SONS3: A Network Data Driven Simulation Framework for Mobility Robustness Optimization

Full text

SONS3: A Network Data Driven Simulation
Framework for Mobility Robustness Optimization
Frans Laakso, Jani Puttonen, Janne Kurjenniemi,
Biying Wang, Di Zhang
Magister Solutions Ltd.
Jyväskylä, Finland
Email: {firstname.lastname@magister.fi}
Furqan Ahmed, Jarno Niemelä
Elisa Corporation
Helsinki, Finland
Email: {firstname.lastname@elisa.fi}
Abstract—Self-Organizing Networks (SON) aim at automation
of network parameters to make the planning, configuration,
management, optimization and healing of mobile radio access
networks simpler and faster, and thus also more cost efficient.
SON use case in this article is Mobility Robustness Optimization
(MRO), which targets automatic detection and correction of
handover (HO) parameterization. To this end we present novel
simulation framework called Self-Organizing Simulator 3 (SonS3)
that makes use of real network data to enable an accurate
performance analysis of mobility related SON use cases. In
addition, we present an MRO algorithm which adapts dynamically
the Cell Individual Offset (CIO), and a performance analysis
within a realistic urban scenario. The proposed MRO algorithm
leads to an overall reduction of more than 30% in mobility
problems without causing a significant increase in the number of
HOs.
Keywords—Self-organizing network, SON, mobility robustness
optimization, MRO, ns-3, Network Simulator, performance.
I. INTRODUCTION
While wireless networks are evolving, they are also
becoming increasingly complex. The deployments involve
several overlapping frequency and technology layers, and
different kinds of hybrid macro / indoor scenarios. The
technologies involve configuring large number of parameters,
which can be difficult to optimize individually, let alone jointly.
Self-Organizing Network (SON) is an automation technology
designed to make the planning, configuration, management,
optimization and healing of mobile radio access networks
simpler and faster. SON has been specified already in 3rd
Generation Partnership Project (3GPP) Releases 8 and 9.
However, there have been updates to it in the following
releases. In addition, upcoming 5G will bring its own
challenges with variety of use cases with possibly conflicting
targets [1].
SON solutions can be divided into three categories: Self-
Configuration, Self-Optimization and Self-Healing. Self-
configuration stands for dynamic plug-and-play configuration
of newly deployed evolved Node Bs (eNBs). Self-healing
consists of a set of features for automatic detection and removal
of failures and automatic adjustment of parameters. Self-
optimization consists of optimization of coverage, capacity,
handovers (HO) and interference. Its primary use cases include
Mobility Robustness Optimization (MRO), Mobility Load
Balancing (MLB), Energy Efficiency (EE), Coverage and
Capacity Optimization (CCO), and Random Access Channel
optimization (RACH Optimization) [2]. In this article we focus
on MRO, which targets automatic detection and correction of
errors in the mobility configuration.
MRO has been studied to reduce both radio link failures
(RLFs) and unnecessary handovers due to improper handover
parameter settings. Generally, Hysteresis (Hys) and Time-To-
Trigger (TTT) are the two main MRO -tunable parameters
which govern HO performance. Most studies have focused on
optimizing these parameters to achieve a specific system gain.
A fuzzy logic-based Hys adjusting scheme was proposed in [3]
to reduce radio link failures (RLFs) in extremely dense small
cell networks. To promote Quality of Experience (QoE) in
high-speed railways, an adaptive TTT handover mechanism
was shown in [4]. The authors in [5] proposed a joint Hys and
TTT optimization scheme to enhance network energy
efficiency in heterogeneous network. However, as Hys and
TTT parameters apply to cell level, this approach is not taking
cell pair specific (CPS) handover optimization into account. [6]
introduced a CPS-based algorithm for the optimization of inter-
radio access technologies (RATs) handover parameters of LTE
and 3G systems. By conditionally adjusting CPS parameters,
[7] improved the performance of inter-cell interference
migration and [8] overcome the unbalanced load issues.
There exist several open source simulators focusing on LTE
technology, e.g., the LTE module of Network Simulator 3 (ns-
3) [9], Vienna LTE simulators [10] and LTE-SIM [11].
However, typical SON use cases have somewhat different
requirements when compared to these LTE system simulators.
For example, with SON the simulation time must be in the order
of several hours or even days. Simulation run time in a full-
blown system simulator may be several hours per simulated
minute. Thus, a new module for ns-3 called SonS3 (Self-
Organizing Network Simulator 3) has been developed to
support the SON use cases. The simulator framework has a
streamlined focus on mobility and handovers, and it has been
developed in co-operation with Finnish network operator Elisa.
In this article, to optimize MRO performance, we focus on
three issues, 1) present a dedicated system simulator for SON
research activities, 2) propose a dynamic CPS MRO algorithm
for optimizing Cell Individual Offset (CIO), and 3) present
MRO performance analysis in realistic urban scenario.
978-1-5386-6358-5/18/$31.00 ©2018 IEEE

II. MOBILITY ROBUSTNESS OPTIMIZATION
MRO is a solution for automatic detection and correction of
errors in the mobility configuration. The target is to minimize
mobility problems, such as HO failures, ping-pong HOs and
RLFs, by optimizing the HO related parameters, e.g., Hys, TTT
or CIO. Specifically, MRO intends to minimize the number of
HOs while keeping the amount of failures at acceptable level.
A. MRO classification
3GPP MRO focus is on RLFs due to too late, too early HOs,
or HOs to wrong cell. [2] Note, coverage (COV) problems are
out of the scope of MRO.
• Too late handover (TLH): RLF occurs in the source
cell before a HO is initiated OR during handover
procedure, and the UE attempts to reestablish its radio
link to another cell.
• Too early handover (TEH): RLF occurs in the target
cell shortly after a HO has been completed or during the
HO process, and the UE attempts to re-establish its radio
link back to the source cell.
• Handover to wrong cell (HWC): RLF occurs in the
target cell after a handover has been completed or during
the HO process, and the UE attempts to re-establish its
radio link to a cell which is not the source cell nor the
target cell.
In addition to RLF related MRO problems, there are also
less severe mobility problems, which do not involve RLFs.
• Ping-pong handover: HO occurs back-and-forth
between source and target cells within a short period of
time.
• Short stay handover: HO occurs to from a source cell
to the target cell, followed by to a third cell within a
short period of time.
However, we do not consider the ping-pong nor the short
stay HOs in this study. The MRO classification rules are
presented in more detail in TABLE I, where the A, B and C
stand for different eNBs involved in the HO procedure. The
time interval related to the HO completed in the past is
parameterized, but in the context of this paper a value of 3
seconds is used.
B. MRO algorithm
CIO has been shown to have a noticeable effect on the HO
performance in MRO [7][8]. In this article, the objective is to
study CIO based MRO algorithm (MRO A) and test its
performance by means of simulations.
The mobility problems are collected by the SON/MRO
module per cell pair, and the problems are classified to be too
late HOs (TLH), too early HOs (TEH) or HOs to wrong cell
(HWC), according to TABLE I. At the end of each MRO
optimization cycle, the amount of too slow (ntlh) and too fast
HOs (nteh + nhwc) are summed up with MRO problem specific
weight (wtle, wteh, whwc) per cell pair. This comparison is
illustrated in (1).
           (1)
If there are sufficient number of samples per cell pair to
make a reliable optimization decision (ntlh,cp + nteh,cp + nhwc,cp),
and there is enough difference in the weighed sums of the too
fast and too slow problems, the algorithm may trigger the HO
parameter optimization. The selected limit for the imbalance
between too fast and too slow event types is two thirds (0.67)
either way.
If there are more too slow HOs, the algorithm decreases the
cell specific offset (Ocn) value by a step-up value. On the other
hand, if there are more too fast HOs, the algorithm is increasing
the cell specific offset with a step-down value. The cell specific
offset is a part of the A3 event, which is assumed to be used in
the HO procedure. The A3 entering condition is presented in
(2): [5]
OffOcsOfsMsHysOcnOfnMn 
(2)
where
• Mn is the measurement result of the neighboring cell
• Ofn is the frequency specific offset of the frequency of
the neighbor cell
TABLE I. MRO PROBLEM CLASSIFICATION TABLE
HO completed in the
near past
HO in process
Cell at the time of
RLF
Reconnect to cell
after RLF
MRO problem
source cell
MRO problem target
cell
MRO problem
category
Irrelevant
-
A
A
A
A
COV
-
-
A
B
A
B
TLH
B to A
-
A
B
B
A
TEH
B to A
-
A
C
B
A
HWC
B to A
A to B
A|B
C
B
A
HWC
Irrelevant
A to B
A|B
A
A
B
TEH
-
A to B
A|B
B
A
B
TLH
B to A
A to B|C
A|B|C
B
B
A
TEH
-
A to B
A|B
C
A
B
HWC
B to A
A to C
A|C
C
B
A
TEH

• Ocn is the cell specific offset of the neighbor cell
• Ms is the measurement result of the serving cell, which
is not considered
• Ofs is the frequency specific offset of the serving
frequency
• Ocs is the cell specific offset of the serving cell
• Hys is the hysteresis parameter for this event
• Off is the offset parameter for this event
Thus, the algorithm aims to do local optimization per cell
pair be either advancing or delaying the HO based on the
classified MRO events.
III. SONS3 SYSTEM SIMULATOR
As mentioned before, there exist several open source
simulators, which could have been used as a baseline for SON
R&D studies [9] [10] [11]. However, as these simulators are not
dedicated SON simulators, a new module for ns-3 called Self-
Organizing Network Simulator 3 (SonS3) has been developed
to support the SON use cases.
The main idea is that the simulator is kept as simple as
possible, but still as realistic as possible to support the SON use
cases. In practice this means that the simulator focuses on a
couple of essential features:
• Realistic mobility models – the users are moving within
defined paths, e.g., roads, sidewalks or railways.
• Base station (BS) positions/orientations and matching
geographical propagation data from, e.g., a network
planning tool.
• Physical layer measurements e.g. Reference Symbol
Received Power (RSRP) and Channel Quality Indicator
(CQI).
• PHY layer measurement filtering, i.e,. L1 sliding window
and L3 filtering.
• RRC mobility events (esp. A3), and handover model at the
BS side
• Radio Link Failure model based on filtered wideband CQI
• MRO problem classification and MRO algorithm
However, the features, which are not in the focus, are either
simplified or omitted entirely. Such feature is, for example, the
LTE data plane (e.g. user traffic, scheduling). Omitting these
features improves the simulator performance drastically, since
the simulator can be run with a longer simulation resolution,
e.g., 10 or 50 ms.
IV. RESULTS
In this paper, we present a performance comparison in a
realistic urban scenario. A small city of Järvenpää in Finland,
was selected for the study with real BS orientations and
locations, propagation data from a network planning tool (see
Fig.1) and road and railroad grid from Open Street Maps (OSM)
(see Fig.2). The simulations have been run with the parameters
presented in TABLE II. UE velocity is road type specific: UEs
at railroads and highways move at 100 kmph (black lines), UEs
at intermediate roads move at 50 kmph (blue lines), and UEs in
residential area move at 20 kmph (magenta lines).
TABLE II. SIMULATION PARAMETERS
Parameter
Value
Simulation duration
24 hours
Number of cells
83
Number of active UEs
100
UE velocities
20, 50, 100 kmph
Load
50%
RSRP measurement bandwidth
6 RBs
RSRP measurement interval
50 ms
L1 filter samples
4
L3 filter coefficient
4
A3 hysteresis
3 dB
A3 time-to-trigger
200 ms
RLF Qout
-8 dB
RLF Qin
-6 dB
MRO optimization cycle
2 h
MRO required imbalance
2/3
MRO step-up / step-down
1 dB
MRO event type weights
1
MRO classification timer
3 s
Fig.1 and Fig.2 and present the RLF locations and related
MRO problem classifications on top of both the propagation
environment and waypoint street grid without MRO in active
use. MRO problem are colored as follows: red dots indicate
TLH, green dots indicate TEH, blue dots indicate HWC and
magenta dots indicate COV problems. As seen in the figures,
the red dots are clearly dominating, thus most of the problems
are caused due to too late HOs. In addition, most of the MRO
problems occur on the railroads and highways, where the UE
velocity is 100 kmph. As the velocity increases, the probability
of a UE experiencing a too late HO increases. Esp. the western
railroad seems to be problematic, since it goes through the city
center, where there are smaller cells resulting in a bit higher
interference and more frequent handovers.
The number of classified MRO events is presented with and
without MRO algorithm in Fig.3. It can be seen, that the total
amount of MRO problems (i.e. RLFs) is decreased about 25%
with the usage of the MRO algorithm. The most dominant
reason for RLFs is still the too late HO, but there is also a slight
relative increase in other types of MRO problems. Here, the
MRO algorithm speeds up the handover process which may at
the same time increase the too fast type of MRO problems.
MRO algorithm is run in periodical fashion, e.g., within this
study the MRO algorithm is run once per 2 hours resulting in a
total of 12 MRO iterations within a 24-hour simulation. Next,
we present the amount of MRO problems per iteration without
and with MRO in Fig.4, Fig.5, respectively. The MRO
problems decrease from a level of ~525 per iteration to about
~350 MRO problems per iteration. Thus, MRO algorithm
decreases the amount of MRO problems about 33%.
MRO algorithm tends to increase also the number of HOs.
This may be undesirable from the operator perspective, but, in
this case, the change is also a result of reduced RLFs. Thus, the
MRO problems have transformed into successful HOs. The
number of successful HOs per iteration are shown in Fig.6. The

number of successful HOs has increased about 6% when MRO
is enabled, which is in this case a quite moderate increase.
Fig.1. MRO problems on top of a Rx power map.
Fig.2. MRO problems on top of mobility waypoint grid.
Fig.3. Number of classified MRO problems.
The MRO algorithm works by changing dynamically the
cell pair specific offset, which affects directly the A3 events.
The default CIO is set to be 0 dB, which may be dynamically
changed, if the MRO optimization criteria are met, i.e., there
are statistically enough samples and significant difference in
too late and too fast samples, per cell pair.
Fig.4. Number of classified MRO problems per iteration without MRO.
Fig.5. Number of classified MRO problems per iteration with MRO.
Fig.6. Number of successful HOs per iteration.

In these simulations, we used a minimum number of 20
samples per cell pair, and a required an imbalance of 2/3s, either
way. These are quite moderate parameters, which should trigger
MRO case only in quite clear misconfigurations. The cell
individual offset per iteration have been presented in Fig.7.
Only the cell pairs where at least one MRO adaptation has been
done are shown. The CIO always starts at 0 dB, and it is
changed maximum once per MRO iteration (2 hours), and 1 dB
at a time.
It can be seen, that all CIO changes are towards faster HO
parameterization, which quite evident since most of the
problems are too late HOs. Only 10 cell pairs are modified
during the simulation run. This may be a result of quite
moderate parameterization for MRO algorithm, which causes
adaptation only in quite extreme cases. The changes occur
within the railroad and highway lines where the UEs are
moving with high velocity.
One important aspect in the simulation framework is the
visualization of the selected use case and related simulation
results. SonS3 supports an in-house developed visualization
tool called Network Event File (NEF) player. It enables
building of visual and configurable demonstrations for different
use cases and detailed inspection simulation run-time statistics.
A snapshot of MRO demonstration via NEF player is presented
in Fig.8.
Fig.7. Cell individual offset per iteration.
V. CONCLUSIONS
This article presented a dedicated simulator for Self-
Organizing Networks (SON) Mobility Robustness
Optimization (MRO) purposes, baseline MRO algorithm and
simulation-based performance analysis in realistic urban
scenario. The results indicate that 1) the developed simulator is
capable to handle simulation times of several hours to even
days, and 2) there is a clear gain for the MRO algorithm, even
though the algorithm targets optimization of a single parameter
at cell-pair level. The objective is to continue the work towards
more performant MRO algorithms and extending the simulator
framework capabilities in regard of other SON use cases. In
addition, even though the scope for this work has been in 3GPP
LTE networks, SON capabilities are assumed to be even more
important with the introduction of more capable, but also
complex networks, such as 5G / New Radio (NR). As such, one
of the goals has been to aim to ensure that the developed
simulator is relatively straightforward to be modified for the
future use cases.
Fig.8. NEF player visualization.
ACKNOWLEDGEMENTS
The authors would like to acknowledge Dr. Henrik
Martikainen from Magister Solutions Ltd. for valuable
comments and feedback.
REFERENCES
[1] 3GPP TR 36.902, Self-configuring and Self-optimizing Network (SON)
Use Cases and Solutions, Release 9, v9.3.1, 2011.
[2] S. Hämäläinen, H. Sanneck, C. Sartori, ”LTE Self-Organizing Networks
(SON)”, John Wiley & Sons Ltd., 2012.
[3] Silva, Ketyllen C., Zdenek Becvar, and Carlos RL Francês. "Adaptive
Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile
Networks with Dense Small Cells." IEEE Access (2018).
[4] Cho, Hanbyeog, et al. "Handover based on Maximum Cell Residence
Time and Adaptive TTT for LTE-R High-Speed Railways." KSII
Transactions on Internet & Information Systems 11.8 (2017).
[5] Zhang, Baoling, Weijie Qi, and Jie Zhang. "An Energy Efficiency and
Ping-pong Handover Ratio Optimization in Two-tier Heterogeneous
Networks." Computing and Communication Workshop and Conference
(CCWC), 2018 IEEE 8th Annual. IEEE, 2018.
[6] Awada, Ahmad, et al. "A SON-based Algorithm for the Optimization of
Inter-RAT Handover Parameters." IEEE Transactions on Vehicular
Technology 62.5 (2013): 1906-1923.
[7] Tuncel, Nur Oyku, and Mutlu Koca. "Joint Mobility Load Balancing and
Inter-Cell Interference Coordination for Self-Organizing OFDMA
Networks." Vehicular Technology Conference (VTC Spring), 2015 IEEE
81st. IEEE, 2015.
[8] Hasan, Md Mehedi, Sungoh Kwon, and Jee-Hyeon Na. "Adaptive
Mobility Load Balancing Algorithm for LTE Small-Cell Networks."
IEEE Transactions on Wireless Communications (2018).
[9] Ns-3, Network Simulator 3, https://www.nsnam.org/
[10] Vienna Simulators, https://www.nt.tuwien.ac.at/research/mobile-
communications/vccs/.
[11] LTE-Sim, LTE Simulator,
https://telematics.poliba.it/index.php?option=com_content&view=article
&id=28&Itemid=203&lang=en
[12] 3GPP, TS 36.331, Evolved Universal Terrestrial Radio Access (E-UTRA)
Radio Resource Control (RRC), Protocol specification, Release 9,
v.9.18.0, 2014