Transfer Learning Applied to Deep Reinforcement Learning for 6G Resource Management in Intra- and Inter-Slice RAN-Edge Domains

Leveraging the power of deep reinforcement learning (DRL) and strategic knowledge transfer, our study introduces PIRA-DRL-DTRL, a novel approach to optimizing resource allocation in emerging 6G networks. Central to this research is the innovative application of artificial intelligence (AI) at the network’s edge, enabling efficient management of resources across diverse timeframes while enhancing overall network performance. Implemented within an Open Radio Access Network (O-RAN) architecture, PIRA-DRL-DTRL employs a two-tiered decision-making system to dynamically adapt to varying network demands, ensuring optimal resource allocation for enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC). Our proposed algorithm achieves significant performance gains, providing a 14.28% and 10.67% improvement in throughput for eMBB slices compared to DRL-only and state-of-the-art (SOTA) methods, respectively. Additionally, it reduces delay by 23.57% and 7.48% compared to baseline and SOTA approaches for eMBB slices. By predicting and adapting to network slice demands, PIRA-DRL-DTRL ensures seamless service delivery. This research lays the groundwork for smarter, more efficient 6G networks capable of meeting the dynamic needs of users and applications.