CONCLUSION
This paper presented a novel multiple factors analysis
method for analyzing the cell coverage in GSM network.
The approach consists of 1) collecting mobile measure
data from OMC system, 2) analyzing the measure data
and the basic data of cells, and 3) drawing the cell
coverage chart on GIS and identifying the problem cells
which have deviat ion coverage. Unlike the previous
approaches based on DT or prediction, this new approach
is based on the real mobile phone measure data collected
from OMC, thus the raw data are comprehensive, real-
time and costless. The analysis considers mult iple factors
including the signal level distribution and sample point
distribution etc. This paper first proposed sample point
distribution analysis to describe the contact tightness of
cells in cell coverage identificat ion. So the approach is
more comprehensive. The approach also provides the
visualizat ion chart of cell coverage analysis based GIS
which has practical advantages over the many existing
techniques whose UI only display results as data table. It
improves the availability and efficiency of the analysis
tool for optimization engineers.
The approach is evaluated by large amounts of
measure data from a real GSM network. Over 80%
average problems can be discovered and the accurate is
satisfied.
There are two immediate goals for future work. First,
the accuracy should be increased higher by improving the
analysis algorithm. Second, what reason caused the
deviation occurred and how to resolve the deviat ion can
not be pointed out in the method. These are the direct ions
for further research.
ACKNOWLEDGMENT
School of Network Education, Beijing University of
Posts and Telecommunications (BUPT) support this work
in funding and research environment. Any opinions
expressed in this paper are those of the author and do not
necessarily reflect the views of School of Network
Education, BUPT. The anonymous reviewers provided
useful feedback that helped improve the quality of this
paper.
REFERENCES
[1] Deng Yuren, Wireless Cover and Network Optimization,
Shanxi Electronics Technology, 2005(03), pp.45-46.
[2] Jin Xiao-jia, Pan Yang-fa, and Song Jun-de. The status
and development trend of the mobile communication
network optimization technology, Telecom Technology,
2003(12), pp.1-3.
[3] Ma Ming-ming, Xu Xi, and Wang An-yu, A Analysis
Method of Road Test Datas Simulation and Prediction,
The proceeding of 2006 Seminar of Mobile
Communication Network Planning and Optimization,
2006(09), pp.305-309.
[4] Li Ru-xin, Wang Dao-heng, Xu Ji-sheng, and Yu
Shengbing, The Research for Field Intensity Prediction
Software for Radio Signals in Mobile Networks, Tianjin
Communication Technology, 2003, 3(1), pp. 26-32.
[5] Zhao Ting-bing, and Liu Shang-hong. GSM wireless
network optimization, Science Consult, 2008(03), pp. 49-
49
[6] LIU Ming-chuan, and XU yang, Study on capacity and
cover of the WCDMA ,Journal of Chongqing University
of Posts and Telecommunications, 2004,16(4), pp.121-125
[7] Liu Zhi-ping, Qiu Hong, and Yang Da-cheng, A GIS
Based Forward Coverage Analysis Method of CDMA
Systems, Journal of Beijing University of Posts and
Telecommunications , 2004, 27(4), pp.31-35.
[8] Han Bin-jie, GSM Theory and Network Optimization,
Machinery Industry Press: Beijing, China, 2001.
[9] Shao Chun-li, Hu Peng, Huang Cheng-yi, and Peng Qi,
The Expatiation of Delaunay Algorithms and a Promising
Direction in Application, Science of Surveying and
Mapping, 2004, 29(6), pp.68-71.
[10] V.J.D.Tsai. Delaunay Triangulations in TIN Creation: an
Overview and a Linear-time Algorithm. Int. J .of GIS.
1993.7(6), pp.501-524.
[11] Chen yu, Wang Xianghai, Research on constraint
triangulation. Computer Science, 2008(08), Vol.135
No.18, pp.6-9.
[12] Jia Jinzhang, Liu Jian, and Song Shousen, Connectivity
criteria based on adjacency matrix graphs, Journal of
Liaoning university of engineering technology, 2003,22
(02), pp.158-161.



Lina Lan received the BS degree in computer communication
from BUPT (Beijing University of Posts and
Telecommunications) in 1994, and the MS degree in Computer
application from BUPT in 1997. She is currently working as a
lecturer in BUPT. Her main research interest is software
architecture, network service, and mobile network optimization.

Xuerong Gou received the BS degree in communication
engineering from BUPT in 1976, and the MS degree in
Management Science from Norwegian University in 2001. She
was a visiting researcher in SFU (Simon Fraser University) in
Canada in 2005. She is currently working as a professor in
BUPT. Her main research interest is NGN, network education,
education technology, and mobile network optimization.

Yunhan Xie received the MS degree in communication and
information system from BUPT in 2009. Her research work is
mobile network optimization and GIS application development.

Meng Wu is currently working toward the MS degree in
communication and information system in BUPT. His research
work is mobile network optimization and GIS application
development.



904 JOURNAL OF COMPUTERS, VOL. 6, NO. 5, MAY 2011
2011 ACADEMY PUBLISHER