Abstract  - The rapid evolution of sixth-generation (6G) wireless communication is introducing ultra-dense heterogeneous networks characterized by massive device connectivity, dynamic traffic patterns, and stringent latency requirements. Efficient resource allocation and network optimization are therefore become fundamental challenges for maintaining reliable communication and quality of service. The conventional centralized optimization and deep reinforcement learning approaches suffer from the scalability limitations, excessive communication overhead, privacy concerns, and limited interpretability, which is making them unsuitable for real-time 6G deployments. To address these limitations, this paper proposes an Explainable Federated Deep Reinforcement Learning with the Hierarchical Attention Optimization (XFed-HADRO) framework for an intelligent resource allocation and network optimization in ultra-dense 6G wireless networks. The proposed framework is combining the federated learning to preserve data privacy across the distributed edge nodes, hierarchical attention-based deep reinforcement learning in improving the adaptive resource allocation, and an explainable artificial intelligence module , which is based on Shapley feature attribution in providing transparent decision explanations for spectrum allocation, power control, and user association. A multi-objective reward function jointly optimizes throughput, latency, energy efficiency, fairness, and spectrum utilization while it is reducing signaling overhead. Experimental evaluation is showing that XFed-HADRO is achieving 98.2% resource allocation accuracy, 97.8% spectrum utilization, 31.5 Mbps/W energy efficiency, a fairness index of 0.99, and is reducing the average network latency to 7.4 ms, consistently outperforming Conventional DRL, FLRO, and HARL methods under the identical ultra-dense 6G network simulation conditions.
Abstract  - The rapid evolution of sixth-generation (6G) wireless communication is introducing ultra-dense heterogeneous networks characterized by massive device connectivity, dynamic traffic patterns, and stringent latency requirements. Efficient resource allocation and network optimization are therefore become fundamental challenges for maintaining reliable communication and quality of service. The conventional centralized optimization and deep reinforcement learning approaches suffer from the scalability limitations, excessive communication overhead, privacy concerns, and limited interpretability, which is making them unsuitable for real-time 6G deployments. To address these limitations, this paper proposes an Explainable Federated Deep Reinforcement Learning with the Hierarchical Attention Optimization (XFed-HADRO) framework for an intelligent resource allocation and network optimization in ultra-dense 6G wireless networks. The proposed framework is combining the federated learning to preserve data privacy across the distributed edge nodes, hierarchical attention-based deep reinforcement learning in improving the adaptive resource allocation, and an explainable artificial intelligence module , which is based on Shapley feature attribution in providing transparent decision explanations for spectrum allocation, power control, and user association. A multi-objective reward function jointly optimizes throughput, latency, energy efficiency, fairness, and spectrum utilization while it is reducing signaling overhead. Experimental evaluation is showing that XFed-HADRO is achieving 98.2% resource allocation accuracy, 97.8% spectrum utilization, 31.5 Mbps/W energy efficiency, a fairness index of 0.99, and is reducing the average network latency to 7.4 ms, consistently outperforming Conventional DRL, FLRO, and HARL methods under the identical ultra-dense 6G network simulation conditions.