Multi-dimensional joint resource scheduling for 5G eMBB by a PER-branch dual deep Q-network

The large-scale deployment of fifth-generation (5G) networks has increased the complexity of resource management in enhanced mobile broadband (eMBB) scenarios. Network operators must jointly optimize throughput, energy efficiency, and user fairness. However, conventional scheduling algorithms often struggle to adapt to dynamic channel conditions and heterogeneous service requirements. To address these challenges, this study proposed a multidimensional resource scheduling framework based on a Prioritized Experience Replay Branching Double Deep Q-Network (PER-BDQ). The proposed model employed an action-branching architecture to independently optimize subcarrier allocation, discrete transmit power levels, and computational resource allocation. This decoupled decision mechanism reduced action-space complexity and improved training convergence. Experiments were conducted in a massive multiple-input multiple-output (MIMO) multi-stream transmission environment. The proposed method achieved an aggregate system throughput of 11.34 Gbps and an energy efficiency of 53.9 bits/J, while maintaining stable learning behavior with a cumulative reward standard deviation of 22.3. Compared with baseline methods, the proposed framework consistently improved performance across simulation scenarios. Overall, the results demonstrated the effectiveness of multidimensional joint resource scheduling in simulated 5G eMBB environments and provided a useful reference for intelligent scheduling design.