Deep Reinforcement Learning-Based Resource Allocation for UAV-GAP Downlink Cooperative NOMA in IIoT Systems.

Area of Science:

• Wireless Communication
• Artificial Intelligence
• Optimization

Background:

• Industrial Internet of Things (IIoT) systems require robust communication solutions.
• Unmanned Aerial Vehicle (UAV) Ground Access Point (GAP) cooperative networks offer flexible deployment.
• Non-Orthogonal Multiple Access (NOMA) enhances spectral efficiency but poses allocation challenges.

Purpose of the Study:

• To develop a joint resource allocation and 3D trajectory optimization strategy for UAV-GAP cooperative NOMA systems in IIoT.
• To address diverse Quality-of-Service (QoS) requirements of cooperative and non-cooperative users.
• To improve system throughput, spectral efficiency, and interference management.

Main Methods:

• A Deep Reinforcement Learning (DRL) framework utilizing the Soft Actor-Critic algorithm for joint optimization.
• Optimization of user scheduling, power allocation, and UAV trajectory in continuous action spaces.
• Integration of closed-form power allocation and maximum weight bipartite matching for efficient resource management.

Main Results:

• Significant enhancements in system throughput and spectral efficiency demonstrated through simulations.
• Effective interference management achieved in dynamic IIoT environments.
• Robustness against channel uncertainties confirmed for the proposed scheme.

Conclusions:

• The combination of model-free reinforcement learning and conventional optimization is a viable approach for adaptive resource management.
• The proposed DRL-based scheme effectively handles dynamic UAV-GAP cooperative communication scenarios.
• The study provides a practical solution for optimizing communication performance in IIoT.