Pith. sign in

REVIEW 4 major objections 5 minor 11 references

Enabling Site-Specific Cellular Network Simulation Through Ray-Tracing-Driven ns-3

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A new ns-3 module, SioLENA, feeds ray-traced multipath components into 5G-LENA's unchanged PHY/MAC/beamforming stack, giving site-specific beam steering and end-to-end performance that the standard 3GPP statistical model smooths away.

desk verdict A genuinely useful trace-driven channel module for ns-3 that preserves AoA/AoD and runs the standard LENA stack, with validation gaps around the frequency-domain construction and the statistical baseline. read the letter →

arxiv 2508.04004 v1 pith:XTFWERCB submitted 2025-08-06 cs.NI eess.SP

Tanguy Ropitault , Matteo Bordin , Paolo Testolina , Michele Polese , Pedram Johari , Nada Golmie , Tommaso Melodia This is my paper · ORCID
classification cs.NIeess.SP
keywords raytracingns-35G-LENAsite-specificchannelmodelbeammanagementdigitaltwinmultipathcomponentsmillimeterwave
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Standard system-level cellular simulation uses statistical 3GPP channel models that reproduce average behavior but blur the geometry of a real site. This paper claims that deterministic multipath-component traces — from a ray tracer or from measurements — can be turned into frequency-domain channel matrices inside ns-3 and fed to the existing 5G-LENA PHY/MAC/scheduler/beamforming pipeline unchanged, giving the simulation true site-specific fidelity. In the reported scenarios, the trace-driven channel recovers the true line-of-sight steering direction with sub-degree mean error, and end-to-end SINR/MCS/throughput show sharp inflections at corners, blockage, and diffraction that the 3GPP statistical model largely smooths out. The module is compatible with any trace source in the defined CSV format and with any ns-3 antenna or beam-management configuration, and is presented as a building block for digital-twin simulations.

What carries the argument

Trace-based channel model: two classes, TracesChannelModel and TracesSpectrumPropagationLossModel, replace the 3GPP cluster-generation and spectrum-propagation blocks. They consume external MPC traces in CSV form, construct frequency-domain channel matrices from antenna-array response vectors, apply Doppler shifts per path from node velocities, and compute beamforming gains by multiplying the channel matrix with transmit and receive steering vectors. This keeps 5G-LENA's native PHY abstraction, scheduler, HARQ, link adaptation, and beam management running unmodified.

What would settle it

Run a channel-sounding measurement campaign in the same Boston street-canyon geometry, with the same routes and array configurations; if the measured SINR, MCS, and selected-beam trajectories do not show corner-diffraction and blockage inflections at the same locations, or if the measured beam-steering error exceeds the reported roughly 0.7° and 0.5° RMSE in LoS regions, the central fidelity claim is falsified. A quicker sensitivity check: regenerate the traces with no diffraction and with more than four reflections and observe whether the end-to-end inflections survive.

Watch

Extended reading notes

Core claim

The central claim is that the geometric information needed for directional 5G communication can survive intact into a system-level simulator without rewriting the radio stack. SioLENA parses time-indexed multipath records (delay, amplitude, phase, departure and arrival angles) and assembles the full frequency-domain channel matrix per link at runtime, using array response vectors with phase shifts from element positions; Doppler shifts are computed from node velocity projected onto each MPC direction, and beamforming gains from the channel matrix times steering vectors. The paper demonstrates that this preserves angular fidelity — ideal beamforming selects the true LoS azimuth with 0.74° and

Load-bearing premise

The load-bearing premise is that the multipath traces — here generated by a ray tracer limited to diffraction plus at most four specular reflections, or by measurements — accurately represent the real propagation channel, including its angles and phases; the paper validates the simulator against the ray tracer's own geometry, not against field measurements.

Editorial extensions

If this is right

  • Beamforming and beam-training algorithms can be validated against a specific 3D environment: the trace-based channel recovers true LoS directions and shows the expected end-fire degradation of planar arrays, which a statistical channel cannot generate.
  • End-to-end performance becomes geometry-aware: corner diffraction, street-canyon blockage, and LoS transitions translate directly into MCS, throughput, and delay inflections (for example, a roughly 20 dB SINR drop and MCS collapse during blockage) rather than being averaged out.
  • The approach is trace-source agnostic and frequency-agnostic, so any ray tracer or measurement campaign exporting MPCs in the CSV schema can drive the same ns-3 pipeline in FR1, FR2, or FR3.
  • Statistical and trace-based models are complementary within one simulator: 3GPP models remain available for Monte-Carlo sweeps and large layouts, while the trace engine pinpoints worst-case corners, indoor-outdoor transitions, and moving blockers.
  • Because full channel matrices are exposed at runtime, hybrid or learning-based beam selection and the generation of ML training data for channel estimation become feasible without re-running the ray tracer.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper stops short of validating the ray tracer's own accuracy; a natural next step is to compare SioLENA output against channel-sounding measurements on the same mesh. Until that comparison exists, the module's fidelity is bounded by the fidelity of the trace source.
  • Because full channel matrices and MPC angles are available inside ns-3, the same pipeline could support environment-aware sensing and positioning studies beyond communication metrics, which the paper lists as future work but does not demonstrate.
  • If trace-accurate angular labels are paired with beam-training logs, the simulator can generate ML training sets with geometrically consistent steering directions, an application the paper mentions but does not quantify.
  • The CSV time-indexed trace format also leaves room for dynamic updates mid-simulation, so moving blockers or changing mesh geometry could be imported during a run to study blockage prediction.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents SioLENA, an ns-3 extension to 5G-LENA that replaces the stochastic 3GPP TR 38.901 channel model with a trace-driven channel model. The module takes multipath component (MPC) traces—delay, complex gain, AoD/AoA, and Doppler—from external ray tracers such as Sionna RT or from measurement campaigns, constructs frequency-domain channel matrices, and feeds them into the existing 5G-LENA PHY/MAC stack without modifying that stack. The authors claim this preserves full compatibility with 5G-LENA while adding site-specific geometric fidelity. The evaluation has two parts: a beamforming validation in a Place de l'Étoile LoS scenario, where ideal beamforming recovers the UE azimuth with sub-degree RMSE, and an end-to-end comparison in a Boston street-canyon scenario, where trace-based SINR, MCS, throughput, and delay are compared against one realization of the TR 38.901 UMi model. The trace-based model exhibits sharp inflections (NLoS-to-LoS transitions, blockage, corner diffraction) that the statistical model does not show.

Significance. If the channel-matrix construction is correct, SioLENA is a valuable open-source contribution: it provides a path to site-specific, full-stack ns-3 simulations for beam-management, blockage, and digital-twin studies while reusing the mature 5G-LENA MAC/PHY pipeline. The explicit release of the module and trace-generation pipeline, and the architectural choice to keep the existing scheduler and beamforming stack untouched, are practical strengths. However, the current validation is largely an internal-consistency check against the ray tracer's own geometry, and the end-to-end comparison uses a single stochastic realization. The central claim that the trace-driven engine exposes inflections that the statistical model does not exhibit is therefore plausible but not yet established with the statistical rigor required for a definitive comparison.

major comments (4)
  1. [Section III-B] The core transformation from MPC traces to per-subcarrier channel matrices is described only qualitatively ("appropriate phase shifts ... according to element positions and carrier wavelength"). No explicit equation is given for H(f) or the per-subcarrier channel coefficient; the delay phase exp(-j2π f τ) across the OFDM band, the mapping of Sionna's exported complex gain to a path coefficient, and the normalization of the array response are not specified. Every end-to-end result in Section IV flows through this matrix, so a sign or scaling error in the phase/array manifold would manufacture exactly the reported inflections. The beamforming test in Fig. 2 is insensitive to absolute phase and delay-phase slope because it only recovers the LoS azimuth. Please add the closed-form channel-matrix equation, define all symbols, and add a small two-path frequency-selective validation against a k
  2. [Section IV-B] The end-to-end comparison with TR 38.901 uses a single realization of the stochastic model. A statistical channel model produces different LoS/NLoS draws and small-scale fading in every realization; one draw cannot support the claims that the statistical model "erroneously reports LoS" (Fig. 4a) or that it "smooths" fluctuations. Run multiple seeds/realizations for the 3GPP model, report the range or confidence intervals of SINR/MCS/throughput, and show that the trace-based inflections lie outside that variability. Without this, the paper's strongest claim—that the trace-driven engine exposes performance inflections that the statistical model does not exhibit—is not statistically supported.
  3. [Section IV-A] The beamforming validation is a self-consistency check: idealBeamforming selects the steering vector that maximizes received power in a channel matrix built from the same MPC traces that contain the AoD being recovered. The sub-degree RMSE therefore certifies internal consistency of the trace import and the array response, not that the traced AoDs match a real-world channel. The text should present this as an integration test, not as "precise beam-steering validation" in the abstract. A complementing test using an independently generated angle reference (e.g., a measured or analytically specified trace) would make the validation meaningful.
  4. [Section III-C and Section IV] The site-specific fidelity claim rests entirely on the assumption that Sionna RT, with the Boston Twin mesh and the configured propagation physics, accurately represents a real propagation channel. No comparison with field measurements is provided, so ray-tracer or mesh errors propagate directly into the reported results. The paper should explicitly state this limitation; ideally, at least one comparison with a measured channel dataset or public channel-sounding measurement in a similar environment would be included. This is a correctness-risk concern, not a rejection of the approach.
minor comments (5)
  1. [Fig. 4a] Legend typo: "Los: 3GPP only" should be "LoS: 3GPP only."
  2. [Section IV-B] Minor formatting issues: "10 000 packets/s" and "100 msusing" lack proper spacing; please clean up the text.
  3. [Section III-B] The CSV schema expected by TracesChannelModel is never shown. A small table listing columns, units, and an example row would make the interface reproducible and tool-agnostic.
  4. [Figure 2] The dashed line is described as "true UE azimuth" but it is presumably the geometric azimuth computed from positions, not the AoD of the LoS MPC. Define it precisely in the caption or text.
  5. [References] Reference [1] is incomplete: "p. 101933" lacks the journal/proceedings name and volume. Please correct.

Circularity Check

1 steps flagged · score 2.0 of 10

Beam-steering validation is a transparent self-consistency check; end-to-end results are not circular.

  1. self definitional [Section III-B (channel-matrix construction) and Section IV-A / Fig. 2 (beamforming validation)]
    "For each MPC, the antenna array response vectors are computed based on the provided angles, with appropriate phase shifts applied according to element positions and carrier wavelength. ... [W]e verify that the trace-based channel faithfully preserves the geometric information needed for beamforming, i.e., that the steering angles selected by 5G-LENA closely match the true LoS directions embedded in the RT trace."

    The channel matrix H is built by evaluating the array response at the angles provided in the trace. idealBeamforming then selects the steering vector that maximizes received power, i.e., θ̂ = argmax |w(θ)^H H|². For a LoS-dominant MPC set, this is maximized when w(θ) is aligned with the array response at the embedded AoD, so θ̂ equals the input AoD up to scan-grid quantization. Therefore the 'prediction' that the beam angle matches the LoS direction is an identity: the output angle is the input angle passed through the array-manifold construction. This is a useful implementation consistency test, but it does not independently validate the trace-based channel model against any external ground truth; the same MPC information is both the input and the benchmark.

full rationale

Score 2. The only load-bearing circularity candidate is the beam-steering validation in Fig. 2. Because TracesChannelModel builds H from the trace's AoD/AoA angles and idealBeamforming maximizes received power over steering vectors, the selected angle is, by construction, the angle embedded in the trace (up to the 1° scan grid). The paper is explicit that the comparison is against 'true LoS directions embedded in the RT trace,' so this is a transparent self-consistency check rather than an external validation. It is flagged as a minor self-definitional step, but it does not affect the paper's central claim: the end-to-end SINR/MCS/throughput curves are true outputs of ns-3's unmodified PHY/MAC stack fed by the constructed matrices; no parameter is fit to make those curves match a target. The comparison with TR 38.901 is a model-vs-model demonstration using the ray-traced geometry as reference, which is an assumption (unvalidated against field data) but not a circular derivation. The Boston Twin mesh [11] is a self-citation, but it is used only as an input scenario, not as justification for a theorem. The Sec. III-B description lacks closed-form equations for H(f), per-subcarrier delay phases, and Sionna gain mapping; that is a correctness/robustness risk, not a circularity, because no result is assumed from itself. Overall: no significant circularity beyond the transparent self-consistency test.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the fidelity of the input MPC traces: the ray tracer's accuracy, the environment mesh correctness, and the plane-wave superposition model. No parameters are fitted to data; the only hand-chosen settings are the ray-tracing configuration (e.g., four reflection orders). The validation uses the same geometry as the input, so it is a consistency check rather than an external measurement.

free parameters (1)
  • Maximum number of specular reflections in ray tracer = 4 (and diffraction enabled)
    Set to 4 in the Sionna RT configuration (Sec. IV: 'configured to include diffraction and up to four orders of specular reflection'); higher orders would change the MPC set and all downstream outputs. It is a hand-chosen simulation setting, not fitted to measurement.
assumptions (4)
  • domain assumption Sionna RT MPC traces accurately represent the physical propagation channel in the simulated scenarios
    The whole validation takes the ray-tracer output as ground truth (Sec. IV). No comparison against measurements is provided, so site-specific fidelity is conditional on the ray tracer's correctness.
  • domain assumption The 3D environment meshes (Boston Twin, Place de l'Etoile) are geometrically accurate to the real sites
    Used to define LoS/NLoS ground truth and expected beam directions (Sec. IV-A, IV-B). Errors in the mesh would propagate directly into the claimed site-specific effects.
  • domain assumption A channel can be represented as a finite sum of discrete MPCs with delay, amplitude, phase, AoA, AoD
    Section III-B constructs H by summing array responses weighted by complex MPC gains. This omits diffuse scattering and assumes plane waves, a standard but nontrivial modeling choice.
  • domain assumption Doppler shift is captured by projecting node velocity onto each MPC direction
    Section III-B: 'Doppler shifts are computed directly from node velocities projected onto the MPC propagation directions.' This neglects other time-varying effects such as moving scatterers or phase evolution between trace samples.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enabling Site-Specific Cellular Network Simulation Through Ray-Tracing-Driven ns-3." pith.science (2026). https://pith.science/paper/XTFWERCB

@misc{pith2026250804004,
  author       = {Pith},
  title        = {Pith review of: Enabling Site-Specific Cellular Network Simulation Through Ray-Tracing-Driven ns-3},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XTFWERCB}},
  note         = {Machine review of arXiv:2508.04004}
}
read the original abstract

Evaluating cellular systems, from 5G New Radio (NR) and 5G-Advanced to 6G, is challenging because the performance emerges from the tight coupling of propagation, beam management, scheduling, and higher-layer interactions. System-level simulation is therefore indispensable, yet the vast majority of studies rely on the statistical 3GPP channel models. These are well suited to capture average behavior across many statistical realizations, but cannot reproduce site-specific phenomena such as corner diffraction, street-canyon blockage, or deterministic line-of-sight conditions and angle-of-departure/arrival relationships that drive directional links. This paper extends 5G-LENA, an NR module for the system-level Network Simulator 3 (ns-3), with a trace-based channel model that processes the Multipath Components (MPCs) obtained from external ray-tracers (e.g., Sionna Ray Tracer (RT)) or measurement campaigns. Our module constructs frequency-domain channel matrices and feeds them to the existing Physical (PHY)/Medium Access Control (MAC) stack without any further modifications. The result is a geometry-based channel model that remains fully compatible with the standard 3GPP implementation in 5G-LENA, while delivering site-specific geometric fidelity. This new module provides a key building block toward Digital Twin (DT) capabilities by offering realistic site-specific channel modeling, unlocking studies that require site awareness, including beam management, blockage mitigation, and environment-aware sensing. We demonstrate its capabilities for precise beam-steering validation and end-to-end metric analysis. In both cases, the trace-driven engine exposes performance inflections that the statistical model does not exhibit, confirming its value for high-fidelity system-level cellular networks research and as a step toward DT applications.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

11 extracted references · 9 canonical work pages

  1. [1]

    An E2E simulator for 5G NR networks,

    N. Patriciello, S. Lagen, B. Bojovic, and L. Giupponi, “An E2E simulator for 5G NR networks,” p. 101933, 2019

    work page 2019

  2. [2]

    Study on channel model for frequencies from 0.5 to 100 GHz,

    3GPP, “Study on channel model for frequencies from 0.5 to 100 GHz,” Technical Report (TR) 38.901, 2024

    work page 2024

  3. [3]

    Implementation of a Spatial Channel Model for ns-3,

    T. Zugno, M. Polese, N. Patriciello, B. Bojovi ´c, S. Lagen, and M. Zorzi, “Implementation of a Spatial Channel Model for ns-3,” in Proceedings of the 2020 Workshop on Ns-3 , ser. WNS3 ’20, 2020, p. 49–56

    work page 2020

  4. [4]

    Sionna RT: Differentiable ray tracing for radio propagation modeling,

    J. Hoydis, F. A ¨ıt Aoudia, S. Cammerer, M. Nimier-David, N. Binder, G. Marcus, and A. Keller, “Sionna RT: Differentiable ray tracing for radio propagation modeling,” in 2023 IEEE Globecom Workshops (GC Wkshps). IEEE, 2023, pp. 317–321

    work page 2023

  5. [5]

    Investigation and Comparison of 3GPP and NYUSIM Channel Models for 5G Wireless Communications,

    T. S. Rappaport, S. Sun, and M. Shafi, “Investigation and Comparison of 3GPP and NYUSIM Channel Models for 5G Wireless Communications,” in IEEE 86th V ehicular Technology Conference (VTC-Fall) , 2017

    work page 2017

  6. [6]

    Accuracy Versus Complexity for mmWave Ray-Tracing: A Full Stack Perspective,

    M. Lecci, P. Testolina, M. Polese, M. Giordani, and M. Zorzi, “Accuracy Versus Complexity for mmWave Ray-Tracing: A Full Stack Perspective,” IEEE Transactions on Wireless Communications , vol. 20, no. 12, pp. 7826–7841, 2021

    work page 2021

  7. [7]

    High Fi- delity Simulation of IEEE 802.11ad in ns-3 Using a Quasi-deterministic Channel Model,

    H. Assasa, J. Widmer, T. Ropitault, A. Bodi, and N. Golmie, “High Fi- delity Simulation of IEEE 802.11ad in ns-3 Using a Quasi-deterministic Channel Model,” in Proceedings of the 2019 Workshop on Next- Generation Wireless with Ns-3 , ser. WNGW 2019, 2019, p. 22–25

    work page 2019

  8. [8]

    Quasi- deterministic channel model parameters for a data center at 60 GHz,

    C. Gentile, P. B. Papazian, R. Sun, J. Senic, and J. Wang, “Quasi- deterministic channel model parameters for a data center at 60 GHz,” IEEE Antennas and Wireless Propagation Letters , vol. 17, no. 5, pp. 808–812, 2018

    work page 2018

Show all 11 references
  1. [9]

    VaN3Twin: the Multi-Technology V2X Digital Twin with Ray-Tracing in the Loop,

    R. Pegurri, D. Gasco, F. Linsalata, M. Rapelli, E. Moro, F. Raviglione, and C. Casetti, “VaN3Twin: the Multi-Technology V2X Digital Twin with Ray-Tracing in the Loop,” 2025. [Online]. Available: https://arxiv.org/abs/2505.14184

    2025 arXiv

  2. [10]

    Ns3 meets Sionna: Using Realistic Channels in Network Simulation,

    A. Zubow, Y . Pilz, S. R ¨osler, and F. Dressler, “Ns3 meets Sionna: Using Realistic Channels in Network Simulation,” 2024. [Online]. Available: https://arxiv.org/abs/2412.20524

    2024 arXiv

  3. [11]

    Boston Twin: the Boston Digital Twin for Ray-Tracing in 6G Networks,

    P. Testolina, M. Polese, P. Johari, and T. Melodia, “Boston Twin: the Boston Digital Twin for Ray-Tracing in 6G Networks,” in Proceedings of the ACM Multimedia Systems Conference 2024 , ser. MMSys ’24. ACM, Apr. 2024, p. 441–447

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.