Methods for Self-Healing based on traces and unsupervised learning in Self-Organizing Networks

Abstract

With the advent of Long-Term Evolution (LTE) networks and the spread of a highly varied range of

services, mobile operators are increasingly aware of the need to strengthen their maintenance and

operational tasks in order to ensure a quality and positive user experience. Furthermore, the co-

existence of multiple Radio Access Technologies (RAT), the increase in the traffic demand and the need

to provide a great variety of services are steering the cellular network toward a new scenario where

management tasks are becoming increasingly complex. As a result, mobile operators are focusing their

efforts to deal with the maintenance of their networks without increasing either operational

expenditures (OPEX) or capital expenditures (CAPEX). In this context, it is becoming necessary to

effectively automate the management tasks through the concept of the Self-Organizing Networks (SON).

In particular, SON functions cover three different areas: Self-Configuration, Self-Optimization and Self-

Healing. Self-Configuration automates the deployment of new network elements and their parameter

configuration. Self-Optimization is in charge of modifying the configuration of the parameters in order to

enhance user experience. Finally, Self-Healing aims reduce the impact that failures and services

degradation have on the end-user. To that end, Self-Healing (SH) systems monitor the network elements

through several alarms, measurements and indicators in order to detect outage and degraded cells,

then, diagnose the cause of their problem and, finally, execute the compensation or recovery actions.

Even though mobile networks are become more prone to failures due to their huge increase in

complexity, the automation of the troubleshooting tasks through the SH functionality has not been fully

realized. Traditionally, both the research and the development of SON networks have been related to

Self-Configuration and Self-Optimization. This has been mainly due to the challenges that need to be

faced when SH systems are studied and implemented. This is especially relevant in the case of fault

diagnosis. However, mobile operators are paying increasingly more attention to self-healing systems,

which entails creating options to face those challenges that allow the development of SH functions.

On the one hand, currently, the diagnosis continues to be manually done since it requires considerable

hard-earned experience in order to be able to effectively identify the fault cause. In particular,

troubleshooting experts thoroughly analyze the performance of the degraded network elements by

means of measurements and indicators in order to identify the cause of the detected anomalies and

symptoms. Therefore, automating the diagnosis tasks means knowing what specific performance

indicators have to be analyzed and how to map the identified symptoms with the associate fault cause.

This knowledge is acquired over time and it is characterized by being operator-specific based on their

policies and network features. Furthermore, troubleshooting experts typically solve the failures in a

network without either documenting the troubleshooting process or recording the analyzed indicators

along with the label of the identified fault cause. In addition, because there is no specific regulation on

documentation, the few documented faults are neither properly defined nor described in a standard

way (e.g. the same fault cause may be appointed with different labels), making it even more difficult to

automate the extraction of the expert knowledge. As a result, this a lack of documentation and lack of

historical reported faults makes automation of diagnosis process more challenging.

On the other hand, when the exact root cause cannot be remotely identified through the statistical

information gathered at cell level, drive test are scheduled for further information. These drive tests aim

to monitor mobile network performance by using vehicles to personally measure the radio interface

quality along a predefined route. In particular, the troubleshooting experts use specialized test

equipment in order to manually collect user-level measurements. Consequently, drive test entail a hefty

expense for mobile operators, since it involves considerable investment in time and costly resources

(such as personal, vehicles and complex test equipment). In this context, the Third Generation

Partnership Project (3GPP) has standardized the automatic collection of field measurements (e.g.

signaling messages, radio measurements and location information) through the mobile traces features

and its extended functionality, the Minimization of Drive Tests (MDT). In particular, those features allow

to automatically monitor the network performance in detail, reaching areas that cannot be covered by

drive testing (e.g. indoor or private zones). Thus, mobile traces are regarded as an important enabler for

SON since they avoid operators to rely on those expensive drive tests while, at the same time, provide

greater details than the traditional cell-level indicators. As a result, enhancing the SH functionalities

through the mobile traces increases the potential cost savings and the granularity of the analysis. Hence,

in this thesis, several solutions are proposed to overcome the limitations that prevent the development

of SH with special emphasis on the diagnosis phase. To that end, the lack of historical labeled databases

has been addressed in two main ways. First, unsupervised techniques have been used to automatically

design diagnosis system from real data without requiring either documentation or historical reports

about fault cases. Second, a group of significant faults have been modeled and implemented in a

dynamic system level simulator in order to generate an artificial labeled database, which is extremely

important in evaluating and comparing the proposed solutions with the state-of- the-art algorithm. Then,

the diagnosis of those faults that cannot be identified through the statistical performance indicators

gathered at cell level is automated by the analysis of the mobile traces avoiding the costly drive test. In

particular, in this thesis, the mobile traces have been used to automatically identify the cause of each

unexpected user disconnection, to geo-localize RF problems that affect the cell performance and to

identify the impact of a fault depending on the availability of legacy systems (e.g. Third Generation, 3G).

Finally, the proposed techniques have been validated using real and simulated LTE data by analyzing its

performance and comparing it with reference mechanisms.