Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy-Efficient Intra-Domain Network Slicing for Multi-Layer Orchestration in Intelligent-Driven Distributed 6G Networks: Learning Generic Assignment Skills with Unsupervised Reinforcement Learning

Navideh Ghafouri, John S. Vardakas, Kostas Ramantas, Christos Verikoukis

TL;DR

A pre-train phase is proposed to discover useful assignment skills in network edge domains by utilizing unsupervised Reinforcement Learning (unsupervised RL), which enables scalability and decentralization, but it also delivers efficiency by assisting domain controllers to provide various service types.

Abstract

Since the 6th Generation (6G) of wireless networks is expected to provide a new level of network services and meet the emerging expectations of the future, it will be a complex and intricate networking system. 6Gs sophistication and robustness will be accompanied by complexities, which will require novel strategies to tackle them. This research work focuses on decentralized and multi-level system models for 6G networks and proposes an energy efficient automation strategy for edge domain management and Network Slicing (NS) with the main objective of reducing the networks complexity by leveraging scalability, efficiency, and generalization. Accordingly, we propose a pre-train phase to discover useful assignment skills in network edge domains by utilizing unsupervised Reinforcement Learning (unsupervised RL). The suggested technique does not depend on the domain specifications and thus is applicable to all the edge domains. Our proposed approach not only enables scalability and decentralization, but it also delivers efficiency by assisting domain controllers to provide various service types. We implemented the pre-training phase, and monitored that the discovered assignment skills span the entire interval of possible resource assignment portions for every service type.

Energy-Efficient Intra-Domain Network Slicing for Multi-Layer Orchestration in Intelligent-Driven Distributed 6G Networks: Learning Generic Assignment Skills with Unsupervised Reinforcement Learning

TL;DR

A pre-train phase is proposed to discover useful assignment skills in network edge domains by utilizing unsupervised Reinforcement Learning (unsupervised RL), which enables scalability and decentralization, but it also delivers efficiency by assisting domain controllers to provide various service types.

Abstract

Since the 6th Generation (6G) of wireless networks is expected to provide a new level of network services and meet the emerging expectations of the future, it will be a complex and intricate networking system. 6Gs sophistication and robustness will be accompanied by complexities, which will require novel strategies to tackle them. This research work focuses on decentralized and multi-level system models for 6G networks and proposes an energy efficient automation strategy for edge domain management and Network Slicing (NS) with the main objective of reducing the networks complexity by leveraging scalability, efficiency, and generalization. Accordingly, we propose a pre-train phase to discover useful assignment skills in network edge domains by utilizing unsupervised Reinforcement Learning (unsupervised RL). The suggested technique does not depend on the domain specifications and thus is applicable to all the edge domains. Our proposed approach not only enables scalability and decentralization, but it also delivers efficiency by assisting domain controllers to provide various service types. We implemented the pre-training phase, and monitored that the discovered assignment skills span the entire interval of possible resource assignment portions for every service type.

Paper Structure

This paper contains 9 sections, 4 figures.

Table of Contents

  1. Introduction
  2. from network softvarization and conventional communication-oriented management to fully intelligent networks with task-oriented approaches
  3. A multi-layer and decentralized AI-driven network system
  4. Enabling technologies: RL, FL, TL and Unsupervised RL
  5. Our Methodology: Unsupervised RL-based Assignment Skill Learning
  6. Generic unsupervised skill discovery for automation of distributed network domains: An energy-efficient approach
  7. Prospective implementations and Advantages: HRL Orchestration
  8. Energy Efficiency
  9. Conclusions and future steps

Figures (4)

  • Figure 1: Network´s general framework with a close look to the edge domains
  • Figure 2: Automation of edge domains: (a) Domains' management scheme in the system model, (b) A closer look to the offline and Online phases
  • Figure 3: Discovered Skills
  • Figure 4: Coverage of the discovered skills across the resources of each domain