Pith. sign in

REVIEW 3 major objections 4 minor 35 references

This paper claims that adding an adaptive momentum term, gated by cosine similarity between the current mixed gradient and the previous update, lets distributed multichannel active noise control converge faster when communication delays for

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

2026-08-01 18:48 UTC pith:3CKOYACS

load-bearing objection A plausible incremental ANC algorithm undermined by a sign inconsistency: Eq. (14) applies maximum momentum exactly when the text says it should attenuate. the 3 major comments →

arxiv 2607.17165 v1 pith:3CKOYACS submitted 2026-07-19 eess.AS eess.SP

Adaptive Momentum Enhanced Distributed Multichannel Active Noise Control for Faster Convergence under Communication Delays

Junwei Ji , Woon-Seng Gan , Boxiang Wang , Ziyi Yang , Haowen Li This is my paper
classification eess.AS eess.SP
keywords distributed multichannel active noise controlFxLMSadaptive momentumcosine similaritycommunication delaysauto-shrink step sizeconvergence acceleration
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.

Distributed active noise control systems exchange gradient information between nodes, but communication delays can make them unstable. A prior fix, the auto-shrink step size (ASSS-MGDFxLMS), guarantees stability by shrinking the step size as delays grow, at the cost of slow convergence. This paper introduces an adaptive momentum term that re-accelerates convergence: each node measures the cosine similarity between its instantaneous mixed gradient and its previous weight update, and scales the momentum factor by a power of that similarity. When the two directions align, the momentum factor rises toward its maximum to speed learning; when they do not, it weakens to preserve stability. Simulation results with sudden and gradually varying delays show the new algorithm converging faster than ASSS-MGDFxLMS and avoiding the divergence that fixed-momentum variants can exhibit.

Core claim

The central claim is that AMAS-MGDFxLMS—formed by adding an adaptive momentum term to the ASSS-MGDFxLMS update—provides faster convergence than its base algorithm while maintaining stable noise reduction under communication delays. The momentum factor at node k is βk(n) = min(β0 |ρk(n)|^p, β0), where ρk(n) is the cosine similarity between the mixed gradient gk(n) and the previous weight increment vk(n) = wk(n) − wk(n−1), with 0 < p < 1. Because |ρk| is typically small for high-dimensional vectors, the power-law scaling lifts small similarities, so the momentum is active when directions agree and attenuated when they disagree. The paper demonstrates in two simulated scenarios—sudden jumps in

What carries the argument

The key mechanism is the adaptive momentum factor βk(n) = min(β0 |ρk(n)|^p, β0), where ρk(n) is the cosine similarity between the mixed gradient and the momentum (previous weight update). This factor dynamically gates how much of the previous update is injected into the current weight update. The cosine similarity acts as a directional agreement probe: strong alignment (|ρ| near 1) enables maximal acceleration, while weak or anti-aligned directions reduce the momentum contribution, preventing the instability that a fixed momentum factor would cause under delayed gradients.

Load-bearing premise

The paper assumes that the adaptive momentum factor, especially when the gradient and previous update point in opposite directions (|ρ| = 1, giving the maximum momentum), preserves the stability guarantee that the auto-shrink step size provides, but this is not proven analytically and is only demonstrated in one simulation setup with hand-chosen hyperparameters (p = 0.25, β0 not reported).

What would settle it

A stability analysis or a simulation with a different delay pattern—for example, delays alternating rapidly between 0 and 1 second, or a larger network with 10 or more nodes and measured acoustic paths—that produces divergence of AMAS-MGDFxLMS (ANSE growing unboundedly) where ASSS-MGDFxLMS remains stable, or a measurement showing no convergence-speed improvement over ASSS-MGDFxLMS in such a setting, would refute the paper's central claim.

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

If this is right

  • When communication delays change suddenly, AMAS-MGDFxLMS converges faster than ASSS-MGDFxLMS while keeping the error bounded, whereas the unmodified MGDFxLMS diverges and fixed-momentum variants can over-accelerate into divergence.
  • Under a gradually fluctuating network delay, the adaptive momentum maintains stable and faster noise reduction, eliminating the need to manually tune a momentum factor for a given delay condition.
  • The extra computation is modest—about 4Lw+2 multiplications and 6Lw−3 additions per node per iteration—making the algorithm practical for real-time distributed ANC.
  • The design is a direct plug-in modification to ASSS-MGDFxLMS, so systems already using that algorithm can upgrade without changing the distributed gradient-exchange protocol.

Where Pith is reading between the lines

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

  • The same cosine-similarity gating could be applied to other distributed adaptive filters (e.g., diffusion LMS or augmented diffusion FxLMS) that suffer from delayed information exchange, since the gating only uses local gradient and momentum vectors.
  • The exponent p appears to control how aggressively small directional agreements are amplified; we infer that p likely needs to scale with the ratio of the delay spread to the filter length, though the paper does not investigate this relationship.
  • The anti-aligned case (ρ ≈ −1) deserves scrutiny: because |ρ|^p = 1, the momentum is applied at full strength in the opposite direction of the gradient—a configuration that could destabilize the update if the delayed gradient is stale; the paper does not provide a theoretical bound for this case.
  • One testable extension would be to replace βk(n) with a version that only engages momentum when ρk(n) > 0, ignoring negative similarities, and compare convergence and stability; this would isolate whether the anti-aligned full-momentum case is benign or harmful.

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

3 major / 4 minor

Summary. The paper proposes AMAS-MGDFxLMS, an extension of the ASSS-MGDFxLMS distributed multichannel active noise control algorithm. The method adds an adaptive momentum term whose strength is set from the cosine similarity between the current mixed gradient and the previous update. The stated goal is to accelerate convergence under communication delays while preserving stability. The paper reports simulation results for a six-node ANC window setup showing that AMAS-MGDFxLMS converges faster than ASSS-MGDFxLMS and avoids the divergence seen with a fixed momentum factor.

Significance. If the claims hold, the algorithm would address a real practical limitation of delay-robust DMCANC, where the auto-shrink step size trades convergence speed for stability. The incremental contribution is modest but relevant to the ANC community. The paper gives an explicit cost estimate for the added computations, which is useful. However, the central adaptive mechanism is internally inconsistent with its stated motivation, and the empirical evidence is thin: no theoretical stability analysis, unstated hyperparameter values, and single-trace simulations. The significance is therefore conditional on correcting these issues.

major comments (3)
  1. [III, Eq. (14)] The adaptive momentum rule contradicts the stated purpose. The text says that when gradient and momentum directions are inconsistent, the momentum contribution is attenuated. But beta_k(n) = min(beta0 |rho_k(n)|^p, beta0) uses the absolute value: as rho approaches -1, |rho|^p approaches 1 and the momentum factor is at its maximum. Anti-aligned updates, which are the main risk under communication delays, therefore receive full momentum. The ASSS step-size scaling in Eq. (9) is not shown to compensate, and no stability bound is given. The central claim of stable faster convergence is not established for this case. Please either change the rule to a signed measure that actually attenuates anti-aligned momentum, or provide a rigorous argument and targeted simulations showing that applying maximum momentum for anti-aligned updates is stable.
  2. [IV A/B and Table I] The two new hyperparameters are not fully reported. p is fixed to 0.25 for all simulations, and beta0 is said to be 'estimated under ideal network conditions [35]' but the value and estimation procedure are not given. The fixed-momentum comparator FMAS-MGDFxLMS is also not specified (which beta0 value was chosen and why). The reported acceleration may therefore depend on hand-tuned or unreported parameter choices. Please report the actual beta0 value, give the estimation procedure, and include a sensitivity study over p and beta0 to show the improvement is not a tuning artifact.
  3. [IV, Figs. 3 and 4] The simulation evidence consists of single traces from one configuration, with no statistical replicates, error bars, or quantitative convergence metrics. The visual claim of faster convergence is not robustly supported. Please run multiple independent trials (different noise realizations, initializations, and delay profiles) and report mean/median curves with spread, or provide quantitative values such as time to reach a target ANSE and final steady-state ANSE for each algorithm.
minor comments (4)
  1. [II A, around Fig. 1] The text says 'shown in 1' but should say 'shown in Fig. 1'.
  2. [IV] The ANSE metric is mentioned but not defined. Please provide the formula or a citation with the exact definition used.
  3. [III] The claim that cosine similarity values are 'typically small' because the vectors are high-dimensional needs justification; v_k(n) is a difference of consecutive filters and may have structured small norm. A remark with a bound or reference would help.
  4. [III, after Eq. (15)] The computational cost statement ('approximately 4L_w+2 multiplications and 6L_w-3 additions per node') is not derived. A short breakdown of the operations would improve verifiability.

Circularity Check

0 steps flagged

No significant circularity; the absolute-value momentum formula is a correctness concern, not a circularity.

full rationale

The paper's core algorithm is a direct modification of the previously proposed ASSS-MGDFxLMS update: Eq. (15) adds an adaptive momentum term v_k(n) scaled by beta_k(n) from Eq. (14). The claimed faster convergence is supported by simulations comparing AMAS-MGDFxLMS against MGDFxLMS, ASSS-MGDFxLMS, and FMAS-MGDFxLMS, rather than being derived by construction from the adaptive-momentum definition. The self-citations (e.g., [31], [35]) supply the baseline algorithm and momentum-factor estimation, but the reported acceleration is not forced by those citations; it is an empirical observation over the tested scenarios. No parameter is fitted to a subset of data and then presented as a prediction of a closely related quantity; beta0 and p are chosen hyperparameters, and their values (or the omission of beta0) create a reproducibility/rigor gap, not circularity. One genuine internal inconsistency warrants note but is not circular: Eq. (14) uses |rho_k(n)|^p, so when rho is near -1 (anti-aligned gradient and momentum), the momentum factor is at its maximum, contradicting the text's statement that 'otherwise, the momentum contribution is attenuated to enhance stability.' This threatens the stability claim under delayed communication, but the update rule is not equivalent to its inputs by construction. Overall, the derivation chain is not circular.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim rests on two hand-chosen hyperparameters (beta0, p) and on the inherited assumptions of the ASSS-MGDFxLMS framework. No new physical entities are introduced. The adaptive momentum mechanism is heuristic and lacks theoretical grounding, increasing the dependence on the specific simulation setup.

free parameters (3)
  • beta0 (maximum momentum factor) = not reported
    Upper bound in beta_k(n)=min(beta0|rho|^p, beta0); said to be estimated under ideal network conditions [35], but the simulation value is never stated, blocking reproduction.
  • p (power-law exponent) = 0.25
    Controls scaling of cosine similarity; 'selected according to the order of magnitude of rho' but fixed to 0.25 for all simulations without sensitivity analysis.
  • mu0 (initial step size) = 1e-7
    Standard LMS step size used in all simulations; chosen by hand and inherited from the ASSS-MGDFxLMS baseline.
axioms (4)
  • domain assumption The estimated self-secondary path s_hat_kk(n) and compensation filters c_mk(n) are accurate enough that the mixed-gradient update approximates centralized MEFxLMS performance.
    Invoked in Section II-A; the entire distributed algorithm assumes these path estimates are available and reliable.
  • ad hoc to paper The stability and performance guarantees of ASSS-MGDFxLMS under communication delays carry over when the adaptive momentum term is added.
    The paper provides no stability analysis for the momentum-modified update (15); it relies on the prior ASSS result [31] plus the simulation to claim stability.
  • ad hoc to paper Cosine similarity values between the high-dimensional gradient and momentum vectors are small, so raising |rho| to a power p<1 is a valid way to amplify the signal.
    Section III states this to justify the power-law scaling; no formal justification or empirical distribution is given.
  • domain assumption Communication delays are integer sample counts and their maximum per node is known to the node.
    Used in equations (6)-(8) to define the auto-shrink step size.

pith-pipeline@v1.3.0-alltime-deepseek · 7566 in / 10890 out tokens · 103664 ms · 2026-08-01T18:48:27.790808+00:00 · methodology

0 comments
read the original abstract

Distributed multichannel active noise control (DMCANC) reduces the computational burden of centralized ANC systems by distributing processing tasks across multiple nodes, while requiring information exchange to achieve satisfactory global noise reduction. To improve robustness under communication delays, the auto-shrink step size mixed-gradients filtered reference LMS (ASSS-MGDFxLMS) algorithm has been proposed. However, the reduced step size inevitably slows convergence. In this work, an adaptive momentum term is introduced to accelerate convergence, where cosine similarity is used to evaluate the alignment between the instantaneous gradient and the momentum component and dynamically adjust the momentum parameter. This design accelerates convergence when the directions are consistent while preserving stability under delayed communication. Simulation results demonstrate that the proposed adaptive momentum ASSS-MGDFxLMS (AMAS-MGDFxLMS) algorithm achieves faster convergence than ASSS-MGDFxLMS while maintaining stable and effective noise reduction performance.

Figures

Figures reproduced from arXiv: 2607.17165 by Boxiang Wang, Haowen Li, Junwei Ji, Woon-Seng Gan, Ziyi Yang.

Figure 1
Figure 1. Figure 1: A general DMCANC network, where each ANC node consists of a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Block diagram of the ASSS-MGDFxLMS algorithm for the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Noise reduction performance using different algorithms when com [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] 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

35 extracted references · 2 canonical work pages

  1. [1]

    S. J. Elliott,Signal processing for active control. San Diego, Calif.: Academic, 2001, includes bibliographical references and index

    2001

  2. [2]

    An analysis of multiple correlation cancellation loops with a filter in the auxiliary path,

    D. Morgan, “An analysis of multiple correlation cancellation loops with a filter in the auxiliary path,”IEEE Transactions on acoustics, speech, and signal processing, vol. 28, no. 4, pp. 454–467, 1980. [Online]. Available: https://dx.doi.org/10.1109/tassp.1980.1163430

    arXiv 1980

  3. [3]

    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

    1999

  4. [4]

    Ten questions concerning active noise control in the built environment,

    B. Lam, W.-S. Gan, D. Shi, M. Nishimura, and S. Elliott, “Ten questions concerning active noise control in the built environment,”Building and Environment, vol. 200, p. 107928, 2021

    2021

  5. [5]

    A survey on adaptive active noise control algorithms overcoming the output saturation effect,

    Y . Guo, D. Shi, X. Shen, J. Ji, and W.-S. Gan, “A survey on adaptive active noise control algorithms overcoming the output saturation effect,” Signal Processing, vol. 222, p. 109525, 2024

    2024

  6. [6]

    Co-forecasting of time- varying spatial-frequency map for selective fixed-filter multichannel 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 multichannel anc based on dynamic factor graph,”IEEE Transactions on Audio, Speech and Language Processing, 2025

    2025

  7. [7]

    A spherical-harmonic domain selective spatial active noise control system based on sound field reproduction,

    H. Zhang, H. J. Sun, J. A. Zhang, P. Samarasinghe, and Y . A. Zhang, “A spherical-harmonic domain selective spatial active noise control system based on sound field reproduction,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  8. [8]

    A novel hybrid active noise control with auxiliary filter-based virtual sensing for mixed unwanted noise reduction,

    I. Tsuji, S. Toyooka, and Y . Kajikawa, “A novel hybrid active noise control with auxiliary filter-based virtual sensing for mixed unwanted noise reduction,”Applied Acoustics, vol. 239, p. 110808, 2025

    2025

  9. [9]

    Transferable 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, “Transferable selective virtual sensing active noise control technique based on metric learning,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  10. [10]

    Deep anc: A deep learning approach to active noise control,

    H. Zhang and D. Wang, “Deep anc: A deep learning approach to active noise control,”Neural Networks, vol. 141, pp. 1–10, 2021

    2021

  11. [11]

    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,” inICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 981–985

    2024

  12. [12]

    Dynamic neural network switching for active control of nonlinear systems,

    X. Pike and J. Cheer, “Dynamic neural network switching for active control of nonlinear systems,”The Journal of the Acoustical Society of America, vol. 158, no. 1, pp. 154–163, 2025

    2025

  13. [13]

    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

    2025

  14. [14]

    A multiple error lms algorithm and its application to the active control of sound and vibration,

    S. Elliott, I. Stothers, and P. Nelson, “A multiple error lms algorithm and its application to the active control of sound and vibration,”IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 35, no. 10, pp. 1423–1434, 1987. [Online]. Available: https://dx.doi.org/10.1109/tassp.1987.1165044

    arXiv 1987

  15. [15]

    Fast implementations of the filtered-x lms and lms algorithms for multichannel active noise control,

    S. C. Douglas, “Fast implementations of the filtered-x lms and lms algorithms for multichannel active noise control,”IEEE Transactions on speech and audio processing, vol. 7, no. 4, pp. 454–465, 1999

    1999

  16. [16]

    Computation- efficient solution for fully-connected active noise control window: Analysis and implementation of multichannel adjoint least mean square algorithm,

    D. Shi, B. Lam, J. Ji, X. Shen, C. K. Lai, and W.-S. Gan, “Computation- efficient solution for fully-connected active noise control window: Analysis and implementation of multichannel adjoint least mean square algorithm,”Mechanical Systems and Signal Processing, vol. 199, p. 110444, 2023

    2023

  17. [17]

    A partial-update minimax algorithm for prac- tical implementation of multi-channel feedforward active noise control,

    C. Shi and Y . Kajikawa, “A partial-update minimax algorithm for prac- tical implementation of multi-channel feedforward active noise control,” in2018 16th International Workshop on Acoustic Signal Enhancement (IWAENC). IEEE, 2018, pp. 1–15

    2018

  18. [18]

    Mixed-error approach for multi-channel active noise control of open windows,

    T. Murao, C. Shi, W.-S. Gan, and M. Nishimura, “Mixed-error approach for multi-channel active noise control of open windows,”Applied Acoustics, vol. 127, pp. 305–315, 2017

    2017

  19. [19]

    A computational-efficient adaptive algorithm based on frequency point selection in multichannel active noise control systems,

    S. Lian, T. Li, S. Sun, S. Wang, and J. Lu, “A computational-efficient adaptive algorithm based on frequency point selection in multichannel active noise control systems,”Mechanical Systems and Signal Process- ing, vol. 238, p. 113176, 2025

    2025

  20. [20]

    On eigenvalue shaping for two- channel decentralized active noise control systems,

    S. Pradhan, G. Zhang, S. Zhao, K. Niwa, and W. Bastiaan Kleijn, “On eigenvalue shaping for two- channel decentralized active noise control systems,”Applied Acoustics, vol. 205, p. 109260, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0003682X23000580

    2023

  21. [21]

    Decentralized two- channel active noise control for single frequency by shaping matrix eigenvalues,

    G. Zhang, J. Tao, X. Qiu, and I. Burnett, “Decentralized two- channel active noise control for single frequency by shaping matrix eigenvalues,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 27, no. 1, pp. 44–52, 2019. [Online]. Available: https://dx.doi.org/10.1109/taslp.2018.2869686

    arXiv 2019

  22. [22]

    Interaction between multiple feedforward active control systems,

    S. J. Elliott and C. C. Boucher, “Interaction between multiple feedforward active control systems,”IEEE Transactions on speech and audio processing, vol. 2, no. 4, pp. 521–530, 1994. [Online]. Available: https://dx.doi.org/10.1109/89.326611

    work page doi:10.1109/89.326611 1994

  23. [23]

    Active noise control over adaptive distributed networks,

    M. Ferrer, M. de Diego, G. Pi ˜nero, and A. Gonzalez, “Active noise control over adaptive distributed networks,”Signal Processing, vol. 107, pp. 82–95, 2015

    2015

  24. [24]

    Diffusion algorithm for active noise control in distributed networks,

    C. Anto ˜nanzas, M. Ferrer, A. Gonzalez, M. de Diego, and G. Pi ˜nero, “Diffusion algorithm for active noise control in distributed networks,” in22nd Int. Congr. Sound Vib. 2015 ICSV 2015, 2015, Conference Proceedings, pp. 1–8

    2015

  25. [25]

    Performance analysis of diffusion filtered-x algorithms in multitask anc systems,

    Y . J. Chu, S. C. Chan, Y . Zhao, and M. Wu, “Performance analysis of diffusion filtered-x algorithms in multitask anc systems,” in27th Eur. Signal Process. Conf (EUSIPCO). IEEE, 2019, Conference Proceedings, pp. 1–5. [Online]. Available: https://dx.doi.org/10.23919/eusipco.2019.8903076

    arXiv 2019

  26. [26]

    Performance analysis of a diffusion control method for anc systems and the network design,

    Y . J. Chu, C. M. Mak, Y . Zhao, S. C. Chan, and M. Wu, “Performance analysis of a diffusion control method for anc systems and the network design,”Journal of Sound and Vibration, vol. 475, p. 115273, 2020

    2020

  27. [27]

    A diffusion fxlms algorithm for multi-channel active noise control and variable spatial smoothing,

    Y . J. Chu, S. C. Chan, C. M. Mak, and M. Wu, “A diffusion fxlms algorithm for multi-channel active noise control and variable spatial smoothing,” in2021 Proc. - ICASSP IEEE Int. Conf. Acoust. Speech Signal Process.IEEE, 2021, Conference Proceedings, pp. 4695–4699. [Online]. Available: https://dx.doi.org/10.1109/icassp39728.2021.9414609

    arXiv 2021

  28. [28]

    Distributed active noise control based on an augmented diffusion f¡i¿x¡/i¿lms algorithm,

    T. Li, S. Lian, S. Zhao, J. Lu, and I. S. Burnett, “Distributed active noise control based on an augmented diffusion f¡i¿x¡/i¿lms algorithm,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, pp. 1–15, 2023. [Online]. Available: https://dx.doi.org/10.1109/taslp.2023.3261742

    arXiv 2023

  29. [29]

    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. [Online]. Available: https://dx.doi.org/10.1121/10.0022573

    work page doi:10.1121/10.0022573 2023

  30. [30]

    Experimental study of a distributed active noise control system with multi-device nodes based on augmented diffusion strategy,

    T. Li, S. Zhao, L. Rao, H. Zou, K. Chen, J. Lu, and I. S. Burnett, “Experimental study of a distributed active noise control system with multi-device nodes based on augmented diffusion strategy,”The Journal of the Acoustical Society of America, vol. 156, no. 5, pp. 3246–3259, 2024

    2024

  31. [31]

    Mixed-gradients distributed filtered reference least mean square algorithm – a robust distributed multichannel active noise control algorithm,

    J. Ji, D. Shi, and W.-S. Gan, “Mixed-gradients distributed filtered reference least mean square algorithm – a robust distributed multichannel active noise control algorithm,”IEEE Transactions on Audio, Speech and Language Processing, vol. 33, pp. 1563–1575, 2025. [Online]. Available: https://dx.doi.org/10.1109/taslpro.2025.3552932

    arXiv 2025

  32. [32]

    Adaptive momentum coefficient for neural network optimization,

    Z. Rashidi, K. Ahmadi KA, A. An, and X. Wang, “Adaptive momentum coefficient for neural network optimization,” inJoint European Con- ference on Machine Learning and Knowledge Discovery in Databases. Springer, 2020, pp. 35–51

    2020

  33. [33]

    A practical distributed active noise control algorithm overcoming communication restrictions,

    J. Ji, D. Shi, Z. Luo, X. Shen, and W.-S. Gan, “A practical distributed active noise control algorithm overcoming communication restrictions,” in2023 Proc. - ICASSP IEEE Int. Conf. Acoust. Speech Signal Process. IEEE, 2023, Conference Proceedings, pp. 1–5

    2023

  34. [34]

    Grams: Gradient descent with adaptive momentum scaling,

    Y . Cao, X. Li, and Z. Song, “Grams: Gradient descent with adaptive momentum scaling,” inFirst Workshop on Scalable Optimization for Efficient and Adaptive Foundation Models, 2024

    2024

  35. [35]

    Multi-channel adjoint least mean square algorithm with momentum factor applied on active noise control,

    X. Shen, Y . Guo, J. Ji, D. Shi, and W.-S. Gan, “Multi-channel adjoint least mean square algorithm with momentum factor applied on active noise control,”Mechanical Systems and Signal Processing, vol. 228, p. 112415, 2025

    2025