Scholarlink

The explosive growth of 5G-enabling services, including self-driving cars, telemedicine, and augmented reality experiences, has heightened the need for dynamic and responsive spectrum allocation methods. In multi-operator 5G ecosystems, where multiple competing network providers share limited spectral resources, conventional and rule-based allocation mechanisms are insufficient. A legacy system is not agile enough to adapt in real-time to user demand rates, traffic volume and interference patterns. Consequently, underutilization of the spectrum, inter-operator collisions, and a lack of high-quality service have become a common predicament within urban and congested areas. To overcome these inefficiencies, we develop RL-SAGE (Reinforcement Learning-based Spectrum Allocation using Game-theoretic Environment). This decentralised, real-time engine dynamically assigns and re-assigns frequency bands to multiple operators using deep reinforcement learning. We model each operator as a separate agent, trained with the Deep Q-Network (DQN) to find the best allocation strategies, given real-time network conditions within the networks the operator operates on — e.g., the interference level, traffic load, and user mobility. The game-theoretic layer of coordination enables the resolution of conflicts between agents fairly and equitably, thereby maximising achievable spectrum utility. The performance is evaluated through simulation results in a synthetic multi-operator 5G setup, indicating that RL-SAGE outperforms conventional heuristic and static allocation systems. It achieves a 26% gain in spectrum utilisation, a 32% reduction in interference, and more than 90% accuracy in allocation choices. Additionally, the system has low decision latency, making it applicable for real-time deployment in next-generation networks. In conclusion, intelligent and versatile spectrum management can be done revolutionarily when using the RL-SAGE. Combining reinforcement learning strength in support of the decentralisation of games-theoretic coordination features, the engine constructed in the current paper that will effectively and fairly facilitate sharing of locales available in the spectrum among competing operators in a proactive, responsive, and cost-sensitive way, will eventually lead towards exceptionally reliable and performance-still 5G network services.

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 Real-Time Reinforcement Learning Engine for Spectrum Allocation in Multi-operator 5G Environments

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mohammed I. Habelalmateen

摘要

The explosive growth of 5G-enabling services, including self-driving cars, telemedicine, and augmented reality experiences, has heightened the need for dynamic and responsive spectrum allocation methods. In multi-operator 5G ecosystems, where multiple competing network providers share limited spectral resources, conventional and rule-based allocation mechanisms are insufficient. A legacy system is not agile enough to adapt in real-time to user demand rates, traffic volume and interference patterns. Consequently, underutilization of the spectrum, inter-operator collisions, and a lack of high-quality service have become a common predicament within urban and congested areas. To overcome these inefficiencies, we develop RL-SAGE (Reinforcement Learning-based Spectrum Allocation using Game-theoretic Environment). This decentralised, real-time engine dynamically assigns and re-assigns frequency bands to multiple operators using deep reinforcement learning. We model each operator as a separate agent, trained with the Deep Q-Network (DQN) to find the best allocation strategies, given real-time network conditions within the networks the operator operates on — e.g., the interference level, traffic load, and user mobility. The game-theoretic layer of coordination enables the resolution of conflicts between agents fairly and equitably, thereby maximising achievable spectrum utility. The performance is evaluated through simulation results in a synthetic multi-operator 5G setup, indicating that RL-SAGE outperforms conventional heuristic and static allocation systems. It achieves a 26% gain in spectrum utilisation, a 32% reduction in interference, and more than 90% accuracy in allocation choices. Additionally, the system has low decision latency, making it applicable for real-time deployment in next-generation networks. In conclusion, intelligent and versatile spectrum management can be done revolutionarily when using the RL-SAGE. Combining reinforcement learning strength in support of the decentralisation of games-theoretic coordination features, the engine constructed in the current paper that will effectively and fairly facilitate sharing of locales available in the spectrum among competing operators in a proactive, responsive, and cost-sensitive way, will eventually lead towards exceptionally reliable and performance-still 5G network services.