Dynamic Spectrum Sharing Based on Deep Reinforcement Learning in Mobile Communication Systems

Area of Science:

• Wireless Communication
• Artificial Intelligence
• Resource Management

Background:

• Increasing demand for mobile services leads to spectrum scarcity.
• Cognitive radio systems offer a solution for efficient spectrum utilization.
• Multi-dimensional resource allocation remains a challenge in cognitive radio.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) strategy for secondary users (SUs) in cognitive radio systems.
• To optimize spectrum sharing and transmission power control for SUs.
• To address the problem of multi-dimensional resource allocation.

Main Methods:

• Utilized Deep Q-Network (DQN) and Deep Recurrent Q-Network (DRQN) for neural network construction.
• Implemented a DRL-based training approach for SU strategy design.
• Conducted simulation experiments to evaluate the proposed method.

Main Results:

• The proposed DRL method effectively improves user reward and reduces collisions.
• Outperformed opportunistic multichannel ALOHA by approximately 10% (single SU) and 30% (multi-SU) in terms of reward.
• Analyzed algorithm complexity and parameter influence on DRL training.

Conclusions:

• DRL provides an effective strategy for spectrum sharing and power control in cognitive radio.
• The proposed DRL approach offers significant performance improvements over traditional methods.
• Further exploration of DRL algorithm parameters is beneficial for optimization.