Causality aware explainable deep reinforcement learning with adaptive attention mechanisms for scalable resource orchestration in 6G wireless networks

The emerging sixth generation (6G) wireless networks are expected to enable ultra dense, heterogeneous and AI native environments, supporting mission critical applications with stringent requirements on latency, scalability and interpretability. However, conventional Deep Reinforcement Learning (DRL) based resource management frameworks lack transparency, struggle with high-dimensional state spaces and fail to adapt effectively under dynamic network conditions. To address these challenges, this article proposed a novel Causality Aware Explainable Deep Reinforcement Learning (CE-DRL) framework that integrates causal inference with adaptive attention mechanisms for scalable and interpretable resource orchestration in 6G environments. Besides that the proposed framework also construct dynamic causal graphs to identify and prune inactive features to reduce the complexity and enhance convergence ratio. Additionally, a dual layer adaptive attention mechanism is added to refine both the temporal and spatial features under varying network loads. The framework further embeds explainable AI (XAI) components such as saliency maps and layer wise relevance propagation to offer transparent decision traces. Extensive simulations were performed on a virtualized 6G testbed to demonstrate the performance of proposed framework. The obtained results signify that CE-DRL outperforms other baseline DRL algorithms to reduce the latency upto 35%, energy efficiency upto 28% and increase the throughput by 14% and explainability factor upto 170% as a percentage ratio and ensure a reliable and scalable AI driven resource orchestration for real-time 6G wireless infrastructures.