Scholarlink

<p>Dynamic resource allocation is a critical challenge in 5G heterogeneous networks (HetNets) due to rapidly varying traffic demands, user mobility, and channel conditions. Traditional Time Division Duplex (TDD) resource allocation methods are often static, unable to adapt in real-time to varying network conditions and traffic demands. To address these limitations, this paper proposes a modified Deep Reinforcement Learning (DRL)-based intelligent TDD configuration framework for adaptive radio resource allocation in 5G HetNets. The proposed approach integrates an enhanced Partial Least Squares Regression (PLSR) fine-tuning mechanism and an improved activation function within a modified Deep Belief Network (DBN) to improve learning efficiency and allocation accuracy. Furthermore, a novel hybrid optimization algorithm, namely the Secretary Bird and Pufferfish Optimization Algorithm (SBPFOA), is introduced to optimize the activation weights of the modified DBN. The framework also employs dynamic Q-value iteration and experience replay memory to adaptively adjust uplink/downlink TDD configurations according to network conditions. Simulation results demonstrate that the proposed Modified-DRL model significantly outperforms existing methods, achieving higher channel usability (mean = 0.6656), lower delay (4490.99&#xa0;ms), reduced packet loss ratio (0.2233), and improved throughput (3600.303 Mbps). In particular, the proposed method achieved the lowest packet loss ratio of 0.148 at round 1500 compared to conventional DRL and deep learning models. The obtained results confirm that the proposed framework effectively enhances network reliability, resource utilization, and communication efficiency in dynamic 5G HetNet environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advanced deep reinforcement learning techniques for dynamic resource allocation in 5G heterogeneous networks

  • 期刊论文

  • 发表日期：

    2026年7月4日

查看全文
  • G. Arul Dalton,
  • A. Bamila Virgin Louis,
  • P. Sankar,
  • A. Ramachandran

摘要

Dynamic resource allocation is a critical challenge in 5G heterogeneous networks (HetNets) due to rapidly varying traffic demands, user mobility, and channel conditions. Traditional Time Division Duplex (TDD) resource allocation methods are often static, unable to adapt in real-time to varying network conditions and traffic demands. To address these limitations, this paper proposes a modified Deep Reinforcement Learning (DRL)-based intelligent TDD configuration framework for adaptive radio resource allocation in 5G HetNets. The proposed approach integrates an enhanced Partial Least Squares Regression (PLSR) fine-tuning mechanism and an improved activation function within a modified Deep Belief Network (DBN) to improve learning efficiency and allocation accuracy. Furthermore, a novel hybrid optimization algorithm, namely the Secretary Bird and Pufferfish Optimization Algorithm (SBPFOA), is introduced to optimize the activation weights of the modified DBN. The framework also employs dynamic Q-value iteration and experience replay memory to adaptively adjust uplink/downlink TDD configurations according to network conditions. Simulation results demonstrate that the proposed Modified-DRL model significantly outperforms existing methods, achieving higher channel usability (mean = 0.6656), lower delay (4490.99 ms), reduced packet loss ratio (0.2233), and improved throughput (3600.303 Mbps). In particular, the proposed method achieved the lowest packet loss ratio of 0.148 at round 1500 compared to conventional DRL and deep learning models. The obtained results confirm that the proposed framework effectively enhances network reliability, resource utilization, and communication efficiency in dynamic 5G HetNet environments.