Data analysis methods for cellular network performance optimization

Abstract

Modern cellular networks including GSM/GPRS and UMTS networks offer faster and more versatile communication services for the network subscribers. As a result, it becomes more and more challenging for the cellular network operators to enhance the usage of available radio resources in order to meet the expectations of the customers.

Cellular networks collect vast amounts of measurement information that can be used to monitor and analyze the network performance as well as the quality of service.

In this thesis, the application of various data-analysis methods for the processing of the available measurement information is studied in order to provide more efficient methods for performance optimization. In this thesis, expert-based methods have been presented for the monitoring and analysis of multivariate cellular network performance data. These methods allow the analysis of performance bottlenecks having an effect in multiple performance indicators.

In addition, methods for more advanced failure diagnosis have been presented aiming in identification of the causes of the performance bottlenecks. This is important in the analysis of failures having effect on multiple performance indicators in several network elements.

Finally, the use of measurement information in selection of most useful optimization action have been studied. In order to obtain good network performance efficiently, the expected performance of the alternative optimization actions must be possible to evaluate. In this thesis, methods to combine measurement information and application domain models are presented in order to build predictive regression models that can be used to select the optimization actions providing the best network performance.