Simultaneous Localization and Mapping Using Active mmWave Sensing in 5G NR

Chao-Kai Wen; Fan Liu; Jiaxiang Guo; Jie Yang; Shi Jin; Shuqiang Xia; Tao Du

REVIEW 3 major objections 6 minor 13 references

Simultaneous Localization and Mapping Using Active mmWave Sensing in 5G NR

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A 5G terminal can map a room and locate itself by listening to its own beams.

desk verdict An honest prototype paper whose SLAM claim is only simulated; the one experimental number is calibration-in-sample. read the letter →

arxiv 2507.04662 v1 pith:OLEWGKXL submitted 2025-07-07 eess.SP

Tao Du , Jie Yang , Fan Liu , Jiaxiang Guo , Shuqiang Xia , Chao-Kai Wen , Shi Jin This is my paper · ORCID

classification eess.SP

keywords 5GNRmmWavesensingactivepointcloudsimultaneouslocalizationandmappingposegraphoptimizationpowerdelayprofilehardwarecalibration

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

This paper tries to establish that a 5G New Radio terminal can actively sense its surroundings and estimate its own trajectory without relying on base stations, effectively turning its millimeter-wave beams into a laser scanner. It claims that point clouds extracted from power-delay-profile peaks in each beam direction, fused with lidar-style scan matching and pose graph optimization, yield both an accurate global trajectory and a radio map of the environment. The experimental prototype achieves a ranging root-mean-square error of 0.34 m on a metal plate, and a simulation of a circular trajectory shows the pose-graph-optimized map preserving the room's shape. If the claim holds, future 5G devices could map and navigate using only their own transmissions and echoes.

What carries the argument

The load-bearing mechanism is the conversion of each beam direction's power delay profile (PDP) into one point in a radio point cloud. For a beam steered by a DFT codebook, the PDP is the IFFT of the estimated frequency-domain channel; under the assumption of a single reflector per direction, the PDP peak gives the reflector's round-trip range and the beam's maximum-gain angle gives its bearing. A fractional IFFT bisection search refines the peak, matched filtering with coherent integration over 12 OFDM symbols suppresses sidelobes, and a fixed hardware delay is calibrated using predefined target points. The resulting point clouds are then fused by lidar-style scan matching, submap-based loop closure, and pose graph optimization, which converts local relative pose estimates into a globally consistent trajectory and map.

What would settle it

Measure in a room with two reflectors of comparable strength separated by less than a meter in the same beam and inspect whether the point cloud separates them or merges them into a false peak; if the resulting map loses the room's walls or the trajectory error grows well beyond the reported 0.34 m, the single-reflector-per-beam premise does not carry to realistic rooms.

Watch

Extended reading notes

Core claim

The central claim is that a monostatic 5G NR millimeter-wave terminal can act as its own laser scanner: by transmitting OFDM symbols through DFT-codebook beams and receiving echoes on a dedicated array, the terminal builds a radio point cloud from the peaks of the power delay profile in each beam direction. The paper argues that this point cloud, after fixed hardware-delay calibration and fusion along a continuous trajectory with scan matching, loop closure, and pose graph optimization, yields both an accurate terminal trajectory and a detailed environmental map. Experimental support comes from a 28 GHz prototype that achieves 0.34 m ranging RMSE on a single metal plate and reconstructs a glass door and a concrete wall in a room, while simulation shows the optimized trajectory preserving the environment's shape, including a central obstacle. The author would state that active mmWave sensing on standard 5G NR hardware can deliver lidar-like simultaneous localization and mapping from the device's own transmissions.

Load-bearing premise

Each beam direction yields at most one true reflector, and the strongest power-delay-profile peak together with the beam-center angle is a faithful point-cloud sample.

Editorial extensions

If this is right

A 5G NR terminal could navigate and map without any base-station cooperation, since all sensing uses its own transmitted signals and echoes.
Sub-meter positioning with 0.34 m ranging error is achievable in single-reflection conditions, making room-scale localization feasible for indoor agents.
Loop closure plus pose graph optimization removes the drift that would otherwise accumulate over a continuous trajectory, as shown in the circular-trajectory simulation.
The approach works within the standard 5G NR frame structure by reserving sensing OFDM symbols, so it can be ported to other NR terminals with shared codebooks.

Reading between the lines

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

The single-reflector-per-beam assumption means the published accuracy is a best case; adapting the detector to resolve multiple returns per beam, for example via successive cancellation, is a natural next step the paper does not demonstrate.
The heatmap in the paper already shows distortion at large off-normal angles, so a fair test of real-room mapping would compare the reconstructed map against a ground-truth floor plan rather than only a metal plate.
The reported 0.34 m figure depends on the 95 MHz bandwidth and 28 GHz arrays; porting the same pipeline to different bandwidths or carrier frequencies would change the range resolution and likely the achieved map quality.

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.

Desk Editor's Note

read the letter

Here's my take. The paper is a competent system demonstration, not a validated SLAM result. The genuinely new piece is coupling a monostatic 5G NR mmWave prototype with a lidar-style point cloud SLAM pipeline, and getting a real radio point cloud of a room at 28 GHz. The binary-search refinement over the PDP peak is a sensible, small improvement, and the 0.34 m RMSE on a single metal plate is a nice number, but it is not an independent prediction: the same 16 target positions are used to estimate the hardware delay in Eq. (7) and then to compute the error. So that accuracy is partly fitted. The MF coherent integration result is credible; about 5 dB sidelobe reduction with randomized data matches intuition.

The soft spot is bigger. The SLAM section is evaluated only in simulation, where Sionna ray tracing generates one reflector per beam direction—the paper states 'assuming a single target per direction.' That clean assumption is violated in the measured room, as Fig. 2 shows off-normal distortion, and the -13 dB cleaning threshold is chosen from the same simulation used to show SLAM success. So the experimental evidence never exercises the SLAM loop, and there are no trajectory or map error metrics from measurements. The conclusion's 'accurate global trajectory and environmental mapping' is therefore supported only by a simulation that shares the same sensor model assumptions.

That said, the paper is transparent about the single-target assumption and the distortion, and the hardware work is real. The calibration issue is a common pitfall, not a fraud. If the authors ran a held-out experiment—say, move the plate to new positions after calibration, or run the actual terminal along a short trajectory and report pose error—the claim would be much stronger.

Who gets value: ISAC and radio SLAM researchers, and people building mmWave sensing prototypes. It deserves a serious referee because the integration is useful and the experimental platform is valuable, but it should come back with at least one independent accuracy test and quantitative SLAM metrics.

Referee Report

3 major / 6 minor

Summary. The paper proposes an active mmWave 5G NR sensing scheme in which a terminal with separate TX/RX and RX antenna arrays transmits OFDM signals, estimates the power delay profile (PDP) per DFT beam direction, extracts range-angle point clouds via a bisection search, and then applies a laser-style SLAM pipeline (scan matching, loop closure, pose graph optimization) to estimate the terminal trajectory and reconstruct an environmental map. The system is validated in two ways: OTA experiments on a prototype report a single-target ranging RMSE of 0.34 m after hardware-delay calibration, and a Sionna ray-tracing simulation demonstrates SLAM with and without PGO. The headline claim is that a 5G NR terminal can build a point-cloud map and localize itself using only its own active mmWave transmissions.

Significance. If the claim holds, the paper would demonstrate a practical active-sensing SLAM capability on a 5G NR prototype, extending passive radio SLAM to detailed point-cloud mapping without external anchors. The strengths are the OTA prototype implementation, the use of standard NR frame structures, and the explicit treatment of hardware delay calibration and sidelobe suppression via coherent integration of OFDM symbols. However, the experimental evidence supports only single-target ranging, not the full SLAM loop, and the simulation used to demonstrate SLAM relies on an idealized one-reflector-per-beam assumption and a filtering threshold tuned on the same simulation. These gaps currently limit the strength of the central claim.

major comments (3)

[§V-A, Eq. (7), Table II] The hardware delay TH is estimated from the same 16 target positions that are then used to compute the ranging errors in Table II. After subtracting a constant bias fitted on those exact points, the reported RMSE of 0.34 m is an in-sample residual, not an independent prediction of ranging accuracy. The paper should either use held-out target positions for the RMSE evaluation, report cross-validated errors, or explicitly state that Table II measures calibration residual rather than absolute ranging accuracy.
[§IV-B, Fig. 5] The SLAM results are obtained only from a Sionna simulation in which the point cloud is generated by casting one ray per beam direction and taking the closest target, as stated in §IV-A ('assuming a single target per direction'). The experimental system, by contrast, produces off-normal distortion and sidelobe artifacts that the paper itself acknowledges in §V-A and Fig. 2. Because the simulated point clouds do not reproduce the measured sensor behavior, the simulation does not establish that the SLAM pipeline works on real radio point clouds. An experimental SLAM experiment with ground-truth trajectory and map evaluation, or a simulation driven by the measured point-cloud statistics, is needed to support the claimed trajectory and mapping accuracy.
[§IV-A, §V-A] The single-target-per-beam assumption is central to the point-cloud extraction step, yet the experimental environment contains multiple reflecting surfaces (glass door, concrete wall, side walls), and Fig. 2 and §V-A report visible distortion at large angular offsets. The paper does not quantify how many beams in the measured data violate the single-target assumption, nor how those violations affect the subsequent registration and mapping. Without this characterization, the link between the demonstrated single-target ranging accuracy and the claimed environmental mapping performance remains unsubstantiated.

minor comments (6)

[§III-B] There is a typo in 'muliplying' and in 'intergation'; these should be corrected to 'multiplying' and 'integration'.
[§III-C, Eq. (7)] The notation T(i)_O, T(i)_1, and T(i)_2 is used before the sentence defining them; the definitions should be moved before Eq. (7) or the equation should be introduced after the variables are defined.
[§V-B] The simulation reports an SNR of 10 dB but does not state how SNR is defined (per-beam, per-symbol, or with respect to the strongest target). This should be clarified for reproducibility.
[§V-A] The sentence stating that the hardware delay is approximately 32 µs is surprising given the sub-meter ranging errors reported later; the authors might clarify whether this is a one-way or round-trip delay and whether it includes cable/processing delays.
[References] Reference [6] contains a formatting error in the author initials ('Y . Ge'), and reference [8] is cited as the source of hardware details while the new contributions relative to that prototype paper are not explicitly delineated.
[Fig. 5] The qualitative description of the maps (distortion, drift, resolution of the 'CNV' structure) is not accompanied by quantitative map or trajectory error metrics, such as RMSE against ground truth or map precision/recall, which would strengthen the SLAM evaluation.

Circularity Check

1 steps flagged · score 6.0 of 10

The only quantitative experimental result, a 0.34 m ranging RMSE, is an in-sample residual on the same 16 target positions used to fit the hardware delay in Eq. (7); the SLAM trajectory/map claim is demonstrated only with idealized single-target-ray point clouds.

fitted input called prediction [Section V-A (Experimental Setup and Results), Eq. (7) and Table II]
"Initially, the system’s hardware delay was estimated to be approximately 32 µs using 16 target points spaced at 0.8m intervals, calculated according to (7). After compensating for this delay, the ranging errors using Nc-point IFFT ( ε1) and bisection search ( ε2) are summarized in Table II."

TH in Eq. (7) is the average over the 16 calibration targets of (measured delay minus OTA delay), and Table II reports errors on the same 16 positions spaced 0.8 m apart. Subtracting this fitted scalar forces the mean residual error over those positions to be zero by construction, so the reported 0.34 m RMSE is the in-sample residual scatter of the calibration fit, not an independent ranging prediction. The conclusion nonetheless presents it as 'Experimental validation ... demonstrated an RMSE of 0.34 m', making the only quantitative experimental support for the central claim a fitting residual.

full rationale

The only quantitative experimental prediction—single-target ranging RMSE of 0.34 m—is computed on the identical 16 plate positions used to estimate the hardware delay in Eq. (7), so the mean error is zeroed by construction and the RMSE is a fit residual rather than a held-out validation. The SLAM trajectory/map claim is supported only by the simulation in Sections IV-A and V-B, where each ray is parameterized by the DFT codebook angle and the delay of the closest target at that angle under the single-target-per-direction assumption; the real room experiment (Fig. 2) reports off-normal distortion, so the simulation does not exercise the sensing errors present in practice. This is a validation gap, not an additional circular reduction. The self-citations in the paper ([5], [8], [11]) are used for hardware specifications and related-work context rather than as the load-bearing justification of the main result. Because one 'prediction' reduces to a fitted residual, the circularity score is 6.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central contribution is a system integration, not a new physical model. The load-bearing fitted quantities are the hardware delay and the point-cloud threshold; the principal domain assumptions are point-reflector, single-target-per-beam, far-field, and monostatic modeling, plus reliance on a ray-tracing simulator for SLAM validation.

free parameters (2)

Hardware delay bias T_H = approximately 32 microseconds
Estimated from 16 target positions via Eq. (7); all experimental range estimates are corrected by this value, and the same target positions are used to report RMSE.
Point cloud energy threshold = -13 dB
Chosen in Section IV-A from Sionna simulations to discard sidelobe-induced false points; applied to both simulation and experiments without independent optimization or validation.

assumptions (4)

domain assumption Single-target-per-beam assumption: the strongest PDP peak corresponds to the nearest physical reflector along the beam center direction.
Stated in Section IV-A as 'assuming a single target per direction'; breaks for extended or diffuse surfaces, such as the off-normal walls where the paper itself reports point cloud distortion.
domain assumption Far-field plane-wave channel model with point reflectors (Eq. 2).
Used to derive steering vectors and delays; ignores near-field effects, diffuse scattering, and the physical extent of reflectors, which are not validated against the experimental heatmap.
domain assumption Monostatic geometry with negligible baseline between TX/RX and RX arrays.
The two arrays are physically separated; the paper says increased spacing may lead to ranging errors 'to be calibrated in future work', yet the current pipeline treats the geometry as monostatic.
domain assumption The Sionna ray-tracing model with a known Blender geometry is an accurate representation of a real environment.
The SLAM validation in Section V-B uses point clouds generated from this model, so the simulated SLAM accuracy inherits any inaccuracies of the model and does not test real-world multipath complexity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Simultaneous Localization and Mapping Using Active mmWave Sensing in 5G NR." pith.science (2026). https://pith.science/paper/OLEWGKXL

@misc{pith2026250704662,
  author       = {Pith},
  title        = {Pith review of: Simultaneous Localization and Mapping Using Active mmWave Sensing in 5G NR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OLEWGKXL}},
  note         = {Machine review of arXiv:2507.04662}
}

read the original abstract

Millimeter-wave (mmWave) 5G New Radio (NR) communication systems, with their high-resolution antenna arrays and extensive bandwidth, offer a transformative opportunity for high-throughput data transmission and advanced environmental sensing. Although passive sensing-based SLAM techniques can estimate user locations and environmental reflections simultaneously, their effectiveness is often constrained by assumptions of specular reflections and oversimplified map representations. To overcome these limitations, this work employs a mmWave 5G NR system for active sensing, enabling it to function similarly to a laser scanner for point cloud generation. Specifically, point clouds are extracted from the power delay profile estimated from each beam direction using a binary search approach. To ensure accuracy, hardware delays are calibrated with multiple predefined target points. Pose variations of the terminal are then estimated from point cloud data gathered along continuous trajectory viewpoints using point cloud registration algorithms. Loop closure detection and pose graph optimization are subsequently applied to refine the sensing results, achieving precise terminal localization and detailed radio map reconstruction. The system is implemented and validated through both simulations and experiments, confirming the effectiveness of the proposed approach.

Figures

Figures reproduced from arXiv: 2507.04662 by the authors.

**Figure 1.** System Architecture Design. To mitigate interference with existing 5G NR communication systems [8], sensing OFDM symbols are inserted within the reserved symbols allocated to the first 5 ms of each frame, as depicted by the light blue region in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Experimental results of radio point cloud for a glass door. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The impact of PDP sidelobes and beam sidelobes on target detection. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Comparison of MF sidelobe levels with experimental results [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 5.** Figure 5: Simulation results of radio point cloud imaging-based SLAM. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 11 canonical work pages

[1]

Integrated sensing and communications: Recent advances and ten open challenges,

S. Lu et al ., “Integrated sensing and communications: Recent advances and ten open challenges,” IEEE Internet Things J. , vol. 11, no. 11, pp. 19094–19120, Feb. 2024

work page 2024
[2]

Radio SLAM: A review on radio-based simultaneous localization and mapping,

B. Amjad et al ., “Radio SLAM: A review on radio-based simultaneous localization and mapping,” IEEE Access , vol. 11, pp. 9260–9278, Jan. 2023

work page 2023
[3]

A belief propagation algorithm for multipath-based SLAM,

E. Leitinger, F. Meyer, F. Hlawatsch, K. Witrisal, F. Tufvesson, and M. Z. Win, “A belief propagation algorithm for multipath-based SLAM,” IEEE Trans. Wireless Commun. , vol. 18, no. 12, pp. 5613–5629, Sep. 2019

work page 2019
[4]

5G mmWave cooperative positioning and mapping using multi- model PHD filter and map fusion,

H. Kim, K. Granstr ¨om, L. Gao, G. Battistelli, S. Kim, and H. Wymeer- sch, “5G mmWave cooperative positioning and mapping using multi- model PHD filter and map fusion,” IEEE Trans. Wireless Commun. , vol. 19, no. 6, pp. 3782–3795, Mar. 2020

2020
[5]

Hybrid active and passive sensing for SLAM in wireless communication systems,

J. Yang, C.-K. Wen, and S. Jin, “Hybrid active and passive sensing for SLAM in wireless communication systems,” IEEE J. Sel. Areas Commun., vol. 40, no. 7, pp. 2146–2163, Jul. 2022

work page 2022
[6]

Integrated Monostatic and Bistatic mmWave Sensing,

Y . Ge et al. , “Integrated Monostatic and Bistatic mmWave Sensing,” in Proc. IEEE Global Communications Conference , Dec. 2023, pp. 3897– 3903

work page 2023
[7]

Millimeter-wave radio SLAM: End-to-end processing methods and experimental validation,

E. Rastorgueva-Foi et al. , “Millimeter-wave radio SLAM: End-to-end processing methods and experimental validation,” IEEE J. Sel. Areas Commun., vol. 42, no. 9, pp. 2550–2567, Sep. 2024

work page 2024
[8]

ISAC Prototype System for Multi-Domain Cooperative Communication Networks

J. Yang, H. Que, T. Du, L. Liang, X. Li, C.-K. Wen, and S. Jin, “ISAC prototype system for multi-domain cooperative communication net- works,” Oct. 2024. [Online]. Available: https://arxiv.org/abs/2410.22956

work page Pith review arXiv 2024

Show all 13 references

[9]

Indoor environment-adaptive mapping with beamsteer- ing massive arrays,

F. Guidi et al. , “Indoor environment-adaptive mapping with beamsteer- ing massive arrays,” IEEE Trans. V eh. Technol. , vol. 67, no. 10, pp. 10139–10143, Oct. 2018

2018
[10]

Radio-based sensing and indoor mapping with millimeter-wave 5G NR signals,

C. B. Barneto et al. , “Radio-based sensing and indoor mapping with millimeter-wave 5G NR signals,” in Proc. Int. Conf. Localization GNSS (ICL-GNSS), Jun. 2020, pp. 1–5

2020
[11]

Angle-based SLAM on 5G mmWave systems: Design, implementation, and measurement,

J. Yang et al. , “Angle-based SLAM on 5G mmWave systems: Design, implementation, and measurement,” IEEE Internet of Things J., pp. 1– 17, 2023

2023
[12]

On designing sequences with impulse-like periodic correlation,

P. Stoica, H. He, and J. Li, “On designing sequences with impulse-like periodic correlation,” IEEE Signal Process. Lett. , vol. 16, no. 8, pp. 703–706, Aug. 2009

2009
[13]

Sionna: An open-source library for next-generation physical layer research,

J. Hoydis et al. , “Sionna: An open-source library for next-generation physical layer research,” Mar. 2022, arXiv:2203.11854

2022 arXiv

Pith tools

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

[1] [1]

Integrated sensing and communications: Recent advances and ten open challenges,

S. Lu et al ., “Integrated sensing and communications: Recent advances and ten open challenges,” IEEE Internet Things J. , vol. 11, no. 11, pp. 19094–19120, Feb. 2024

work page 2024

[2] [2]

Radio SLAM: A review on radio-based simultaneous localization and mapping,

B. Amjad et al ., “Radio SLAM: A review on radio-based simultaneous localization and mapping,” IEEE Access , vol. 11, pp. 9260–9278, Jan. 2023

work page 2023

[3] [3]

A belief propagation algorithm for multipath-based SLAM,

E. Leitinger, F. Meyer, F. Hlawatsch, K. Witrisal, F. Tufvesson, and M. Z. Win, “A belief propagation algorithm for multipath-based SLAM,” IEEE Trans. Wireless Commun. , vol. 18, no. 12, pp. 5613–5629, Sep. 2019

work page 2019

[4] [4]

5G mmWave cooperative positioning and mapping using multi- model PHD filter and map fusion,

H. Kim, K. Granstr ¨om, L. Gao, G. Battistelli, S. Kim, and H. Wymeer- sch, “5G mmWave cooperative positioning and mapping using multi- model PHD filter and map fusion,” IEEE Trans. Wireless Commun. , vol. 19, no. 6, pp. 3782–3795, Mar. 2020

2020

[5] [5]

Hybrid active and passive sensing for SLAM in wireless communication systems,

J. Yang, C.-K. Wen, and S. Jin, “Hybrid active and passive sensing for SLAM in wireless communication systems,” IEEE J. Sel. Areas Commun., vol. 40, no. 7, pp. 2146–2163, Jul. 2022

work page 2022

[6] [6]

Integrated Monostatic and Bistatic mmWave Sensing,

Y . Ge et al. , “Integrated Monostatic and Bistatic mmWave Sensing,” in Proc. IEEE Global Communications Conference , Dec. 2023, pp. 3897– 3903

work page 2023

[7] [7]

Millimeter-wave radio SLAM: End-to-end processing methods and experimental validation,

E. Rastorgueva-Foi et al. , “Millimeter-wave radio SLAM: End-to-end processing methods and experimental validation,” IEEE J. Sel. Areas Commun., vol. 42, no. 9, pp. 2550–2567, Sep. 2024

work page 2024

[8] [8]

ISAC Prototype System for Multi-Domain Cooperative Communication Networks

J. Yang, H. Que, T. Du, L. Liang, X. Li, C.-K. Wen, and S. Jin, “ISAC prototype system for multi-domain cooperative communication net- works,” Oct. 2024. [Online]. Available: https://arxiv.org/abs/2410.22956

work page Pith review arXiv 2024