Multi-dimensional joint resource scheduling for 5G eMBB by a PER-branch dual deep Q-network

Area of Science:

• Telecommunications Engineering
• Artificial Intelligence
• Network Resource Management

Background:

• Fifth-generation (5G) networks present complex resource management challenges in enhanced mobile broadband (eMBB) scenarios.
• Existing scheduling algorithms struggle with dynamic channel conditions and diverse service needs.
• Optimizing throughput, energy efficiency, and user fairness is crucial for network operators.

Purpose of the Study:

• To develop a multidimensional resource scheduling framework for 5G eMBB.
• To address the limitations of conventional scheduling algorithms in dynamic environments.
• To improve system performance by jointly optimizing key network parameters.

Main Methods:

• Proposed a Prioritized Experience Replay Branching Double Deep Q-Network (PER-BDQ) framework.
• Employed an action-branching architecture for independent optimization of subcarrier, power, and computational resources.
• Reduced action-space complexity and improved training convergence through decoupled decision-making.

Main Results:

• Achieved an aggregate system throughput of 11.34 Gbps in a massive MIMO environment.
• Reached an energy efficiency of 53.9 bits/J.
• Demonstrated stable learning with a cumulative reward standard deviation of 22.3, outperforming baseline methods.

Conclusions:

• The proposed PER-BDQ framework effectively manages multidimensional resources in 5G eMBB.
• The approach offers a significant improvement in throughput and energy efficiency.
• This study provides a valuable reference for intelligent scheduling in future wireless networks.