An AI-Driven Closed-Loop Assurance Framework for the Cloud-Native 5G Core Network: Autonomous Anomaly Detection and Self-Healing of Multi-Vendor Network Functions

Siddiq Bin Salam
Additional contact information
Siddiq Bin Salam: Saudi Telecom Company, Saudi Arabia

Technium, 2026, vol. 31, 508-522

Abstract: The migration of the mobile core to a cloud-native, service-based 5G Standalone (SA) architecture replaces a small number of monolithic nodes with dozens of dynamically scaled, containerised network functions (NFs) communicating over service-based interfaces. This elasticity is the source of 5G's agility, but it also defeats the manual, threshold-based operational practices that assured previous core generations: faults now propagate across NFs in seconds, and human-paced remediation cannot protect the service-level agreements (SLAs) of latency-critical slices. This paper proposes Autonomous Core Assurance and Remediation (ACAR), an AI-driven, closed-loop assurance framework that brings the zero-touch principle to the multi-vendor 5G core. ACAR is organised as an Observe–Analyse–Decide–Act control loop across three planes: an observability plane that collects NF KPIs, signalling rates, resource utilisation, and slice-SLA metrics over standardised interfaces; an AI analytics plane in which an autoencoder scores reconstruction error to detect anomalies on unlabelled telemetry and a classifier maps each fault to a remediation policy; and a remediation orchestration plane that executes policy-guarded self-healing actions — scale-out, restart, reroute, isolate — through NFV-MANO and the service mesh. The framework is aligned with the ETSI Zero-touch network and Service Management (ZSM) reference architecture and is designed for a sovereign, in-Kingdom 5G core. A numerical evaluation injects a signalling-storm anomaly against a multi-vendor SMF and shows that ACAR detects within a few seconds and reduces SLA-violation duration by roughly 80% and mean time to recovery by roughly two-thirds relative to a reactive baseline. All metrics are expressed against modelled baselines rather than proprietary operator data.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://techniumscience.com/index.php/technium/article/view/13738 (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:tec:techni:ojs:13738

DOI: 10.47577/technium.v31i.13738

Access Statistics for this article

Technium is currently edited by Scurtu Ionut Cristian

More articles in Technium from Technium Science
Bibliographic data for series maintained by Ana Maria Golita ().