GenNet: Computing-Efficient Generative AI for Deterministic Transmission Scheduling in 6G Networks

作者：Weiting Zhang, Jiadong Ren, Tao Zheng, Ruibin Guo, Han Zhang, Shiwen Mao, Hongke Zhang · 发表于：IEEE Communications Magazine · 年份：2025 · DOI：10.1109/mcom.001.2400588 · 被引用次数：4 · 研究领域：Advanced MIMO Systems Optimization、Advanced Wireless Communication Technologies、IoT Networks and Protocols

With the development of the sixth generation (6G) networks, ubiquitous intelligence, computing, and networking integration, low energy consumption, and low delay have become the key characteristics. These characteristics pose significant challenges for traffic scheduling, particularly in the context of intelligent computing services. In recent years, generative artificial intelligence (AI) has demonstrated remarkable performance not only in natural language processing and image generation but also in network optimization. This article presents a generative AI-endogenous three-layer network architecture, named GenNet, to support more efficient scheduling of intelligent computing services in 6G networks by the dynamic adaptation of heterogeneous computing resources and diversified service requirements. Moreover, we propose a dueling double deep Q-network (D3QN)-based transmission scheduling algorithm that leverages diffusion models to achieve cross-domain end-to-end deterministic transmission. The proposed algorithm facilitates efficient interaction between edge devices and intelligent computing centers while reducing system cost and delay, and utilizes the denoising capability of the diffusion model to adaptively configure networks and optimize resource allocation. Finally, we present a case study, followed by a discussion of open research issues that are essential for generative AI and future networks.