pith. machine review for the scientific record. sign in

arxiv: 2605.11875 · v1 · submitted 2026-05-12 · 📡 eess.SP · cs.AI

Recognition: unknown

Modulation Consistency-based Contrastive Learning for Self-Supervised Automatic Modulation Classification

Chenxu Wang, Hanlin Mo, Hantong Xing, Licheng Jiao, Lirong Han, Shuang Wang, Xinyu Hu

Pith reviewed 2026-05-13 05:23 UTC · model grok-4.3

classification 📡 eess.SP cs.AI
keywords automatic modulation classificationself-supervised learningcontrastive learningmodulation consistencyradio signalssignal processingdeep learninglow-label regimes
0
0 comments X

The pith

Mod-CL learns shared modulation types from unlabeled radio signals by contrasting different temporal segments of the same instance.

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

The paper establishes that automatic modulation classification suffers when self-supervised methods use generic pretext tasks that entangle modulation information with channel noise and symbol variations. It identifies that different time segments within one transmitted signal reliably share the modulation scheme as a structural prior that aligns directly with the classification goal. The proposed Mod-CL framework builds positive pairs from these segments, applies data augmentations, and uses a custom contrastive objective to pull representations together only on modulation semantics while pushing apart nuisance factors. This matters because practical radio systems generate vast unlabeled data yet labeling each modulation type is costly, so a method that extracts task-relevant features without labels could cut annotation needs substantially. If the approach holds, models pretrained this way should require fewer labels to reach high accuracy on downstream classification.

Core claim

Intra-instance modulation consistency is a task-aware prior in which different temporal segments of one signal instance preserve the same modulation type while varying in waveform details, channel effects, and noise. Mod-CL exploits this by sampling positive pairs from segmented views of the identical instance, combined with augmentations, inside a contrastive loss that avoids intra-instance supervisory conflicts and forces the encoder to retain only the invariant modulation information.

What carries the argument

The intra-instance modulation consistency prior together with the Mod-CL contrastive objective that constructs positive pairs from temporal segments of the same signal.

If this is right

  • Mod-CL produces representations that outperform existing self-supervised baselines on RadioML datasets for automatic modulation classification.
  • The gains are largest when only a small fraction of labels is available for the final linear probe.
  • The learned features suppress entanglement with symbol, channel, and noise variations compared with task-agnostic SSL methods.
  • A single pretraining stage on unlabeled data yields improved downstream accuracy without changes to the classifier architecture.

Where Pith is reading between the lines

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

  • The same segment-consistency idea could be tested on other time-series classification problems where intra-instance labels remain constant, such as speech or vibration analysis.
  • If the prior generalizes, it would reduce reliance on simulated channel models by allowing direct use of real over-the-air recordings for pretraining.
  • One could measure whether the contrastive objective also improves robustness to unseen modulation variants or new noise distributions not present in RadioML.

Load-bearing premise

Different temporal segments of the same transmitted signal always share the identical modulation type and differ only in unrelated nuisance factors such as noise or channel effects.

What would settle it

Run Mod-CL on a dataset where positive pairs are deliberately drawn from segments carrying different modulations; if linear probing accuracy then falls to or below that of standard augmentation-based contrastive baselines, the utility of the consistency prior is refuted.

Figures

Figures reproduced from arXiv: 2605.11875 by Chenxu Wang, Hanlin Mo, Hantong Xing, Licheng Jiao, Lirong Han, Shuang Wang, Xinyu Hu.

Figure 1
Figure 1. Figure 1: Schematic illustration of intra-instance modulation [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The architecture of the proposed Mod-CL framework. The pre-training stage begins with modulation-consistent positive [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Per-SNR test accuracy (%) under the linear probing protocol with [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Mechanism validation on RML 2016.10A under the [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: t-SNE visualization of encoder features on the test set [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation studies on RML 2016.10A under the linear [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
read the original abstract

Deep learning-based AMC methods have achieved remarkable performance, but their practical deployment remains constrained by the high cost of labeled data. Although self-supervised learning (SSL) reduces the reliance on labels, existing SSL-based AMC methods often rely on task-agnostic pretext objectives misaligned with modulation classification, leading to representations entangled with nuisance factors such as symbol, channel, and noise. In this paper, we identify intra-instance modulation consistency as a task-aware structural prior, whereby different temporal segments of the same signal may differ in waveform while preserving the same modulation type, thus providing a principled cue for task-aligned self-supervision. Based on this prior, we propose Mod-CL, a Modulation consistency-based Contrastive Learning framework that constructs positive pairs from different temporal segments of the same signal instance, to encourage the model to learn shared modulation information while suppressing nuisance variations. We further develop a contrastive objective tailored to Mod-CL, which jointly exploits temporal segmentation and data augmentation to pull together views sharing the same modulation semantics while avoiding supervisory conflicts within each signal instance. Extensive experiments on RadioML datasets show that Mod-CL consistently outperforms strong baselines, especially in low-label regimes, achieving substantial improvements in linear probing accuracy.

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

1 major / 2 minor

Summary. The manuscript proposes Mod-CL, a self-supervised contrastive learning framework for automatic modulation classification (AMC). It introduces intra-instance modulation consistency as a task-aware prior, forming positive pairs from different temporal segments of the same RadioML-style signal instance. A tailored contrastive objective combines this segmentation with data augmentation to encourage learning of shared modulation semantics while suppressing nuisance factors such as channel and noise. The paper reports that Mod-CL outperforms strong baselines on RadioML datasets, with particularly strong gains in low-label regimes via linear probing accuracy.

Significance. If the central claim holds and the method successfully achieves nuisance-invariant representations aligned with modulation classification, this could meaningfully advance label-efficient SSL approaches for AMC. The use of a domain-specific structural prior (modulation consistency) rather than generic pretext tasks is a clear strength, offering a more principled alternative to existing SSL-AMC methods and potentially improving robustness in practical wireless scenarios with scarce labels.

major comments (1)
  1. Section 3 (Method): The positive-pair construction from temporal segments of the same instance (detailed around the description of Mod-CL) assumes these segments differ primarily in waveform while sharing modulation type. However, under the standard RadioML signal model, such segments share identical channel realizations (fading, phase offset) and exhibit correlated noise. The paper invokes 'data augmentation' to jointly exploit segmentation, but does not specify whether independent per-view channel randomization (e.g., distinct fading or phase draws) is applied to each segment. Without this, the contrastive pull provides no explicit gradient signal for invariance to channel/noise, which is load-bearing for the claimed suppression of nuisance factors and the reported linear-probing gains in low-label regimes.
minor comments (2)
  1. Abstract: The claims of 'substantial improvements' and 'consistent outperformance' would be strengthened by including at least one or two key quantitative results (e.g., accuracy deltas on specific RadioML datasets and label fractions) to allow immediate assessment of effect size.
  2. Notation and figures: Ensure all augmentation operations (segmentation, noise addition, etc.) are explicitly labeled in the framework diagram and that the contrastive loss formulation (likely Eq. in Section 3) clearly distinguishes the positive-pair sampling strategy from standard SimCLR-style objectives.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive summary and significance assessment of our work, as well as for the detailed major comment. We address the point below and will incorporate the necessary clarification in the revised manuscript.

read point-by-point responses

  1. Referee: Section 3 (Method): The positive-pair construction from temporal segments of the same instance (detailed around the description of Mod-CL) assumes these segments differ primarily in waveform while sharing modulation type. However, under the standard RadioML signal model, such segments share identical channel realizations (fading, phase offset) and exhibit correlated noise. The paper invokes 'data augmentation' to jointly exploit segmentation, but does not specify whether independent per-view channel randomization (e.g., distinct fading or phase draws) is applied to each segment. Without this, the contrastive pull provides no explicit gradient signal for invariance to channel/noise, which is load-bearing for the claimed suppression of nuisance factors and the reported linear-probing gains in low-label regimes.

    Authors: We appreciate this observation and agree that the current description in Section 3 lacks sufficient detail on the augmentation pipeline for the positive pairs. In the revised version, we will explicitly state that independent data augmentations—including distinct channel realizations such as independently drawn fading coefficients, phase offsets, and noise—are applied to each temporal segment. This ensures the contrastive objective generates gradients that promote invariance to channel and noise variations while aligning on shared modulation semantics. We will also include a brief description of the augmentation implementation to make this aspect unambiguous. revision: yes

Circularity Check

0 steps flagged

No circularity: Mod-CL follows from stated signal prior without reduction to inputs

full rationale

The paper grounds its method in the observable property that temporal segments of one signal instance share modulation type while differing in waveform details. It then defines positive pairs for contrastive learning directly from this prior and augments with standard data transforms. No equation or claim reduces the learned representation or objective back to a fitted parameter, self-citation chain, or renamed input; the contrastive loss is a standard InfoNCE variant applied to the constructed pairs. Experiments on RadioML supply independent empirical checks rather than tautological validation. The derivation therefore remains self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the domain assumption that temporal segments preserve modulation identity and on the empirical claim that the resulting contrastive training yields better representations; no free parameters or invented entities are stated in the abstract.

axioms (1)
  • domain assumption Different temporal segments of the same signal instance share the same modulation type while differing in waveform, symbol, channel, and noise.
    This is explicitly identified as the task-aware structural prior that enables the positive-pair construction.

pith-pipeline@v0.9.0 · 5527 in / 1323 out tokens · 80434 ms · 2026-05-13T05:23:35.599329+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

40 extracted references · 40 canonical work pages · 1 internal anchor

  1. [1]

    Revolution of wireless signal recognition for 6g: Recent advances, challenges and future directions,

    H. Zhang, F. Zhou, H. Du, Q. Wu, and C. Yuen, “Revolution of wireless signal recognition for 6g: Recent advances, challenges and future directions,”IEEE Communications Surveys & Tutorials, vol. 28, pp. 3521–3563, 2026

    work page 2026

  2. [2]

    Recent advances in automatic modulation classification technology: Methods, results, and prospects,

    Q. Zheng, X. Tian, L. Yu, A. Elhanashi, and S. Saponara, “Recent advances in automatic modulation classification technology: Methods, results, and prospects,”International Journal of Intelligent Systems, vol. 2025, no. 1, p. 4067323, 2025

    work page 2025

  3. [3]

    Boosting robustness in automatic modulation recognition for wireless communications,

    Y . Zhao, Y . Wang, C. Zhang, C. Li, Z. Xiong, L. Zhu, and D. Niyato, “Boosting robustness in automatic modulation recognition for wireless communications,”IEEE Transactions on Cognitive Communications and Networking, vol. 11, no. 3, pp. 1635–1648, 2025

    work page 2025

  4. [4]

    A hybrid approach for cross-dataset modulation recognition of wireless inter- ference,

    Z. Zhang, H. Li, Y . Li, Z. Chen, S. Wang, and T. Luo, “A hybrid approach for cross-dataset modulation recognition of wireless inter- ference,”IEEE Transactions on Communications, vol. 73, no. 12, pp. 13 677–13 690, 2025

    work page 2025

  5. [5]

    Deep learning models for wireless signal classification with distributed low- cost spectrum sensors,

    S. Rajendran, W. Meert, D. Giustiniano, V . Lenders, and S. Pollin, “Deep learning models for wireless signal classification with distributed low- cost spectrum sensors,”IEEE Transactions on Cognitive Communica- tions and Networking, vol. 4, no. 3, pp. 433–445, 2018

    work page 2018

  6. [6]

    A ultra-low cost and accurate amc algorithm and its hardware implementation,

    Y . Zhao, T. Deng, B. Gavin, E. A. Ball, and L. Seed, “A ultra-low cost and accurate amc algorithm and its hardware implementation,”IEEE Open Journal of the Computer Society, vol. 6, pp. 460–467, 2025

    work page 2025

  7. [7]

    Contrastive self- supervised clustering for specific emitter identification,

    X. Hao, Z. Feng, R. Liu, S. Yang, L. Jiao, and R. Luo, “Contrastive self- supervised clustering for specific emitter identification,”IEEE Internet of Things Journal, vol. 10, no. 23, pp. 20 803–20 818, 2023

    work page 2023

  8. [8]

    Learn to defend: Adversarial multi-distillation for automatic modulation recognition models,

    Z. Chen, Z. Wang, D. Xu, J. Zhu, W. Shen, S. Zheng, Q. Xuan, and X. Yang, “Learn to defend: Adversarial multi-distillation for automatic modulation recognition models,”IEEE Transactions on Information Forensics and Security, vol. 19, pp. 3690–3702, 2024

    work page 2024

  9. [9]

    Generalized automatic modulation classification for ofdm systems under unseen synthetic channels,

    S. Huang, J. He, Z. Yang, Y . Chen, S. Chang, Y . Zhang, and Z. Feng, “Generalized automatic modulation classification for ofdm systems under unseen synthetic channels,”IEEE Transactions on Wireless Com- munications, vol. 23, no. 9, pp. 11 931–11 941, 2024

    work page 2024

  10. [10]

    Automatic composite-modulation classification using cyclic-paw-print features for cognitive aerospace communications,

    X. Yan, X. Zhong, H.-C. Wu, P. Yang, Q. Wang, and Y . Chen, “Automatic composite-modulation classification using cyclic-paw-print features for cognitive aerospace communications,”IEEE Transactions on Communications, vol. 72, no. 9, pp. 5486–5502, 2024

    work page 2024

  11. [11]

    Online hybrid likelihood based modulation classification us- ing multiple sensors,

    B. Dulek, “Online hybrid likelihood based modulation classification us- ing multiple sensors,”IEEE Transactions on Wireless Communications, vol. 16, no. 8, pp. 4984–5000, 2017

    work page 2017

  12. [12]

    A likelihood-based algo- rithm for blind identification of qam and psk signals,

    D. Zhu, V . J. Mathews, and D. H. Detienne, “A likelihood-based algo- rithm for blind identification of qam and psk signals,”IEEE Transactions on Wireless Communications, vol. 17, no. 5, pp. 3417–3430, 2018

    work page 2018

  13. [13]

    Deep learning for modulation recognition: A survey with a demonstration,

    R. Zhou, F. Liu, and C. W. Gravelle, “Deep learning for modulation recognition: A survey with a demonstration,”IEEE Access, vol. 8, pp. 67 366–67 376, 2020

    work page 2020

  14. [14]

    Automatic modulation classification: A deep architecture survey,

    T. Huynh-The, Q.-V . Pham, T.-V . Nguyen, T. T. Nguyen, R. Ruby, M. Zeng, and D.-S. Kim, “Automatic modulation classification: A deep architecture survey,”IEEE Access, vol. 9, pp. 142 950–142 971, 2021

    work page 2021

  15. [15]

    Mclhn: Toward automatic modulation classification via masked contrastive learning with hard negatives,

    C. Xiao, S. Yang, Z. Feng, and L. Jiao, “Mclhn: Toward automatic modulation classification via masked contrastive learning with hard negatives,”IEEE Transactions on Wireless Communications, vol. 23, no. 10, pp. 14 304–14 319, 2024

    work page 2024

  16. [16]

    A transformer-based contrastive semi-supervised learning framework for automatic modula- tion recognition,

    W. Kong, X. Jiao, Y . Xu, B. Zhang, and Q. Yang, “A transformer-based contrastive semi-supervised learning framework for automatic modula- tion recognition,”IEEE Transactions on Cognitive Communications and Networking, vol. 9, no. 4, pp. 950–962, 2023

    work page 2023

  17. [17]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” inProceedings of the 37th International Conference on Machine Learning, H. D. III and A. Singh, Eds., 2020, pp. 1597–1607

    work page 2020

  18. [18]

    Masked au- toencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick, “Masked au- toencoders are scalable vision learners,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 16 000–16 009

    work page 2022

  19. [19]

    Self-contrastive learning based semi-supervised radio modulation classification,

    D. Liu, P. Wang, T. Wang, and T. Abdelzaher, “Self-contrastive learning based semi-supervised radio modulation classification,” inMILCOM 2021 - 2021 IEEE Military Communications Conference (MILCOM), 2021, pp. 777–782

    work page 2021

  20. [20]

    Gaf-mae: A self- supervised automatic modulation classification method based on gramian angular field and masked autoencoder,

    Y . Shi, H. Xu, Y . Zhang, Z. Qi, and D. Wang, “Gaf-mae: A self- supervised automatic modulation classification method based on gramian angular field and masked autoencoder,”IEEE Transactions on Cognitive Communications and Networking, vol. 10, no. 1, pp. 94–106, 2024

    work page 2024

  21. [21]

    Predicting spectral information for self-supervised signal classification,

    Y . Xu, S. W. 0001, H. Xing, C. Wang, D. Quan, R. Y . 0038, D. Zhao, and L. Mei, “Predicting spectral information for self-supervised signal classification,” inProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI 2025, Montreal, Canada, August 16-22, 2025, 2025, pp. 6758–6766

    work page 2025

  22. [22]

    An introduction to deep learning for the physical layer,

    T. O’Shea and J. Hoydis, “An introduction to deep learning for the physical layer,”IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 4, pp. 563–575, 2017

    work page 2017

  23. [23]

    Rml22: Realistic dataset generation for wireless modulation classification,

    V . Sathyanarayanan, P. Gerstoft, and A. E. Gamal, “Rml22: Realistic dataset generation for wireless modulation classification,”IEEE Trans- actions on Wireless Communications, vol. 22, no. 11, pp. 7663–7675, 2023

    work page 2023

  24. [24]

    S. S. Haykin,Digital communications. Wiley New York, 1988

    work page 1988

  25. [25]

    Radio machine learning dataset generation with gnu radio,

    T. J. O’shea and N. West, “Radio machine learning dataset generation with gnu radio,” inProceedings of the GNU radio conference, vol. 1, no. 1, 2016

    work page 2016

  26. [26]

    Convolutional radio modula- tion recognition networks,

    T. J. O’Shea, J. Corgan, and T. C. Clancy, “Convolutional radio modula- tion recognition networks,” inInternational conference on engineering applications of neural networks. Springer, 2016, pp. 213–226

    work page 2016

  27. [27]

    Automatic modulation classification algorithm using higher-order cumulants under real-world channel con- ditions,

    V . D. Orlic and M. L. Dukic, “Automatic modulation classification algorithm using higher-order cumulants under real-world channel con- ditions,”IEEE Communications Letters, vol. 13, no. 12, pp. 917–919, 2009

    work page 2009

  28. [28]

    Automatic modulation classification of overlapped sources using multiple cumulants,

    S. Huang, Y . Yao, Z. Wei, Z. Feng, and P. Zhang, “Automatic modulation classification of overlapped sources using multiple cumulants,”IEEE Transactions on Vehicular Technology, vol. 66, no. 7, pp. 6089–6101, 2017

    work page 2017

  29. [29]

    Performance study of cyclostationary based digital modulation classification schemes,

    U. Satija, M. S. Manikandan, and B. Ramkumar, “Performance study of cyclostationary based digital modulation classification schemes,” in2014 9th International Conference on Industrial and Information Systems (ICIIS), 2014, pp. 1–5

    work page 2014

  30. [30]

    Automatic modulation classification architectures based on cyclostationary features in impulsive environments,

    T. V . R. O. C ˆamara, A. D. L. Lima, B. M. M. Lima, A. I. R. Fontes, A. D. M. Martins, and L. F. Q. Silveira, “Automatic modulation classification architectures based on cyclostationary features in impulsive environments,”IEEE Access, vol. 7, pp. 138 512–138 527, 2019

    work page 2019

  31. [31]

    Cnn-based automatic modulation classification under phase imperfections,

    T. K. Oikonomou, N. G. Evgenidis, D. G. Nixarlidis, D. Tyrovolas, S. A. Tegos, P. D. Diamantoulakis, P. G. Sarigiannidis, and G. K. Karagiannidis, “Cnn-based automatic modulation classification under phase imperfections,”IEEE Wireless Communications Letters, vol. 13, no. 5, pp. 1508–1512, 2024

    work page 2024

  32. [32]

    Signet: A novel deep learning framework for radio signal classification,

    Z. Chen, H. Cui, J. Xiang, K. Qiu, L. Huang, S. Zheng, S. Chen, Q. Xuan, and X. Yang, “Signet: A novel deep learning framework for radio signal classification,”IEEE Transactions on Cognitive Communi- cations and Networking, vol. 8, no. 2, pp. 529–541, 2022

    work page 2022

  33. [33]

    Real-time radio technology and modulation classification via an lstm auto-encoder,

    Z. Ke and H. Vikalo, “Real-time radio technology and modulation classification via an lstm auto-encoder,”IEEE Transactions on Wireless Communications, vol. 21, no. 1, pp. 370–382, 2022

    work page 2022

  34. [34]

    Mcformer: A transformer based deep neural network for automatic modulation classification,

    S. Hamidi-Rad and S. Jain, “Mcformer: A transformer based deep neural network for automatic modulation classification,” in2021 IEEE Global Communications Conference (GLOBECOM), 2021, pp. 1–6

    work page 2021

  35. [35]

    Amc-transformer: Automatic modulation classification based on enhanced attention model,

    Y . Xu, “Amc-transformer: Automatic modulation classification based on enhanced attention model,”INFOCOMMUNICATIONS JOURNAL, vol. 17, no. 4, pp. 32–40, 2025

    work page 2025

  36. [36]

    Multi- representation domain attentive contrastive learning based unsupervised automatic modulation recognition,

    Y . Li, X. Shi, H. Tan, Z. Zhang, X. Yang, and F. Zhou, “Multi- representation domain attentive contrastive learning based unsupervised automatic modulation recognition,”Nature Communications, vol. 16, no. 1, p. 5951, 2025

    work page 2025

  37. [37]

    Aflnet: Auxiliary feature learning-guided cross-channel automatic modulation classification,

    H. Xing, S. Wang, C. Wang, D. Quan, H. Mo, L. Mei, H. Zhou, and L. Jiao, “Aflnet: Auxiliary feature learning-guided cross-channel automatic modulation classification,”IEEE Transactions on Communi- cations, vol. 73, no. 12, pp. 13 519–13 534, 2025

    work page 2025

  38. [38]

    Momentum contrast for unsupervised visual representation learning,

    K. He, H. Fan, Y . Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised visual representation learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 9729–9738

    work page 2020

  39. [39]

    Representation Learning with Contrastive Predictive Coding

    A. v. d. Oord, Y . Li, and O. Vinyals, “Representation learning with contrastive predictive coding,”arXiv preprint arXiv:1807.03748, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  40. [40]

    Sigda: A superimposed domain adaptation framework for automatic modulation classification,

    S. Wang, H. Xing, C. Wang, H. Zhou, B. Hou, and L. Jiao, “Sigda: A superimposed domain adaptation framework for automatic modulation classification,”IEEE Transactions on Wireless Communications, vol. 23, no. 10, pp. 13 159–13 172, 2024

    work page 2024