An Experimental System for SON Verification

Tsvetko Tsvetkov∗ , Henning Sanneck† , and Georg Carle∗

∗ Department of Computer Science, Technische Universität München
Email: {tsvetko.tsvetkov, carle}@in.tum.de
† Nokia, Munich, Germany

Email: henning.sanneck@nsn.com

Abstract—In a mobile Self-Organizing Network (SON) the SON Verification Function

CM Data
usage of a SON coordinator is necessary to preclude the exe- Performance Report
cution of conflicting SON function instances. Typically, such a Area Analyzer for Verification Area Cell Level
coordinator is accountable for conflict prevention and resolution Function Function

and does not take into consideration that the activity of function Verification
Area
KPI Anomaly
Levels
instances may induce an undesired network behavior. In this
paper, we propose an experimental system for the verification of Area Resolver Anomaly Level
CM Undo
Request Function Function
Configuration Management (CM) changes induced by the activity
PM Data
of SON function instances. The main part of this system is the CM
Verification
Request
Undo
SON verification function which is triggered when CM change Execution
SON Coordinator ACK / NACK SONSON
Function
requests get acknowledged by the SON coordinator. It analyses Function
SON Function
Execution
the resulting Performance Management (PM) data and in case it Request PM / CM
Data
detects an undesired network behavior it sends a request to the
coordinator to undo the responsible changes. CM Change
LTE Network Simulator

I. I NTRODUCTION
Figure 1. SON Simulation System (S3) Components
In a Self-Organizing Network (SON), several SON function
instances may get active at the same time and try to perform
conflicting Configuration Management (CM) changes. In such
Motivated by the ideas of anomaly detection and diagnosis
a case, a SON coordinator [1] is typically employed to reject
we have developed the SON verification function. The purpose
the requests which would engage in conflicts and allow those
of our function is to assess the impact of SON-induced CM
which would guarantee a flawless network operation. Usually,
changes and provide the corrective action in case they have
the decisions of a coordinator are based on rules used to
caused an undesired or unusual network behavior. When action
anticipate and avoid known conflicts between SON function
requests get acknowledged, the SON coordinator forwards the
instances. In literature, such an approach is typically referred
task of observing the performance impact of those changes
to as pre-action SON coordination.
to our verification function. The coordinator sends for this
Unfortunately, approved CM changes may not necessarily purpose a verification request message identifying the cells
lead to the performance targeted by the corresponding SON that have been reconfigured by a SON function instance.
functions or by the operator himself. A SON coordinator does Moreover, the coordinator informs our function about the area
not consider the fact that the actions of a large number of influenced by that reconfiguration as well as the time the
deployed SON functions may cause an undesired network be- change has an effect on other running function instances. Based
havior [2]. Usually, a SON coordinator is designed for conflict on this information, our function determines where to look
prevention and resolution between SON function instances, for an anomaly in order to find the changes responsible for
e.g., taking care that two instances are not modifying the an undesired behavior to occur. The approach our function is
same CM parameter on the same cell. However, it does not following we also classify as post-action verification.
observe whether such changes have had a negative impact on
the network performance. II. SON S IMULATION S YSTEM
An anomaly detection and diagnosis mechanism [3] ad- To evaluate the behavior of the our concept, we have
dresses such kind of problems. The purpose of such a mech- developed the SON Simulation System (S3), as outlined in
anism is to detect an undesired network behavior, perform Figure 1. It consists of four components as described below.
root cause analysis and provide a corrective action. In order 1) LTE Network Simulator: We employ an Long Term
to assess the impact of a set of configuration changes, an Evolution (LTE) radio network simulator that allows us to
anomaly detector learns the faultless behavior of the network reconfigure 12 Evolved NodeBs (eNBs) at run-time. The
and later uses it as a basis of comparison to identify significant simulated network area equals to 50 km2 . The simulator
deviations from the usual behavior. To provide the corrective periodically computes and exports Performance Management
action, the diagnosis part typically learns the impact of faults (PM) data for every cell in so-called simulation rounds, each
on different performance indicators. corresponding to approximately 100 minutes in real time. At
978-1-4799-5863-4/14/$31.00
c 2014 IEEE the beginning of a round, the simulator configures the network
 100 1
as defined by the CM parameter setup. During a round, 1500
uniformly distributed mobile users follow a random walk 99

mobility model (speed 6 km/h) and use the network. The 98 0.9
constant bit rate requirement of each user is set to 256 kpbs.
Furthermore, the simulator makes use of the path loss radio 97

propagation model. At the end of a round, the simulator exports 96 0.8

PM data for each cell which is forwarded to the SON Function

HOSR

CQI
95
Engine (SFE) and our verification function as well.
94 0.7
2) SON Function Engine: The SFE is a runtime environ-
ment for SON functions which handles their communication 93

and configuration. Every time the LTE network simulator

92 0.6
completes a round, the SFE triggers the monitoring phase of
HOSR, enabled SON verification fuction
all SON functions. Should a CM change request be generated, 91 HOSR, disabled SON verification fuction
CQI, enabled SON verification fuction
it is immediately forwarded to the SON coordinator. Based CQI, disabled SON verification fuction
90 0.5
on the coordinator’s decision, the SFE deploys the requested 0 2 4 6 8 10 12 14 16 18
Simulation Round
CM parameter changes to the simulator. In total, the SFE
includes three optimization functions: the Mobility Robustness Figure 2. Impact on the HOSR and CQI
Optimization (MRO), Remote Electrical Tilt (RET), and Trans-
mission Power (TXP) function, as defined in [4]. Furthermore,
an instance of all three functions is running for each cell in The area analyzer function determines whether the cells within
the network. that scope are showing subnormal behavior and triggers a CM
undo for the CM change base if such a state can be confirmed.
3) SON Coordinator: The SON coordinator performs pre-
action coordination by employing the batch coordination con- III. E VALUATION
cept with dynamic priorities, as defined in [1]. The concept is
designed for batch processing of SON function requests. More Let us assume the following scenario: should TXP change
precisely, every SON function instance has an assigned bucket the transmission power in such a way that it induces a negative
and dynamic priority. The bucket initially contains a number of impact on the KPIs monitored by the RET function, the
tokens which are reduced every time a request by the function latter one may try to provide a corrective action which may
instance is accepted and increased if a request is rejected. If the not necessarily lead to an improvement. Moreover, since we
bucket gets empty, the priority is set to minimum. The priority employ a pre-action coordination mechanism that dynamically
can be increased again if a request is rejected. The coordinator adapts the priority of the running function instances, a high
collects all requests for a simulation round, determines the prioritized RET function may even prevent a low prioritized
conflicts and sends an Acknowledgment (Ack) for the requests TXP from undoing its change. Figure 2 outlines the result
with the highest priority and a Non-Acknowledgment (Nack) of the Handover Success Rate (HOSR) and Channel Quality
for the others. Indicator (CQI) captured by five cells in such a situation. Note
that the latter one is computed as the weighted harmonic mean
4) SON Verification Function: Our post-action verification of the CQI channel efficiency. Up to round 5, the observed
approach involves a tight integration with SON coordination. set of five cells are showing a normal and usual behavior. At
The SON verification function we propose analyzes the net- the time where the TXP function makes a wrong decision by
work performance for acknowledged action request of SON decreasing the transmission power of just two cells, all five
function instances. In case the activity of a given instance cells begin to experience an anomalous behavior. At this point,
causes an undesired network behavior, our function requests the advantage of employing our function can be seen. Instead
permission to undo the responsible CM changes from the SON of letting the functions slowly provide a solution, i.e., let
coordinator for the affected area. To achieve its task, though, RET and TXP adapt the coverage and the let MRO adjust the
the SON verification function function makes use of four Handover (HO) parameters, our function triggers a CM undo
helper functions: (1) an anomaly level, (2) a cell level, (3) an which returns the set of cells to the normal state experienced
area resolver, and (4) an area analyzer function. The anomaly before.
level function allows us to differentiate between normal and
subnormal cell Key Performance Indicator (KPI) values. The R EFERENCES
output is a KPI anomaly level which depicts the deviation of
[1] R. Romeikat, H. Sanneck, and T. Bandh, “Efficient , Dynamic Coordi-
a KPI from its expectation. The cell level function creates an nation of Request Batches in C-SON Systems,” in IEEE Veh. Technol.
overall performance metric of individual cells. The output is Conf. (VTC Spring). Dresden, Germany: IEEE, Jun. 2013.
the sum of the weighted anomaly levels which corresponds to [2] T. Kürner, M. Amirijoo, I. Balan, H. van den Berg, A. Eisenblätter
the cell level. The area resolver function defines the spatial et al., “Final Report on Self-Organisation and its Implications in Wireless
scope we are going to observe for anomalies. It consists of Access Networks,” Self-optimisation and self-configuration in wireless
a set Σ that includes the cells that have been reconfigured networks (SOCRATES), Deliverable D5.9, Jan. 2010.
by a SON function instance and a set of cells Ω that have [3] S. Nováczki, “An Improved Anomaly Detection and Diagnosis Frame-
work for Mobile Network Operators,” in 9th International Conference
been possibly influenced by that reconfiguration process. We on Design of Reliable Communication Networks (DRCN), Mar. 2013.
call Σ the CM change base and Ω the CM change extension [4] S. Hämäläinen, H. Sanneck, and C. Sartori, Eds., LTE Self-Organising
area. Altogether (i.e., Σ ∪ Ω) they compose the verification Networks (SON): Network Management Automation for Operational
area V which is the spatial scope we observe for anomaly. Efficiency. Chichester, UK: John Wiley & Sons, Dec. 2011.