PvFL-RA: Private Federated Learning for D2D Resource Allocation in 6G Communication

Abstract

Next-generation mobile networks (NGMN) have ushered in unprecedented demands for efficient data transfer and low-latency applications. Within the realm of 6G technology, device-to-device (D2D) communication emerges as a pivotal solution to address these challenges. Despite the promise of D2D in enabling massive data transfer with ultra-low latency, efficient radio resource allocation is required for D2D communication. This research introduces a novel decentralized private federated learning mechanism for D2D resource allocation (PvFL-RA) to tackle the privacy and resource management challenges in 6G. PvFL-RA integrates intelligent resource management methods with private federated learning, aiming to optimize the allocation of resources in a privacy-preserving manner. A novel D2D underlay communication is proposed, incorporating non-interference channel state information (CSI). PvFL-RA leverages CSI to extract channel gain concerning beam direction and distance, accurately determining non-interference zones for user equipment (UE) communication. The constructed CSI dataset trains a federated learning model incorporating local differential privacy (LDP) for predicting transmission power. Comparative analysis with the traditional centralized resource allocation model (CN-RA) demonstrates PvFL-RA’s ability to accurately predict data rate and transmission power. Additionally, an abolition study contrasts PvFL-RA with federated learning-based resource allocation (FL-RA), revealing a privacy and accuracy tradeoff between the two models. Results underscore that the proposed decentralized PvFL-RA model significantly diminishes the overhead associated with centralized nodes while efficiently allocating near-optimal resources with enhanced privacy. This research contributes valuable insights into the evolving landscape of 6G D2D communication and resource allocation.