5G/5G+ network management employing AI-based continuous deployment.

Purpose: The paper proposes a novel concept of deployment process dedicated to continuous development, integration, testing, and upgrading the software-defined AI/ML-based functionalities in 5G/5G+ cellular networks, reliable and minimally invasive for network infrastructure, users, and service quality, automating DevOps and performance management towards the zero-touch networks. The idea is demonstrated by the continuous deployment framework, where a virtual cluster of representative cellular network resources coexists synchronously with the operator network serving the thousands-to-millions of end-users. Performance KPIs measured in this twinning subnetwork is the time series exhibiting the complex interrelations governing a temporal state of the whole system. Methods: One of the crucial problems is the detection when the changes of configuration of individual base station impacted its performance, which was realized by change points identification using an enhanced-PELT (e-PELT) algorithm. This yielded a sequence of complex data patterns (between successive change points) into sections of unique waveforms corresponding to stable network conditions. The e-PELT scheme includes a peak removal and an ACPO module for automatic penalty search. Results: e-PELT algorithm, validated against experimental data in comparative studies, outperforms the AMOC, Binary Segmentation, and BOCPD schemes, increasing the F1-score by more than 12% and the Accuracy by 10% over the PELT baseline. Conclusion: The results obtained in a CD cluster with fixed and automated penalty search show that e-PELT provides outstanding results, proving that the proposed approach can be successfully applied to analyze data sequences from different 5G network configurations. The concept of AI/ML-featured Continuous Deployment Framework dedicated to cellular networks can impact the development and integration of new telecommunication features, including possible extension for the role of standardized NWDAF module in customer networks, enhanced with the application of methodology outlined in the paper. (c) 2023 Elsevier B.V. All rights reserved.