Link Scheduling Using Graph Neural Networks

Efficient scheduling of transmissions is a key problem in wireless networks. The main challenge stems from the fact that optimal link involves solving maximum weighted independent set (MWIS) problem, which known to be NP-hard. In practical schedulers, centralized and distributed greedy heuristics are commonly used approximately solve MWIS problem. However, most these ignore important topological information network. To overcome this limitation, we propose fast based on graph convolutional networks (GCNs) can implemented manners. Our heuristic tree search guided by GCN 1-step rollout. our solver, generates topology-aware node embeddings combined with per-link utilities before invoking solver. Moreover, novel reinforcement learning scheme developed train non-differentiable pipeline. Test results medium-sized show reach near-optimal solution quickly, shallow reduce nearly half suboptimality gap solver minimal increase complexity. proposed schedulers also exhibit good generalizability across weight distributions.