Comparing Ray Tracers¶

Under Construction

This document is currently under active development. Content may be incomplete or subject to change as we continue to refine the comparison and gather additional information.

This document provides a comprehensive comparison of three prominent ray tracing tools: Wireless Insite, Sionna, and AODT (Aerial Omniverse Digital Twin). The focus is purely on the ray tracing component rather than end-to-end wireless system simulation capabilities.

Overview¶

What This Comparison Covers¶

This comparison examines the similarities and differences across three ray tracers:

Wireless Insite (X3D) - Commercial solution from Remcom
Sionna RT - Open-source ray tracer from NVIDIA
AODT - NVIDIA's Aerial Omniverse Digital Twin

Key Philosophy¶

While all three tools have evolved to become more than simple ray tracers, this comparison focuses purely on the ray tracing component. Interface with wireless system simulation tools is beyond the scope.

Scope of Comparison¶

Ray Tracing Only, Not Wireless System Simulation¶

Corollary 1: Beamforming, standard compliance, hardware imperfections modeling, scheduler types, and other system-level capabilities don't matter beyond the possibility of performing all simulation in a single platform.

Corollary 2: Antenna patterns aren't that relevant as they can be applied in post-processing. It's always more general to do ray tracing with the isotropic element. However, antenna arrays matter for element-to-element ray tracing and whether the response is synthetic or ground truth per element.

Corollary 3: Waveforms don't matter. In all systems, the signal bandwidth is used purely for system simulations, not for ray tracing. The specific waveform matters only in the context of sensing and communications, not for finding rays and computing EM fields.

Simulation Method vs Scale¶

Method	Use Case	Simulation Scale	Complexity	Time for 1-meter scene
Maxwell	Antenna and Circuit Design	< 10 cm	Very high	27 centuries
Optics	Light Propagation	10 cm - 10 m	High	1 day
Geometric	Radio Propagation	> 1 m	Medium	5 seconds

Logical Distribution of Functions¶

Some decisions can't be made with just two ray tracers. Having three ray tracers helped bring clarity regarding the distribution of roles in the environment:

1. txrx (antenna panel)¶

Mainly bool(polarization / tx / rx) and antenna specifics
Assumes isotropic radiation pattern

2. paths¶

rx/tx locations, angles of arrival/departure, powers/phases/delays
Interaction types & locations
Assumes electromagnetic propagation law of geometric raytracing
Assumes orientations in post

3. scene¶

Geometry, objects, movement → stuff that has a material

4. materials¶

EM properties of materials

Ray Tracing Capabilities¶

Supported Propagation Mechanisms¶

Ray Tracer	LoS	Reflection (max)	Diffraction (wedge/edge/max)	Diffuse Scattering (last bounce/any/keep all)	Refraction/Transmission (max)	Max No. of Paths	Max Depth	Algorithm	Angular Ray Launching Limits	Frequencies
Wireless Insite (X3D)	✅	✅ (30)	✅ / ✅ / 3	✅ / ✅ / ❌	✅ (8)	~100?	~30	SBR*	✅	100 MHz - 100 GHz
Sionna	✅	✅ (any)	✅ / ✅✅ / 1*	✅ / ❌ / ✅	❌ (0)	~50?	any?	SBR & Exhaustive	❌	100 MHz - 100 GHz
AODT	✅	✅ (5)	✅ / ✅ / 1	✅ / ❌ / ✅	✅ (5*)	~256,000	10*	SBR	❌	100 MHz - 100 GHz

Legend: - ✅✅ = configurable - *SBR = Shooting and Bouncing Rays

Note: Frequencies are mostly bound by ITU materials

Interaction Capabilities¶

Additional supported features:

Feature	Wireless Insite	Sionna	AODT
`max_reflections`	5	`max_reflections`	`reflection` (= max depth if no scat, else = depth-1)
`max_diffractions`	1	`max_diffractions`	`int(diffraction)`
`max_scattering`	1	`max_scattering`	`int(scattering)`
`max_transmissions`	0	`max_transmissions`	-
`edge_diffraction`	-	-	`edge_diffraction`
`diffuse_reflections`	-	`diffuse_reflections`	`(depth - 1)`
`diffuse_diffractions`	-	`diffuse_diffractions`	`(False)`
`diffuse_transmissions`	-	`diffuse_transmissions`	-
`diffuse_final_interaction_only`	-	`diffuse_final_interaction_only`	`(True)`
`diffuse_random_phases`	-	`diffuse_random_phases`	`scat_random_phases`
`terrain_reflection`	TRUE	`terrain_reflection`	`(True)`
`terrain_diffraction`	TRUE	`terrain_diffraction`	`(False)`
`terrain_scattering`	FALSE	`terrain_scattering`	`(True)`
`scat_keep_prob`	-	-	`0.001`
`synthetic_array`	TRUE	-	`synthetic_array`

Antenna Support¶

Antenna Types¶

Ray Tracer	Isotropic	Omnidirectional	Dipole (Halfwave/infinitesimal/Config.)	Monopole	microstrip patch	tr38901	Horn	Parabolic	Other	Array (Rectangular/Flexible)	Mutual coupling	Max size (single pol.)	Per element orientation
Wireless Insite	✅	✅	✅ / ✅* / ✅	✅	✅	❌	✅	✅	✅	✅ / ✅	❌	-	❌
Sionna	✅*	❌	✅ / ✅ / ❌	❌	❌	✅	❌	❌	❌	✅ / ✅	❌	-	❌
AODT	✅	❌	✅ / ✅ / ✅	❌	✅	❌	❌	❌	❌	✅ / ❌	✅*	64 RU / 8 UE	❌

Additional Antenna Features¶

Feature	Wireless Insite	Sionna	AODT
Custom Element Patterns	✅	✅	✅
Antenna Design Helper	✅	❌	❌
Antenna Array Visualizer	✅	✅	✅
Multi antenna elements	✅	✅	✅
Multiple txrx pairs per sim. (>2 diff panels)	✅	❌	✅
Different panels for UL/DL	✅	❌	❌
Synthetic array mode	❌	✅	❌

Polarization Support¶

Polarization Type	Wireless Insite	Sionna	AODT
Single (V/H)	✅	✅	✅
Single (flexible)	✅	✅	✅
Dual (V/H)	✅	✅	✅
Dual (cross)	✅	✅	✅

Notes: - In Wireless Insite, the infinitesimal dipole can be obtained by configuring the length of the dipole to be very small - In Sionna, the isotropic pattern is not exactly the idealized pattern as in literature - it still accounts for orientation and causes changes in vertical polarization - In AODT, mutual coupling is only supported for halfwave dipoles - Sionna has 2 polarization models: one only accounts for slant/tilt (model 2, simpler), the other accounts for both tilt and azimuth (model 1, more precise). Both follow 3GPP

Waveform Support¶

Ray Tracer	Sinusoid	Sinc	Chirp	Raised Cosine	Root RC	Tukey	Hanning	Hamming	Blackman	Gaussian	Custom	OFDM waveform & grid	Response per subcarrier	Use of Bandwidth in RT
Wireless Insite	✅	❌	✅	✅	✅	✅	✅	✅	✅	✅	❌	❌	❌	❌
Sionna	✅	✅	❌	✅	✅	❌	✅	✅	✅	❌	✅	✅	❌	❌
AODT	✅	❌	❌	❌	❌	❌	❌	❌	❌	❌	❌	✅	?	❌

Note: The waveforms don't really matter for ray tracing. We only care about: - The ray tracing / output - Whether bandwidth matters - Whether any better computation is done by considering a resource grid (like adjustment of the responses at each subcarrier's frequency)

Materials & Scattering¶

Material Properties Support¶

Both ray tracers support ITU-R P.2040 materials with varying parameter naming conventions (see Parameter Equivalence section).

Decoupling Ray Tracing and EM Phenomena¶

The issue is that some paths depend on the material, namely for scattering and refraction/transmission. This is good for research but not for maximum realism.

Diffuse Scattering: Key Differences¶

The maximum number of scattering events supported by any ray tracer is 1. All parties agree that no further depth than 1 is necessary. However, where they differ is at what stage in a path this scattering event can happen:

Sionna: Only allows scattering in the last bounce after reflections
AODT: Only allows one scattering, nothing else on the path profile
Wireless Insite: Has an option for putting scattering in the last bounce only, or not. Plus, lets us configure how many of each event can happen until scattering. The number of possible combinations becomes really large when the scattering event is not enforced on the last bounce.

Conclusion: Wireless Insite is the most realistic and flexible in this regard. However, in THz, 2 scattering events may become relevant since scattering may become the main form of propagation.

Software & Platform Details¶

Platform Compatibility¶

Feature	Wireless Insite	Sionna RT	AODT
OS Support
Windows	✅	✅*	✅*
MacOS	❌	✅	❌
Linux	✅*	✅	✅*
Hardware
GPU supported	✅	❌*	✅
Headless mode	✅	✅	❌
Min. Num. GPU	1	1	2
Min. GPU RAM	2 GB	N/A	12GB & 48GB
Development
Version	3.3?	0.19.1	1.2
Open-source	❌	✅	✅*
Language	❔	Python	C++ / Python
Active Development	❔	✅	✅
Release Cycle	❔	4 / year	3 / year
Forum	❌	✅	✅
First release	2002	2023	2024
Support latency	~days	~days	<12h
Documentation
Ray tracing algorithm	✅	❌	❌
EM principles	✅	✅	❌
Tips	✅	❌	❌

Notes on Compatibility: - Wireless Insite UI only works on Windows - Sionna works on Windows via WSL2 - AODT: UI runs on Windows and Linux; Backend runs only on Linux; For single-GPU capabilities, the machine must use Linux currently - AODT open-source: Yes, after getting accepted in NVIDIA's developer program (primarily a mechanism to understand who is working with the software; all developers from universities and industry get accepted)

Language Details¶

AODT: - UI: Python using the Omniverse Framework - Backend: One backend in C++ and another in Python - EM Solver (performance critical): C++

Sionna: - Python leveraging GPU-accelerated frameworks like TensorFlow and DrJit - Doesn't have a UI per se, but supports interactive ray tracing visualizations in Jupyter notebooks - Requires knowledge of Python to use

Wireless Insite: - Has comprehensive GUI tools including antenna design helpers and array visualizers - Includes tips and heuristic guidelines on parameter setting, e.g.: - "0.2° ray spacing for areas smaller than 500m x 500m" - "6 reflections are usually sufficient, except in narrow streets or areas with highly reflective buildings"

Parameter Equivalence¶

Material Properties¶

Category	Parameter	Typical Value	Wireless Insite	Sionna	AODT
Material	name	ITU ...	`name`	`name`	`label`
	thickness	0.03	`thickness`	-	-
	permittivity		`permittivity`	`relative_permittivity`	-
	conductivity		`conductivity`	`conductivity`	-
	a,b,c,d		-	-	`itu_r_p2040_{a/b/c/d}`

Scattering Model Properties¶

Parameter	Wireless Insite	Sionna	AODT
scattering model	`diffuse_scattering_model`	`scattering_pattern`	(UI)
roughness	`roughness` (std = RMS)	-	`rms_roughness`
scattering coefficient	`fields_diffusively_scattered`	`scattering_coefficient`	`scattering_coeff`
cross polarization factor	`cross_polarized_power`	`xpd_coefficient`	`scattering_xpd`
alpha_i	`directive_alpha`	`alpha_i`	`exponent_alpha_i`
alpha_r	`directive_beta`	`alpha_r`	`exponent_alpha_r`

Ray Tracer Caveats¶

Sionna Caveats¶

Diffraction: Only first-order diffraction is available. And, when accounting for diffractions, no other interaction along the path is possible (i.e., single-interaction path only)
Scattering: Only happens in the last bounce, and previous interactions can only be reflections
Ray Tracing Methods: Sionna has 2 types:
Uniform SBR in a Fibonacci sphere
Exhaustive (not compatible with scattering) but finds "all possible combinations of primitives" → might be possible to use it as a ground truth!

AODT Caveats¶

Only specular reflection interactions along a ray with transmissions
Only direct diffuse scattering (single bounce and nothing else on the path - like Sionna with diffractions)
Unbelievable scale: Maximum number of paths limited to: 500 × num_ru_ele × num_ue_ele
If RU and UE have max elements, 64 × 8 = 512
500 × 512 = 256,000 paths per user
This assumes 1M rays emitted per RU and 10k users

Wireless Insite Caveats¶

Algorithm: X3D uses SBR with "exact path calculation algorithm"
Adjacent Path Generation (APG): Uses ray interpolation to determine exact paths to receiver based on rays that land close to it
This method relies on interpolation, so it can result in accuracy errors in some cases
For the same computational load, may help result in more accurate results
Reduces total computation load necessary to find a path that intersects the receiver sphere
Enablement is optional
APG Parameters:
Adjacency distance: Max distance between true receiver position and where coarse ray landed
Number of paths: Number of paths required in approximation (higher = more accurate interpolation)
Path Validation: Algorithm validates angles and blocking situations, discarding non-compliant paths
This can be a place for possible errors if diffuse scattering is ignored

General Ray Tracing Characteristics¶

SBR requires intersecting rays with a receiver sphere: Bigger spheres collect rays faster but may be less accurate

Performance Comparison¶

Key Observations¶

Wireless Insite: Tends to put numbers to maximum parameters (doesn't necessarily mean less capable - may be purely a development practice)
AODT shines on: Massive resolution per user (i.e., many, many paths)
Wireless Insite shines on: Massive depth (many diffractions, scattering, reflections)
In terms of realism: Depth beyond a certain point won't lead to more resolution, so AODT's approach may be better, although still lacking a bit on the basics

Sources of Error: Digital Twin vs Real World¶

Even if a digital twin is PERFECT, there will be differences from the real world:

Scene (Geometry)¶

Geometry (e.g., size, shape, exact location of objects)
LOD errors (not capturing all the right details and angles)

Materials (EM Properties)¶

Materials (conductivity, permittivity, roughness, thickness)

Paths (EM Propagation)¶

Imperfect ray tracing
Imperfect propagation modeling

Hardware¶

Time offset
Frequency offset
Distortion in amplifiers
Quantization in ADC
IQ imbalance
Filter imperfections

Hard-to-Model Factors¶

Human: - User device position and orientation error (and mobility) - Non-modeling of blockages by other users - Non-modeling of self-user blockage by hand, head, or body

Channel Measurements: - Imperfect measurement (includes unknown noise) - Channel asymmetry (most devices have more receive antennas than transmit antennas) - Antenna channel mapping problem - Frequency mapping problem

Note: The necessary realism of a digital twin should always be discussed with its purpose/utility in mind - what wireless task is the DT intended to help?

Future Work & Open Questions¶

Papers to Review¶

Toward Real-Time Digital Twins of EM Environments: Computational Benchmark for Ray Launching Software
WiThRay: A Versatile Ray-Tracing Simulator for Smart Wireless Environments
The Design and Applications of High-Performance Ray-Tracing Simulation Platform for 5G and Beyond Wireless Communications: A Tutorial
A Survey of 5G Channel Measurements and Models
A Comprehensive Review of Efficient Ray-Tracing Techniques for Wireless Communication

Topics for Further Analysis¶

Performance (runtime) across: - Number of users in scene - Number of triangles in scene

Accuracy (comparison between ray tracers): - Per-path comparisons for different path types: - Phase diff, Amplitude diff, delay diff - Number of paths found - Resultant Channel MSE - How does MSE change as we increase simulation parameters? (randomness off) - Should tell us which one is "closer" to ground truth - Check Sionna's "exhaustive" method - Check if Wireless Insite without APG is more precise

Features analysis: - RT Method - Paths discarded - Open-source? GIS included? RAN included?

Other considerations: - List other ray tracers and speculate why they are or aren't widely used - Ask Sionna, AODT, Wireless Insite maintainers to review for accuracy - List stuff not modelled by any ray tracer: - A) Because it's not included in ITU (e.g., magnetic losses) - B) Just because they didn't (e.g., frequency response along the band for a signal with given bandwidth)

Disclaimers¶

This comparison was created to help researchers and practitioners understand the trade-offs between different ray tracing solutions. It should be noted that this is a pragmatic modeling choice by all ray tracing engines. For transparency: information herein is accurate for the ray tracer versions considered at the time of writing.

Version information: Wireless Insite 3.3(?), Sionna 0.19.1, AODT 1.2

Last Updated: 2026-01