A Hybrid Neighbor Optimization Algorithm for SON based on Network

Topology, Handover Counters and RF Measurements

D. Duarte

, A. Martins

1,2

, P. Vieira

1,3

and A. Rodrigues

1,4

Instituto de Telecomunicac¸

ooes (IT), Lisbon, Portugal

CELFINET, Consultoria em Telecomunicac¸

oes Lda., Lisbon, Portugal

Instituto Superior de Engenharia de Lisboa (ISEL), ADEETC, Lisbon, Portugal

Instituto Superior T

ecnico (IST), Lisbon, Portugal

Keywords:

Self-Organizing Networks, Neighbor Cell List, Optimization, Radio Frequency Measurement, Handover

Statistics.

Abstract:

With the increasing complexity of current wireless networks, it became evident the need for Self-Organizing

Networks (SON), which aims to automate most of the associated radio planning and optimization tasks. Within

SON, this paper aims to optimize the Neighbor Cell List (NCL) for radio network cells. An algorithm com-

posed by three decision criteria was developed: geographic localization and orientation, according network

topology, Radio Frequency (RF) measurements collected by drive-tests or traces and Performance Manage-

ment (PM) counters from Handover (HO) statistics. The ﬁrst decision, proposes a new NCL taking into

account the Base Station (BS) location and interference tiers, based on the quadrant method. The last two

decision criteria consider signal strength and interference level measurements and HO statistics in a time pe-

riod, respectively. They also deﬁne a priority to each cell and added, kept or removed neighbor relation,

based on user deﬁned constraints. The algorithms were developed and implemented over new radio network

optimization professional tool. Several case studies were produced using real data from a mobile operator.

1 INTRODUCTION

According to a recent trafﬁc report published by

Cisco



, global mobile data trafﬁc reached 3.7 ex-

abytes per month at 2015, and up to 2.1 exabytes per

month at 2014. 4

Generation (4G) trafﬁc exceeded

Generation (3G) trafﬁc for the ﬁrst time in 2015.

Although 4G connections represented only 14 % of

mobile connections in 2015, they already account for

47 % of mobile data trafﬁc, while 3G connections rep-

resented 34 % of mobile connections and 43 % of the

trafﬁc. In 2015, a 4G connection generated six times

more trafﬁc, on average, than a non 4G connection

(Cisco, 2016).

With the already existing 3G/4G standards, espe-

cially with the Long Term Evolution (LTE) (3GPP,

2010), more complexity is added to current networks.

The co-existence of multiple standards, mostly from

different suppliers, and the increasing demand of 3

party services, asks for more management effort from

mobile network operators. The main goal for SON

implementation is to automate most of the common

planning, optimization and operational tasks, reduc-

ing operators operational and capital costs. Crossed

sectors detection is an example of automatic opera-

tional tasks. It is one of the most common implemen-

tation errors when deploying base-stations (Duarte

et al., 2015).

The increasing network intensiﬁcation, even with

more small cells and more Heterogeneous Net-

works (HetNets) is also coming rapidly (Ramiro and

Hamied, 2012). This ongoing change will bring more

multi-standard demands and multi-vendor challenges.

Efﬁcient and effective operations must overcome such

complexity. Hence, the only way these challenges

can be cost-effectively, efﬁciently and humanely over-

come is through the use of more automated and au-

tonomous systems.

The SON use cases can be structured as Self-

Planning, Self-Deployment, Self-Optimization and

Self-Healing. Network optimization is a continuous

closed-loop process encompassing periodic perfor-

mance evaluation, parameter optimization and rede-

ployment of the optimized parameter values into the

network. The optimization decisions can be carried

out by human subjects or computerized systems, lead-

Duarte, D., Martins, A., Vieira, P. and Rodrigues, A.

A Hybrid Neighbor Optimization Algorithm for SON based on Network Topology, Handover Counters and RF Measurements.

DOI: 10.5220/0005958201450151

In Proceedings of the 13th International Joint Conference on e-Business and Telecommunications (ICETE 2016) - Volume 6: WINSYS, pages 145-151

ISBN: 978-989-758-196-0

145

ing the last to self-optimization. The optimization al-

gorithms will aim at tuning parameter values in order

to achieve a well-deﬁned goal, expressed in terms of

coverage, capacity, quality or even a controlled com-

bination of them, taking into consideration that any

optimization activity always involves implicit trade-

offs between these key variables.

Within self-optimization, this paper aims to opti-

mize the NCL for radio network cells. The paper con-

tribution is incremental to previous published work

(Duarte et al., 2014), incorporating recent research

leading to new results.

The paper is organized as follows. Section 2

overviews the NCL issue. Section 3 presents the pro-

posed algorithm, followed by the simulation results

and analysis presented in Section 4. Finally, conclu-

sions are drawn in the ﬁnal section.

2 NEIGHBOR CELL LIST

OVERVIEW

NCL optimization pertains to the optimization of the

existing neighbor list of a cell plus the neighbor lists

that are applied to the neighboring cells. This con-

siders the cells that are already in a production en-

vironment, as well as the new cells, for which com-

pletely new neighbor lists have to be generated. The

target scope covers intra-frequency, inter-frequency

and inter-technology neighbors. The HO is one of

the most critical issues in cellular networks. It en-

ables connection continuity for mobiles during mo-

bility, while allowing the efﬁcient use of resources,

like time and frequency reuse between cells. Most

of todays cellular standards use mobile-assisted HO,

in which the mobile measures the signal quality of

neighbor cells and reports the measurement result to

the network. If the signal quality from a neighbor cell

is better than the serving cell by a handover margin,

the network initiates an HO to that cell (Nguyen and

Claussen, 2010). The conﬁguration and NCL man-

agement are one of the most important issues to pro-

vide seamless mobility for User Equipment (UE)s in

the radio network. Between handover candidate cells,

each base-station has a connection of direct interface

with other base-stations and messages related to the

HO procedure are exchanged through the radio in-

terface in order to prepare HO execution. Each cell

maintains a list of cells that are the targets for con-

nection.

In LTE without SON mode of operation, a human

system operator determines the NCL and uploads it

into the evolved NodeB (eNB). On the other hand,

in LTE systems under SON mode of operation, also

called Autumatic Neighbor Relation (ANR), each

eNB is supposed to conﬁgure and manage the NCL

autonomously. Firstly, the initial NCL has to be con-

ﬁgured without any cooperation with the UE when

a new eNB has been deployed since, initially, there

are not any activated UEs belonging to the new eNB.

Secondly, the NCL needs to be constantly updated

based on the change of network topology by collect-

ing the information from the UE (Kim et al., 2010).

The basic condition of deﬁning a neighbor relation

between cells is the coverage overlap with each other,

and forming a HO region, as presented in Fig.1. In

a sectored cell environment, sector cells belonging to

the same eNB are neighbors. The measurements re-

ported by the UE, such as Reference Signal Received

Power (RSRP) and Reference Signal Received Qual-

ity (RSRQ) are then used by an optimization algo-

rithm implemented in the eNB, in order to continu-

ously update the list of neighboring cells.

Figure 1: Missing neighbor relation (Duarte et al., 2014).

It is important to note that a delay is introduced

when the HO is attempted into a neighbor cell that

is not included in the NCL. If the NCL includes

any cells which are not supposed to be the neighbor

cells, the system overhead is wasted due to unnec-

essary management and maintenance of direct inter-

face between neighbor cells. Therefore, with or with-

out ANR, accurate NCL conﬁguration is demanded

to prevent HO delay and save system overhead (Kim

et al., 2010), which motivates the current research is-

sue.

3 THE ENHANCED NCL

ALGORITHM

This section presents the proposed algorithm. A de-

scription of the algorithm is made along with the used

line of thought in NCL optimization.

WINSYS 2016 - International Conference on Wireless Networks and Mobile Systems

146

3.1 Algorithm Overview

The NCL optimization ﬂowchart is presented in Fig-

ure 2. The network topology is one of the inputs.

It contains the BS geographic location, antenna con-

ﬁguration and orientation, as well as the cell identi-

ﬁers and Primary Scrambling Code (PSC) or Physical

Cell Identity (PCI). Another considered input is the

initial NCL conﬁguration of each cell, with all the

deﬁned neighbor relations. Finally, geo-positioned

traces (Vieira et al., 2014) and drive-test data are

available, along with PM statistics, focusing on the

number of HO for each deﬁned neighbor relation.

The initial NCL is the starting point, if it exists. In

order to optimize the NCL, three distinct criteria were

developed in the algorithm, which will be described

in the following: planning, through network topology,

HO statistics and RF measurements. They also deﬁne

a priority to each cell considering several calculation

constraints, introduced by the operating radio engi-

neer.

Thus, the ﬁnal goal is to prepare a global prior-

itized list for each cell considering the three crite-

ria, simultaneously. The output will be an optimized

NCL that maximizes the HO target, based on several

aspects of the implemented network, and user con-

strains (Duarte et al., 2014). Finally, cells of higher

Figure 2: Proposed algorithm ﬂowchart.

priority are added, according to Table 1, only lim-

ited by the maximum NCL size. With this approach,

the co-located cells are always added to new NCL

and neighbor relations with good HO performance

are kept. Nearby cells must be on the NCL. It’s also

possible to identify missing neighbors in overlap con-

ditions, through the use of drive-test measurements

(Sousa et al., 2015). These candidate cells will be

added to the NCL. If the list is yet not full, then low

priority candidate cells will be added or kept, by plan-

ning criteria, the last with HO performance below

thresholds.

Table 1: Priority ordered cells.

Criteria Priority

Co-Loc 1

HO 2

Distance 3

DT 4

Plan 5

HObellowTh 6

3.2 Network Topology Decision

It is based on the distance from the source cell to the

target cell and antenna orientation. After processing

the network topology, the base-station’s cell informa-

tion is available. Hence, the distance between cells

is calculated and one proximity ranking is deﬁned for

each target cell. Thereafter, the several tiers are de-

ﬁned, using classical hexagonal radio planning the-

ory, where the six closer sites are considered as the

ﬁrst tier. The ﬁrst tier and adjacent cells are manda-

tory, and must appear in the new NCL. These cells are

classiﬁed with priority 3. This calculation is based in

the source/target cell location and orientation. The

addition of the target cells to the NCL should follow

the scheme illustrated in Figure 3 as the rules in Ta-

ble 2. Firstly, the source cell orientation is veriﬁed

by its azimuth. Assuming a source cell, represented

in orange, belonging to the ﬁrst quadrant (1

◦

Q), i.e.,

between 0

◦

and 90

◦

, every target cells directed there,

must be added to the NCL. For the remaining quad-

rants, only those sectors directed to the source cell

area service (in this case, the 1

◦

Q), are added to NCL

with priority 5. In order to know the position of each

target cell, i.e., its quadrant, it is calculated the an-

gle between cells, by own location and using its geo-

graphic coordinates (Duarte et al., 2014).

A Hybrid Neighbor Optimization Algorithm for SON based on Network Topology, Handover Counters and RF Measurements

147

1º Q2º Q

4º Q3º Q

Figure 3: The quadrant theory.

Table 2: The quadrant theory bounds.

Target Quadrant Eligible Target

Location Sector

[ 0

◦

; 90

◦

] [ 0

◦

; 360

◦

]

[ 90

◦

; 180

◦

] [ 0

◦

; 90

◦

] & [ 270

◦

; 360

◦

]

[ 180

◦

; 270

◦

] [ 0

◦

; 90

◦

]

[ 270

◦

; 360

◦

] [ 0

◦

; 180

◦

]

3.3 Handover Statistics Decision

The HO statistics decision criteria aims to keep cells

belonging to the initial NCL. The PM counters are

collected to each cell for a large period, e.g. one

month. If the existing HO number, NumHO, or the

HO ratio, HoWeight, in the neighbor relation is above

a certain thresholds introduced by the user, the cell is

marked with priority 2 to be kept, see Figure 4. This

is useful, since it allows to remove cells with low HO

count from the source cell. These criteria does not

allow to add new cells.

Figure 4: Handover statistics decision ﬂowchart.

3.4 RF Measurements Decision

The RF measurement decision criteria adds or keeps

target cells to the new NCL, if its signal strength level

is above a certain threshold, minimum signal strength

SSMin, collected by drive-test or network traces (Car-

valho and Vieira, 2011). Additionally, it has to satisfy

all overlap conditions, see Equation 1 and Equation 2

for 3G networks:

RSCP

− RSCP

< OverlapSS

T h

(1)

Ec/Io

− Ec/Io

< OverlapQual

T h

(2)

where RSCP

and RSCP

are the source and tar-

get cell received signal strength, respectively, in each

sample. Ec/Io is the quality level.

The algorithm uses the source cell corresponding

measurements and, for each sample, checks the sig-

nal and quality levels for the target cells. The target

cell is marked with priority 4 if it satisﬁes Equation

1 and Equation 2, for a minimum number of sam-

ples NumSamples. This approach can also be made

for quality measurements. The RF Measurement de-

cision ﬂowchart is presented in Figure 5.

Figure 5: RF Measurements decision ﬂowchart.

4 APPLYING THE ALGORITHM

TO LIVE NETWORKS

This section presents the main results considering the

neighbor cell list optimization. The algorithm has

been tested in several cities for large cell clusters. For

one of the clusters, 1238 cells were optimized, ap-

proximately 152 sites with three 3G carriers (F1, F2

and F3), for an intra-frequency analysis. This itera-

tion, as mentioned before, was performed using scan-

ner data information collected by drive test, see Fig-

ure 6, and Soft-HO statistics. In this test case, the fol-

WINSYS 2016 - International Conference on Wireless Networks and Mobile Systems

148

lowing thresholds in Table 3 were used. These thresh-

olds are empirical values estimated from several sim-

ulations. By changing the thresholds, more or less

cells are added to the new list. If we increase the HO

parameter, more cells will be removed. On the other

hand, increasing the SSMin and NumMinSamples,

fewer cells are added by drive test.

Figure 6: Drive test in cluster analysis.

Table 3: Input parameters.

Parameter Value

NCL Size 28

NumMinHO 1000

HO Weight [%] 0.1

SS Min [dBm] -95

OverlapSSTh [dB] 6

OverlapQualTh [dB] 8

NumMinSamples 5

Table 4 shows all the algorithm suggestions.

Table 4: Cluster results.

Action / Frequency F1 F2 F3

Add 3410 3674 2446

Co-Loc - - -

DT 30 27 22

Distance 274 303 231

Plan 3106 3344 2193

Keep 5653 5306 5697

Co-Loc 646 646 572

HO 2927 2788 3783

DT 83 77 48

Distance 351 344 207

Plan 1492 13339 994

HOBelowTh 154 112 93

Remove 1957 1652 1917

Directly from the results, we can ﬁnd that there

isn’t any intra-frequency co-located neighbor relation

missing. 9498 neighbor relations are being kept due

to HO statistics thresholds, representing 30% of all

the suggestions. Scanner data samples provided 287

suggestions, both existent neighbor relations, to keep,

and new neighbor relations to be added. With the HO

statistics information, RF measurements and network

topology, the algorithm suggests 5526 neighbors to

be removed, since they are below the thresholds de-

ﬁned for input. In total, there are 30% suggestions

of additions, 52.6% to keep and 17.4% of removals.

This means that most of neighbor relations are well

deﬁned, but there are a large percentage of missing

neighbors.

4.1 Case Study - Network Topology

(Planning)

Using only the network topology, site location, dis-

tance to the source and orientation, it is possible to

get a new neighbor cell list. This generated list, to

the source cell (blue), does not take into account the

old list, only suggests adding neighbor relations. This

criteria, used individually, allows to plan the neighbor

list for a new site, when neither statistics nor scanner

data exists. Applying the quadrant theory described

above, the algorithm added (green) 26 neighbor rela-

tions, as seen in Figure 7.

Figure 7: Neighbor cell list planning.

4.2 Case Study - HO Statistics (PM

Counters)

With only HO statistics, the algorithm proposes to re-

move all neighbor relations which aren’t verifying the

conditions, according to user input thresholds. In this

test case, the cells with less than 1000 HOs and neigh-

bor relations ration below 0.1% in one month, are re-

moved. The remaining relations are kept, see Figure

8 and Table 5.

Table 5: Test Case - HO Statistics actions.

Action Number Cells Color

Kept 18 Yellow

Removed 9 Red

A Hybrid Neighbor Optimization Algorithm for SON based on Network Topology, Handover Counters and RF Measurements

149

Figure 8: Neighbor cell list based in HO statistics.

4.3 Case Study - RF Measurements

(Scanner Data)

With this criteria, the algorithm only ﬁnds possible

missing neighbor relations. The Figure 9 shows that

were added 12 neighbors in overlap conditions. It

means that at least ﬁve samples, the Received Signal

Code Power (RSCP) level between source and each

target cell is bellow to 6 dB. The difference between

Ec/Io is less than 8 dB. All samples (blue circles) are

-95 dBm of minimum RSCP level.

Figure 9: Neighbor cell list based in scanner data.

4.4 Case Study - All Decision Criteria

Finally, all criteria are joint. Thus, it is possible to

keep neighbor relations with a large HO number and

add missing neighbors with overlap, serving in the

same area, through scanner data. It is also allowed to

add or keep cells from network topology. The quad-

rant method, planning algorithm based on azimuth

and site location, is important to complete the new

neighbor list. Table 6 shows all the actions proposed

for the new neighbor cell list.

With high priority, co-located cells were kept . Af-

ter this, HO priority allows keeping the neighbor rela-

tions with good HO performance (16 cells). Only one

Figure 10: Neighbor cell list optimized.

Table 6: Neighbor Cell List actions.

Action Number of Cells Color

Add 6 Green

Co-Loc -

DT -

Plan 6

Keep 22 Yellow

Co-Loc 2

HO 16

DT 1

Plan 3

Remove 5 Red

Neighbor Relation (NR) was kept through RF mea-

surement, priority 4. As seen, there are more cells in

overlap conditions, but HO statistics are more prior-

itized. Finally, 9 NRs were added/kept due to plan-

ning priority. Five neighbors were removed from the

old list. These last cells fail the input thresholds or

the NCL size is full and own priority is below other

candidates. Table 7, in appendix, shows the optimized

cell list generated by the algorithm and the ﬁnal action

to apply to each NR.

5 CONCLUSIONS

In this paper, a neighbor cell list optimization algo-

rithm for radio access networks is proposed. Three

distinct criteria were developed: network topology,

HO statistics and RF measurements. The ﬁrst criteria,

proposes a new NCL taking account the site location,

azimuths and interference tiers, based on the quad-

rant method. The last two consider signal strength

measurements and HO statistics, respectively. They

also deﬁne a priority to each cell enabling neighbor

relation addition/removal based on user deﬁned con-

straints. The algorithms were implemented over an

already existing radio network optimization profes-

WINSYS 2016 - International Conference on Wireless Networks and Mobile Systems

150

sional tool. Several case studies were produced using

real data from a mobile operator. These algorithms

should be combined, in order to produce a globally

optimized NCL that maximizes the HO performance,

based on network speciﬁcs, and user constrains.

ACKNOWLEDGEMENTS

This work was supported by the Instituto de

Telecomunicac¸

oes and the Portuguese Foundation for

the Science and Technology (FCT) under GOLD

(PEst-OE/EEI/LA0008/2013).

REFERENCES

3GPP (2010). LTE; Evolved Universal Terrestrial Radio

Access (E-UTRA) and Evolved Universal Terrestrial

Radio Access Network (E-UTRAN); Overall descrip-

tion; Stage 2. TR 36.300, 3rd Generation Partnership

Project (3GPP).

Carvalho, M. and Vieira, P. (2011). Simulating long term

evolution self-optimizing based networks. In Conf.

on Electronics, Telecommunications and Computers -

CETC, volume 1, pages 1–1.

Cisco (2016). Cisco Visual Networking Index: Global Mo-

bile Data Trafﬁc Forecast Update, 2015-2020. Tech-

nical report, Cisco.

Duarte, D., Vieira, P., Rodrigues, A., Martins, A., Silva,

N., and Varela, L. (2014). Neighbour List Optimiza-

tion for Real LTE Radio Networks. In Wireless and

Mobile, 2014 IEEE Asia Paciﬁc Conference on, pages

183–187.

Duarte, D., Vieira, P., Rodrigues, A. J., and Silva, N. (2015).

A New Approach for Crossed Sector Detection in Live

Mobile Networks based on Radio Measurements. In

Wireless Personal Multimedia Communications Symp.

- WPMC, volume 1.

Kim, D., Shin, B., Hong, D., and Lim, J. (2010). Self-

conﬁguration of neighbor cell list utilizing e-utran

nodeb scanning in lte systems. In Consumer Commu-

nications and Networking Conference (CCNC), 2010

7th IEEE.

Nguyen, V. M. and Claussen, H. (2010). Efﬁcient Self-

Optimization of Neighbour Cell Lists in Macrocellu-

lar Networks. In Personal Indoor and Mobile Radio

Communications (PIMRC), 2010 IEEE 21st Interna-

tional Symposium on, pages 1923–1928.

Ramiro, J. and Hamied, K. (2012). Self-Organizing Net-

works (SON): Self-Planning, Self-Optimization and

Self-Healing for GSM, UMTS and LTE. Wiley Pub-

lishing, 1st edition.

Sousa, M., Martins, A., Vieira, P., Oliveira, N., and Ro-

drigues, A. (2015). Caracterizacao da Fiabilidade de

Medidas R

adio em Larga Escala para Redes Auto-

Otimizadas. In 9. Congresso do Comit

e Portugu

es da

URSI - ”5G e a Internet do futuro”.

Vieira, P., Silva, N., Fernandes, N., Rodrigues, A. J., and

Varela, L. (2014). Improving Accuracy for OTD

Based 3G Geolocation in Real Urban/Suburban En-

vironments. In Wireless Personal Multimedia Com-

munications Symp. - WPMC, volume 1.

APPENDIX

In this appendix, a optimized neighbor cell list is show

for the test case present above. The Table 7 indicates

all NR proposals ordered by algorithm priorities.

Table 7: Optimized neighbor cell list.

Cell HO Overlap Priority Action

SUCC Samples

A 90485 18 CoLoc Keep

B 23785 8 CoLoc Keep

C 65735 54 HO Keep

D 55432 15 HO Keep

E 48584 25 HO Keep

F 41501 50 HO Keep

G 41259 51 HO Keep

H 33856 30 HO Keep

I 18061 8 HO Keep

J 17138 8 HO Keep

K 16601 0 HO Keep

L 11078 5 HO Keep

M 10924 2 HO Keep

N 8682 3 HO Keep

O 2864 0 HO Keep

P 1950 0 HO Keep

Q 949 0 HO Keep

R 724 0 HO Keep

S 0 16 DT Keep

T 0 1 Plan Add

U 0 0 Plan Add

V 85 0 Plan Keep

W 182 0 Plan Keep

X 87 0 Plan Keep

Y 0 0 Plan Add

Z 0 0 Plan Add

AA 0 0 Plan Add

AB 0 0 Plan Add

AC 331 0 HObelowTh Remove

AD 185 0 HObelowTh Remove

AE 58 0 HObelowTh Remove

AF 42 0 HObelowTh Remove

AG 0 0 HObelowTh Remove

A Hybrid Neighbor Optimization Algorithm for SON based on Network Topology, Handover Counters and RF Measurements

151