Unsupervised Performance Functions for Wireless Self-Organising Networks

Traditionally, in cellular networks, troubleshooting experts have manually analyzed Key Performance Indicators (KPI), so that they could identify the cause of problems and fix them. With the emergence of Self-Organizing Networks, Self-Healing systems are designed to automate those troubleshooting tasks. With that aim, the behavior of the KPIs (i.e. their profile under normal and abnormal conditions) needs to be modeled. Since the behavior of the KPIs is network-dependent and it changes as the network evolves, their profile should be automatically defined and readjusted depending on the characteristics of the network. Therefore, in this letter, an automatic process to model the KPIs based on the real data taken from the network is proposed. In particular, this method is characterized by designing a pair of functions (named performance functions) from the statistical behavior of real data without requiring any information about the existence of faults (i.e. unsupervised learning). Results have shown the reliability and effectiveness of the proposed method in comparison to reference approaches.