Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatial-Temporal Graph Representation Learning for Tactical Networks Future State Prediction

Junhua Liu, Justin Albrethsen, Lincoln Goh, David Yau, Kwan Hui Lim

TL;DR

The paper addresses predicting future connectivity in tactical communication networks by introducing STGED, a Spatial-Temporal Graph Encoder-Decoder framework that learns node-edge representations from past network states and forecasts next-state links. It combines a Graph Transformer Convolution-based spatial encoder with an LSTM temporal encoder and a Multi-Layer Perceptron decoder to produce next-state connectivity scores, trained via Binary Cross Entropy with a 0.5 threshold. The work contributes two Anglova-based datasets (CNTM and CNCM), a comprehensive experimental comparison against multiple baselines, and an ablation analysis showing the superiority of the GTC-LSTM configuration, achieving up to 99.2% accuracy in five-step predictions. This approach advances QoS-enabled decision making in dynamic, multi-hop tactical networks by providing accurate, scalable short-term connectivity forecasting. The results demonstrate STGED’s potential for proactive resource allocation in complex terrain and mobility scenarios, with practical impact on network reliability in military operations.

Abstract

Resource allocation in tactical ad-hoc networks presents unique challenges due to their dynamic and multi-hop nature. Accurate prediction of future network connectivity is essential for effective resource allocation in such environments. In this paper, we introduce the Spatial-Temporal Graph Encoder-Decoder (STGED) framework for Tactical Communication Networks that leverages both spatial and temporal features of network states to learn latent tactical behaviors effectively. STGED hierarchically utilizes graph-based attention mechanism to spatially encode a series of communication network states, leverages a recurrent neural network to temporally encode the evolution of states, and a fully-connected feed-forward network to decode the connectivity in the future state. Through extensive experiments, we demonstrate that STGED consistently outperforms baseline models by large margins across different time-steps input, achieving an accuracy of up to 99.2\% for the future state prediction task of tactical communication networks.

Spatial-Temporal Graph Representation Learning for Tactical Networks Future State Prediction

TL;DR

The paper addresses predicting future connectivity in tactical communication networks by introducing STGED, a Spatial-Temporal Graph Encoder-Decoder framework that learns node-edge representations from past network states and forecasts next-state links. It combines a Graph Transformer Convolution-based spatial encoder with an LSTM temporal encoder and a Multi-Layer Perceptron decoder to produce next-state connectivity scores, trained via Binary Cross Entropy with a 0.5 threshold. The work contributes two Anglova-based datasets (CNTM and CNCM), a comprehensive experimental comparison against multiple baselines, and an ablation analysis showing the superiority of the GTC-LSTM configuration, achieving up to 99.2% accuracy in five-step predictions. This approach advances QoS-enabled decision making in dynamic, multi-hop tactical networks by providing accurate, scalable short-term connectivity forecasting. The results demonstrate STGED’s potential for proactive resource allocation in complex terrain and mobility scenarios, with practical impact on network reliability in military operations.

Abstract

Resource allocation in tactical ad-hoc networks presents unique challenges due to their dynamic and multi-hop nature. Accurate prediction of future network connectivity is essential for effective resource allocation in such environments. In this paper, we introduce the Spatial-Temporal Graph Encoder-Decoder (STGED) framework for Tactical Communication Networks that leverages both spatial and temporal features of network states to learn latent tactical behaviors effectively. STGED hierarchically utilizes graph-based attention mechanism to spatially encode a series of communication network states, leverages a recurrent neural network to temporally encode the evolution of states, and a fully-connected feed-forward network to decode the connectivity in the future state. Through extensive experiments, we demonstrate that STGED consistently outperforms baseline models by large margins across different time-steps input, achieving an accuracy of up to 99.2\% for the future state prediction task of tactical communication networks.
Paper Structure (23 sections, 9 equations, 4 figures, 4 tables)

This paper contains 23 sections, 9 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Tactical Communication Networks
  4. Link Prediction in MANETs
  5. GNNs in Networking
  6. Spatial-Temporal Graph Neural Networks (STGNN)
  7. Summary
  8. Methodology
  9. Preliminary
  10. Spatial-Temporal Graph Encoder-Decoder Framework
  11. Spatial-Temporal Graph Encoder
  12. Multi-Layer Perceptron Decoder
  13. Experiments
  14. Datasets
  15. Communication Network with Tactical Movements (CNTM)
  16. ...and 8 more sections

Figures (4)

  • Figure 1: A Spatial-Temporal Graph Encoder-Decoder (STGED) framework for modelling Tactical Communication Networks (TCN). A series of TCN states pass through a graph-based neural network to learn the spatial features for each state. Subsequently, the states form a time-series input and passes through a recurrent neural network (RNN) to learn the temporal features, i.e. the tactical movement patterns. The last hidden layer of the RNN acts as the embeddings of nodes in the TCN. The decoder takes pairs of node embeddings as input and predicts their connectivity at the next time step.
  • Figure 2: Node positions over time (in minutes) for CNTM and CNCM datasets. The top row shows the objective-oriented mobility of the CNTM dataset, the tank company deploys from the upper left to bottom right. The bottom row shows the random mobility of the CNCM dataset. Nodes start off in 3 groups and randomly move toward and away from each other.
  • Figure 3: CNTM t=60m magnified
  • Figure 4: Node hop count over time for the CNTM and CNCM datasets. This refers to the amount of nodes needed to transmit from the source node (y-axis) to the destination node (x-axis). If a node can send to another node directly, the hop count is 1. If a node has no route to another node, the hop count is 0 and the color is white. The top row of figures shows the number of hops needed to connect nodes of the CNTM dataset, and has a maximum of 4 hops at t=30m. The bottom row of figures illustrates number of hops needed to connect nodes of the CNCM dataset, and has a maximum of 6 hops at t=30m.