Multi-Agent Deep Reinforcement Learning for Dynamic Spectrum Access
Dr. Kai Zhang, Dr. Jun Li, Prof. Zhu Han
University of Houston
Abstract
We develop a multi-agent deep reinforcement learning (MARL) framework for dynamic spectrum access in cognitive radio networks. Each secondary user acts as an independent agent learning to access spectrum holes without explicit coordination. Our cooperative MARL approach achieves 92% spectrum utilization while reducing collisions by 60% compared to non-cooperative approaches.
AI Summary
- Multi-agent RL framework for decentralized dynamic spectrum access.
- Achieves 92% spectrum utilization with 60% fewer collisions.
- Agents learn cooperative behavior without explicit communication.
- Converges within 500 episodes in realistic simulation environments.
Key Findings
- 1Cooperative reward shaping enables implicit coordination without message passing.
- 2The framework handles dynamic numbers of agents entering and leaving the system.
- 3Outperforms both centralized and independent Q-learning baselines.
Industry Implications
Enables efficient spectrum sharing critical for 6G multi-band operation.
Applicable to unlicensed band coexistence and shared spectrum regimes.
Provides a framework for autonomous spectrum management in future networks.
Read the Original Paper
Access the full paper on arXiv for complete methodology, results, and references.
Open on arXivRelated Papers
Vision Transformers for RF Spectrum Monitoring and Classification
DeepSig Inc. — 8 citations
AI/ML PapersTransformer-Based Channel Estimation for Massive MIMO Systems
Tsinghua University — 12 citations
AI/ML PapersFederated Reinforcement Learning for Distributed Network Optimization
Stanford University — 8 citations