Scholarlink

The exponential growth of mobile data traffic and the proliferation of connected devices have positioned energy efficiency as a critical challenge in fifth-generation (5G) networks. While significant research efforts have addressed individual energy-saving pillars—Advanced Sleep Modes (ASM), Dynamic Spectrum Management (DSM), Network Densification (ND), and Energy Harvesting (EH)—these approaches operate in isolation, leading to suboptimal energy utilization and potential conflicts between optimization objectives. This paper introduces a comprehensive framework that combines four energy-efficiency pillars into a single integrated machine learning-based Self-Organizing Network (SON) model. The proposed framework employs Multi-Agent Reinforcement Learning (MARL) in a hierarchical manner to cooperate at different scales, which allows flexible decisions with respect to whether to enter sleep mode, how to share spectrum resources, control cell density and use infrastructure. The framework fills an important gap in the related literature, which is dominated by positioning optimization within the context of a single pillar and fails to consider multi-pillar cooperation. This paper provides an awareness of six crucial pillar interdependencies, a four-tier coordinating architecture and the formal complexity analysis of the mechanisms proposed. In which future work will verify the framework through MATLAB simulation with actual network conditions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Unified ML Framework for Cross-Pillar Energy Optimization in 5G Self-organizing Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sharmin Sharmin,
  • Ismail Ahmedy,
  • Bryan Raj Peter Jabaraj,
  • Rafidah Md Noor,
  • Muhammad Umair Munir

摘要

The exponential growth of mobile data traffic and the proliferation of connected devices have positioned energy efficiency as a critical challenge in fifth-generation (5G) networks. While significant research efforts have addressed individual energy-saving pillars—Advanced Sleep Modes (ASM), Dynamic Spectrum Management (DSM), Network Densification (ND), and Energy Harvesting (EH)—these approaches operate in isolation, leading to suboptimal energy utilization and potential conflicts between optimization objectives. This paper introduces a comprehensive framework that combines four energy-efficiency pillars into a single integrated machine learning-based Self-Organizing Network (SON) model. The proposed framework employs Multi-Agent Reinforcement Learning (MARL) in a hierarchical manner to cooperate at different scales, which allows flexible decisions with respect to whether to enter sleep mode, how to share spectrum resources, control cell density and use infrastructure. The framework fills an important gap in the related literature, which is dominated by positioning optimization within the context of a single pillar and fails to consider multi-pillar cooperation. This paper provides an awareness of six crucial pillar interdependencies, a four-tier coordinating architecture and the formal complexity analysis of the mechanisms proposed. In which future work will verify the framework through MATLAB simulation with actual network conditions.