Aspects of knowledge mining on minimizing drive tests in self-organizing cellular networks
Research output: Book/Report › Doctoral thesis › Collection of Articles
|Place of Publication||Tampere|
|Publisher||Tampere University of Technology|
|Number of pages||127|
|Publication status||Published - 22 Aug 2014|
|Publication type||G5 Doctoral dissertation (article)|
|Name||Tampere University of Techology. Publication|
|Publisher||Tampere University of Technology|
The demand for mobile data traffic is about to explode and this drives operators to find ways to further increase the offered capacity in their networks. If networks are deployed in the traditional way, this traffic explosion will be addressed by increasing the number of network elements significantly. This is expected to increase the costs and the complexity of planning, operating and optimizing the networks. To ensure effective and cost-efficient operations, a higher degree of automation and self-organization is needed in the next generation networks. For this reason, the concept of self-organizing networks was introduced in LTE covering multitude of use cases. This was specifically done in the areas of self-configuration, self-optimization and selfhealing of networks. From an operator’s perspective, automated collection and analysis of field measurements while complementing the traditional drive test campaigns is one of the top use cases that can provide significant cost savings in self-organizing networks.
This thesis studies the Minimization of Drive Tests in self-organizing cellular networks from three different aspects. The first aspect is network operations, and particularly the network fault management process, as the traditional drive tests are often conducted for troubleshooting purposes. The second aspect is network functionality, and particularly the technical details about the specified measurement and signaling procedures in different network elements that are needed for automating the collection of the field measurement data. The third aspect concerns the analysis of the measurement databases that is a process used for increasing the degree of automation and self-awareness in the networks, and particularly the mathematical means for autonomously finding meaningful patterns of knowledge from huge amounts of data. Although the above mentioned technical areas have been widely discussed in previous literature, it has been done separately and only a few papers discuss how for example, knowledge mining is employed for processing field measurement data in a way that minimizes the drive tests in self-organizing LTE networks.
The objective of the thesis is to use well known knowledge mining principles to develop novel self-healing and self-optimization algorithms. These algorithms analyze MDT databases to detect coverage holes, sleeping cells and other geographical areas of anomalous network behavior. The results of the research suggest that by employing knowledge mining in processing the MDT databases, one can acquire knowledge for discriminating between different network problems and detecting anomalous network behavior. For example, downlink coverage optimization is enhanced by classifying RLF reports into coverage, interference and handover problems. Moreover, by incorporating a normalized power headroom report with the MDT reports, better discrimination between uplink coverage problems and the parameterization problems is obtained. Knowledge mining is also used to detect sleeping cells by means of supervised and unsupervised learning. The detection framework is based on a novel approach where diffusion mapping is used to learn about network behavior in its healthy state. The sleeping cells are detected by observing an increase in the number of anomalous reports associated with a certain cell. The association is formed by correlating the geographical location of anomalous reports with the estimated dominance areas of the cells.
Moreover, RF fingerprint positioning of the MDT reports is studied and the results suggest that RF fingerprinting can provide a quite detailed location estimation in dense heterogeneous networks. In addition, self-optimization of the mobility state estimation parameters is studied in heterogeneous LTE networks and the results suggest that by gathering MDT measurements and constructing statistical velocity profiles, MSE parameters can be adjusted autonomously, thus resulting in reasonably good classification accuracy.
The overall outcome of the thesis is as follows. By automating the classification of the measurement reports between certain problems, network engineers can acquire knowledge about the root causes of the performance degradation in the networks. This saves time and resources and results in a faster decision making process. Due to the faster decision making process the duration of network breaks become shorter and the quality of the network is improved. By taking into account the geographical locations of the anomalous field measurements in the network performance analysis, finer granularity for estimating the location of the problem areas can be achieved. This can further improve the operational decision making that guides the corresponding actions for example, where to start the network optimization. Moreover, by automating the time and resource consuming task of tuning the mobility state estimation parameters, operators can enhance the mobility performance of the high velocity UEs in heterogeneous radio networks in a cost-efficient and backward compatible manner.