ARCADE: uma metodologia orientada a dados para modelagem e aprimoramento da cobertura de redes celulares

Abstract

The evolution of mobile networks and the increasing complexity of their management have made fine-tuning the RF environment progressively more challenging. As a result, standardization bodies and industry players have been investing in the automation of labor-intensive and complex processes, particularly the improvement of RAN coverage in cellular systems. Some proposals, such as the MDT functionality introduced in recent 3GPP releases, aim to collect raw network data to enable the development of tools that automate tasks previously carried out by specialized engineers. However, as observed in practice, RF coverage data are often sparse and do not by themselves provide comprehensive information about cell coverage, making systematic and thorough analysis difficult. The main problems leading to coverage efficiency loss are known as overshooting, where a cell covers more area than intended, interfering with and reducing the capacity of neighboring cells, and undershooting, where coverage falls well short of the intended area, causing service gaps and underutilization of network resources. This work proposes the Adaptative Radio Coverage Analysis through Data-driven Evolution (ARCADE) methodology, which, drawing on RF data obtained by any means (such as MDT, drive tests, or crowdsourcing), enables the determination of an improved network coverage configuration through an evolutionary process, designed to be eventually automated within the system itself. The proposal also assumes that no design data are available (such as tower heights, antenna orientation, or antenna models, or terrain databases), allowing a self-contained implementation within the cellular system itself. This eliminates dependence on the reliability and accuracy of such data and the need to acquire and integrate these databases, simplifying solution deployment, while also avoiding the deviations inherent in the various theoretical mathematical propagation prediction models commonly used today. This work can be divided into two distinct problems to be investigated sequentially. The first arises from the need for adequate modeling of the RF environment based on data that are often sparse, depending on the volume of users participating in the feature or application. It must also be considered that the data will contain outlier points, stemming from reading or processing errors, or from highly sporadic measurements outside the intended context (such as those from user devices located on high rooftops), which can radically distort coverage modeling if not discarded. Furthermore, the samples may also contain anomalies arising from design faults, such as overshooting or undershooting cells, and these anomalies must be correctly represented in the model. The second problem, after the RF environment has been mapped, is the determination of a set of network parameter adjustments in order to achieve better coverage efficiency with quality -- that is, carrier coverage with effective dominance over interference -- in the system. In this context, this thesis proposes a methodology encompassing all stages of a network coverage improvement process that is self-contained within the network context and follows a data-driven approach -- that is, one that does not rely on mathematical modeling (e.g., radio frequency propagation models). This methodology is proposed end-to-end, beginning with an examination of the data acquisition stage (focused on a practical crowdsourcing implementation as an alternative to drive test), proceeding through RF environment modeling via data extrapolation methods that exclude outliers while preserving representations of network anomalies, and finally proposing a cluster coverage improvement method that yields a network configuration with superior performance relative to the initial configuration. Results show that both the modeling methodology is well-suited to the proposed problem and the coverage improvement method achieved significant gains in coverage quality, with cell power adjustments (increases or reductions) within the cluster improving both the covered area and the area with single-server dominance -- the ideal condition for interference reduction in a cellular system.