Deep Reinforcement Learning Assisted Adaptive Weighting for Slice-Aware Multipath Routing in 5G Networks

Network slicing lets Ultra Reliable Low Latency Communications (URLLC), enhanced Mobile Broadband (eMBB), and massive Machine Type Communica tions (mMTC) services share the same fifth generation (5G) infrastructure, but their routing requirements conflict. Fixed packet duplication improves URLLC reliability, yet it consumes redundant resources and may select correlated paths with common failure risks. This paper proposes Deep Q-Network assisted Slice Aware Adaptive Multipath Routing (DQN-SA-AMR), an adaptive weighting framework for slice aware multipath routing. The Slice Aware Adaptive Multi path Routing (SA-AMR) utility evaluates candidate routes using reliability gain, delay gain, copy overhead, shared risk overlap, and congestion cost. Instead of selecting routes directly, the Deep Q-Network (DQN) selects a high level operating mode that adjusts the utility weights and the URLLC route budget. Simulation results show a favourable trade-off between reliability and overhead compared with fixed weight SA-AMR and fixed route count duplication.