Multi-Agent Reinforcement Learning for Secure and Adaptive Spectrum Allocation in 6G Networks

The rapid evolution of 6G networks necessitates intelligent and secure spectrum allocation to meet the demands of ultra-dense connectivity, low latency, and high data throughput. Multi-Agent Reinforcement Learning (MARL) offers a promising direction for enabling adaptive decision-making across distributed network agents. However, existing spectrum allocation methods often suffer from limited adaptability to dynamic environments and are vulnerable to security threats such as jamming and unauthorized access. To address these limitations, this study proposes a novel framework based on Multi-Agent Deep Q-Network (MADQN), where multiple agents learn collaborative and secure spectrum allocation strategies through deep reinforcement learning. The framework integrates adversarial training to detect and mitigate malicious behaviors while ensuring efficient and conflict-free spectrum sharing. The proposed method is particularly suitable for dynamic spectrum sharing in dense urban 6G environments, enabling real-time adaptation to traffic load, user mobility, and interference patterns. Experimental results demonstrate that MADQN significantly improves spectrum utilization efficiency by 97.6%, minimizes interference by 35%, and enhances system security by 98.4%, as well as scalability by 96.5%, compared to traditional centralized and rule-based approaches. This framework represents a scalable and robust solution for future 6G networks requiring autonomous and secure spectrum management.

Hali tarjima qilinmagan