Q-Learning Based Joint Energy-Spectral Efficiency Optimization in Multi-Hop Device-to-Device Communication.

Area of Science:

• Wireless Communication Networks
• Machine Learning Applications
• Network Optimization

Background:

• Critical communication networks face partial connectivity challenges for On-Scene Available (OSA) user equipment (UE).
• Disseminating information rapidly and maintaining UE power is crucial in hybrid infrastructure scenarios.

Purpose of the Study:

• To develop a dynamic adaptation approach for multi-hop Device-to-Device (D2D) communication.
• To improve joint energy-spectral efficiency (ESE) while minimizing latency for critical information dissemination.

Main Methods:

• A hybrid Q-learning scheme was applied to learner agents (OSA UEs) and scheduler agents (Remote Radio Heads - RRHs).
• Algorithms for next hop and RRH selection were proposed to manage heterogeneous UE characteristics.
• A dynamic adaptation approach based on machine learning was implemented.

Main Results:

• The proposed approach achieved approximately 67% improvement in joint energy-spectral efficiency compared to baseline systems.
• Energy efficiency for OSA UEs saw a gain of approximately 30%.
• Latency was reduced by approximately 50% with the proposed framework utilizing Cloud Radio Access Network (C-RAN).

Conclusions:

• The dynamic adaptation approach effectively balances energy preservation and rapid information delivery in partially connected networks.
• Machine learning, specifically Q-learning, offers a viable solution for optimizing performance in critical communication systems.
• The integration of C-RAN significantly enhances the efficiency and responsiveness of D2D communication for public safety.