Efficient resource allocation in Device-to-Device (D2D) communication within 6G networks is crucial for enhancing overall network performance and efficiency. This paper presents a novel Deep Learning (DL) based approach for Radio Resource Allocation (RRA), leveraging Distributed Artificial Intelligence (DAI) using Belief-Desire-Intention eXtended (BDIx) agents, dynamic feedback allocation, and a Deep Feedback Neural Network (DFBNN). Additionally, Federated Learning (FL) is integrated to enable distributed training across BDIx agents, serving as D2D Relays (D2DR) or D2D Multihop Relays (D2DMHR), ensuring data privacy and reducing communication overhead. The proposed method is thoroughly evaluated against traditional graph-based and game-theoretic algorithms and Deep Feedforward Neural Networks (DFNN). Results demonstrate significant improvements in interference management, data rate, and execution time. By providing scalable, adaptive, and resilient resource allocation, this proposed method meets the stringent requirements of 6G applications, paving the way for more efficient and reliable network operations.

6G networks, D2D communication, radio resource allocation, Deep Learning, DFBNN, Federated Learning, DAI