Semi-supervised learning-based coverage hole detection in cellular networks

Abstract

For any time-critical mobile network-dependent applications and services, coverage is one of the prominent factors for providing the best Quality of Service (QoS) and Quality of Experience (QoE). A simple Coverage Hole (CH) may degrade the performance and the reputation of any operator by reducing the Key Performance Indicators (KPIs). This is one of the important aspects which need to be planned from the phase of network deployment throughout the whole operational stage. Many factors can cause CH such as attenuation, obstacles and improper network planning. Traditionally, a Drive Test (DT) used to be carried out in order to assess the quality of the mobile network signal. With technological advancement, DT has been replaced by the Minimization of Drive Test (MDT) and included as a part of Self-Organizing Networkss (SONs). The MDT process is applicable to networks that operate in 3G, 4G and 5G technologies. With this method, operators are able to measure network performance with the help of end users' devices. Thus, the network can be managed more conveniently, performance is improved, quality is increased, and maintenance costs are reduced for the network. However, the processing of MDT at the operators' side remains time-consuming and complex especially for CH analysis and detection from mobile network data. Therefore, we present a method by utilising Semi-Supervised Learning (SSL) in this paper so that this task becomes uncomplicated with improved accuracy. Our results show that the proposed method achieves better accuracy than the usual classification algorithm.