Channel and Power Allocation for Multi-Cell NOMA Using Multi-Agent Deep Reinforcement Learning and Unsupervised...

Area of Science:

• Wireless communication systems
• Telecommunications engineering
• Artificial intelligence in networking

Background:

• Non-orthogonal multiple access (NOMA) offers enhanced data throughput for 5G and 6G technologies.
• Optimizing channel and power allocation in multi-cell NOMA systems is complex due to dynamic environments and lack of optimal labels.
• Conventional supervised learning methods are insufficient for NOMA resource allocation.

Purpose of the Study:

• To propose an efficient framework for near-optimal channel and power resource allocation in multi-cell NOMA systems.
• To enhance the energy efficiency and transmission rates of NOMA systems.
• To address the limitations of supervised learning in dynamic wireless environments.

Main Methods:

• Integration of Multi-Agent Deep Reinforcement Learning (MDRL) for channel allocation and Unsupervised Learning (UL) for power allocation.
• Development of a Multi-Agent Deep Reinforcement Learning Neural Network (MDRLNN) for channel allocation.
• Utilization of an attention-based Unsupervised Learning Neural Network (ULNN) for power allocation, using joint actions from MDRLNN as input.
• Training both MDRLNN and ULNN using the expectation of energy efficiency.

Main Results:

• The proposed MDRL-UL framework achieves near-optimal channel and power allocation.
• Simulation results demonstrate superior performance compared to existing algorithms.
• The system exhibits significantly higher energy efficiency and improved transmission rates.

Conclusions:

• The MDRL-UL framework effectively overcomes challenges in NOMA resource allocation.
• This approach offers a promising solution for maximizing performance in advanced wireless systems.
• The integration of MDRL and UL provides a robust method for dynamic resource management.