PaperlayerView exact revision on arXiv

Source-linked AI summary

An Open-Source Hybrid Ray-Tracing and 3GPP Channel Simulator for Non-Terrestrial Network

Yu-Chien Lin, Cheng-Yi Chang, Yen-Cheng Lai, Ming-Chun Lee, Ta-Sung Lee

arXiv:2610.37536v1eess.SYcs.SY

TL;DR

NTN channel simulation must account for propagation conditions that differ from terrestrial links, while statistical models lack site-specific realizations and RT can produce sparse multipath. The paper combines RT-generated dominant paths with 3GPP statistical paths and subpaths in a unified simulator. Its results show preservation of site-specific characteristics alongside richer multipath, and the simulator is open-sourced.

  • Problem

    Statistical NTN models do not provide site-specific realizations, while RT may generate too few multipath components under feasible complexity.

    Source: §I · p. 1

  • Method

    The framework combines RT-generated dominant site-specific paths with statistically generated paths and subpaths, alongside Sionna RT and 3GPP simulators in a unified workflow.

    Source: §I · p. 2

  • Results

    The hybrid simulator preserves site-specific propagation characteristics while providing sufficiently rich and statistically meaningful multipath components.

    Source: §I · p. 2, §VI · pp. 6–7

  • Takeaways & Limitations

    The simulator is released as an open-source project to support reproducibility and further NTN research.

    Source: §I · p. 2, §VI · p. 7

Abstract

from arXiv · show

This paper proposes a hybrid channel simulation framework that combines site-specific ray tracing (RT) with 3GPP-compliant statistical channel simulation for non-terrestrial networks (NTNs). The proposed approach uses RT to generate dominant propagation paths and augments them with statistically generated paths and subpaths to provide richer multipath realizations under practical simulation complexity. To provide a complete simulation framework, we integrate Sionna RT, 3GPP, and hybrid RT-3GPP simulators within a unified workflow. Simulation results based on key channel statistics and power delay profiles show that the proposed method preserves site-specific propagation characteristics while providing sufficiently rich multipath components, thereby addressing the complementary limitations of conventional RT-based and purely statistical NTN channel simulators. The developed simulator is also released as an open-source project to support reproducibility and further NTN research.

I. INTRODUCTION

NTN channels pose modeling challenges that motivate combining site-specific ray tracing with statistical channel simulation. The paper presents a unified workflow integrating Sionna RT, 3GPP, and hybrid simulation, using ephemeris-based satellite geometry.

I. INTRODUCTION

  • NTN satellite links involve long propagation distances, elevation-dependent blockage and fading, and rapidly varying delay and Doppler from satellite motion.These conditions motivate channel simulation methodologies distinct from conventional terrestrial modeling.Source: §I · p. 1
  • Statistical NTN simulators provide efficient, reproducible realizations across scenarios, but their results are driven by scenarios and parameters rather than explicit site geometry.RT can provide site-specific results, yet finite sampling over long satellite-ground distances can leave receivers with few multipath components and distorted channel statistics.Source: §I · p. 1
  • The hybrid approach combines RT-generated dominant site-specific paths with statistically generated paths and subpaths to enrich multipath under practical complexity.It is intended to preserve RT's site-specific propagation characteristics while addressing sparse RT multipath and the lack of site-specific realizations in statistical models.Source: §I · pp. 1–2

II. SIMULATOR ARCHITECTURE AND SIMULATION FLOW

  • The framework unifies Sionna RT, 3GPP statistical, and hybrid RT-3GPP channel generation in an executable workflow.It retrieves simulation context, runs the selected simulator, establishes spatial consistency, and outputs the generated channels.Source: §II · p. 2

A. Scene and Ephemeris-Based Satellite Geometry

  • Users can customize the scene with a Sionna-compatible XML scene and referenced PLY meshes; external preprocessing may avoid added simulator runtime.The Manhattan scene is the default scene used for discussion.Source: A. Scene and Ephemeris-Based Satellite… · p. 2
  • Satellite geometry is computed from TLE data using Skyfield and a WGS84 observer, yielding time-varying azimuth, elevation, slant range, and Doppler.The resulting ephemeris trajectory is applied at each time step and can drive RT, statistical channel evolution, and their combination.Source: A. Scene and Ephemeris-Based Satellite… · p. 2

B. Channel Generation Modes

The framework offers RT-only, 3GPP statistical, and hybrid channel-generation modes. In the hybrid mode, RT paths anchor the channel and 3GPP-generated components add multipath; the procedure first identifies dominant RT paths and classifies LoS conditions.

B. Channel Generation Modes

  • Hybrid mode uses deterministic RT paths as site-specific anchors and augments them with 3GPP statistical multipath components.RT paths act as dominant clusters; weaker components and statistical clusters add richer multipath and small-scale randomness that RT alone may miss at feasible complexity.Source: B. Channel Generation Modes · p. 2

III. PROPOSED HYBRID RT-3GPP CHANNEL SIMULATOR

  • The simulator workflow includes RT path extraction, correlated LSP sampling, statistical cluster generation, cluster merging, subpath expansion, and channel-coefficient construction.The paper discusses the downlink LoS case and notes that NLoS and uplink extensions can follow a similar procedure.Source: §III · p. 2

A. Site-Specific Multipath Components Generation

  • At each satellite position, RT determines whether a physical LoS path is present, selecting the LoS or NLoS procedure for subsequent 3GPP subpath generation.In an LoS scenario, the earliest-delay path is LoS, and paths within 1 ns of it are also treated as LoS.Source: A. Site-Specific Multipath Components… · p. 3
  • Each retained RT path is represented by excess delay, power, and angular parameters; some statistically generated NLoS paths are augmented, then each NLoS path receives associated subpaths, while LoS paths need no subpath augmentation.Source: A. Site-Specific Multipath Components… · p. 3

B. Correlated LSP Sampling

The hybrid simulator samples six correlated large-scale parameters using scenario-specific statistics from 3GPP NTN tables.

B. Correlated LSP Sampling

  • Six LSPs—K, DS, ASD, ASA, ZSD, and ZSA—are sampled as a correlated vector using scenario-specific means, standard deviations, and a correlation matrix.The parameters are derived using a standard-normal vector and a factorization of the correlation matrix; the scenario statistics come from 3GPP TR 38.811, Sec. 6.7.2.Source: B. Correlated LSP Sampling · pp. 3–4

C. RT-Conditioned Statistical-Cluster Delays Generation

Statistical-cluster delays are generated in relation to the retained RT clusters, using the 3GPP delay-scaling parameter and a filtering step to avoid near-overlap.

C. RT-Conditioned Statistical-Cluster Delays Generation

  • The required statistical-cluster mean delay is computed following Step 5 of Sec. 8.4 in 3GPP TR 38.901; the passage defines LRT, τRT, N′, and rτ.LRT is retained RT paths; τRT their mean excess delay; N′ candidate statistical clusters; and rτ the delay-scaling factor specified in 3GPP TR 38.811.Source: C. RT-Conditioned Statistical-Cluster… · p. 4
  • Candidate statistical clusters are removed when their delays are too close to an RT cluster, with p0 set to 0.2 following 3GPP TR 38.901.The procedure completes delay generation for dominant clusters obtained either from RT or statistical generation.Source: C. RT-Conditioned Statistical-Cluster… · p. 4

D. Powers and Angles of Statistical Clusters

This section describes how statistical-cluster powers are determined using virtual powers, followed by the generation of cluster angles. RT-cluster powers remain those supplied by the ray tracer.

D. Powers and Angles of Statistical Clusters

E. Sub-Path Generation

The procedure expands clusters into subpaths while retaining cluster-level delay and power, then differentiates subpath directions using angular offsets. It constructs subpath channel coefficients and sums their contributions to form the frequency response.

E. Sub-Path Generation

  • In the LoS case, the first LoS cluster remains a single path, while each remaining cluster expands into M subpaths.Source: E. Sub-Path Generation · p. 4
  • Subpaths within a cluster share its delay and are assigned equal power, while their directions are differentiated by intra-cluster angular offsets.Source: E. Sub-Path Generation · p. 4
  • Subpath angles follow the 3GPP procedure, after which random coupling and channel-coefficient construction are performed; the frequency response sums contributions from all subpaths.Source: E. Sub-Path Generation · p. 5

IV. SPATIAL CONSISTENCY FOR TIME-VARYING NTN CHANNELS

For moving-satellite channels, the method carries statistical clusters across time using inferred scatterer geometry rather than independently redrawing channels at every step. Fixed scatterers support spatial consistency, but their geometry may become unrepresentative over time.

IV. SPATIAL CONSISTENCY FOR TIME-VARYING NTN CHANNELS

  • Independent channel redraws at each time step may destroy temporal continuity, motivating the frozen-scatterer statistical-consistency module.Source: §IV · p. 5
  • The method uses the same inferred scattering point across time slots, updating cluster delay, power, departure direction, and propagation as satellite geometry changes.RT clusters follow the RT simulator; statistical clusters at later times are inferred from the initial channels and geometry.Source: §IV · p. 5
  • Because statistical clusters assume fixed scattering points, accuracy may gradually degrade as simulation evolves; regenerate scatterers when their geometry no longer represents the propagation environment sufficiently.Source: §IV · p. 5

V. SIMULATION RESULTS AND VALIDATION

The section reports validation through comparisons of large-scale parameters and illustrations of power delay profiles.

V. SIMULATION RESULTS AND VALIDATION

  • Validation is presented through large-scale parameter comparisons and power delay profile illustrations.Source: §V · p. 5

A. Simulation Setup and Evaluation Metrics

The evaluation uses a 28-GHz Manhattan-scene setup with receiver positions at two satellite elevations. It compares delay, Rician, and angular-spread statistics.

A. Simulation Setup and Evaluation Metrics

  • The evaluation uses a 28-GHz carrier, a 600-km satellite distance, and 50 receiver positions at 40° and 70° elevations.The RT setup uses 3 × 10^6 rays, maximum depth five, diffuse scattering coefficient 0.25, and disables diffraction.Source: A. Simulation Setup and Evaluation… · p. 5
  • The configuration includes LoS, specular and diffuse reflection, and refraction, with a five-sample trajectory over 0.2 seconds.Source: A. Simulation Setup and Evaluation… · p. 5
  • The evaluation measures RMS delay spread, Rician K-factor, and angular spreads, comparing delay spread, AOA, ADO, ZOA, and ZOD on a log scale.Azimuth AOA spread uses a power-weighted circular mean and circular RMS spread across subpaths.Source: A. Simulation Setup and Evaluation… · pp. 5–6
  • Table I summarizes the simulation configuration.Source: Table I · p. 6

B. Simulation Results

The hybrid simulator’s channel statistics are closer to 3GPP’s than Sionna RT’s, while its power-delay profile retains deterministic paths and adds statistical multipath. The reported LSP comparisons are limited to LoS scenarios.

B. Simulation Results

  • The hybrid simulator’s delay spread and Rician factor agree well with 3GPP, while its angle spreads are close but slightly different; Sionna RT reports different statistics at both elevations.The authors attribute the RT differences to limited subpath representation under reasonable complexity, making LoS more prominent.Source: B. Simulation Results · p. 6
  • Sionna RT pathloss is 176.95 dB, versus 176.96 dB for theoretical free-space pathloss.Source: B. Simulation Results · p. 6
  • The reported LSP statistics are for LoS scenarios only, and LoS probability decreases as satellite elevation increases.Source: B. Simulation Results · p. 6
  • The 40° power-delay profiles show Sionna RT with a dominant LoS path and few delayed components, 3GPP with richer multipath, and Hybrid between them.Hybrid retains dominant deterministic components while adding statistical multipath.Source: B. Simulation Results · p. 6

VI. CONCLUSIONS

The framework combines Sionna RT, 3GPP, and hybrid RT-3GPP simulation. Its hybrid approach preserves site-specific propagation characteristics while overcoming conventional RT multipath sparsity.

VI. CONCLUSIONS

  • The framework integrates Sionna RT, 3GPP, and hybrid RT-3GPP simulators, combining RT-generated dominant paths with statistical paths and subpaths.The approach aims to provide richer multipath realizations under feasible complexity.Source: §VI · p. 6
  • The proposed method preserves site-specific propagation characteristics while overcoming the multipath sparsity of conventional RT-based simulation.Source: §VI · p. 7
  • The complete simulator has been open-sourced on GitHub.Source: §VI · p. 7
Loading 2610.37536v1…