Reverb Networks has developed a system that enables Self Optimizing Networks (SON). The system aims to maximize network performance by changing the tilts of antennas to shift load between them. SON can significantly reduce the need for labor intensive manual network optimization.Changes are currently applied with the assumption that the same tilt value will be valid at all times of the day. This project investigates whether the existing solution can be modified, with little effort, to find appropriate times to apply different tilts, using Artificial Intelligence and Machine Learning techniques, to possibly achieve a more optimal outcome.The given solution groups (clusters) segments in time using unsupervised learning. In addition, nearby cells (a cell is an area that normally has more than one antenna serving it) are grouped (clustered) to form zones with similar time segments. These individual time segments are then optimized by separate instances of the existing solution.The learning algorithms are evaluated with a dataset that contains information from 195 cells, which has been collected from a live network over the course of one month. The results show that several time segments can be identified for most cells. In many cases, these time segments are different in length, start time, and end time.A prototype is presented, which shows that it is possible to integrate the suggested approach with the existing solution and that the necessary changes are not extensive.While it is not yet possible to show that the proposed solution surely leads to a more optimized network (this will require testing in a live network or extensive network simulation), the fact that multiple time segments could be identified for most cells is encouraging. And it does seem reasonable to expect that more frequent changes will provide gains in network performance.
[1]
J. H. Ward.
Hierarchical Grouping to Optimize an Objective Function
,
1963
.
[2]
J. MacQueen.
Some methods for classification and analysis of multivariate observations
,
1967
.
[3]
Martin Odersky,et al.
An Overview of the Scala Programming Language
,
2004
.
[4]
George Nagy,et al.
In search of meaning for time series subsequence clustering: matching algorithms based on a new distance measure
,
2006,
CIKM '06.
[5]
Ecole Doctorale Edite.
Automated RRM Optimization of LTE networks using Statistical Learning
,
2010
.
[6]
Samson Lasaulce,et al.
Learning equilibria with partial information in decentralized wireless networks
,
2011,
IEEE Communications Magazine.
[7]
Eamonn J. Keogh,et al.
Probabilistic discovery of time series motifs
,
2003,
KDD '03.
[8]
Samir Ranjan Das,et al.
Understanding traffic dynamics in cellular data networks
,
2011,
2011 Proceedings IEEE INFOCOM.
[9]
Erik Dahlman,et al.
3G Evolution, Second Edition: HSPA and LTE for Mobile Broadband
,
2008
.
[10]
Ian H. Witten,et al.
The WEKA data mining software: an update
,
2009,
SKDD.
[11]
Eamonn J. Keogh,et al.
Clustering of time-series subsequences is meaningless: implications for previous and future research
,
2003,
Third IEEE International Conference on Data Mining.
[12]
Jessica Lin,et al.
Finding Motifs in Time Series
,
2002,
KDD 2002.