Mobile network monitoring is not a trivial task to perform. The complexity originates mainly from the usage of complex technologies and a large number of users connected to the network. This thesis deals with setting a baseline for mobile network measurement data analysis and anomaly detection.
Firstly, a few significant mechanisms of LTE technology are presented. The usage of these mechanisms is important for providing high data throughput and is reflected in measurement results. I continue with the description of various types of measurements in mobile networks. In addition to this, key challenges that should be considered when performing these measurements are presented. Furthermore, a new field of measurements with user-equipment (i.e. mobile phones) acting as a measurement device is introduced. qMON is a measurement system for mobile networks which enables user-equipment measurements. I present the existing qMON architecture and explain how an anomaly detection system can be added to it.
Four experiments were performed using the qMON system and a hypothesis was formed for each of the experiments. I explain, why each of the hypotheses is important for a successful implementation of anomaly detection system. Analysis of measurement results is performed with Python in order to test the hypotheses.
Initial part of data analylisis is carried out with the intention of showing the importance of expert knowledge in mobile network analysis. This importance is demonstrated through a feature subset selection process which is followed by the definition of the problem – we are dealing with multivariate and unevenly sampled time series. I explain the procedure which was used in order to convert the problem to an evenly sampled time series problem. The converted data, which is the result of the aforementioned procedure, is used to test the hypotheses. In the testing phase I focus on a subset of attributes, namely the quality of the received reference radio signal (RSRQ) and the measured network data throughput. A significant 24-hour seasonal component is discovered in the data, as well as the effect that the stress test of the base station has on the measurement results. I also show that the number of active devices in a cell cannot be assessed through RSRQ measurements. Key differences between LTE and UMTS technologies are presented as the final analysis result. The thesis is concluded with an estimate of how close we are to a functioning anomaly detection system.
|
