Diffusion-RL for Scalable Resource Allocation for 6G Networks

2025-06-09

AI Summary

To address the joint physical resource block (PRB) and power allocation challenge under coexisting eMBB, URLLC, and mMTC services in O-RAN, this paper proposes a diffusion-enhanced deep reinforcement learning (DRL) framework. It innovatively integrates a diffusion-based generative mechanism into the DRL policy network, leveraging controllable noise to guide exploration and thereby enhancing global optimization capability and adaptability to dynamic environments while satisfying multi-dimensional QoS constraints. The framework is tightly coupled with network slicing and the O-RAN RAN Intelligent Controller (RIC) architecture to enable real-time online scheduling. Experimental results demonstrate that, compared to DQN, SS-VAE, and exhaustive search, the proposed method achieves a 37% reduction in end-to-end latency and a 29% improvement in weighted throughput for scenarios with over one thousand UEs—significantly improving scalability and spectral–energy resource efficiency.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Learning to Search
• Machine Learning: Distributed Machine Learning & Federated Learning

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

This paper presents a novel approach to resource allocation in Open Radio Access Networks (O-RAN), leveraging a Generative AI technique with network slicing to address the diverse demands of 5G and 6G service types such as Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communications (URLLC), and Massive Machine-Type Communications (mMTC). Additionally, we provide a comprehensive analysis and comparison of machine learning (ML) techniques for resource allocation within O-RAN, evaluating their effectiveness in optimizing network performance. We introduce a diffusion-based reinforcement learning (Diffusion-RL) algorithm designed to optimize the allocation of physical resource blocks (PRBs) and power consumption, thereby maximizing weighted throughput and minimizing the delay for user equipment (UE). The Diffusion-RL model incorporates controlled noise and perturbations to explore optimal resource distribution while meeting each service type's Quality of Service (QoS) requirements. We evaluate the performance of our proposed method against several benchmarks, including an exhaustive search algorithm, deep Q-networks (DQN), and the Semi-Supervised Variational Autoencoder (SS-VAE). Comprehensive metrics, such as throughput and latency, are presented for each service type. Experimental results demonstrate that the Diffusion-based RL approach outperforms existing methods in efficiency, scalability, and robustness, offering a promising solution for resource allocation in dynamic and heterogeneous O-RAN environments with significant implications for future 6G networks.

Problem

• Optimizing resource allocation in 6G O-RAN for diverse service demands
• Enhancing network performance via Diffusion-RL for throughput and latency
• Comparing ML techniques for scalable QoS-aware resource distribution

Innovation

• Diffusion-RL optimizes PRBs and power consumption
• Controlled noise meets diverse QoS requirements
• Outperforms DQN and SS-VAE in efficiency