For a variety of reasons (including, for example, increasing cyber security threats, increased network heterogeneity, the increased use of virtualization technologies, etc.) maintaining the fifth generation (5G) networks of tomorrow will be challenging. To address such challenges techniques are presented herein that support a multivariate time series unsupervised method for anomaly detection using Key Performance Indicators (KPIs) that are derived from various infrastructure level metrics collected from all kinds of networking nodes deployed in 5G networks. Multivariate time series unsupervised anomaly detection is the future of anomaly detection with systems generating real-time time series data but this is an area that has not yet been explored with 5G. This invention introduces anomaly detection in 5G networks at the node level or deployment level instead of simply monitoring anomalous behavior in a particular KPI on a particular node. Additionally, aspects of the techniques presented herein provide a semi-automated method for anomaly diagnosis and deep-dive anomaly analytics in support of better systems of the future. The main focus herein lies in anomaly detection in modern 5G network deployments at the complete node or deployment level using infrastructure level data from 5G nodes to detect and prevent network failures. We present a multivariate time-series unsupervised AI algorithm to solve this problem which helps detect anomalous behavior and in turn facilitates better network troubleshooting and capacity forecasting.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Ghosh, Sreyan; ., Sakshi; and Kataria, Vijay, "MULTIVARIATE TIME SERIES UNSUPERVISED ANOMALY DETECTION AND DIAGNOSIS IN 5G NETWORKS", Technical Disclosure Commons, (December 14, 2020)