Joint Optimization of Bandwidth and Power Allocation in Uplink Systems with Deep Reinforcement Learning

Area of Science:

• Wireless communication systems
• Artificial intelligence in telecommunications
• Resource management in networks

Background:

• Increasing numbers of users in multi-cell systems cause explosive interference, degrading communication quality.
• Inter-cell interference is a major challenge in optimizing wireless resource utilization.
• Existing methods struggle to balance throughput maximization with quality of service constraints.

Purpose of the Study:

• To propose a novel approach for joint optimization of bandwidth and transmit power allocation.
• To enhance system throughput while suppressing co-channel interference.
• To guarantee quality of service (QoS) constraints in wireless networks.

Main Methods:

• Developed a joint-priority-based reinforcement learning (JPRL) approach.
• Decoupled the joint problem into bandwidth assignment and power allocation sub-problems.
• Utilized multi-agent double deep Q network (MADDQN) for bandwidth allocation and prioritized multi-agent deep deterministic policy gradient (P-MADDPG) for power allocation.

Main Results:

• The JPRL method demonstrated accelerated model training.
• Achieved superior system throughput compared to alternative methods.
• Average throughput was 10.4-15.5% higher than homogeneous-learning benchmarks and 17.3% higher than genetic algorithms.

Conclusions:

• The proposed JPRL approach effectively optimizes wireless resource utilization.
• JPRL significantly improves system throughput and mitigates interference.
• This method offers a promising solution for future wireless communication systems.