Pith. sign in

REVIEW 5 major objections 5 minor 28 references

This paper claims that a domain-adversarial ConvNeXt autoencoder, aligning channel-impulse-response features across two room layouts, restores UWB jammer localization to 34.67 cm mean error—down from 207.99 cm for the best source-trained mo

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 00:14 UTC pith:L5D346SN

load-bearing objection New UWB jammer localization datasets and solid baselines, but the headline 34.67 cm result is confounded by labeled target fine-tuning and a missing target-only supervised baseline. the 5 major comments →

arxiv 2511.01819 v1 pith:L5D346SN submitted 2025-11-03 cs.LG cs.AI

Machine and Deep Learning for Indoor UWB Jammer Localization

Hamed Fard , Mahsa Kholghi , Benedikt Gro{\ss} , Gerhard Wunder This is my paper
classification cs.LG cs.AI MSC 68T07
keywords UWB jammer localizationdomain shiftadversarial domain adaptationgradient reversal layerConvNeXt autoencoderchannel impulse responseindoor localizationregression
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that machine learning can locate a malicious UWB jammer inside a single room, but that models trained in one layout fail badly when furniture is moved. To fix this, the authors propose a domain-adversarial autoencoder that aligns the raw channel-impulse-response features between old and new layouts, and they show it reduces the mean localization error from about 208 cm to about 35 cm. This matters because UWB-based asset tracking and intrusion detection need to keep working when the environment changes, and jammers are a realistic threat to such systems.

Core claim

On a new dataset collected in the same test room before and after a layout change, source-trained baseline models degrade drastically—XGBoost's mean Euclidean error rises from 20.16 cm to 207.99 cm. A denoising ConvNeXt autoencoder with a gradient-reversal layer, called A-CNT, aligns CIR-derived features across domains and restores the mean error to 34.67 cm, with 56% of predictions within 30 cm, compared to 3% for the source-only baselines. The paper attributes this recovery specifically to adversarial feature alignment, since a non-adversarial version of the same autoencoder only reaches 148.02 cm.

What carries the argument

The central mechanism is a gradient-reversal layer (GRL) inserted between the encoder and a domain classifier, which inverts the domain-discrimination gradients during backpropagation, forcing the encoder to learn features that are uninformative about which room layout a sample came from. This is coupled with a compact ConvNeXt autoencoder that takes the first 100 taps of the UWB channel impulse response (as magnitude, sine phase, and cosine phase) and reconstructs the input while jointly training a regression head to predict jammer coordinates.

Load-bearing premise

The fine-tuning phase requires labeled target-domain samples with known jammer coordinates, so the method is weakly supervised at the final step, not fully unsupervised as the framing suggests.

What would settle it

Run the A-CNT pipeline with no target labels at all (only domain labels for the adversarial step) and evaluate on a second target layout with more than 16 jammer positions; if the mean error stays above ~150 cm or the 34.67 cm figure fails to replicate across multiple seeds and room rearrangements, the central claim of label-free adversarial recovery is false.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the result holds, adversarial domain alignment could be a practical recipe for keeping UWB jammer localization accurate when indoor furniture, partitions, or reflectors change, without retraining from scratch on the new layout.
  • The proposed A-CNT pipeline outperforms both classical UDA methods (CORAL, MMD) and a non-adversarial autoencoder, suggesting that gradient reversal adds value beyond simple feature reweighting or reconstruction-based transfer.
  • The source-domain regression baseline (XGBoost at 20.16 cm mean error) shows that classical ML on diagnostic registers is strong in a fixed environment, so future transfer methods can use these lightweight models as a reference point.
  • The paper's 3,000-sample bottleneck-embedding experiment, where a linear classifier separates five spatial zones with ~99.4% ROC-AUC, suggests the aligned features preserve spatial structure even after domain confusion, pointing toward interpretable and transferable representations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper fine-tunes with labeled target-domain coordinates (Eq. 5), so the final step is weakly supervised rather than fully label-free; a true test would be to see whether the 34.67 cm result holds when only the source has labels and the target only provides domain identity.
  • Because the target layout includes just 16 jammer positions and a single configuration, the 34.67 cm figure is a demonstration on one small grid, not a certified accuracy; repeated evaluations across new layouts and more positions would be the natural way to gauge real-world robustness.
  • The result hints that alignment at the level of raw CIR taps—not just diagnostic features—is what carries transferable information, which could motivate similar representations for other RF-based localization problems facing environment shifts.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. This paper studies indoor localization of a malicious UWB jammer under environmental domain shift. It introduces two datasets collected in the same room before and after a layout change (52 and 16 jammer positions, 461,795 and 28,793 samples respectively), benchmarks classical ML and deep models on classification and regression in the source domain, and proposes a domain-adversarial ConvNeXt autoencoder (A-CNT) with a gradient-reversal layer for CIR-based domain adaptation. The headline result is that A-CNT reduces mean Euclidean error on the target layout from 207.99 cm (source-trained XGBoost) to 34.67 cm, restoring F≤30cm to 0.56.

Significance. The datasets and promised public code are potentially valuable to the UWB localization and security communities; the benchmark study is systematic, includes multiple metrics and interpretability analyses, and the domain-shift problem is practically important. However, the central claim that adversarial feature alignment is the active ingredient that enables transferable localization is not established by the experiments as reported, because labeled target samples are used in the fine-tuning phase and no target-only supervised baseline is provided. If the missing controls confirm the ablation, the contribution would be solid and useful.

major comments (5)
  1. [Section 4.3, Eq. (5), Table 5] The pipeline is framed as unsupervised domain adaptation but fine-tuning uses labeled target samples (x_i, y_i) via L_reg. The final 34.67 cm result is therefore obtained after supervised training on the target domain. This confounds the effect of adversarial alignment with the trivial effect of training on target labels. Please add (i) a target-only supervised baseline (e.g., XGBoost on diagnostic features or the same regression head on the same CIR features, trained only on the 28,793 target samples) and (ii) an ablation of A-CNT without GRL but with the same fine-tuning procedure. Without these, the 83% improvement claim is not evidence for domain-adversarial alignment.
  2. [Section 4.3, Eq. (6)] The fine-tuning domain loss labels target-domain samples x_i with BCE(...,0), while Eq. (2) uses 0 for source and 1 for target. If this is not a typo, the domain classifier is being trained to label target samples as source during fine-tuning, which would be the opposite of the intended adversarial alignment; if it is a typo, it should be corrected to label 1. Please clarify and ensure the implementation matches the formalism.
  3. [Section 5, Table 5 and Section 4.3] The CNT vs A-CNT comparison is described as differing only by the GRL, but the text does not state whether CNT also receives labeled-target fine-tuning. If CNT is evaluated after the same fine-tuning protocol, the comparison isolates the GRL; if CNT does not include fine-tuning, the 148.02 vs 34.67 cm gap conflates the GRL with the fine-tuning stage. Please report the exact training protocol for each row of Table 5.
  4. [Section 5, Table 5, Figure 4] The target evaluation uses a single layout with 16 positions and reports point estimates without repeated-seed or cross-validation intervals. In addition, hyperparameters (α, λ_ft, σ, tap truncation) are described as chosen after 'preliminary experiments' and 'performance plateaus... evaluated on dedicated hold-out set' using target-domain behavior. This selection on the target hold-out makes the headline numbers optimistically biased. Report mean ± std over at least 5 independent runs (or nested validation) and state which target samples were used for early stopping vs final evaluation.
  5. [Section 4.1] CIR features for the DANN pipeline are 'normalized using a standard scaler fitted jointly to source and target datasets.' If target test samples are included in the scaler fit, the target evaluation is not independent; feature statistics from the evaluation set leak into model inputs. Please clarify the exact split used for fitting the scaler and, if target was included, re-run with a scaler fitted only on source (and target training, if separated) data.
minor comments (5)
  1. [Section 4.3, first paragraph of Fine-Tuning] Typo: 'reconstruction loss, regression loss and and adversarial domain loss' should read 'and adversarial domain loss'.
  2. [Table 3, RF row] The Time column for RF shows '2' without units or formatting consistent with other rows; clarify whether this is 2 minutes and add a decimal for readability.
  3. [Figure 6 caption] The caption states that overlap in the projected space is reduced after adaptation while also arguing for domain confusion; this apparent contradiction should be explained in the text.
  4. [Eq. (3) and surrounding text] The symbol λ is used both in the formal loss and in the schedule '0.05 to 0.2' without a subscript, making it ambiguous whether it is the same as λ_ft. Use λ_align for the joint-adversarial phase to avoid confusion.
  5. [Section 3.2, Data Collection Protocol] The number of samples per position is not reported; providing the per-position sample count would help assess class balance in the source domain (52 positions) and target domain (16 positions).

Circularity Check

0 steps flagged

No significant circularity; results are empirical benchmarks, not derivations.

full rationale

The paper is an empirical benchmarking study rather than a derivation chain: the reported quantities (20.16 cm, 207.99 cm, 34.67 cm) are measured errors of trained models, not predictions derived from assumptions. Eqs. (1)-(7) only define training losses. The fine-tuning phase explicitly uses labeled target-domain data (Sec. 4.3, Eq. (5)), and the paper's own comparison of A-CNT to the non-adversarial CNT isolates the gradient-reversal contribution, so the central claim about adversarial alignment is not simply an input renamed as an output. The lack of a target-only supervised baseline and the use of target-domain hold-out behavior for early stopping are legitimate experimental-design concerns (external validity, attribution of the gain to fine-tuning vs. alignment), but they are not circularity: no result is equivalent to an input by construction. There are no load-bearing self-citations, imported uniqueness theorems, or ansatz-by-citation. Hence score 0.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

No invented physical entities. Free parameters are mostly training hyperparameters chosen by hand; the most consequential fitted element is the use of labeled target samples during fine-tuning, which drives much of the reported gain.

free parameters (5)
  • CIR tap truncation at 100 taps = 100 taps
    Chosen after preliminary experiments (§4.1); longer tap sequences degraded performance.
  • Denoising noise sigma = 0.6
    Gaussian noise injected into encoder activations to prevent trivial identity mappings (§4.3).
  • GRL reversal strength schedule = 0.05 to 0.2 over 40 epochs
    Sigmoid schedule; early stopping at epoch 18 based on domain classifier AUC (§4.3).
  • Fine-tuning loss weights alpha and lambda_ft = alpha 0.5→0.1, lambda_ft 0.0→0.5, beta=1
    Progressively adjusted during 200 fine-tuning epochs (§4.3).
  • Autoencoder architecture hyperparameters = 3→128 channels; 782,211 parameters
    Chosen by hand; not optimized with Optuna (§4.3).
axioms (4)
  • domain assumption CIR taps and diagnostic features contain sufficient information to localize a jammer in a room.
    Baseline classification and regression rely on this; no physical model is provided to justify it beyond empirical performance (§4).
  • domain assumption Domain shift between room layouts manifests as a measurable feature-distribution shift.
    Visualized with Wasserstein distance in Appendix A.1; the visualization itself does not prove that adversarial alignment is the correct remedy.
  • domain assumption Gradient-reversal adversarial training learns features that transfer across domains.
    Core assumption of A-CNT, borrowed from DANN [7]; the paper does not derive conditions under which this holds for CIR data.
  • domain assumption Random splitting of samples yields independent train/test observations.
    The split strategy is not described; repeated samples from the same jammer position during TurtleBot rotation could inflate metrics if split randomly rather than by position.

pith-pipeline@v1.3.0-alltime-deepseek · 11758 in / 10370 out tokens · 113856 ms · 2026-08-04T00:14:27.243230+00:00 · methodology

0 comments
read the original abstract

Ultra-wideband (UWB) localization delivers centimeter-scale accuracy but is vulnerable to jamming attacks, creating security risks for asset tracking and intrusion detection in smart buildings. Although machine learning (ML) and deep learning (DL) methods have improved tag localization, localizing malicious jammers within a single room and across changing indoor layouts remains largely unexplored. Two novel UWB datasets, collected under original and modified room configurations, are introduced to establish comprehensive ML/DL baselines. Performance is rigorously evaluated using a variety of classification and regression metrics. On the source dataset with the collected UWB features, Random Forest achieves the highest F1-macro score of 0.95 and XGBoost achieves the lowest mean Euclidean error of 20.16 cm. However, deploying these source-trained models in the modified room layout led to severe performance degradation, with XGBoost's mean Euclidean error increasing tenfold to 207.99 cm, demonstrating significant domain shift. To mitigate this degradation, a domain-adversarial ConvNeXt autoencoder (A-CNT) is proposed that leverages a gradient-reversal layer to align CIR-derived features across domains. The A-CNT framework restores localization performance by reducing the mean Euclidean error to 34.67 cm. This represents a 77 percent improvement over non-adversarial transfer learning and an 83 percent improvement over the best baseline, restoring the fraction of samples within 30 cm to 0.56. Overall, the results demonstrate that adversarial feature alignment enables robust and transferable indoor jammer localization despite environmental changes. Code and dataset available at https://github.com/afbf4c8996f/Jammer-Loc

Figures

Figures reproduced from arXiv: 2511.01819 by Benedikt Gro{\ss}, Gerhard Wunder, Hamed Fard, Mahsa Kholghi.

Figure 1
Figure 1. Figure 1: Left: Schematic layout of the experimental environment showing the positions of the four UWB receivers (RX 1–4), the transmitter (TX) mounted on a TurtleBot robot with a 1 m extension arm, and the 52 distinct jammer positions (red dots) within a 3 m×5 m area. Right: The actual indoor setup, with the sender (TX), receivers (RX), and marked jammer locations on the floor corresponding to the schematic. Enviro… view at source ↗
Figure 2
Figure 2. Figure 2: Fine-tuning convergence over 200 epochs: (a) loss components Lrec, Lreg, Ldom; (b) weight schedules for α and λft Formalization of Loss Components. To summarize the losses utilized at each training phase mathematically, the following equations define reconstruc￾tion, regression, and adversarial losses employed (Eqs.1-7). In these equations, x t i denotes unlabeled samples from the target domain, x s j deno… view at source ↗
Figure 3
Figure 3. Figure 3: Confusion matrices of the three deep learning models evaluated on the source dataset (52 classes). Regression on source dataset. The goal of this experiment is to assess the regression performance of various ML and DL models for spatial localiza￾tion when trained, validated, and tested solely on the source dataset [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Left: K-means clustering of the 3000 hold-out ground-truth (x, y) points into five spatial zones. Marker “×” shows each zone centroid and area is proportional to visit count. Right: Schematic of the modified experimental layout with four UWB receivers (RX 1–4), the TurtleBot-mounted transmitter (TX) on a 1 m arm, and the 16 jammer positions (red dots) in a 3 m×5 m area. 6 Conclusion The challenge of UWB ja… view at source ↗
Figure 5
Figure 5. Figure 5: Side-by-side visualization of domain shift across taps. Left: Per-tap Wasserstein distance (EMD), with shaded regions indicating multiple intervals where EMD ≥ 0.1. Right: Mean feature values for source and target domains, with shaded regions where the absolute difference |∆mean| ≥ 0.1. A dashed green curve shows |∆mean| per tap. Shaded areas are bounded by the actual metric values for visual fidelity. A.2… view at source ↗
Figure 6
Figure 6. Figure 6: t-SNE visualization of feature representations before and after adversarial do￾main adaptation. Pretraining results in initial domain overlap, while GRL reshapes the feature space to promote domain confusion. Although overlap in the projected space is reduced after adaptation, the learned representation leads to improved target generalization, demonstrating that improved generalization can occur even as fe… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 1 linked inside Pith

  1. [1]

    Proceedings of the IEEE106(7), 1183–1203 (2018)

    Aditya, S., Molisch, A.F., Behairy, H.M.: A survey on the impact of multipath on wideband time-of-arrival based localization. Proceedings of the IEEE106(7), 1183–1203 (2018)

    2018

  2. [2]

    In: Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining

    Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A next- generation hyperparameter optimization framework. In: Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. pp. 2623–2631 (2019)

    2019

  3. [3]

    IEEE Access6, 17429–17441 (2018)

    Bregar, K., Mohorčič, M.: Improving indoor localization using convolutional neu- ral networks on computationally restricted devices. IEEE Access6, 17429–17441 (2018)

    2018

  4. [4]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Chollet, F.: Xception: Deep learning with depthwise separable convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1251–1258 (2017)

    2017

  5. [5]

    Advances in data science and information engineering: proceedings from ICDATA 2020 and IKE 2020 pp

    Farahani, A., Voghoei, S., Rasheed, K., Arabnia, H.R.: A brief review of domain adaptation. Advances in data science and information engineering: proceedings from ICDATA 2020 and IKE 2020 pp. 877–894 (2021)

    2020

  6. [6]

    IEEE Communications Surveys & Tutorials (2025)

    Fischer, G.K., Schaechtle, T., Gabbrielli, A., Bordoy, J., Häring, I., Höflinger, F., Rupitsch, S.J.: A systematic survey and comparative analysis of angular-based indoor localization and positioning technologies. IEEE Communications Surveys & Tutorials (2025)

    2025

  7. [7]

    Journal of machine learning research17(59), 1–35 (2016)

    Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., March, M., Lempitsky, V.: Domain-adversarial training of neural networks. Journal of machine learning research17(59), 1–35 (2016)

    2016

  8. [8]

    Neural Processing Letters55(3), 2429–2469 (2023)

    HassanPour Zonoozi, M., Seydi, V.: A survey on adversarial domain adaptation. Neural Processing Letters55(3), 2429–2469 (2023)

    2023

  9. [9]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    2016

  10. [10]

    IEEE Communications Letters24(10), 2226–2230 (2020)

    Jiang, C., Shen, J., Chen, S., Chen, Y., Liu, D., Bo, Y.: Uwb nlos/los classification using deep learning method. IEEE Communications Letters24(10), 2226–2230 (2020)

    2020

  11. [11]

    In: 2022 19th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP)

    Jun, L., Ruizhi, Z., Xiaofeng, S.: Precise positioning method of uwb in signal jamming. In: 2022 19th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP). pp. 1–4. IEEE (2022) 15

    2022

  12. [12]

    IEEE Transactions on Cognitive Commu- nications and Networking (2025)

    Li, L., Xing, Z., Guo, X., Zheng, H.: D-2mloc: Dual-domain mlp-mixer framework for csi-fingerprinting indoor localization. IEEE Transactions on Cognitive Commu- nications and Networking (2025)

    2025

  13. [13]

    APSIPA Transactions on Signal and Information Processing11(1) (2022)

    Liu, X., Yoo, C., Xing, F., Oh, H., El Fakhri, G., Kang, J.W., Woo, J., et al.: Deep unsupervised domain adaptation: A review of recent advances and perspectives. APSIPA Transactions on Signal and Information Processing11(1) (2022)

    2022

  14. [14]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Liu, Z., Mao, H., Wu, C.Y., Feichtenhofer, C., Darrell, T., Xie, S.: A convnet for the 2020s. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 11976–11986 (2022)

    2022

  15. [15]

    In: International conference on machine learning

    Long, M., Cao, Y., Wang, J., Jordan, M.: Learning transferable features with deep adaptation networks. In: International conference on machine learning. pp. 97–105. PMLR (2015)

    2015

  16. [16]

    Advances in neural information processing systems30(2017)

    Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. Advances in neural information processing systems30(2017)

    2017

  17. [17]

    IEEE Internet of Things Journal (2025)

    Nkrow, R.E., Boshoff, D., Silva, B., Hancke, G.P.: Adalos: A domain adaptive uwb nlos identification for dynamic settings. IEEE Internet of Things Journal (2025)

    2025

  18. [18]

    In: GLOBECOM 2023-2023 IEEE Global Communica- tions Conference

    Peterseil, P., Etzlinger, B., Khanzadeh, R., Springer, A.: Trustworthiness score for uwb indoor localization. In: GLOBECOM 2023-2023 IEEE Global Communica- tions Conference. pp. 189–194. IEEE (2023)

    2023

  19. [19]

    Foundations and Trends®in Machine Learning11(5-6), 355–607 (2019)

    Peyré, G., Cuturi, M., et al.: Computational optimal transport: With applications to data science. Foundations and Trends®in Machine Learning11(5-6), 355–607 (2019)

    2019

  20. [20]

    In: 2010 IEEE International Conference on Ultra-Wideband

    Poturalski, M., Flury, M., Papadimitratos, P., Hubaux, J.P., Le Boudec, J.Y.: The cicada attack: Degradation and denial of service in ir ranging. In: 2010 IEEE International Conference on Ultra-Wideband. vol. 2, pp. 1–4. IEEE (2010)

    2010

  21. [21]

    IEEE access11, 6973–7020 (2023)

    Singhal, P., Walambe, R., Ramanna, S., Kotecha, K.: Domain adaptation: chal- lenges, methods, datasets, and applications. IEEE access11, 6973–7020 (2023)

    2023

  22. [22]

    In: Proceedings of the AAAI conference on artificial intelligence

    Sun, B., Feng, J., Saenko, K.: Return of frustratingly easy domain adaptation. In: Proceedings of the AAAI conference on artificial intelligence. vol. 30 (2016)

    2016

  23. [23]

    Advances in neural information pro- cessing systems30(2017)

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. Advances in neural information pro- cessing systems30(2017)

    2017

  24. [24]

    ACM Transactions on Intelligent Systems and Technology (TIST)11(5), 1–46 (2020)

    Wilson, G., Cook, D.J.: A survey of unsupervised deep domain adaptation. ACM Transactions on Intelligent Systems and Technology (TIST)11(5), 1–46 (2020)

    2020

  25. [25]

    arXiv preprint arXiv:2209.09943 (2022)

    Xia, H., Wang, P.P., Koike-Akino, T., Wang, Y., Orlik, P., Ding, Z.: Adversarial bi- regressor network for domain adaptive regression. arXiv preprint arXiv:2209.09943 (2022)

    Pith/arXiv arXiv 2022

  26. [26]

    IEEE Communications Letters (2025)

    Xue, M., Xu, Z., Zhang, J., Wang, H., Shen, Y.: A generalization method for indoor localization via domain-invariant feature learning. IEEE Communications Letters (2025)

    2025

  27. [27]

    In: Proceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security

    Yang, Y., Wu, Z., Zhang, Y., Chen, T., Li, J., Yang, J., Liu, W., Zhang, X., Shi, R., Li, J., et al.: Uwbad: Towards effective and imperceptible jamming attacks against uwb ranging systems with cots chips. In: Proceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security. pp. 3376–3390 (2024)

    2024

  28. [28]

    ACM Computing Surveys (CSUR)46(2), 1–32 (2013) 16

    Yang, Z., Zhou, Z., Liu, Y.: From rssi to csi: Indoor localization via channel re- sponse. ACM Computing Surveys (CSUR)46(2), 1–32 (2013) 16

    2013