Airborne Network Packet Delay Estimation Dataset for Spatio Temporal Graph Neural Networks, Graph Neural Network and Neural Network

1. Simulation Overview
This dataset comprises processed simulation results from an Airborne Network. The simulations modeled networks of 25 to 50 aerial vehicles with different network parameters and node velocities ranging from 0 to 520 km/h.
In each simulation, 50 to 100 data flows with diverse source-destination pairs were scheduled. Each flow consisted of 5 to 50 packets (sizes: 300 to 1200 bytes) transmitted at 0.5-second intervals. During initialization:
- Node Distance: Followed a normal distribution (e.g., μ=160 km), bounded between 1 km and 300 km.
- Connectivity: Links were established based on hop-to-hop distance, with a maximum reception range of 250 km.
- The resulting PyTorch dataset facilitates the development of models to estimate experienced delay based on current network states and determined routes. Supported architectures include Spatio-Temporal Graph Neural Networks (STGNNs), Graph Neural Networks (GNNs), and Feed-Forward Neural Networks (FFNs).
2. GNN-Based Data Representation
For GNN-based algorithms, the input utilizes PyTorch Geometric data consisting of node hidden states and adjacency indices. In this representation, GNN nodes represent network links and data flows. Each GNN node contains three specific features:
- Links: Data rate, propagation delay, and link velocity.
- Flows: Packet size, number of hops, and number of scheduled packets.
3. Dataset Variants
A. dataset_STGNN
Designed for STGNN and GNN training, this PyTorch Geometric dataset includes:
- x: A 2D feature matrix. The final row contains flow features [packet size, hops, packets], while the preceding rows contain link features [data rate, propagation delay, velocity].
- edge_index: The adjacency matrix defining connectivity between GNN nodes.
- past_edge_states: A 3D tensor [Edges, Time, Features] representing historical states.
- Target Labels & Metadata: Includes meanDelay, first_delay, last_delay, mean_velocity_of_nodes, sim_id, flow_start, number_of_nodes, and min_bottleneck_data_rate.
B. dataset_NN.
A standard PyTorch tensor for FFNs containing:
- Average data rate of links in the path.
- Minimum data rate (bottleneck) in the path.
- Total propagation delay.
- Number of hops.
- Packet size and number of packets.
- Maximum distance between two hops and the corresponding link velocity.