Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Wireless Interference Patterns: Decoupled GNN for Throughput Prediction in Heterogeneous Multi-Hop p-CSMA Networks

Faezeh Dehghan Tarzjani, Bhaskar Krishnamachari

TL;DR

This work tackles the hard problem of predicting per-node saturation throughput in heterogeneous p-CSMA networks with multihop interference. It introduces a Decoupled Graph Convolutional Network (D-GCN) that separates self-transmission processing from neighbor interference, uses unnormalized attention, and stacks multiple layers to capture $k$-hop interference, achieving a low $3.3\%$ normalised MAE. The approach provides a differentiable, scalable surrogate for exact Markov-chain analyses, enabling rapid gradient-based network optimization with utilities within $1\%$ of theoretical optima and speedups up to $\sim\!10^4$–$10^5$×. This architecture delivers strong practical impact for real-time MAC optimization in diverse, topology-driven wireless deployments.

Abstract

The p-persistent CSMA protocol is central to random-access MAC analysis, but predicting saturation throughput in heterogeneous multi-hop wireless networks remains a hard problem. Simplified models that assume a single, shared interference domain can underestimate throughput by 48-62% in sparse topologies. Exact Markov-chain analyses are accurate but scale exponentially in computation time, making them impractical for large networks. These computational barriers motivate structural machine learning approaches like GNNs for scalable throughput prediction in general network topologies. Yet off-the-shelf GNNs struggle here: a standard GCN yields 63.94% normalized mean absolute error (NMAE) on heterogeneous networks because symmetric normalization conflates a node's direct interference with higher-order, cascading effects that pertain to how interference propagates over the network graph. Building on these insights, we propose the Decoupled Graph Convolutional Network (D-GCN), a novel architecture that explicitly separates processing of a node's own transmission probability from neighbor interference effects. D-GCN replaces mean aggregation with learnable attention, yielding interpretable, per-neighbor contribution weights while capturing complex multihop interference patterns. D-GCN attains 3.3% NMAE, outperforms strong baselines, remains tractable even when exact analytical methods become computationally infeasible, and enables gradient-based network optimization that achieves within 1% of theoretical optima.

Learning Wireless Interference Patterns: Decoupled GNN for Throughput Prediction in Heterogeneous Multi-Hop p-CSMA Networks

TL;DR

This work tackles the hard problem of predicting per-node saturation throughput in heterogeneous p-CSMA networks with multihop interference. It introduces a Decoupled Graph Convolutional Network (D-GCN) that separates self-transmission processing from neighbor interference, uses unnormalized attention, and stacks multiple layers to capture -hop interference, achieving a low normalised MAE. The approach provides a differentiable, scalable surrogate for exact Markov-chain analyses, enabling rapid gradient-based network optimization with utilities within of theoretical optima and speedups up to –×. This architecture delivers strong practical impact for real-time MAC optimization in diverse, topology-driven wireless deployments.

Abstract

The p-persistent CSMA protocol is central to random-access MAC analysis, but predicting saturation throughput in heterogeneous multi-hop wireless networks remains a hard problem. Simplified models that assume a single, shared interference domain can underestimate throughput by 48-62% in sparse topologies. Exact Markov-chain analyses are accurate but scale exponentially in computation time, making them impractical for large networks. These computational barriers motivate structural machine learning approaches like GNNs for scalable throughput prediction in general network topologies. Yet off-the-shelf GNNs struggle here: a standard GCN yields 63.94% normalized mean absolute error (NMAE) on heterogeneous networks because symmetric normalization conflates a node's direct interference with higher-order, cascading effects that pertain to how interference propagates over the network graph. Building on these insights, we propose the Decoupled Graph Convolutional Network (D-GCN), a novel architecture that explicitly separates processing of a node's own transmission probability from neighbor interference effects. D-GCN replaces mean aggregation with learnable attention, yielding interpretable, per-neighbor contribution weights while capturing complex multihop interference patterns. D-GCN attains 3.3% NMAE, outperforms strong baselines, remains tractable even when exact analytical methods become computationally infeasible, and enables gradient-based network optimization that achieves within 1% of theoretical optima.
Paper Structure (23 sections, 12 equations, 8 figures, 7 tables)

This paper contains 23 sections, 12 equations, 8 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Definition
  4. Proposed Methodology
  5. Baseline GNN Architectures for Heterogeneous Interference Modeling
  6. GCN (Kipf & Welling)
  7. GraphSAGE (Hamilton et al.)
  8. GIN (Xu et al.)
  9. GINE (Hu et al.)
  10. Decoupled Graph Convolutional Network (D-GCN)
  11. Summary of Architectural Differences
  12. Dataset Generation and Evaluation Metrics
  13. Graph-Neural Network Configuration
  14. Evaluation Metrics
  15. Experimental Results & Performance Evaluation
  16. ...and 8 more sections

Figures (8)

  • Figure 1: Example of heterogeneous p-persistent CSMA in a 10-node wireless network
  • Figure 2: Example time slot assignment for Figure \ref{['fig:wirelessnetwork']} p-CSMA network with transmission duration T=2. Each row represents a node's channel access pattern over 8 time slots. Green boxes (S) indicate successful transmission, red boxes (C) denote conflicts where transmission attempts fail due to neighbor interference, and white boxes (I) represent idle slots.
  • Figure 3: Proposed Graph Neural Network architecture for heterogeneous p-CSMA throughput
  • Figure 4: Architecture of the Decoupled Graph Convolutional Network (D-GCN) for p-CSMA throughput prediction.
  • Figure 5: Comparison of Normalized Mean Absolute Error (NMAE %) between D-GCN (Decoupled GCN) and GINE architectures across different transmission durations (T = 2, 3, 4, 5) and network sizes ($N\in\{4,6,8,10,12,14,16,18,20\}$)
  • ...and 3 more figures