⚡ electrical engineering

Digital-Twin-Aided Dynamic Spectrum Sharing and Resource Management in Integrated Satellite-Terrestrial Networks

This paper proposes a digital-twin-aided dynamic spectrum sharing framework for integrated satellite-terrestrial networks that utilizes compressed sensing and successive convex approximation to optimize joint long-term and short-term resource management, thereby significantly reducing system congestion as validated by simulations using real-world traffic data and 3D maps.

Original authors: Hung Nguyen-Kha, Vu Nguyen Ha, Ti Nguyen, Eva Lagunas, Joel Grotz, Symeon Chatzinotas, Björn Ottersten

Published 2026-07-16

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Hung Nguyen-Kha, Vu Nguyen Ha, Ti Nguyen, Eva Lagunas, Joel Grotz, Symeon Chatzinotas, Björn Ottersten

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world's internet as a massive, bustling city where billions of people are trying to talk to each other at the same time. For years, we've built more roads (cell towers) on the ground to handle the traffic, but in crowded cities, the roads are jammed, and in remote villages, there are no roads at all. To fix this, scientists are looking up to the sky, adding a new layer of "highways" using Low Earth Orbit (LEO) satellites that zip around the planet. The big idea is to let these satellites and ground towers share the same radio frequencies, like two neighbors sharing a single driveway to get to the grocery store. This is called "Dynamic Spectrum Sharing." But here's the catch: the satellites are moving fast, the ground towers are stuck in place, and the buildings in cities bounce signals around like pinballs, creating a chaotic mess of interference. Trying to manage this traffic in real-time is like trying to direct a stampede of elephants while riding a unicycle; by the time you shout an instruction, the elephants have already moved.

This is where a "Digital Twin" comes in. Think of a Digital Twin not as a robot, but as a hyper-realistic, living video game simulation of the real world. It's a virtual copy of the city, the satellites, and the people, running on a supercomputer. Because it's a simulation, it can peek into the future. It can predict where a satellite will be in ten minutes or where a crowd of people will gather, allowing the network managers to set up the traffic lights before the jam happens. The paper you are about to read explores how to use this "crystal ball" to manage the shared radio waves between satellites and ground towers, ensuring everyone gets a fair slice of the internet pie without the whole system crashing.

The Paper's Story: A Two-Step Dance

The authors of this paper, a team of researchers from Luxembourg and industry partners, propose a clever two-step strategy to solve the congestion problem in these integrated satellite-ground networks. They call their system "Digital-Twin-Aided Dynamic Spectrum Sharing."

Step 1: The Crystal Ball Strategy (DT-JointRA)
First, the system uses the Digital Twin to look ahead. Instead of reacting to traffic jams as they happen, the "Network Management System" (the brain of the operation) uses the twin to predict what the network will look like in the next cycle. It simulates the movement of satellites, the location of users (like cars on Google Maps), and even how buildings will block or bounce signals. Based on this future view, it makes "long-term decisions." It decides which satellite or ground tower should serve which group of people, how much radio frequency space (bandwidth) to give to each service, and how to split the traffic. It's like a traffic controller who, seeing a storm approaching on the radar, pre-emptively reroutes planes to avoid the turbulence before they even take off.

Step 2: The Reality Check (RT-Reﬁne)
However, a simulation is never perfect. The real world is messy; a user might suddenly stop their car, or a building might reflect a signal differently than the map predicted. If the system only relied on the crystal ball, it might make mistakes. So, the authors add a second step: "Real-Time Refinement." Once the system starts running in the real world, it constantly checks the actual conditions. It takes the plan made by the Digital Twin and tweaks it on the fly, specifically for the ground towers. It's like a GPS app that gives you a route based on traffic predictions, but then instantly reroutes you when it sees a real-time accident ahead. This step adjusts the power and connections of the ground towers to fix any errors caused by the imperfect prediction, ensuring the plan actually works in reality.

What They Found (and What They Didn't)

The researchers tested their idea using a very realistic setup. They didn't just use simple math; they used a 3D map of London to simulate how signals bounce off real buildings. They used actual traffic data from real people and real satellite orbit data (from Starlink's TLE data). They ran their algorithms on a powerful computer to see how well they could minimize "queue lengths"—a fancy way of saying "how long people have to wait for their data."

The results from their simulations were quite promising. They found that their two-step method (predicting first, then refining) was significantly better than other methods that tried to manage traffic without a Digital Twin or without the real-time adjustment.

The "Full Information" Benchmark: They compared their method to a "Full Information Algorithm" (FIA), which is a theoretical perfect system that knows everything about the future and the present instantly. Even though their method uses predictions, it came very close to this perfect system, with only a tiny gap in performance (about 0.27 MB difference in queue length in their tests).
The "Heuristic" Benchmark: Compared to simpler, "greedy" methods that just grab the best signal available without thinking ahead, their method reduced the waiting time (queue length) by a massive amount—up to 17 MB in some scenarios.
Speed: One of the coolest findings was about speed. Because the Digital Twin did the heavy lifting of planning ahead, the real-time "refinement" step didn't have to work as hard. It could find a good solution in just a few iterations (about 3), whereas other methods took much longer (around 10 iterations). This suggests the system could actually run fast enough to be used in real life.

What They Explicitly Rule Out

The paper is careful to point out what their system is not doing. They explicitly state that they are not trying to solve the problem by just turning up the power on the satellites or ground towers. In fact, their simulations showed that simply increasing power beyond a certain point (around 36 dBm for ground towers) didn't help much and just wasted energy. They also ruled out the idea of managing the network without a Digital Twin or without looking ahead; their results showed that ignoring the future or the complex 3D environment leads to much worse congestion.

How Sure Are They?

It is important to remember that these results come from computer simulations, not from a live network running in the sky right now. The authors used real-world data (London maps, real traffic patterns) to make the simulation as realistic as possible, but they haven't physically built and tested this on a live satellite network yet. They suggest that their method is "practically feasible" and "efficient," but they frame these as strong findings from their specific test environment. They acknowledge that in the real world, the Digital Twin might not be 100% perfect (they tested this by adding "errors" to the simulation), but their "refinement" step is designed specifically to fix those small mistakes.

In short, the paper suggests that by using a "crystal ball" (Digital Twin) to plan ahead and then doing a quick "reality check" (Refinement) to fix any errors, we can keep the internet flowing smoothly between the ground and the stars, even in the busiest cities. It's a smart, two-step dance that keeps the traffic moving without needing to build a million new roads.

Technical Summary: Digital-Twin-Aided Dynamic Spectrum Sharing and Resource Management in Integrated Satellite-Terrestrial Networks

1. Problem Statement

The paper addresses the challenges of managing resource allocation and dynamic spectrum sharing (DSS) in Integrated Satellite-Terrestrial Networks (ISTNs) supporting 5G-Advanced and 6G requirements. As mobile traffic grows, relying solely on terrestrial networks (TNs) leads to inefficient spectrum utilization in sparse areas and high infrastructure costs. While Low Earth Orbit (LEO) satellites offer complementary coverage, integrating them with TNs introduces critical complexities:

Inter-System Interference (ISyI): Co-primary spectrum sharing between TNs and Satellite Networks (SatNets) creates complex interference patterns, exacerbated by diverse terrestrial environments (e.g., urban canyons causing reflection and blockage).
Dynamic Conditions: High mobility of both LEO satellites and users, combined with long propagation distances, leads to rapidly varying link conditions.
Signaling Overhead: Real-time channel estimation and control signaling over long satellite-to-ground distances are challenging and introduce latency.
Traffic Heterogeneity: The system must support diverse services, including delay-sensitive (uRLLC), satellite communication, and multi-network services, each with distinct latency and reliability requirements.

The core objective is to minimize system congestion (measured by mean queue length) by optimizing joint resource management decisions—specifically bandwidth allocation (BWA), traffic steering, user association, resource block (RB) assignment, and power control—under practical constraints.

2. Methodology

The authors propose a time-window-based Digital Twin (DT)-aided framework that operates on two time scales: long-term (cycle-based) and short-term (sub-frame based).

A. System and Digital Twin Model

Network Architecture: The system comprises $N$ Terrestrial Access Points (TAPs) and one LEO satellite serving $K$ User Equipments (UEs). It utilizes a 5G-NR downlink transmission model with Frequency Division Duplexing (FDD) for the satellite and Time Division Duplexing (TDD) for the terrestrial network.
Digital Twin Construction: A DT is deployed at the Network Management System (NMS) to create a virtual replica of the physical network.
- Static Components: Includes 3D maps of the target area (London), TAP positions, and antenna characteristics.
- Dynamic Components: Models UE mobility, traffic demand, and LEO satellite movement (using Two-Line Element data).
- Channel Prediction: The DT uses a Ray-Tracing (RayT) tool combined with the 3D map to predict Channel State Information (CSI). To account for prediction errors (e.g., missing map details), the model introduces a correlation factor $\xi$ to relate virtual Non-Line-of-Sight (NLoS) components to real-world channels.
Traffic Model: Supports three service types: Delay-sensitive (D, served only by TAPs), SatCom (S, served only by LEO), and Multinet (M, served by both). Traffic steering splits data flows between TN and SatNet based on predicted demand.

B. Optimization Framework

Two optimization problems are formulated to minimize the average queue length (QL):

DT-JointRA (Long-term/Pre-solve):
- Goal: Optimize resource allocation for an upcoming cycle using predicted information (DT output).
- Variables: Bandwidth allocation, traffic steering ratios, user association, RB assignment, and power control.
- Complexity: Formulated as a Mixed-Integer Non-Linear Program (MINLP) due to binary association variables and non-convex SINR/rate constraints.
- Solution Approach:
  - Compressed Sensing: Binary variables are relaxed using $\ell_0$ -norm approximations to represent link activation.
  - Successive Convex Approximation (SCA): Non-convex rate and SINR constraints are convexified using slack variables and first-order Taylor approximations.
  - Algorithm: An iterative algorithm (DT-JointRA) solves the transformed convex problem to obtain a local optimal solution.
RT-Refine (Short-term/Refinement):
- Goal: Refine the terrestrial network's short-term decisions (power control, association, RB assignment) for each sub-frame using actual real-time feedback, while keeping satellite decisions fixed (due to long propagation delays).
- Mechanism: Uses the solution from DT-JointRA as an initial point. It incorporates actual CSI for terrestrial links and predicted CSI for satellite links, applying an interference margin coefficient ( $\kappa$ ) to account for prediction uncertainty in inter-system interference.
- Solution Approach: Similar SCA-based convexification is applied to solve the refinement problem efficiently at the Terrestrial Central Unit (TCU).

3. Key Contributions

The paper claims the following specific contributions:

Novel DT Model: Development of a DT model that goes beyond replicating traffic and position to include realistic terminal antenna patterns, actual 3D maps, and RayT-based CSI emulation for ISTNs.
Joint Optimization Formulation: Formulation of two distinct optimization problems ("Joint-RA" and "Refinement") that leverage DT predictions for long-term planning and real-time feedback for short-term correction, addressing the mixed-integer non-linear nature of the problem.
Efficient Algorithms: Proposal of the DT-JointRA and RT-Refine algorithms utilizing compressed sensing and SCA techniques to solve the complex MINLP problems.
Practical Evaluation: Validation using a realistic 3D map of London, vehicle mobility traces from Google Navigator, and actual traffic datasets. The study demonstrates the system's ability to adapt to real-world conditions.

4. Numerical Results

Simulations were conducted using actual traffic data and a London 3D map. Key findings include:

Congestion Minimization: The proposed RT-Refine algorithm significantly outperforms benchmarks (Greedy, Heuristic, Reference, and SF-RA algorithms) in minimizing mean queue length. It reduces queue lengths by approximately 17 MB compared to Greedy/Heuristic methods and 1.5–4 MB compared to interference-aware benchmarks.
Convergence Speed:
- The DT-JointRA algorithm converges in approximately 25 iterations.
- The RT-Refine phase 2 (real-time refinement) converges in only 3 iterations when initialized with the DT-JointRA output, compared to ~10 iterations for a standard sub-frame algorithm without this initialization.
Robustness to Prediction Errors:
- The system effectively mitigates channel mismatches caused by position errors through the refinement phase.
- An interference margin coefficient ( $\kappa = 1.1$ ) was found optimal, reducing mean queue length by ~0.75 MB compared to no margin, while avoiding the rate degradation caused by overestimating interference.
Performance Gap: The RT-Refine scheme achieves performance very close to a "Full-Info Algorithm" (which uses perfect real-time knowledge for all decisions), with a gap of only 0.27 MB in mean queue length, demonstrating the efficacy of the DT-assisted pre-optimization.
Service Satisfaction: Unlike benchmarks which leave significant portions of delay-sensitive traffic unserved (up to 22.2%), the proposed algorithms satisfy all delay-sensitive service requirements.

5. Significance and Claims

The paper positions its work as a practical solution for the future of 6G ISTNs. Its significance lies in:

Bridging the Gap: It addresses the gap between theoretical resource management and practical deployment by incorporating realistic 3D environmental modeling and handling the latency constraints of satellite communications.
Proactive vs. Reactive: The framework demonstrates that leveraging Digital Twins for proactive long-term decision-making, followed by reactive short-term refinement, is a viable strategy to handle the dynamic and complex nature of ISTNs.
Feasibility: The results highlight the practical feasibility of the solution, showing that the proposed algorithms can meet execution time horizons and adapt to real-world traffic and mobility patterns, significantly reducing system congestion compared to existing state-of-the-art benchmarks.

The authors conclude that the "pre-solve–refinement" strategy is a generalizable approach for DT-based systems, where predicted information supports long-term planning and real-time feedback ensures alignment with actual system conditions.