Neighbor Exclusion-Based Graph Neural Network for Sequential Channel Allocation in Wireless Local Area Networks

Abstract

Dynamic channel allocation in wireless local area networks (WLANs) is essential for mitigating co-channel interference and satisfying time-varying traffic demands under limited spectrum resources. Beyond optimizing each network snapshot, an effective allocation scheme should preserve temporal continuity to avoid excessive channel switching. Graph neural networks (GNNs) have emerged as a promising solution by modeling topology-dependent interference relationships among access points (APs). However, most existing GNN-based schemes treat consecutive allocation steps independently and may suffer from oversmoothing, which can make interfering APs produce overly similar channel decisions. To address these issues, we propose the neighbor exclusion-based graph neural network for sequential channel allocation (NEG-SCA). The proposed framework learns allocation decisions from current traffic demand, previous allocation results, and graph-structured interference relationships. Its neighbor-exclusion aggregation reduces direct representation mixing among interfering APs, while its temporal-differential mechanism adaptively balances allocation preservation and demand-driven reallocation. Extensive simulations demonstrate that NEG-SCA achieves a favorable trade-off among interference reduction, switching-cost control, and demand satisfaction under diverse WLAN configurations.