Low-Latency Relay Selection in NR-V2X Vehicular Communications via Graph Isomorphism Networks with Edge Features

Short summary

This paper uses a Graph Isomorphism Network with Edge Features (GINE) to replace slow MILP solvers for relay selection in NR-V2X vehicular networks. GINE matches MILP link-level decisions with 0.96 accuracy and inference under 5ms. A hybrid GP-MILP strategy preserves optimal solutions while keeping solver runtime below 30ms for 98% of instances, making MILP-grade optimization compatible with V2X latency budgets.

• •GINE network achieves 0.96 accuracy matching MILP relay decisions with sub-5ms inference
• •Hybrid GP-MILP preserves optimal solutions with sub-30ms runtime for 98% of instances
• •End-to-end connectivity gains up to 12% over 1-hop MILP baseline

Generated with AI, which can make mistakes.