A canonical correlation-based framework for performance analysis of radio access networks

Abstract

Data driven optimization and machine learning based performance diagnostics of radio access networks entails significant challenges arising not only from the nature of underlying data sources but also due to complex spatio-temporal relationships and interdependencies between cells due to user mobility and varying traffic patterns. We discuss how to study these configuration and performance management data sets and identify relationships between cells in terms of key performance indicators using multivariate analysis. To this end, we leverage a novel framework based on canonical correlation analysis (CCA), which is a highly effective method for not only dimensionality reduction but also for analyzing relationships across different sets of multivariate data. As a case study, we discuss energy saving use-case based on cell shutdown in commercial cellular networks, where we apply CCA to analyze the impact of capacity cell shutdown on the KPIs of coverage cell in the same sector. Data from LTE Network is used to analyzed example case. We conclude that CCA is a viable approach for identifying key relationships not only between network planning and configuration data, but also dynamic performance data, paving the way for endeavors such as dimensionality reduction, performance analysis, and root cause analysis for performance diagnostics.