Page 58 - Proceedings of the 2018 ITU Kaleidoscope

2018 ITU Kaleidoscope Academic Conference

management automation in 5G: Challenges and
opportunities”, PIMRC, Valencia, 2016

[2] H. Sanneck, S. Hämäläinen, C. Sartori, ”LTE Self-
Organizing Networks”, Wiley 2012.

[3] A. Zolli, A. M. Healy, “Resilience – Why Things
Bounce Back”, Headline Publishing Group, 2012.

[4] A. Berns, S. Ghosh, "Dissecting Self-* Properties",
Third IEEE International Conference on Self-
Figure 6 – Anomaly le1vel and anomalies detected by SEF Adaptive and Self-Organizing Systems, 2009
compared to labelled problem states
[5] S. S. Laster, A. O. Olatunji, "Automic Computing:
We can see that it detected the degradations on the last two Towards a Self-Healing System", Proceedings of the
days, but that in addition an anomaly is raised already during Spring, 2007
the second-last three-day cycle, giving an early warning on
the impeding problem that the QoE-driven backhaul SON is [6] D. Michalopoulos, B. Gajic, B. Crespo, A.
no longer able to resolve. The automated diagnosis of SEF
can distinguish between the radio and transport network Gopalasingham, J. Belschner, “Network Resilience
in Virtualized Architectures”, 11th International
degradations and appropriate corrective actions can be taken.
The deployment of corrective actions will be studied as Conference on Interactive Mobile Communication
Technologies and Learning (IMCL), 2017
future work.
[7] L. Bodrog, M. Kajo, S. Kocsis and B. Schultz, "A
8 CONCLUSIONS AND FUTURE WORK
robust algorithm for anomaly detection in mobile
networks," 2016 IEEE 27th Annual International
In this paper, we argued that to be able to meet the reliability Symposium on Personal, Indoor, and Mobile Radio
requirements of future 5G use cases, such as for the ultra- Communications (PIMRC), Valencia, 2016, pp. 1-6.
reliable communications, the networks need to be resilient
also against unforeseeable problems. We introduced the [8] Caglar Aytekin, Xingyang Ni, Francesco Cricri,
wider context of resilience in 5G networks and presented our
anomaly detection and diagnosis based self-healing concept, Emre Aksu, “Clustering and Unsupervised Anomaly
Detection with L2 Normalized Deep Auto-Encoder
which enables early detection of and reaction to problems in
a dynamic way. The method is generic and can be applied to Representations”, International Joint Conference on
Neural Networks (IJCNN), 2018
other radio technologies as well, e.g. to LTE. We discussed
how active and transfer learning methods can be used to
mitigate the diagnosis knowledgebase collection effort and [9] G. Ciocarlie, U. Lindqvist, S. Nováczki & H.
to combine the insights from both machine learning based Sanneck, “Detecting Anomalies in Cellular
analytics and the human expert. Furthermore, we presented Networks Using an Ensemble Method”, 9th
a holistic self-healing method over several management International Conference on Network and Service
areas as an enabler for a truly over-arching and resilient Management (CNSM 2013), pp. 171–174.
solution for self-healing. Such over-arching methods require
not only the transfer of data, but transfer of knowledge, for [10] G. Ciocarlie et. al., “Diagnosis cloud: Sharing
example of the anomaly patterns and their diagnosis labels knowledge across cellular networks”, 12th
or features indicating the corrective actions executed on International Conference on Network and Service
different management areas, between organizations and Management (CNSM), Montreal, 2016
vendors. Lastly, we introduced the SON Experimental
Framework (SEF), where the self-healing capabilities have [11] L. Bajzik, C. Deak, T. Karasz, P. Szilagi, Z. Vincze,
been implemented and demonstrated using data from real C. Vulkan, “QoE Driven SON for Mobile Backhaul
th
network instances as well as in live integrations. In SEF, we Demo”, 6 International Workshop on Self-
also evaluated the cooperative self-healing between QoE- Organizing Networks (IWSON), 2016
driven backhaul and Radio Access Network Management
SON functions. In our future work, the transfer learning [12] L. Bodrog, M. Kajo, S. Kocsis, B. Schultz,
methods will be also further studied in the context of self- “Demonstrator of KPI Analytics for Anomaly
healing as well as looking into the deployment of corrective Detection and Diagnosis in Mobile Networks”, The
actions. 28th Annual IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications
9 REFERENCES (IEEE PIMRC), Montreal, 2017

[1] S. Mwanje, G. Decarreau, C. Mannweiler, M.
Naseer-ul-Islam, L. C. Schmelz, “Network

– 42 –