Modeling Nonterrestrial Network Traffic

Define NTN Traffic Modeling Objectives

A robust NTN traffic model begins with defining clear design objectives tailored to use cases and user expectations. Each NTN-relevant service type—enhanced mobile broadband (eMBB) and massive machine type communications (mMTC)—has distinct requirements:

• eMBB emphasizes throughput and coverage, requiring efficient use of satellite resources.
• mMTC scales for device density while tolerating variable latency.

The primary focus at this stage is network architecture, coverage, and capacity. Service‑level agreements (SLAs) are more relevant to network operation than to network design and need not be a central concern during the modeling process.

Key design parameters such as latency, throughput, jitter, and jurisdictional routing requirements form the basis for traffic scheduling, capacity design, and performance validation. Incorporating regional regulatory obligations, such as country‑specific routing rules, ensures alignment with 3GPP Releases 17, 18, and 19. A concise reference table of target values helps teams align model outputs with expected service behaviors throughout the design process.

Collect Input Data for NTN Traffic Models

Before simulating, collect the data needed to capture real‑world network behavior and traffic demand. Essential data sets include:

• Satellite ephemeris data: Orbital parameters enabling precise satellite position prediction over time
• Gateway and ground station locations: For computing link budgets and backhaul availability
• RAN measurements: Signal strength, throughput, and interference data
• Link budgets: Accounting for transmitter power, antenna gain, and atmospheric losses
• Policy and operational data: Covering device authentication, priority settings, and internetwork coordination

Creating a structured checklist helps ensure data completeness. For example, aligning ephemeris and RAN data time stamps supports consistent spatiotemporal correlation, improving path and latency estimates across the model’s simulation horizon. When modeling satellite networks, it is critical to accurately model satellite orbits so that dynamic message routes can be established.

As satellites move relative to ground stations and each other, the topology of the network changes continuously. Orbit models determine which nodes are in view, what link latencies apply, and which paths are viable at any given moment.

You can use MATLAB^® with Satellite Communications Toolbox to model satellite orbits from ephemeris data, propagate trajectories over time, and perform link budget calculations that underpin dynamic route planning. Aerospace Toolbox complements this approach with additional coordinate transforms and atmosphere models needed for precise orbit propagation. See Aerospace Toolbox Examples for orbit propagation walk-throughs.

Account for Latency in NTN Traffic Modeling

Latency is a foundational design parameter in NTN modeling and a key driver of user experience. In NTNs, latency is mainly driven by satellite altitude and transmission path dynamics. The table below shows typical round-trip delays and representative applications for LEO, MEO, and GEO systems.

When modeling, include separate uplink and downlink latency components and account for potential signal interruptions during satellite handover or GNSS reacquisition. Incorporating latency variability ensures the traffic model realistically predicts performance under congestion or rerouting events. These key aspects have been addressed across 3GPP Releases 17, 18, and 19 as NTN definitions expanded and matured.

Build and Simulate an NTN Network Model

Combine Physical, Mobility, and RF Models

Accurate NTN modeling blends three lenses—physical, mobility, and RF:

1. Physical models emulate radio propagation, Doppler shifts, and fading unique to satellite‑to‑ground channels.
2. Mobility simulators track satellite trajectories and user motion, handling event‑driven transitions such as handovers.
3. RF models contribute to accurate link budget predictions. They are also essential to designing phased array systems essential to close NTN links.

You can use MATLAB and Simulink^® to virtually propagate satellite orbits, calculate Doppler shifts, impose fading on signals, compute losses through an RF chain, and assess the effects of nonlinearities on RF signals. Relevant starting points include Satellite Communications Toolbox Examples for link-level simulation and Aerospace Toolbox Examples for orbital mechanics and coordinate frame transformations.

Apply AI/ML to Dynamic Routing

As NTN topologies shift continuously with satellite movement, static routing tables quickly become outdated. AI and machine learning algorithms are increasingly being applied to establish and maintain dynamic routing tables that adapt in real time to changing network conditions.

Several algorithm families have emerged as particularly effective for NTN dynamic routing:

• Reinforcement learning (RL): Deep Q-network (DQN) and proximal policy optimization (PPO) agents learn optimal routing policies by interacting with a simulated network environment. The agent observes link states, queue depths, and predicted satellite positions, then selects next-hop decisions that minimize end-to-end latency or packet loss. These approaches have been demonstrated in LEO megaconstellation scenarios where topology changes occur on subminute timescales.
• Graph neural networks (GNNs): Because satellite networks are naturally represented as time-varying graphs, GNN-based routers encode node and edge features (link capacity, Doppler shift, elevation angle) and predict optimal paths across the graph. GNNs generalize well to unseen topologies, making them attractive for constellations that are still being deployed.
• Federated learning: In large constellations, centralizing training data is impractical due to latency and bandwidth constraints. Federated learning allows individual satellites or ground stations to train local routing models and share only model updates, producing a globally improved routing policy without centralizing raw telemetry.
• Predictive handover and prerouting: Time series forecasting models (LSTM, transformer based) predict link availability and quality windows using orbital mechanics and historical channel data. Routes are computed before a satellite rises above the horizon, eliminating reactive rerouting delays.

MATLAB supports the development and validation of these algorithms. Reinforcement Learning Toolbox™ can be used to train and simulate RL-based routing agents against a satellite network environment, while Satellite Communications Toolbox provides the orbit propagation and link geometry needed to build realistic training scenarios. See Scenario Generation and Visualization Examples for practical starting points.

Integrate Policy, Orchestration, and Network Constraints

Policy and orchestration layers govern how traffic flows within NTNs under real constraints. Network orchestration dynamically manages quality of service (QoS), capacity distribution, and resource prioritization through automated policies.

Embedding these controls within simulation ensures compliance and realism. Examples include enforcing QoS tiers, applying geographic routing restrictions, or simulating failover between satellite and terrestrial links. Using standardized architectures such as O‑RAN further ensures alignment with multivendor orchestration frameworks, streamlining future integration and certification.

Deploy Closed-Loop Monitoring and Model Retraining

Closed‑loop monitoring keeps models aligned with evolving network states. These systems automatically ingest telemetry, detect deviations, and retrain machine learning (ML) components when performance drift occurs.

The closed‑loop workflow includes:

1. Monitoring: Capture live key performance indicators (KPIs) such as latency, jitter, and throughput.
2. Analysis: Trigger anomaly detection using predictive ML models.
3. Action: Adjust routing or retrain models automatically.
4. Verification: Confirm that key performance indicators return to expected design targets.

Continuous retraining and model versioning help sustain reliability as new satellites launch or user demand shifts. You can use MATLAB to create predictive ML models that drive corrective action. Satellite Communications Toolbox Examples provides relevant simulation environments for validating these models against realistic NTN scenarios.

See also: Satellite Communications Toolbox, Aerospace Toolbox