Toward Real-Time Digital Twins of EM Environments: Computational Benchmark for Ray Launching Software

Michele Zhu, Lorenzo Cazzella, Francesco Linsalata, Maurizio Magarini, Matteo Matteucci, Umberto Spagnolini

This paper provides a much-needed reality check for the "Digital Twin" (DT) hype in wireless communications. While the industry dreams of real-time, high-fidelity radio maps for 6G, the underlying computational engines—ray-based propagation simulators—are notoriously heavy. The authors establish a standardized benchmark to evaluate how current software handles the trade-off between environmental complexity and simulation speed.

TL;DR

The authors propose a standardized benchmark to evaluate the computational performance of ray-based propagation software, a key bottleneck for real-time Digital Twins. By testing industry-standard tools across varying mesh complexities and link configurations, they highlight how environmental detail and diffuse scattering are the primary hurdles to achieving real-time performance.

What Problem Does This Paper Address?

To realize the vision of 6G, networks need to be "environment-aware." This is achieved through Digital Twins: virtual replicas of the physical world that generate real-time "Radio Maps" to optimize beamforming, handovers, and interference management.

However, generating these maps requires simulating how millions of radio waves (rays) interact with buildings, cars, and terrain. This is traditionally done using Ray Tracing (RT) or Ray Launching (RL). The challenge is that these simulations are computationally expensive. If a simulation takes minutes to run, the Digital Twin is no longer "real-time," rendering it useless for dynamic urban environments. Until now, there hasn't been a standardized way to compare different software engines to see which ones are actually fit for the DT era.

Methods

The authors developed a 5-step benchmark workflow:

  1. Pre-process & Load: Converting 3D meshes into software-compatible formats.
  2. EM Setup: Defining material properties (permittivity, conductivity).
  3. Ray Model Setup: Configuring bounces, reflections, and diffractions.
  4. Load Configuration: Placing Transmitters (Tx) and Receivers (Rx).
  5. Simulation: Measuring the "wall-clock time" (actual elapsed time) for the computation.

They compared three major software solutions:

  • Remcom Wireless InSite (X3D): A commercial tool using an optimized Shooting and Bouncing Rays (SBR) method.
  • MathWorks Ray Tracing (SBR): Part of the MATLAB Antenna Toolbox.
  • NVIDIA Sionna RT: An open-source, GPU-accelerated library that uses a "Fibonacci" sampling method for ray launching.

The benchmark evaluates two configurations: Single Link (SL) (one Tx to one Rx) to test raw algorithm speed, and Multi-Link (ML) (one Tx to many Rx) to test parallelization and GPU acceleration.

Datasets

The authors introduced two distinct urban scenarios to test how "mesh complexity" impacts speed:

  • Scenario 1 (CARLA High Definition): A complex, high-fidelity 3D mesh of a city center (Town 10) exported from the CARLA automotive simulator. It features intricate building geometries and high vertex counts.
  • Scenario 2 (OSM Low Definition): A simplified "2.5D" model of a campus in Milan, Italy, derived from OpenStreetMap. Buildings are represented as simple extruded polygons.

The authors have made these 3D meshes and the benchmark source code available on GitHub to encourage community-wide standardization.

Key Insights and Ideas

  • The Mesh Complexity Bottleneck: Moving from a simplified OSM map to a high-definition CARLA mesh isn't just a visual upgrade; it's a computational killer. In complex scenes, the time required to calculate ray-object intersections scales dramatically, often making real-time updates impossible for traditional CPU-based methods.
  • GPU Parallelization is Non-Negotiable: For Multi-Link scenarios (which are the norm for radio maps), GPU-accelerated tools like NVIDIA Sionna showed "near-constant" time behavior. While CPU-based tools slowed down linearly as more links were added, the GPU handled the parallel workload with ease.
  • The "Diffuse Scattering" Tax: While reflections and diffractions are standard, "diffuse scattering" (rays scattering off rough surfaces) is essential for urban realism. However, the benchmark shows this can increase computation time by 10x to 25x in complex environments, identifying it as a major area for algorithmic optimization.
  • Warm-up and JIT Overhead: The authors noted that GPU-based tools often require a "warm-up" period for Just-In-Time (JIT) compilation. For a single, simple simulation, a CPU might actually be faster, but the GPU wins decisively as the scale of the problem grows.

Key Results

  • Scenario 1 (Complex): Using Remcom’s X3D, a single-link simulation with 1 reflection took ~440 seconds. Increasing to 10 reflections pushed this to ~730 seconds.
  • Scenario 2 (Simple): Most tools completed single-link simulations in under 15 seconds, showing that simplified geometry is currently the only way to approach "real-time" performance on standard hardware.
  • Sionna Efficiency: In Multi-Link tests on Scenario 2, Sionna (GPU) maintained a wall-clock time of roughly 10-20 seconds even as the number of interactions increased, whereas CPU-based methods scaled up to hundreds of seconds.

Limitations and Future Work

  • Accuracy vs. Speed: This benchmark focuses strictly on computational time. It does not evaluate the accuracy of the resulting radio maps against real-world measurements. A fast but inaccurate simulator is just as problematic as a slow, accurate one.
  • Hardware Sensitivity: Results are highly dependent on the specific GPU/CPU used (in this case, a GTX 1070 Ti). Future work could expand this to modern H100 or A100 data center GPUs.
  • Dynamic Objects: The current benchmark uses static meshes. Real-world Digital Twins must account for moving cars and pedestrians, which adds another layer of complexity not yet captured here.

Summary updated

Abstract

Digital Twin has emerged as a promising paradigm for accurately representing wireless communication electromagnetic environments. The resulting virtual representation of reality facilitates comprehensive insights into the propagation environment, empowering multi-layer decision-making processes at the physical communication level. This paper investigates the impact of ray-based model simulation within real-time Digital Twins. A benchmark for ray-based propagation simulations is presented to evaluate computational time, considering two urban scenarios characterized by different mesh complexity, single and multiple wireless link configurations, and simulations with/without diffuse scattering. Exhaustive empirical analyses are performed showing the behavior of different ray-based solutions. By offering standardized simulations and scenarios, this work provides a technical benchmark for practitioners involved in the implementation of real-time Digital Twins and optimization of ray-based propagation models.
BibTeX
@article{zhu2024toward,
  author = {Michele Zhu and Lorenzo Cazzella and Francesco Linsalata and Maurizio Magarini and Matteo Matteucci and Umberto Spagnolini},
  title = {{Toward Real-Time Digital Twins of EM Environments: Computational Benchmark for Ray Launching Software}},
  journal = {IEEE Open J. Commun. Soc.},
  year = {2024},
  url = {https://openreview.net/forum?id=etAWtk87Nl},
}

Loading related content...

Figure