pith. sign in

arxiv: 2604.23144 · v1 · submitted 2026-04-25 · 📡 eess.AS · eess.SP

Predictive Directional Selective Fixed-Filter Active Noise Control for Moving Sources via a Convolutional Recurrent Neural Network

Boxiang Wang , Zhengding Luo , Dongyuan Shi , Junwei Ji , Xiruo Su , Woon-Seng Gan This is my paper

Pith reviewed 2026-05-08 07:06 UTC · model grok-4.3

classification 📡 eess.AS eess.SP
keywords active noise controlmoving sourcesdirectional selectivefixed-filter ANCconvolutional recurrent neural networkdirection of arrival predictionnoise tracking
0
0 comments X

The pith

A neural network forecasts the direction of moving noise sources to select the right fixed control filter ahead of time.

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

The paper proposes Predictive Directional Selective Fixed-Filter Active Noise Control (PD-SFANC), which adds a Convolutional Recurrent Neural Network to standard directional selective fixed-filter ANC. Conventional D-SFANC picks a pre-trained filter based on the current direction of arrival, but this lags when the source moves because selection happens after the direction changes. The CRNN learns temporal patterns in past noise movements to predict the next direction and the matching filter, allowing cancellation to begin before the noise arrives from the new position. Simulations across different movement scenarios show this yields better tracking and overall noise reduction than several baseline ANC methods.

Core claim

The authors establish that a CRNN trained to capture the hidden temporal dynamics of moving noise can reliably predict the future control filter, enabling the PD-SFANC system to improve noise-tracking ability and dynamic noise-reduction performance compared to conventional D-SFANC and other representative ANC baselines, as confirmed in numerical simulations for various movement scenarios.

What carries the argument

The Predictive Directional SFANC (PD-SFANC) method, in which a Convolutional Recurrent Neural Network predicts the next direction-of-arrival and corresponding pre-trained control filter from observed movement patterns.

If this is right

  • The system tracks non-stationary noise from moving sources more effectively than reactive selection.
  • Dynamic noise reduction improves across multiple tested movement scenarios.
  • Pre-trained fixed filters can be used proactively without continuous adaptation.
  • Performance gains hold relative to several established ANC baselines.

Where Pith is reading between the lines

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

  • Real-time deployment in enclosed spaces such as vehicle cabins could reduce audible noise during source transitions if prediction remains accurate amid reflections.
  • The same temporal-prediction idea might be applied to other ANC parameters such as frequency content when sources move.
  • If prediction errors are small, the method could lower computational load by avoiding full adaptive filtering in many cases.

Load-bearing premise

That patterns learned from past noise movements will predict future directions accurately enough for the selected filter to outperform reactive methods in actual acoustic conditions.

What would settle it

An experiment in which a noise source follows a movement trajectory outside the training distribution and the measured noise reduction of PD-SFANC falls below that of standard D-SFANC.

Figures

Figures reproduced from arXiv: 2604.23144 by Boxiang Wang, Dongyuan Shi, Junwei Ji, Woon-Seng Gan, Xiruo Su, Zhengding Luo.

Figure 1
Figure 1. Figure 1: Comparison between (a) directional SFANC and (b) the proposed predictive directional SFANC. spatial characteristics of the noise source, which significantly affect ANC performance [19, 20, 21, 22]. To address this, the Directional SFANC (D-SFANC) method has been proposed, in￾corporating Direction-of-Arrival (DoA) information into the fil￾ter selection process [23, 24, 25]. However, as shown in view at source ↗
Figure 2
Figure 2. Figure 2: Block diagram of the predictive directional SFANC. 2.1. Pre-trained control filter library Prior to the online execution of PD-SFANC, a control filter li￾brary is pre-trained to accommodate noise sources at various DoAs. Assume a discrete grid of DoAs for the noise source, denoted as θv ∈ {θ1, . . . , θV }, where V is the number of can￾didate angles. To alleviate the design complexity, at each DoA θv, a co… view at source ↗
Figure 3
Figure 3. Figure 3: Proposed CRNN architecture for next-frame DoA prediction using multi-frame context view at source ↗
Figure 4
Figure 4. Figure 4: Noise reduction performance in (a) frequency and (b) time domains, and (c) the selected control filter for different ANC methods under vacuum cleaner noise moving at a constant rate, where the DoA varies linearly with an angular velocity of 10◦ /s view at source ↗
Figure 5
Figure 5. Figure 5: Noise reduction performance in (a) frequency and (b) time domains, and (c) the selected control filter for different ANC methods under vacuum cleaner noise moving at a time-varying rate, where the DoA varies sinusoidally between 50◦ and 150◦ . 3.4. Noise reduction performance To evaluate the noise reduction performance of PD-SFANC, a rectangular enclosure of size (11, 9, 3.2) m is simulated. A multi-refere… view at source ↗
read the original abstract

Directional Selective Fixed-Filter Active Noise Control (D-SFANC) can effectively attenuate noise from different directions by selecting the suitable pre-trained control filter based on the Direction-of-Arrival (DoA) of the current noise. However, this method is weak at tracking the direction variations of non-stationary noise, such as that from a moving source. Therefore, this work proposes a Predictive Directional SFANC (PD-SFANC) method that uses a Convolutional Recurrent Neural Network (CRNN) to capture the hidden temporal dynamics of the moving noise and predict the control filter to cancel future noise. Accordingly, the proposed method can significantly improve its noise-tracking ability and dynamic noise-reduction performance. Furthermore, numerical simulations confirm the superiority of the proposed method for handling moving sources across various movement scenarios, compared to several representative ANC baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes Predictive Directional Selective Fixed-Filter Active Noise Control (PD-SFANC), which augments Directional SFANC by using a Convolutional Recurrent Neural Network (CRNN) to capture temporal dynamics of moving noise sources and predict future DoA and the corresponding pre-trained control filter. This is claimed to improve noise-tracking and dynamic noise-reduction performance over reactive D-SFANC and other ANC baselines, with the superiority asserted via numerical simulations across various movement scenarios.

Significance. If the simulation results hold under proper generalization testing, the approach could meaningfully advance fixed-filter ANC for non-stationary sources by shifting from reactive to predictive selection. The CRNN-based prediction of future control filters is a clear technical contribution that leverages temporal modeling, and the provision of simulation-based comparisons against representative baselines supplies concrete evidence for the performance gains.

major comments (2)
  1. [§4] §4 (Numerical Simulations) and abstract: The headline claim that PD-SFANC 'significantly improve[s] its noise-tracking ability' and outperforms baselines 'across various movement scenarios' is load-bearing on the CRNN producing accurate future predictions. The manuscript provides no quantitative prediction-error metrics (e.g., DoA RMSE over the prediction horizon) or ablation showing how prediction errors affect the final noise-reduction level, leaving the net gain after error propagation unverified.
  2. [§4.2] §4.2 (CRNN Training and Testing): No description is given of whether the movement trajectories, speeds, or noise statistics in the test set are disjoint from the training set. Without explicit out-of-distribution or cross-trajectory validation, the reported superiority cannot be taken as evidence of reliable prediction on unseen moving-source conditions, which directly undermines the generalization asserted in the abstract.
minor comments (2)
  1. [Figure 2] Figure 2 (system block diagram): The interface between the CRNN output and the fixed-filter selection block would be clearer if the prediction horizon (in samples or time) were explicitly annotated on the diagram.
  2. [§3.1] §3.1 (CRNN Architecture): The number of convolutional and recurrent layers, filter sizes, and hidden-unit counts are stated, but the exact loss function used for training the DoA predictor (e.g., angular MSE or classification cross-entropy) is not specified.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the detailed and constructive feedback. The comments identify key areas where additional quantitative evidence and documentation would strengthen the presentation of our results. We address each major comment below and will incorporate the suggested improvements in the revised manuscript.

read point-by-point responses

  1. Referee: §4 (Numerical Simulations) and abstract: The headline claim that PD-SFANC 'significantly improve[s] its noise-tracking ability' and outperforms baselines 'across various movement scenarios' is load-bearing on the CRNN producing accurate future predictions. The manuscript provides no quantitative prediction-error metrics (e.g., DoA RMSE over the prediction horizon) or ablation showing how prediction errors affect the final noise-reduction level, leaving the net gain after error propagation unverified.

    Authors: We agree that explicit prediction-error metrics would make the contribution of the CRNN clearer. While the reported noise-reduction gains already reflect the end-to-end effect of any prediction inaccuracies, we will add DoA RMSE curves over the prediction horizon together with an ablation that isolates the impact of prediction error on final attenuation performance. These additions will be placed in the revised Section 4. revision: yes

  2. Referee: §4.2 (CRNN Training and Testing): No description is given of whether the movement trajectories, speeds, or noise statistics in the test set are disjoint from the training set. Without explicit out-of-distribution or cross-trajectory validation, the reported superiority cannot be taken as evidence of reliable prediction on unseen moving-source conditions, which directly undermines the generalization asserted in the abstract.

    Authors: We thank the referee for highlighting this omission. The training and test trajectories were generated with independent random seeds for path, speed, and noise statistics, but the separation was not stated. In the revision we will (i) explicitly describe the disjoint generation procedure in Section 4.2 and (ii) add results on a held-out set of completely unseen trajectories and speeds to demonstrate generalization. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical simulation-based validation

full rationale

The paper's core contribution is a CRNN trained to predict future DoA and select pre-trained control filters for moving noise sources in PD-SFANC. This is not self-definitional: the network learns temporal dynamics from data and is evaluated via numerical simulations against baselines like D-SFANC. No equations or claims reduce to fitted inputs by construction, no load-bearing self-citations appear in the provided text, and the superiority claim rests on reported simulation outcomes rather than renaming or ansatz smuggling. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on the unverified ability of a CRNN to capture and extrapolate temporal dynamics of moving noise; this rests on domain assumptions about noise movement patterns and the sufficiency of simulation scenarios rather than first-principles derivations.

free parameters (1)
  • CRNN weights and hyperparameters
    Neural network parameters are fitted during training on movement data; their specific values and training procedure are not reported in the abstract.
axioms (1)
  • domain assumption Moving noise sources exhibit learnable temporal dynamics that a CRNN can predict ahead of time
    Invoked when proposing the predictive filter selection step.

pith-pipeline@v0.9.0 · 5458 in / 1337 out tokens · 32400 ms · 2026-05-08T07:06:47.983662+00:00 · methodology

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

38 extracted references · 38 canonical work pages · 2 internal anchors

  1. [1]

    Compared with passive methods that rely on bulky barriers, ANC provides a more compact and effective solution

    Introduction Active noise control (ANC) is an advanced technique that can effectively attenuate low-frequency noise through the principle of sound destructive interference, in which a secondary source generates anti-noise with equal amplitude and opposite phase to the unwanted noise [1, 2]. Compared with passive methods that rely on bulky barriers, ANC pr...

    work page

  2. [2]

    Predictive Directional Selective Fixed-Filter Active Noise Control for Moving Sources via a Convolutional Recurrent Neural Network

    Predictive Directional SFANC We introduce the PD-SFANC method to address the core chal- lenge of delayed response for moving source noise control. As shown in Fig. 2, a CRNN running on the co-processor performs DoA prediction and selects the most suitable control filter for theupcomingframe. In parallel, real-time noise control is exe- cuted at the sampli...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  3. [3]

    Simulation setup The simulation parameters are summarized in Table 2

    Numerical Simulations 3.1. Simulation setup The simulation parameters are summarized in Table 2. Specif- ically, the reference microphone array is a tetrahedral arrange- ment of four cardioid microphones (0.025 mdiameter) follow- ing theSennheiser AMBEO VR Micgeometry, which compactly captures the spatial information [30]. Owing to the small array apertur...

    work page

  4. [4]

    By leveraging temporal directional de- pendencies across multiple consecutive frames, the CRNN pre- dicts the source DoA and proactively selects the most suitable control filter

    Conclusion This paper presents a novel PD-SFANC method for handling moving noise sources. By leveraging temporal directional de- pendencies across multiple consecutive frames, the CRNN pre- dicts the source DoA and proactively selects the most suitable control filter. Numerical simulations demonstrate that the pro- posed CRNN achieves accurate DoA predict...

    work page

  5. [5]

    Active noise control,

    S. J. Elliott and P. A. Nelson, “Active noise control,”IEEE signal processing magazine, vol. 10, no. 4, pp. 12–35, 1993

    work page 1993

  6. [6]

    Active noise control: a tutorial review,

    S. M. Kuo and D. R. Morgan, “Active noise control: a tutorial review,”Proceedings of the IEEE, vol. 87, no. 6, pp. 943–973, 1999

    work page 1999

  7. [7]

    A deep learning approach to active noise control

    H. Zhang and D. Wang, “A deep learning approach to active noise control.” inInterspeech, vol. 16, no. 3, 2020, pp. 1141–1145

    work page 2020

  8. [8]

    The application of a multi- reference control strategy to noise cancelling headphones,

    J. Cheer, V . Patel, and S. Fontana, “The application of a multi- reference control strategy to noise cancelling headphones,”The Journal of the Acoustical Society of America, vol. 145, no. 5, pp. 3095–3103, 2019

    work page 2019

  9. [9]

    Recent advances in active noise control inside automobile cabins: Toward quieter cars,

    P. N. Samarasinghe, W. Zhang, and T. D. Abhayapala, “Recent advances in active noise control inside automobile cabins: Toward quieter cars,”IEEE Signal Processing Magazine, vol. 33, no. 6, pp. 61–73, 2016

    work page 2016

  10. [10]

    Spatially selective active noise con- trol systems,

    T. Xiao, B. Xu, and C. Zhao, “Spatially selective active noise con- trol systems,”The Journal of the Acoustical Society of America, vol. 153, no. 5, pp. 2733–2733, 2023

    work page 2023

  11. [11]

    Active noise control: Open problems and challenges,

    S. M. Kuo, K. Kuo, and W. S. Gan, “Active noise control: Open problems and challenges,” inThe 2010 International conference on green circuits and systems. IEEE, 2010, pp. 164–169

    work page 2010

  12. [12]

    Behavior of adaptive algo- rithms in active noise control systems with moving noise sources,

    A. Omoto, D. Morie, and K. Fujiwara, “Behavior of adaptive algo- rithms in active noise control systems with moving noise sources,” Acoustical Science and Technology, vol. 23, no. 2, pp. 84–89, 2002

    work page 2002

  13. [13]

    Least-mean-square weighted parallel iir filters in active-noise-control headphones,

    M. Guldenschuh, “Least-mean-square weighted parallel iir filters in active-noise-control headphones,” in2014 22nd European Sig- nal Processing Conference (EUSIPCO). IEEE, 2014, pp. 1367– 1371

    work page 2014

  14. [14]

    Time- division multiple reference approach for multiple-channel active noise control system,

    C.-Y . Ho, L. Liu, K.-K. Shyu, S. M. Kuo, and C.-Y . Chang, “Time- division multiple reference approach for multiple-channel active noise control system,”Journal of Sound and Vibration, vol. 495, p. 115922, 2021

    work page 2021

  15. [15]

    Active control of sound transmission through a floor-level slit,

    Z. Yang, S. Wang, J. Tao, and X. Qiu, “Active control of sound transmission through a floor-level slit,”The Journal of the Acous- tical Society of America, vol. 154, no. 5, pp. 2746–2756, 2023

    work page 2023

  16. [16]

    An augmented diffusion algorithm with bidirectional communication for a distributed active noise control system,

    T. Li, L. Rao, S. Zhao, H. Duan, J. Lu, and I. S. Burnett, “An augmented diffusion algorithm with bidirectional communication for a distributed active noise control system,”The Journal of the Acoustical Society of America, vol. 154, no. 6, pp. 3568–3579, 2023

    work page 2023

  17. [17]

    A study on multi-channel active sound profiling algorithm for hy- brid control of broadband and narrowband noise inside vehicles,

    X. Liu, X. Zheng, Z. Jia, R. Li, B. Wan, C. Liu, and Y . Qiu, “A study on multi-channel active sound profiling algorithm for hy- brid control of broadband and narrowband noise inside vehicles,” Measurement, vol. 237, p. 115200, 2024

    work page 2024

  18. [18]

    Deep learning-based active trim panels for enhanced air- craft interior noise control,

    B. Wang, M. Misol, Z. Luo, J. Ji, X. Shen, D. Shi, and W.-S. Gan, “Deep learning-based active trim panels for enhanced air- craft interior noise control,” inProceedings of the 31st Interna- tional Congress on Sound and Vibration. The Korean Society for Noise and Vibration Engineering, 2025

    work page 2025

  19. [19]

    Self-boosted weight-constrained fxlms: A robustness distributed active noise control algorithm without internode communication,

    J. Ji, D. Shi, Z. Luo, B. Wang, and W.-S. Gan, “Self-boosted weight-constrained fxlms: A robustness distributed active noise control algorithm without internode communication,”IEEE Sig- nal Processing Letters, 2025

    work page 2025

  20. [20]

    Feedforward selective fixed-filter active noise control: Algorithm and implementation,

    D. Shi, W.-S. Gan, B. Lam, and S. Wen, “Feedforward selective fixed-filter active noise control: Algorithm and implementation,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 28, pp. 1479–1492, 2020

    work page 2020

  21. [21]

    Real-time imple- mentation and explainable ai analysis of delayless cnn-based se- lective fixed-filter active noise control,

    Z. Luo, D. Shi, J. Ji, X. Shen, and W.-S. Gan, “Real-time imple- mentation and explainable ai analysis of delayless cnn-based se- lective fixed-filter active noise control,”Mechanical Systems and Signal Processing, vol. 214, p. 111364, 2024

    work page 2024

  22. [22]

    Trans- ferable selective virtual sensing active noise control technique based on metric learning,

    B. Wang, D. Shi, Z. Luo, X. Shen, J. Ji, and W.-S. Gan, “Trans- ferable selective virtual sensing active noise control technique based on metric learning,” inICASSP 2025-2025 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    work page 2025

  23. [23]

    Direction-of-arrival dependency of active noise cancella- tion headphones,

    S. Liebich, J.-G. Richter, J. Fabry, C. Durand, J. Fels, and P. Jax, “Direction-of-arrival dependency of active noise cancella- tion headphones,” inNoise Control and Acoustics Division Con- ference, vol. 51425. American Society of Mechanical Engineers, 2018, p. V001T08A003

    work page 2018

  24. [24]

    A time-domain multi-channel directional active noise control sys- tem,

    H. Zhang, J. Zhang, F. Ma, P. N. Samarasinghe, and H. Sun, “A time-domain multi-channel directional active noise control sys- tem,” in2023 31st European Signal Processing Conference (EU- SIPCO). IEEE, 2023, pp. 376–380

    work page 2023

  25. [25]

    Cognitive virtual sensing technique for feedforward active noise control,

    R. Xie, A. Tu, C. Shi, S. Elliott, H. Li, and L. Zhang, “Cognitive virtual sensing technique for feedforward active noise control,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 981– 985

    work page 2024

  26. [26]

    Active noise control systems with sound source localization robust to noise source movement,

    S. Toyooka and Y . Kajikawa, “Active noise control systems with sound source localization robust to noise source movement,”IE- ICE Transactions on Fundamentals of Electronics, Communica- tions and Computer Sciences, vol. 108, no. 2, pp. 160–164, 2025

    work page 2025

  27. [27]

    Spatial-frequency- based selective fixed-filter algorithm for multichannel active noise control,

    X. Su, D. Shi, Z. Zhu, W.-S. Gan, and L. Ye, “Spatial-frequency- based selective fixed-filter algorithm for multichannel active noise control,”IEEE Signal Processing Letters, 2024

    work page 2024

  28. [28]

    Frequency-direction aware multichannel selective fixed-filter active noise control based on multi-task learning,

    Z. Luo, D. Shi, X. Su, and W.-S. Gan, “Frequency-direction aware multichannel selective fixed-filter active noise control based on multi-task learning,”IEEE Transactions on Audio, Speech and Language Processing, 2025

    work page 2025

  29. [29]

    Di- rectional selective fixed-filter active noise control based on a con- volutional neural network in reverberant environments,

    B. Wang, Z. Luo, H. Li, D. Shi, J. Ji, Z. Yang, and W.-S. Gan, “Di- rectional selective fixed-filter active noise control based on a con- volutional neural network in reverberant environments,” in2025 Asia Pacific Signal and Information Processing Association An- nual Summit and Conference (APSIPA ASC). IEEE, 2025, pp. 364–369

    work page 2025

  30. [30]

    Co-forecasting of time-varying spatial-frequency map for selective fixed-filter mul- tichannel anc based on dynamic factor graph,

    X. Su, D. Shi, B. Wu, L. Ye, and W.-S. Gan, “Co-forecasting of time-varying spatial-frequency map for selective fixed-filter mul- tichannel anc based on dynamic factor graph,”IEEE Transactions on Audio, Speech and Language Processing, 2025

    work page 2025

  31. [31]

    Localization, detection and tracking of multiple moving sound sources with a convolu- tional recurrent neural network,

    S. Adavanne, A. Politis, and T. Virtanen, “Localization, detection and tracking of multiple moving sound sources with a convolu- tional recurrent neural network,” inWorkshop on Detection and Classification of Acoustic Scenes and Events, 2019, pp. 20–24

    work page 2019

  32. [32]

    Doa estimation of room reflec- tions using nn-based music algorithm,

    H. Li, W. Zhang, and L. Zhang, “Doa estimation of room reflec- tions using nn-based music algorithm,” in2023 Asia Pacific Sig- nal and Information Processing Association Annual Summit and Conference (APSIPA ASC). IEEE, 2023, pp. 1960–1965

    work page 2023

  33. [33]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic opti- mization,”arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review arXiv 2014

  34. [34]

    Sound source distance estimation in diverse and dynamic acoustic con- ditions,

    S. S. Kushwaha, I. R. Roman, M. Fuentes, and J. P. Bello, “Sound source distance estimation in diverse and dynamic acoustic con- ditions,” in2023 IEEE Workshop on Applications of Signal Pro- cessing to Audio and Acoustics (WASPAA). IEEE, 2023, pp. 1–5

    work page 2023

  35. [35]

    A dataset and taxonomy for urban sound research,

    J. Salamon, C. Jacoby, and J. P. Bello, “A dataset and taxonomy for urban sound research,” inProceedings of the 22nd ACM inter- national conference on Multimedia, 2014, pp. 1041–1044

    work page 2014

  36. [36]

    gpurir: A python library for room impulse response simulation with gpu acceler- ation,

    D. Diaz-Guerra, A. Miguel, and J. R. Beltran, “gpurir: A python library for room impulse response simulation with gpu acceler- ation,”Multimedia Tools and Applications, vol. 80, no. 4, pp. 5653–5671, 2021

    work page 2021

  37. [37]

    Exploit- ing temporal context in cnn based multisource doa estimation,

    A. Bohlender, A. Spriet, W. Tirry, and N. Madhu, “Exploit- ing temporal context in cnn based multisource doa estimation,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 29, pp. 1594–1608, 2021

    work page 2021

  38. [38]

    Realman: A real-recorded and an- notated microphone array dataset for dynamic speech enhance- ment and localization,

    B. Yang, C. Quan, Y . Wang, P. Wang, Y . Yang, Y . Fang, N. Shao, H. Bu, X. Xu, and X. Li, “Realman: A real-recorded and an- notated microphone array dataset for dynamic speech enhance- ment and localization,”Advances in Neural Information Process- ing Systems, vol. 37, pp. 105 997–106 019, 2024

    work page 2024