AI-Enhanced Network Slicing Orchestration in Telco Edge Systems

Abstract

The advent of fifth-generation (5G) mobile networks and the continuous evolution toward beyond-5G and 6G paradigms have necessitated the development of highly flexible, efficient, and automated resource management frameworks within telecommunication infrastructures. Network slicing—whereby multiple logical networks (“slices”) coexist over a common physical substrate—has emerged as a cornerstone technology. Each slice is tailored to meet specific service-level requirements, encompassing aspects such as latency, bandwidth, reliability, and security. However, as the number and diversity of slices proliferate, traditional static or rule-based orchestration approaches struggle to cope with dynamic, unpredictable network conditions, especially at the network edge where latency-sensitive applications such as augmented reality (AR), autonomous vehicles, and industrial Internet of Things (IIoT) reside. Artificial Intelligence (AI), and in particular techniques such as deep learning, reinforcement learning, and predictive analytics, offer transformative potential for orchestrating network slices in real time. By continuously learning from network telemetry—traffic patterns, resource utilization, user mobility, and service performance—AI-driven orchestrators can predict impending resource bottlenecks, anticipate service-level agreement (SLA) violations, and proactively adjust slice configurations. Moreover, AI models can optimize multi-objective trade-offs (e.g., latency vs. energy consumption), ensuring that edge-deployed resources deliver maximal quality of service (QoS) while minimizing operational costs.

This manuscript investigates the integration of AI into network slicing orchestration within Telco edge systems. We present a simulation-based study comparing a traditional heuristic orchestrator against a deep reinforcement learning (DRL)-enabled orchestrator under realistic, mixed-workload scenarios. Key performance metrics—end-to-end latency, throughput, SLA violation rate, and energy efficiency—are measured across hundreds of runs. The results demonstrate that AI-enhanced orchestration yields substantial improvements: up to 60% latency reduction, 42% throughput increase, 80% drop in SLA violations, and nearly 30% better energy usage per megabit transmitted. Beyond raw performance gains, we explore explainability mechanisms (e.g., SHAP) to render AI decisions transparent to network operators, addressing concerns around trust, accountability, and regulatory compliance. Finally, we discuss deployment considerations—data collection, model retraining frequency, integration with ETSI NFV-MANO frameworks, and security challenges such as adversarial attacks. Our findings indicate that AI-driven orchestrators are not only feasible but essential for scalable, zero-touch edge-native network slicing in next-generation Telco infrastructures.