Pith. sign in

REVIEW 4 major objections 5 minor 34 references

A 6G Integrated Sensing and Communication Framework for Railway Intrusion Detection and Collision Prediction

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 ↗

Pith's one-line read This paper claims that a machine-learning model reading channel state information from a 6G-style radio link can tell whether an intruder is in a railway danger zone and estimate its position, velocity, and time to collision.

desk verdict A reasonable simulation-only feasibility study for CSI-based railway intrusion detection, but the undefined 'combined MAE' in the abstract undermines the headline quantitative claim. read the letter →

arxiv 2608.04710 v1 pith:5OS2XUJ5 submitted 2026-08-05 cs.LG cs.AIeess.SP

Ajeet Kumar Yadav , Sankaran Balasubramaniam , Aritra Chatterjee , Vinod Aduru , Yogesh Simmhan , Pandarasamy Arjunan This is my paper · ORCID
classification cs.LGcs.AIeess.SP
keywords 6GIntegratedSensingandCommunicationChannelStateInformationRailwayIntrusionDetectionSionna3DCNNBiLSTMCollisionPrediction
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

The paper aims to establish that a 6G radio link can act as a railway safety sensor: by reading channel state information (CSI) from an OFDM waveform, a machine-learning model can tell whether an intruder is inside the danger zone beside the track and can estimate the intruder's position, velocity, and time to collision. The evidence is generated entirely in simulation: a 3D-rendered railway scene in Blender is fed to the Sionna ray-tracing simulator, producing 22,695 CSI matrices with ground-truth labels. A hybrid 3D CNN–BiLSTM model trained on these matrices reaches 99.57% detection accuracy on a balanced test set and a combined mean absolute error of 0.4240 for the three kinematic quantities. The author argues this shows CSI-based integrated sensing and communication is a viable path for railway intrusion detection that goes beyond binary alarm.

What carries the argument

The load-bearing object is the CSI matrix itself, produced by Sionna's physics-based ray tracing. CSI is a complex-valued tensor indexed by transmit antenna, receive antenna, subcarrier, and OFDM symbol; it records how multipath reflections, scattering, and Doppler shifts alter the channel, so moving objects create time-varying patterns that encode both presence and kinematics. The preprocessing pipeline strips pilot and null subcarriers, selects informative subcarriers by an SNR–variance score with coherence-bandwidth decorrelation, removes the static background component, and applies Wiener-filter temporal denoising. The model is a two-stage network: a FrameEncoder made of 3D convolutional and residual blocks learns spatial-frequency features from each CSI frame, and a SequenceEncoder of two BiLSTM layers aggregates them across time, feeding a classification head (intruder present or not) and a regression head (five kinematic values).

What would settle it

Run the trained model on CSI collected from a real railway test track with the same 7x7 or 10x10 antenna configuration and known ground-truth intruder positions; if detection accuracy or kinematic MAE drops far from the reported 99.57% and 0.4240, the simulation-to-real transfer claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that an intruder's presence and motion leave a decodable signature in the complex-valued CSI matrix across antennas, subcarriers, and OFDM symbols, and that the signature can be extracted by a model that combines 3D convolutions for spatial structure with a bidirectional LSTM for temporal structure. On synthetic CSI, the model classifies whether an intruder is in the danger zone with 99.57% accuracy on a balanced test set and predicts relative position, velocity, and time to collision with a combined mean absolute error of 0.4240. The paper also reports that longer input sequences systematically improve accuracy, from 0.90 at a single frame to 0.9999 at 120 frames in one radio configuration, at the cost of latency. The authors present this as evidence that CSI-based sensing can extend conventional binary intrusion detection into collision-risk estimation.

Load-bearing premise

The entire result rests on the assumption that Sionna's ray-traced simulation produces CSI that behaves like real radio propagation on a real railway track, since all training and test data is synthetic; if real-world multipath, noise, weather, or object materials differ substantially, the reported accuracy and error rates would not transfer.

Editorial extensions

If this is right

  • Longer CSI input sequences improve detection accuracy from about 90 percent with a single frame to over 99.9 percent with 120 frames, so system designers can trade latency for reliability.
  • The performance pattern persists across two different radio configurations (small 120 kHz bandwidth and realistic 100 MHz with 833 subcarriers), suggesting the model learns generalizable CSI structure rather than memorizing one setup.
  • The regression head predicts relative position to within roughly 0.42 meters mean absolute error on the test set, and the example inference shows velocity, time-to-collision, and minimum separation distance tracked closely.
  • Because the full codebase is released, the data generation, preprocessing, and model training pipeline can be reproduced, extended to other scenes, or adapted to real-world CSI.

Reading between the lines

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

  • If simulated-to-real transfer holds, existing cell towers near rail corridors could be repurposed as intrusion sensors, complementing cameras and LiDAR that depend on line of sight and lighting.
  • The paper's own limitations imply the strongest gains are for large, reflective, fast-moving intruders; detection of small, absorptive, or nearly stationary objects is likely to degrade, and this is a testable boundary.
  • A natural next step the authors do not run is domain adaptation: use simulated CSI for pretraining and a modest amount of real track data for fine-tuning, then measure how much real data restores the reported accuracy.
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 proposes a CSI-based integrated sensing and communication (ISAC) framework for railway intrusion detection and collision prediction. The authors generate synthetic CSI matrices and ground-truth kinematic labels using a Blender-rendered railway scene and the Sionna ray-tracing simulator, preprocess the CSI through subcarrier selection, static-component removal, and temporal denoising, and train a 3D CNN-BiLSTM model to classify intruder presence and regress relative position, velocity, time-to-collision, and minimum separation distance. On synthetic data they report 99.57% classification accuracy on a balanced test set and a combined regression MAE of 0.4240. The codebase is publicly available, and the paper includes an explicit acknowledgment that real-world validation is still needed.

Significance. If the reported results are reproduced with a well-defined metric, the paper would provide a useful engineering demonstration that machine learning can decode intruder presence and kinematics from simulated 6G channel state information in a railway setting. The contribution is strengthened by the open-source code, the simulation-wise train/test split that avoids trajectory leakage, the sequence-length ablation, and the use of a second, more realistic radio configuration. The significance is limited by the closed-loop nature of the evaluation: both the CSI and the ground-truth labels come from the same simulator, so the results demonstrate the model's ability to invert the simulator's forward model rather than its performance on independently measured data. That limitation is acknowledged in Section V, but it should be reflected more carefully in the abstract's wording.

major comments (4)
  1. [Abstract; §IV-B, Table VII] The headline regression claim, "combined Mean Absolute Error (MAE) of 0.4240," is never defined. Section IV-B reports only the relative-position MAE of 0.4200 m in Table VII; there is no formula for combining position (m), velocity (m/s), and time-to-collision (s), no normalization or weighting scheme, and no per-target MAE table for the full test sets. The example errors in Table VIII (0.001 m/s, 0.012 m/s, 0.182 s, 0.104 m) do not combine with 0.4200 m to yield 0.4240 under any stated rule. The paper's second central quantitative claim is therefore unverifiable as written and must be corrected.
  2. [§III-D, Table III; §IV-B] The regression target vector is not specified. Table III shows a regression head with five outputs, but the paper never enumerates them (relative x/y position? velocity components? time-to-collision? minimum separation distance?), nor does it state how the targets are scaled or weighted in the MSE loss. Without this information, the reported kinematic prediction errors cannot be interpreted. In addition, no regression results are reported for Validation Round 2, so it is unclear whether the claimed combined MAE refers to Round 1, Round 2, or both.
  3. [§IV-A, Table VI; Abstract] The abstract's 99.57% accuracy claim is not tied to a specific validation round or sequence length. Table VI reports 0.9957 accuracy for sequence length 120 on Validation Round 2, but the class distribution of that test set is not given; the text states an 80:20 split of 11,400 samples but does not report the per-class counts. For Round 1 the class counts are given (6,186 class 1 and 5,109 class 0), but the balanced-test-set claim in the abstract is not supported by any reported class counts for the Round 2 test set. Please state which configuration produces the headline number and report the class balance explicitly.
  4. [§III-A; §IV-C, Table VIII] There is an internal inconsistency in the train motion model. Section III-A and Table I state that the train velocity is held constant at 10 m/s, but Section IV-C describes a scenario in which "the train was briefly stopped and then resumed moving," and Figure 11 shows a kink in the relative-distance curve attributed to that stop. If the data generation actually includes non-constant train velocity, Table I and the generation procedure must be updated; if not, the inference example is misdescribed. This matters because time-to-collision and minimum-separation ground truth depend directly on the train trajectory.
minor comments (5)
  1. [§III-C] The subcarrier-selection procedure depends on several unspecified free parameters, including the target keep-ratio, the coherence-bandwidth threshold, and the SNR/variance estimates used in the composite score; please report the chosen values and, ideally, a sensitivity analysis.
  2. [§III-A, Table I] The text states that in 95% of simulations the intruder is at its maximum dimension, but the dataset composition by intruder size is not tabulated; this is relevant because small intruders are described as the difficult detection case and the test split should be shown to cover them.
  3. [§IV-A, Table VI] Please report the exact training/test class counts for Validation Round 2 so the reader can verify the "balanced test data" statement and interpret precision, recall, and F1.
  4. [Figure 10, Table III] The fixed input shape (45, 2, 49, 49, 833) corresponds to Validation Round 2, while Validation Round 1 has a different raw shape (100x100x30x1, 1 OFDM symbol); clarify how the preprocessing maps each round into the model input and whether a separate model was trained per round.
  5. [§IV-C, Figure 11] The caption and text use an 8.0 m danger threshold described as 3 m plus half the train and intruder dimensions; please give the explicit formula so the threshold is reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the synthetic-data evaluation is a self-contained supervised-learning consistency check, not a derivation that reduces to its own inputs.

full rationale

The paper's derivation chain is: (1) generate synthetic CSI with the Sionna ray-tracing simulator over a Blender-rendered railway scene, (2) preprocess the CSI, (3) train a 3D CNN-BiLSTM to classify intruder presence and regress kinematics, and (4) evaluate on a simulation-wise held-out test set. No equation in the paper defines the target in terms of the model output, and no fitted parameter is renamed as a prediction. The ground-truth intruder positions, velocities, and collision times are recorded simulation inputs used as supervised labels; the model is trained to map held-out CSI to those labels. This is a standard closed-loop synthetic evaluation: it demonstrates that the learned model can invert the simulator's forward model on unseen simulated data. That is a meaningful internal consistency check, but it is not a circular derivation because the reported accuracy and MAE are measured on data not used for training, and the labels are not constructed from the model's own outputs. The one load-bearing limitation is external validity, not circularity: Section V states that 'the current study is based on synthetically generated CSI under simulated propagation conditions; therefore, validation using real-world railway measurements is necessary.' This is an honest scoping statement, not a circular step. Separately, the abstract's 'combined MAE of 0.4240' is never defined in Section IV-B, and the per-output errors in Table VIII do not obviously combine to that number; this is a reporting and verifiability defect, but it is not a circularity issue. The paper contains no load-bearing self-citations: references to Sionna, Blender, Optuna, and prior CSI-sensing works are external tools and background, not the authors' own prior claims used to force the present conclusion. Therefore, no circularity step can be exhibited.

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

The central claim rests on the fidelity of the Sionna simulation, the sufficiency of CSI features for kinematic regression, and the preprocessing choices. No new physical entities are introduced. The main free parameters are the subcarrier selection target ratio and the Wiener filter PSD estimates, both of which are unspecified and therefore not auditable from the paper alone.

free parameters (2)
  • Subcarrier selection target keep-ratio
    Section III-C.2 describes an SNR-weighted subcarrier selection with a target keep-ratio, but the paper never reports the ratio value or the final number of selected subcarriers, so the feature dimensionality is a hand-chosen, unaudited parameter.
  • Wiener filter signal and noise PSD estimates
    Section III-C.4 says the filter coefficients are computed from signal and noise autocorrelations that are "estimated from CSI measurements," but no estimator is specified, leaving the denoising strength as an unstated modeling choice that affects the input features.
assumptions (4)
  • domain assumption Sionna ray tracing produces physically faithful CSI for a real railway environment.
    The entire dataset and all conclusions depend on the fidelity of the Blender scene and Sionna simulator in Section III-A. The paper's own Section V acknowledges that real-world validation is required, confirming this is an unverified premise.
  • domain assumption CSI contains sufficient information to recover intruder position, velocity, and time to collision under the tested radio configurations.
    The ML model's success is an empirical test of this premise, but no information-theoretic bound or physical analysis is provided in Section III to establish that the CSI features are sufficient for all three kinematic outputs.
  • domain assumption The preprocessing stages (static component removal, temporal denoising) preserve the intruder's dynamic signature and do not discard the information needed for regression.
    Section III-C applies static component removal and Wiener filtering before training. If these filters suppress the time-varying component caused by the intruder, the reported performance would be an artifact of preprocessing choices.
  • standard math Standard OFDM and ray-tracing mathematics underpin the CSI generation.
    Equations (1)-(4) use standard channel impulse response, OFDM CSI, and delay-Doppler transforms, which are accepted background results in wireless communications.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A 6G Integrated Sensing and Communication Framework for Railway Intrusion Detection and Collision Prediction." pith.science (2026). https://pith.science/paper/5OS2XUJ5

@misc{pith2026260804710,
  author       = {Pith},
  title        = {Pith review of: A 6G Integrated Sensing and Communication Framework for Railway Intrusion Detection and Collision Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5OS2XUJ5}},
  note         = {Machine review of arXiv:2608.04710}
}
read the original abstract

Integrated Sensing and Communication (ISAC) combines sensing and communication to efficiently utilize wireless resources and is emerging as a key paradigm for next-generation wireless networks. By leveraging the wide bandwidth, high frequencies, and massive antenna arrays of 5G-Advanced and 6G systems, ISAC enables physical-layer sensing using Channel State Information (CSI). The 3rd Generation Partnership Project (3GPP) Release 19 identifies 32 potential ISAC use cases, with particular emphasis on detecting and tracking moving objects. In this work, we address the Sensing for Railway Intrusion Detection use case, where intruders, including wildlife, entering a railway track can pose serious collision risks. We generated 22,695 CSI matrices with corresponding ground truth using a 3D-rendered railway environment and the Sionna radio simulator. We developed a machine learning model combining a three-dimensional Convolutional Neural Network (3D CNN) and Bidirectional Long Short-Term Memory (BiLSTM) network to detect intruders in the track danger zone and estimate their real-time position relative to the train, velocity, and time to collision. On synthetic CSI data, the model achieves 99.57% intruder-detection accuracy on a balanced test set and a combined Mean Absolute Error (MAE) of 0.4240 for position, velocity, and time-to-collision prediction. These results demonstrate the potential of CSI-based ISAC sensing with machine learning for reliable railway intrusion detection. The complete codebase for CSI generation, preprocessing, and model development is publicly available at https://github.com/EdgeIntelligenceLab/6g-isac-railway-intrusion-detection.

Figures

Figures reproduced from arXiv: 2608.04710 by the authors.

Figure 2
Figure 2. A visual description of our complete approach. First, we generate CSI [3] data using the Sionna [1] simulation. After that, we pre-process raw CSI [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A 3D rendered scene of a train platform with the help of Blender [2]. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. The 2D layout of the 3D-rendered scene is shown in the diagram. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Physics-based CSI calculation procedure in Sionna. Although sensing is continuous, to generate the CSI, Sionna takes a snapshot of the scene and [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Shape of a standard CSI matrix as given by [9] across Tx and Rx [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: CSI magnitude visualisation using a four-step method when an intruder is present in the danger zone. Current position of moving intruder is [-5.79, -10] [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: CSI phase visualization using the four-step method when an intruder is present in the danger zone. Current position of moving intruder is [-5.79, -10] [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Scattering plot of 1000 CSI data after application of PCA with 3 [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: This is the proposed ML model architecture. It has two major sections: the first extracts spatial-temporal features, and the second is task-specific, [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Predicted trajectory of the relative position of the intruder with [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 31 canonical work pages

  1. [1]

    ”Sionna: An open-source library for next-generation physical layer research.” arXiv preprint arXiv:2203.11854 (2022)

    Hoydis, Jakob, et al. ”Sionna: An open-source library for next-generation physical layer research.” arXiv preprint arXiv:2203.11854 (2022)

    arXiv 2022

  2. [2]

    https://www.blender.org/

  3. [3]

    Foschini, G. J. (1996). Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas. Bell labs technical journal, 1(2), 41-59

    work page 1996

  4. [4]

    https://railsafe.org.au/ media/documents/resources/worksite- protection/Rail-Industry-Safety-Induction-RISI-Handbook.pdf

    work page

  5. [5]

    ”WiFi vision: Sensing, recognition, and detection with commodity MIMO-OFDM WiFi.” IEEE Internet of Things Journal 7.9 (2020): 8296-8317

    He, Ying, et al. ”WiFi vision: Sensing, recognition, and detection with commodity MIMO-OFDM WiFi.” IEEE Internet of Things Journal 7.9 (2020): 8296-8317

    work page 2020

  6. [6]

    Suggested Measures to Mitigate Elephant & Other Wildlife Train Collisions on Vulnerable Railway Stretches in India

    work page

  7. [7]

    Elnashar, Ayman, Marwan Bin Shakar, and Sami Muhaidat. ”Integrated Sensing and Communication in 6G: A Comprehensive Survey of Use Cases, Enabling Technologies, Standardization Roadmap, and Monetiza- tion Frameworks.” IEEE Open Journal of the Communications Society (2026)

    work page 2026

  8. [8]

    Chang, R. W. (1966). Synthesis of band-limited orthogonal signals for multichannel data transmission. Bell system technical journal, 45(10), 1775-1796

    work page 1966

Show all 34 references
  1. [9]

    ”Contactless WiFi sensing and monitoring for future healthcare-emerging trends, challenges, and opportunities.” IEEE Re- views in Biomedical Engineering 16 (2022): 171-191

    Ge, Yao, et al. ”Contactless WiFi sensing and monitoring for future healthcare-emerging trends, challenges, and opportunities.” IEEE Re- views in Biomedical Engineering 16 (2022): 171-191

    2022

  2. [10]

    Pearson, Karl. ”LIII. On lines and planes of closest fit to systems of points in space.” The London, Edinburgh, and Dublin philosophical magazine and journal of science 2.11 (1901): 559-572

    1901

  3. [11]

    ”Long short-term memory.” Neural computation 9.8 (1997): 1735-1780

    Hochreiter, Sepp, and J ¨urgen Schmidhuber. ”Long short-term memory.” Neural computation 9.8 (1997): 1735-1780

    1997

  4. [12]

    ”Integrated sensing and communication (ISAC)—From concept to practice.” Communications of Huawei Re- search (2022): 4-25

    Bayesteh, Alireza, et al. ”Integrated sensing and communication (ISAC)—From concept to practice.” Communications of Huawei Re- search (2022): 4-25

    2022

  5. [13]

    ”A unified future: Integrated sensing and communication (ISAC) in 6G.” IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation (2025)

    Ghosh, Amitava, et al. ”A unified future: Integrated sensing and communication (ISAC) in 6G.” IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation (2025)

    2025

  6. [14]

    Practical Issues and Challenges in CSI-based Integrated Sensing and Communication,

    D. Zhang, D. Wu, K. Niu, X. Wang, F. Zhang, J. Yao, D. Jiang, and F. Qin, “Practical Issues and Challenges in CSI-based Integrated Sensing and Communication,” 2022

    2022

  7. [15]

    ”Architecture Considerations for ISAC in 6G.” 2025 IEEE Conference on Standards for Communications and Networking (CSCN)

    Robitzsch, Sebastian, et al. ”Architecture Considerations for ISAC in 6G.” 2025 IEEE Conference on Standards for Communications and Networking (CSCN). IEEE, 2025

    2025

  8. [16]

    ”A novel high-precision railway obstacle detection algorithm based on 3D LiDAR.” Sensors 24.10 (2024): 3148

    Nan, Zongliang, et al. ”A novel high-precision railway obstacle detection algorithm based on 3D LiDAR.” Sensors 24.10 (2024): 3148

    2024

  9. [17]

    ”Railway Intrusion Risk Quantification with Track Semantic Segmentation and Spatiotemporal Features.” Sensors 25.17 (2025): 5266

    Ning, Shanping, et al. ”Railway Intrusion Risk Quantification with Track Semantic Segmentation and Spatiotemporal Features.” Sensors 25.17 (2025): 5266

    2025

  10. [18]

    ”Railway obstacle intrusion detection and risk assessment based on MSIA-YOLOv8 and DALNet.” Expert Systems with Applications (2025): 130132

    Hu, Tingyi, Feng Gao, and Fuqiang Zhou. ”Railway obstacle intrusion detection and risk assessment based on MSIA-YOLOv8 and DALNet.” Expert Systems with Applications (2025): 130132

    2025

  11. [19]

    3rd Generation Partnership Project; Technical Specification Group TSG SA; Feasibility Study on Integrated Sensing and Communication (Re- lease 19)

  12. [20]

    ”Prediction of degraded infrastructure conditions for railway operation.” Sensors 24.8 (2024): 2456

    Sanz Bobi, Juan de Dios, et al. ”Prediction of degraded infrastructure conditions for railway operation.” Sensors 24.8 (2024): 2456

    2024

  13. [21]

    ”Optuna: A next-generation hyperparameter opti- mization framework.” Proceedings of the 25th ACM SIGKDD interna- tional conference on knowledge discovery & data mining

    Akiba, Takuya, et al. ”Optuna: A next-generation hyperparameter opti- mization framework.” Proceedings of the 25th ACM SIGKDD interna- tional conference on knowledge discovery & data mining. 2019

    2019

  14. [22]

    ”Robust Pedestrian Detection and Intrusion Judgment in Coal Yard Hazard Areas via 3D LiDAR-Based Deep Learning.” Sensors 25.18 (2025): 5908

    Zhao, Anxin, Yekai Zhao, and Qiuhong Zheng. ”Robust Pedestrian Detection and Intrusion Judgment in Coal Yard Hazard Areas via 3D LiDAR-Based Deep Learning.” Sensors 25.18 (2025): 5908

    2025

  15. [23]

    ”Automatic detection of foreign object intrusion along railway tracks based on MACENet.” Plos one 20.8 (2025): e0329303

    Chen, Xichun, et al. ”Automatic detection of foreign object intrusion along railway tracks based on MACENet.” Plos one 20.8 (2025): e0329303

    2025

  16. [24]

    ”AI-based Animal Intrusion Detection System for Human-Wildlife Conflicts in Bhutan.” Zorig Melong— A Technical Journal of Science, Engineering and Technology 8.1 (2025): 46-52

    Dorji, Tsheten, et al. ”AI-based Animal Intrusion Detection System for Human-Wildlife Conflicts in Bhutan.” Zorig Melong— A Technical Journal of Science, Engineering and Technology 8.1 (2025): 46-52

    2025

  17. [25]

    ”BCD-YOLO: a railway perimeter foreign body intrusion detection method based on Yolov8.” Measurement Science and Technology 36.5 (2025): 056007

    Wang, Xiaopeng, Ce Han, and Weidong Jin. ”BCD-YOLO: a railway perimeter foreign body intrusion detection method based on Yolov8.” Measurement Science and Technology 36.5 (2025): 056007

    2025

  18. [26]

    and Wei, S

    Liu, L., Zhang, W., Deng, C., Yin, S. and Wei, S. (2015), Bri- Guard: a lightweight indoor intrusion detection system based on in- frared light spot displacement. IET Sci. Meas. Technol., 9: 306-314. https://doi.org/10.1049/iet-smt.2013.0171

    2015

  19. [27]

    ”6G positioning and sensing through the lens of sustainability, inclusiveness, and trustworthiness.” IEEE Wireless Communications 32.1 (2025): 68-75

    Wymeersch, Henk, et al. ”6G positioning and sensing through the lens of sustainability, inclusiveness, and trustworthiness.” IEEE Wireless Communications 32.1 (2025): 68-75

    2025

  20. [28]

    ”Sionna RT: Differentiable ray tracing for radio propagation modeling.” 2023 IEEE Globecom Workshops (GC Wkshps)

    Hoydis, Jakob, et al. ”Sionna RT: Differentiable ray tracing for radio propagation modeling.” 2023 IEEE Globecom Workshops (GC Wkshps). IEEE, 2023

    2023

  21. [29]

    ”A CSI-based human activity recognition using deep learning.” Sensors 21.21 (2021): 7225

    Moshiri, Parisa Fard, et al. ”A CSI-based human activity recognition using deep learning.” Sensors 21.21 (2021): 7225

    2021

  22. [30]

    ”KAN-Sense: Keypad input recogni- tion via CSI feature clustering and KAN-based classifier.” Electronics 14.15 (2025): 2965

    Koo, Minseok, and Jaesung Park. ”KAN-Sense: Keypad input recogni- tion via CSI feature clustering and KAN-based classifier.” Electronics 14.15 (2025): 2965

    2025

  23. [31]

    ”A Subcarrier Selection Method for Wi-Fi-based Respiration Monitoring using IEEE 802.11 ac/ax Protocols.” 2022 IEEE MTT-S International Microwave Biomedical Conference (IMBioC)

    Wang, Ruilin, et al. ”A Subcarrier Selection Method for Wi-Fi-based Respiration Monitoring using IEEE 802.11 ac/ax Protocols.” 2022 IEEE MTT-S International Microwave Biomedical Conference (IMBioC). IEEE, 2022

    2022

  24. [32]

    ”Integrated sensing and communications: Toward dual- functional wireless networks for 6G and beyond.” IEEE journal on selected areas in communications 40.6 (2022): 1728-1767

    Liu, Fan, et al. ”Integrated sensing and communications: Toward dual- functional wireless networks for 6G and beyond.” IEEE journal on selected areas in communications 40.6 (2022): 1728-1767

    2022

  25. [33]

    ”3D convolutional neural networks for human action recognition.” IEEE transactions on pattern analysis and machine intelligence 35.1 (2012): 221-231

    Ji, Shuiwang, et al. ”3D convolutional neural networks for human action recognition.” IEEE transactions on pattern analysis and machine intelligence 35.1 (2012): 221-231

    2012

  26. [34]

    ”Machine learning in 6G wireless communications.” IEICE Transactions on Communications 106.2 (2023): 75-83

    Ohtsuki, Tomoaki. ”Machine learning in 6G wireless communications.” IEICE Transactions on Communications 106.2 (2023): 75-83

    2023

Pith tools

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