arxiv: 2604.10054 · v1 · submitted 2026-04-11 · 💻 cs.LG · cs.SD

Recognition: unknown

Cross-Validated Cross-Channel Self-Attention and Denoising for Automatic Modulation Classification

Prakash Suman , Yanzhen Qu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:36 UTC · model grok-4.3

classification 💻 cs.LG cs.SD

keywords automatic modulation classificationself-attentiondenoisingdeep learningsignal-to-noise ratioradio signal classificationmachine learning

0 comments

The pith

A deep learning model with cross-channel self-attention and dual-path denoising improves automatic modulation classification accuracy under noisy conditions.

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

The paper addresses the problem that existing deep learning models for automatic modulation classification lose discriminative features when handling noisy signals. It proposes integrating a cross-channel self-attention mechanism to model dependencies between in-phase and quadrature components with dual-path deep residual shrinkage denoising blocks that suppress noise without erasing class-specific structures. Experiments on the RML2018.01a dataset demonstrate accuracy improvements over three benchmark models in the challenging low-to-medium SNR range, and cross-validation supports the model's robustness. This matters because reliable modulation classification is essential for spectrum management and communication systems operating in real-world interference.

Core claim

The central claim is that formalizing baseband modeling as orthogonal subproblems and using cross-channel attention as a generalized complex interaction operator, together with feature-preserving denoising, leads to more robust AMC. The proposed model achieves accuracy increases of 3%, 2.3%, and 14% over PET-CGDNN, MCLDNN, and DAE across -8 dB to +2 dB SNR, with cross-validation yielding a mean accuracy of 62.6%.

What carries the argument

The cross-channel self-attention block that captures dependencies between in-phase and quadrature components, and the dual-path deep residual shrinkage denoising blocks for noise suppression while preserving discriminative structures.

If this is right

Denoising depth strongly influences robustness at low and moderate SNRs.
The model advances interference-aware AMC through orthogonal subproblems and cross-channel attention.
Ablations confirm the critical role of feature-preserving denoising for low-to-medium SNR robustness.
Cross-validation establishes consistent performance metrics including 65.8% macro precision.

Where Pith is reading between the lines

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

This approach may extend to other signal processing tasks involving complex-valued data with noise.
Real-time implementation in hardware could benefit from the residual structure for efficiency.
Further work might explore adaptive denoising depth based on estimated SNR.

Load-bearing premise

The dual-path deep residual shrinkage denoising blocks can suppress noise while preserving the discriminative modulation structures without introducing classification artifacts.

What would settle it

If the proposed model shows no improvement or even lower accuracy than benchmarks when evaluated on a held-out dataset with different noise characteristics or modulation schemes, that would indicate the gains are not general.

Figures

Figures reproduced from arXiv: 2604.10054 by Prakash Suman, Yanzhen Qu.

**Figure 1.** Figure 1: Proposed model architecture This figure presents the overall design of the proposed automatic modulation classification model. The architecture combines two complementary feature extraction pathways. One branch uses a long short-term memory network to learn temporal dependencies across the signal sequence, while the second branch applies cross-channel self-attention to model instantaneous interactions bet… view at source ↗

**Figure 2.** Figure 2: Cross-channel self-attention block This figure illustrates the proposed cross-channel self-attention mechanism for I/Q signal modeling. At each time step, the in-phase and quadrature components are treated as a two-token representation and projected into a latent feature space. Selfattention is then applied across the two tokens to capture their pairwise dependency structure. The block includes query, key… view at source ↗

**Figure 3.** Figure 3: Denoising block A and B This figure shows the two denoising modules used in the proposed network. Denoising Block A performs adaptive thresholding on the incoming feature map using pooled feature statistics. Global average pooling and global max pooling are used to estimate complementary information about feature magnitude and distribution, and these statistics are combined through learnable coefficients … view at source ↗

**Figure 4.** Figure 4: Garrote thresholding and its derivative This figure depicts the garrote thresholding function used in the denoising stages and the corresponding derivative used during optimization. The thresholding function suppresses coefficients with small magnitude while retaining and smoothly shrinking larger coefficients. Compared with more aggressive thresholding methods, garrote thresholding is intended to reduce … view at source ↗

**Figure 6.** Figure 6: Normalized modulation class distribution for train, validation, [PITH_FULL_IMAGE:figures/full_fig_p038_6.png] view at source ↗

**Figure 7.** Figure 7: Normalized SNR level distribution for train, validation, and test [PITH_FULL_IMAGE:figures/full_fig_p038_7.png] view at source ↗

**Figure 8.** Figure 8: Proposed model training and validation loss [PITH_FULL_IMAGE:figures/full_fig_p038_8.png] view at source ↗

**Figure 9.** Figure 9: Proposed model training and validation accuracy [PITH_FULL_IMAGE:figures/full_fig_p038_9.png] view at source ↗

**Figure 10.** Figure 10: Proposed model modulation classification accuracy at each SNR [PITH_FULL_IMAGE:figures/full_fig_p039_10.png] view at source ↗

**Figure 11.** Figure 11: Proposed model vs benchmark models modulation classification [PITH_FULL_IMAGE:figures/full_fig_p039_11.png] view at source ↗

**Figure 12.** Figure 12: Ablation study: Average modulation classification accuracy vs. [PITH_FULL_IMAGE:figures/full_fig_p039_12.png] view at source ↗

**Figure 13.** Figure 13: Ablation study: average modulation classification accuracy per [PITH_FULL_IMAGE:figures/full_fig_p039_13.png] view at source ↗

read the original abstract

This study addresses a key limitation in deep learning Automatic Modulation Classification (AMC) models, which perform well at high signal-to-noise ratios (SNRs) but degrade under noisy conditions due to conventional feature extraction suppressing both discriminative structure and interference. The goal was to develop a feature-preserving denoising method that mitigates the loss of modulation class separation. A deep learning AMC model was proposed, incorporating a cross-channel self-attention block to capture dependencies between in-phase and quadrature components, along with dual-path deep residual shrinkage denoising blocks to suppress noise. Experiments using the RML2018.01a dataset employed stratified sampling across 24 modulation types and 26 SNR levels. Results showed that denoising depth strongly influences robustness at low and moderate SNRs. Compared to benchmark models PET-CGDNN, MCLDNN, and DAE, the proposed model achieved notable accuracy improvements across -8 dB to +2 dB SNR, with increases of 3%, 2.3%, and 14%, respectively. Cross-validation confirmed the model's robustness, yielding a mean accuracy of 62.6%, macro precision of 65.8%, macro-recall of 62.6%, and macro-F1 score of 62.9%. The architecture advances interference-aware AMC by formalizing baseband modeling as orthogonal subproblems and introducing cross-channel attention as a generalized complex interaction operator, with ablations confirming the critical role of feature-preserving denoising for robustness at low-to-medium SNR.

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.

Desk Editor's Note private letter to a colleague

The paper adds cross-channel self-attention and dual-path residual shrinkage denoising to AMC and reports concrete accuracy gains over three baselines at low-to-medium SNR on RML2018.01a.

read the letter

The punchline is that this is a straightforward empirical paper on a neural architecture for automatic modulation classification. It adds cross-channel self-attention to capture I/Q interactions and dual-path residual shrinkage denoising to clean signals without losing class-discriminative features, and it beats three named baselines at low-to-medium SNR on the RML2018.01a dataset. What is new is the particular pairing of those two blocks and the claim that they formalize baseband modeling as orthogonal subproblems. The experiments use stratified sampling across 24 modulation types and 26 SNR levels, report specific accuracy lifts (3% over PET-CGDNN, 2.3% over MCLDNN, 14% over DAE in the -8 to +2 dB range), and give cross-validation aggregates like 62.6% mean accuracy. The abstract notes that ablations highlight the denoising depth as important for robustness. The paper does a good job of focusing on the practical problem of performance drop in noisy conditions and positioning the denoising as feature-preserving rather than destructive. For readers working on signal classification in wireless systems, the numbers and the public dataset make it straightforward to assess. The soft spots are minor but worth noting. The improvements are uneven across baselines, and the overall accuracy level remains in the low 60s, which is typical for those SNRs but limits how much it moves the needle. The abstract lacks details on error bars, training hyperparameters, or how much the model was tuned to this dataset, so it's not yet clear how general the gains are. The conceptual framing is more descriptive than rigorously derived. This work is for specialists in machine learning applied to communications who are looking for architecture ideas to handle noise in AMC. It has enough empirical substance and clear metrics to warrant a serious referee. I would recommend sending it to peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a deep learning architecture for automatic modulation classification (AMC) that integrates a cross-channel self-attention block to capture I/Q dependencies with dual-path deep residual shrinkage denoising blocks to suppress noise while preserving discriminative modulation structures. Experiments on the RML2018.01a dataset employ stratified sampling over 24 modulation types and 26 SNR levels; the model reports accuracy gains of 3%, 2.3%, and 14% over PET-CGDNN, MCLDNN, and DAE respectively in the -8 dB to +2 dB range, together with cross-validation aggregates of 62.6% mean accuracy, 65.8% macro precision, 62.6% macro-recall, and 62.9% macro-F1. Ablations are cited to confirm the critical role of denoising depth for low-to-medium SNR robustness, and the architecture is framed as formalizing baseband modeling via orthogonal subproblems and cross-channel attention as a generalized complex interaction operator.

Significance. If the empirical results prove reproducible, the work offers a concrete, interference-aware advance in AMC by demonstrating that feature-preserving denoising can mitigate the accuracy drop at low-to-medium SNRs that plagues conventional feature extractors. The use of a public dataset, stratified sampling, and cross-validation provides a reproducible baseline, while the reported quantitative gains and ablation evidence on denoising depth supply falsifiable predictions that can be directly tested by the community.

major comments (2)

[Results] Results section (accuracy-vs-SNR curves and cross-validation table): the headline gains (3%, 2.3%, 14%) and CV aggregates are presented without per-run standard deviations, confidence intervals, or statistical significance tests against the three baselines; this weakens the claim that the improvements are robust rather than within experimental variability.
[Proposed Architecture] Method description of dual-path deep residual shrinkage denoising blocks: the central assumption that these blocks suppress noise while preserving discriminative structures without introducing classification artifacts is load-bearing for the low-SNR robustness claim, yet no supporting analysis (e.g., t-SNE visualizations of features before/after denoising or per-class confusion matrices at -8 dB) is provided to verify it.

minor comments (2)

[Introduction] The abstract and introduction use the phrase 'cross-channel self-attention' without an explicit equation or diagram showing how it differs from standard multi-head self-attention applied to the complex baseband signal; a short formal definition would improve clarity.
[Experiments] Training hyperparameters (learning rate schedule, batch size, number of epochs, optimizer) and the exact train/validation/test split ratios after stratified sampling are not stated in the experimental protocol; these details are needed for exact reproduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive recommendation for minor revision. We address each major point below, agreeing that additional statistical rigor and empirical verification will strengthen the manuscript. We will incorporate the suggested analyses in the revised version.

read point-by-point responses

Referee: [Results] Results section (accuracy-vs-SNR curves and cross-validation table): the headline gains (3%, 2.3%, 14%) and CV aggregates are presented without per-run standard deviations, confidence intervals, or statistical significance tests against the three baselines; this weakens the claim that the improvements are robust rather than within experimental variability.

Authors: We agree that reporting variability measures and significance tests would make the robustness claims more convincing. The cross-validation was performed with stratified sampling over 24 modulations and 26 SNRs, but per-run standard deviations were not tabulated. In the revision we will add standard deviations to the CV aggregates (accuracy, precision, recall, F1), include error bars on the accuracy-vs-SNR curves, and report paired statistical tests (e.g., Wilcoxon signed-rank) against the three baselines to quantify whether the observed gains exceed experimental variability. revision: yes
Referee: [Proposed Architecture] Method description of dual-path deep residual shrinkage denoising blocks: the central assumption that these blocks suppress noise while preserving discriminative structures without introducing classification artifacts is load-bearing for the low-SNR robustness claim, yet no supporting analysis (e.g., t-SNE visualizations of features before/after denoising or per-class confusion matrices at -8 dB) is provided to verify it.

Authors: We acknowledge that direct visualization of the denoising effect would better substantiate the feature-preserving claim. While the manuscript already includes ablation studies on denoising depth, we will add t-SNE plots of the feature embeddings immediately before and after the dual-path residual shrinkage blocks at -8 dB SNR, plus per-class confusion matrices at -8 dB for the proposed model versus the baselines. These will be placed in the revised results section (or supplementary material) to demonstrate that discriminative modulation structures are retained without introducing new classification artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper is a standard empirical deep learning architecture paper. It proposes a model with cross-channel self-attention and dual-path residual shrinkage denoising blocks, then reports measured accuracy gains on the fixed public RML2018.01a dataset against named baselines (PET-CGDNN, MCLDNN, DAE) plus cross-validation aggregates. No equations, first-principles derivations, or predictions are present that reduce by construction to fitted parameters, self-definitions, or self-citation chains. All headline numbers are direct experimental outputs, externally falsifiable by reproduction on the same dataset, so the derivation chain is self-contained with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on standard supervised learning assumptions and the representativeness of the RML2018.01a dataset; no new physical entities or ad-hoc constants are introduced beyond typical neural network hyperparameters.

axioms (1)

domain assumption The RML2018.01a dataset with its 24 modulation types and 26 SNR levels is representative of real-world baseband signals for evaluating AMC robustness.
All reported results and cross-validation are performed on this single dataset.

pith-pipeline@v0.9.0 · 5565 in / 1471 out tokens · 50274 ms · 2026-05-10T16:36:13.078969+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

63 extracted references · 58 canonical work pages

[1]

IEEE Access7, 118907–118930 (2019) https://doi.org/10.1109/ ACCESS.2019.2937405

Tayyab, M., Gelabert, X., J¨ antti, R.: A survey on handover management: from LTE to NR. IEEE Access7, 118907–118930 (2019) https://doi.org/10.1109/ ACCESS.2019.2937405

work page arXiv 2019
[2]

Telecom5(1), 255–279 (2024) https://doi.org/10.3390/ telecom5010013

Adamu, P.U., L´ opez-Ben´ ıtez, M.: Analysis of carrier aggregation as a diver- sity technique for improved spectral efficiency and secrecy performance in mobile communications. Telecom5(1), 255–279 (2024) https://doi.org/10.3390/ telecom5010013

2024
[3]

IEEE Open Journal of the Communications Society4, 464–517 (2023) https://doi.org/10.1109/OJCOMS

Parvini, M., Zarif, A.H., Nouruzi, A., Mokari, N., Javan, M.R., Abbasi, B., 29 Ghasemi, A., Yanikomeroglu, H.: Spectrum sharing schemes from 4G to 5G and beyond: protocol flow, regulation, ecosystem, economic. IEEE Open Journal of the Communications Society4, 464–517 (2023) https://doi.org/10.1109/OJCOMS. 2023.3238569

work page doi:10.1109/ojcoms 2023
[4]

Journal of Computational and Graphical Statistics7(4), 469–488 (1998)

Gao, H.-Y.: Wavelet shrinkage denoising using the non-negative garrote. Journal of Computational and Graphical Statistics7(4), 469–488 (1998)

1998
[5]

IEEE Communications Letters25(10), 3287–3290 (2021) https://doi.org/10

Zhang, F., Luo, C., Xu, J., Luo, Y.: An efficient deep learning model for auto- matic modulation recognition based on parameter estimation and transformation. IEEE Communications Letters25(10), 3287–3290 (2021) https://doi.org/10. 1109/LCOMM.2021.3102656

work page arXiv 2021
[6]

IEEE Internet of Things Journal9(3), 2192–2206 (2022) https://doi.org/10.1109/JIOT.2021.3091523

Chang, S., Huang, S., Zhang, R., Feng, Z., Liu, L.: Multitask-learning-based deep neural network for automatic modulation classification. IEEE Internet of Things Journal9(3), 2192–2206 (2022) https://doi.org/10.1109/JIOT.2021.3091523

work page doi:10.1109/jiot.2021.3091523 2022
[7]

IEEE Transactions on Wireless Communications21(1), 370–382 (2022) https://doi.org/10.1109/TWC.2021.3095855

Ke, Z., Vikalo, H.: Real-time radio technology and modulation classification via an lstm auto-encoder. IEEE Transactions on Wireless Communications21(1), 370–382 (2022) https://doi.org/10.1109/TWC.2021.3095855 . Conference Name: IEEE Transactions on Wireless Communications

work page doi:10.1109/twc.2021.3095855 2022
[8]

IEEE Journal of Selected Topics in Signal Processing12(1), 168– 179 (2018) https://doi.org/10.1109/JSTSP.2018.2797022

O’Shea, T.J., Roy, T., Clancy, T.C.: Over-the-air deep learning based radio signal classification. IEEE Journal of Selected Topics in Signal Processing12(1), 168– 179 (2018) https://doi.org/10.1109/JSTSP.2018.2797022

work page doi:10.1109/jstsp.2018.2797022 2018
[9]

In: 2021 5th International Conference on Computer, Communication and Signal Pro- cessing (ICCCSP), pp

Shaik, S., Kirthiga, S.: Automatic modulation classification using densenet. In: 2021 5th International Conference on Computer, Communication and Signal Pro- cessing (ICCCSP), pp. 301–305 (2021). https://doi.org/10.1109/ICCCSP52374. 2021.9465520 . Journal Abbreviation: 2021 5th International Conference on Computer, Communication and Signal Processing (ICCCSP)

work page doi:10.1109/icccsp52374 2021
[10]

Drones7(6), 376 (2023) https://doi.org/10.3390/ drones7060376

Shen, Y., Yuan, H., Zhang, P., Li, Y., Cai, M., Li, J.: A multi-subsampling self-attention network for unmanned aerial vehicle-to-ground automatic mod- ulation recognition system. Drones7(6), 376 (2023) https://doi.org/10.3390/ drones7060376 . Number: 6

2023
[11]

IEEE Transactions on Communications, 1–1 (2024) https://doi.org/10.1109/TCOMM.2024.3477322

Wang, B., Yuan, Z., Lu, J., Zhang, X.: Multitask collaborative learning neural network for radio signal classification. IEEE Transactions on Communications, 1–1 (2024) https://doi.org/10.1109/TCOMM.2024.3477322

work page doi:10.1109/tcomm.2024.3477322 2024
[12]

Physical Communication61, 102170 (2023) https://doi.org/10.1016/j.phycom.2023.102170

Li, Z., Zhang, W., Wang, Y., Li, S., Sun, X.: A lightweight multi-feature fusion structure for automatic modulation classification. Physical Communication61, 102170 (2023) https://doi.org/10.1016/j.phycom.2023.102170

work page doi:10.1016/j.phycom.2023.102170 2023
[13]

In: 2022 IEEE 95th Vehicular Tech- nology Conference: (VTC2022-Spring), pp

Shi, H., Peng, Q., Zhuang, Y.: An improved automatic modulation classification 30 scheme based on adaptive fusion network. In: 2022 IEEE 95th Vehicular Tech- nology Conference: (VTC2022-Spring), pp. 1–5 (2022). https://doi.org/10.1109/ VTC2022-Spring54318.2022.9861036 . Journal Abbreviation: 2022 IEEE 95th Vehicular Technology Conference: (VTC2022-Spring)

work page arXiv 2022
[14]

and Carlsson, Marcel , month = apr, year =

Harper, C., Thornton, M.A., Larson, E.C.: Learnable statistical moments pool- ing for automatic modulation classification. In: ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 8981–8985 (2024). https://doi.org/10.1109/ICASSP48485.2024.10446295 . Journal Abbreviation: ICASSP 2024 - 2024 IEEE Internat...

work page doi:10.1109/icassp48485.2024.10446295 2024
[15]

IEEE Communi- cations Letters24(4), 811–815 (2020) https://doi.org/10.1109/LCOMM.2020

Huynh-The, T., Hua, C.-H., Pham, Q.-V., Kim, D.-S.: Mcnet: an efficient cnn architecture for robust automatic modulation classification. IEEE Communi- cations Letters24(4), 811–815 (2020) https://doi.org/10.1109/LCOMM.2020. 2968030

work page doi:10.1109/lcomm.2020 2020
[16]

In: 2021 International Conference on Wireless Communications and Smart Grid (ICWCSG), pp

Xue, M.L., Huang, M., Yang, J.J., Chen, J.C.: Micnet: a hybrid model based on inception network and cnn for automatic modulation classification. In: 2021 International Conference on Wireless Communications and Smart Grid (ICWCSG), pp. 331–335. https://doi.org/10.1109/ICWCSG53609.2021.00072 . Journal Abbreviation: 2021 International Conference on Wireless ...

work page doi:10.1109/icwcsg53609.2021.00072 2021
[17]

Applied Sciences13(8), 5145 (2023) https://doi.org/ 10.3390/app13085145

Nisar, M.Z., Ibrahim, M.S., Usman, M., Jeong-A, L.: A lightweight deep learn- ing model for automatic modulation classification using residual learning and squeeze–excitation blocks. Applied Sciences13(8), 5145 (2023) https://doi.org/ 10.3390/app13085145 . Place: Basel

work page doi:10.3390/app13085145 2023
[18]

Electronics12(18), 3962 (2023) https://doi.org/10

Harper, C.A., Thornton, M.A., Larson, E.C.: Automatic modulation classification with deep neural networks. Electronics12(18), 3962 (2023) https://doi.org/10. 3390/electronics12183962

2023
[19]

IEEE Signal Processing Letters30, 633–637 (2023) https://doi.org/10.1109/LSP.2023

Xiao, J., Wang, Y., Zhang, D., Ma, Q., Ding, W.: Multiscale correlation net- works based on deep learning for automatic modulation classification. IEEE Signal Processing Letters30, 633–637 (2023) https://doi.org/10.1109/LSP.2023. 3275912

work page doi:10.1109/lsp.2023 2023
[20]

In: 2025 3rd Interna- tional Conference on Device Intelligence, Computing and Communication Tech- nologies (DICCT), pp

Singh, B., Jindal, P., Verma, P., Sharma, V., Prakash, C.: Automatic modulation recognition using modified convolutional neural network. In: 2025 3rd Interna- tional Conference on Device Intelligence, Computing and Communication Tech- nologies (DICCT), pp. 547–550 (2025). https://doi.org/10.1109/DICCT64131. 2025.10986502 . Journal Abbreviation: 2025 3rd I...

work page doi:10.1109/dicct64131 2025
[21]

Electronics12(14), 3175 (2023) https://doi

Duan, R., Li, X., Zhang, H., Yang, G., Li, S., Cheng, P., Li, Y.: A multi-modal modulation recognition method with snr segmentation based on time domain 31 signals and constellation diagrams. Electronics12(14), 3175 (2023) https://doi. org/10.3390/electronics12143175 . Place: Basel

work page doi:10.3390/electronics12143175 2023
[22]

IEEE Communications Letters 27(11), 2968–2972 (2023) https://doi.org/10.1109/LCOMM.2023.3314623

Wei, T., Li, Z., Bi, D., Shao, Z., Gao, J.: Adaptive multi-dimensional shrink- age block for automatic modulation recognition. IEEE Communications Letters 27(11), 2968–2972 (2023) https://doi.org/10.1109/LCOMM.2023.3314623

work page doi:10.1109/lcomm.2023.3314623 2023
[23]

In: Proceedings of the 2023 6th International Confer- ence on Signal Processing and Machine Learning, pp

Cao, J., Zhu, R., Shi, G., Bai, M.: Modulation recognition based on deep learning network. In: Proceedings of the 2023 6th International Confer- ence on Signal Processing and Machine Learning, pp. 357–362. Associa- tion for Computing Machinery,<conf-loc>,<city>Tianjin</city>,<coun- try>China</country>,</conf-loc>(2023). https://doi.org/10.1145/3614008. 36...

work page doi:10.1145/3614008 2023
[24]

In: 2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring), pp

Tekbıyık, K., Ekti, A.R., G¨ or¸ cin, A., Kurt, G.K., Ke¸ ceci, C.: Robust and fast automatic modulation classification with cnn under multipath fading channels. In: 2020 IEEE 91st Vehicular Technology Conference (VTC2020-Spring), pp. 1– 6 (2020). https://doi.org/10.1109/VTC2020-Spring48590.2020.9128408 . Journal Abbreviation: 2020 IEEE 91st Vehicular Tec...

work page doi:10.1109/vtc2020-spring48590.2020.9128408 2020
[25]

Experimental Analysis of Trustworthy In- Vehicle Intrusion Detection System Using eXplainable Artificial Intel- ligence (XAI)

An, T.T., Lee, B.M.: Robust automatic modulation classification in low signal to noise ratio. IEEE Access11, 7860–7872 (2023) https://doi.org/10.1109/ACCESS. 2023.3238995

work page doi:10.1109/access 2023
[26]

EURASIP Journal on Advances in Signal Processing 2023(1), 73 (2023) https://doi.org/10.1186/s13634-023-01036-9

Guo, L., Gao, R., Cong, Y., Yang, L.: Robust automatic modulation classifica- tion under noise mismatch. EURASIP Journal on Advances in Signal Processing 2023(1), 73 (2023) https://doi.org/10.1186/s13634-023-01036-9 . Place: New York

work page doi:10.1186/s13634-023-01036-9 2023
[27]

In: 2022 13th International Conference on Information and Communication Technology Convergence (ICTC), pp

Le, H.-K., Hoang, V.-P., Doan, V.-S., Hoang, M.-T., Dao, N.P.: Performance analysis of convolutional neural networks with different window functions for automatic modulation classification. In: 2022 13th International Conference on Information and Communication Technology Convergence (ICTC), pp. 153–157 (2022). https://doi.org/10.1109/ICTC55196.2022.99527...

work page doi:10.1109/ictc55196.2022.9952750 2022
[28]

IEEE Communications Letters28(2), 328–331 (2024) https://doi.org/10.1109/ LCOMM.2023.3342604

Triaridis, K., Doumanidis, C., Chatzidiamantis, N.D., Karagiannidis, G.K.: Mm-net: a multi-modal approach toward automatic modulation classification. IEEE Communications Letters28(2), 328–331 (2024) https://doi.org/10.1109/ LCOMM.2023.3342604

work page arXiv 2024
[29]

Mathematical Problems in Engineering2020(1), 2678310 (2020) https://doi.org/10.1155/2020/2678310 32

Wu, P., Sun, B., Su, S., Wei, J., Zhao, J., Wen, X.: Automatic modulation classifi- cation based on deep learning for software-defined radio. Mathematical Problems in Engineering2020(1), 2678310 (2020) https://doi.org/10.1155/2020/2678310 32

work page doi:10.1155/2020/2678310 2020
[30]

Sensors23(9) (2023) https://doi.org/10.3390/s23094187

Wang, F., Shang, T., Hu, C., Liu, Q.: Automatic modulation classification using hybrid data augmentation and lightweight neural network. Sensors23(9) (2023) https://doi.org/10.3390/s23094187

work page doi:10.3390/s23094187 2023
[31]

IEEE Transactions on Instrumentation and Measurement72, 1–10 (2023) https://doi.org/10.1109/ TIM.2023.3298657

Xiao, C., Yang, S., Feng, Z.: Complex-valued depthwise separable convolutional neural network for automatic modulation classification. IEEE Transactions on Instrumentation and Measurement72, 1–10 (2023) https://doi.org/10.1109/ TIM.2023.3298657

work page arXiv 2023
[32]

IEEE Access13, 72526–72543 https://doi.org/10.1109/ACCESS.2025.3564032

Ansari, S., Mahmoud, S., Majzoub, S., Almajali, E., Jarndal, A., Bonny, T.: A novel LSTM architecture for automatic modulation recognition: comparative analysis with conventional machine learning and RNN-based approaches. IEEE Access13, 72526–72543 https://doi.org/10.1109/ACCESS.2025.3564032

work page doi:10.1109/access.2025.3564032 2025
[33]

IEEE Internet of Things Journal11(12), 22197–22207 (2024) https://doi.org/10.1109/JIOT.2024.3379429

Li, W., Deng, W., Wang, K., You, L., Huang, Z.: A complex-valued transformer for automatic modulation recognition. IEEE Internet of Things Journal11(12), 22197–22207 (2024) https://doi.org/10.1109/JIOT.2024.3379429 . Conference Name: IEEE Internet of Things Journal

work page doi:10.1109/jiot.2024.3379429 2024
[34]

In: GLOBECOM 2022 - 2022 IEEE Global Communications Conference, pp

Su, H., Fan, X., Liu, H.: Robust and efficient modulation recognition with pyramid signal transformer. In: GLOBECOM 2022 - 2022 IEEE Global Communications Conference, pp. 1868–1874 (2022). https://doi.org/10.1109/GLOBECOM48099. 2022.10001593 . Journal Abbreviation: GLOBECOM 2022 - 2022 IEEE Global Communications Conference

work page doi:10.1109/globecom48099 2022
[35]

Electronics11(10) (2022) https://doi.org/10.3390/electronics11101554

Zheng, Y., Ma, Y., Tian, C.: Tmrn-glu: a transformer-based automatic classifica- tion recognition network improved by gate linear unit. Electronics11(10) (2022) https://doi.org/10.3390/electronics11101554

work page doi:10.3390/electronics11101554 2022
[36]

Memristor- based hardware and algorithms for higher-order Hop- field optimization solver outperforming quadratic Ising machines

Lin, C., Zhang, Z., Wang, L., Wang, Y., Zhao, J., Yang, Z., Xiao, X.: Fast and lightweight automatic modulation recognition using spiking neural network. In: 2024 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1–5 (2024). https://doi.org/10.1109/ISCAS58744.2024.10558051 . Journal Abbreviation: 2024 IEEE International Symposium on Circui...

work page doi:10.1109/iscas58744.2024.10558051 2024
[37]

1109/TAI.2023.3306334

Guo, W., Yang, K., Stratigopoulos, H.-G., Aboushady, H., Salama, K.N.: An end-to-end neuromorphic radio classification system with an efficient sigma- delta-based spike encoding scheme5(4), 1869–1881 (2024) https://doi.org/10. 1109/TAI.2023.3306334 . Conference Name: IEEE Transactions on Artificial Intelligence

work page arXiv 2024
[38]

AI6(8) (2025) https://doi.org/10.3390/ai6080195 33

Suman, P., Qu, Y.: A lightweight deep learning model for automatic modulation classification using dual-path deep residual shrinkage network. AI6(8) (2025) https://doi.org/10.3390/ai6080195 33

work page doi:10.3390/ai6080195 2025
[39]

In: 2022 IEEE 10th International Conference on Information, Communication and Net- works (ICICN), pp

TianShu, C., Zhen, H., Dong, W., Ruike, L.: A iq correlation and long-term fea- tures neural network classifier for automatic modulation classification. In: 2022 IEEE 10th International Conference on Information, Communication and Net- works (ICICN), pp. 396–400 (2022). https://doi.org/10.1109/ICICN56848.2022. 10006621 . Journal Abbreviation: 2022 IEEE 10...

work page doi:10.1109/icicn56848.2022 2022
[40]

In: 2023 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC), pp

Gao, M., Tang, X., Pan, X., Ren, Y., Zhang, B., Dai, J.: Modulation recognition of communication signal with class-imbalance sample based on cnn-lstm dual channel model. In: 2023 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC), pp. 1–6. https://doi.org/10.1109/ ICSPCC59353.2023.10400302

work page arXiv 2023
[41]

Electronics12(22), 4668 https://doi.org/10.3390/electronics12224668

Gao, G., Hu, X., Li, B., Wang, W., Ghannouchi, F.M.: Crosstlnet: a multitask- learning-empowered neural network with temporal convolutional network–long short-term memory for automatic modulation classification. Electronics12(22), 4668 https://doi.org/10.3390/electronics12224668

work page doi:10.3390/electronics12224668
[42]

IEEE Transactions on Wireless Communications, 1–1 (2023) https://doi.org/10

Ding, R., Zhou, F., Wu, Q., Dong, C., Han, Z., Dobre, O.A.: Data and knowledge dual-driven automatic modulation classification for 6g wireless communications. IEEE Transactions on Wireless Communications, 1–1 (2023) https://doi.org/10. 1109/TWC.2023.3316197

work page arXiv 2023
[43]

In: 2021 2nd International Symposium on Computer Engineering and Intelligent Com- munications (ISCEIC), pp

Xue, M.L., Huang, M., Yang, J.J., Wu, J.D.: Mlresnet: an efficient method for automatic modulation classification based on residual neural network. In: 2021 2nd International Symposium on Computer Engineering and Intelligent Com- munications (ISCEIC), pp. 122–126. https://doi.org/10.1109/ISCEIC53685.2021. 00032 . Journal Abbreviation: 2021 2nd Internation...

work page doi:10.1109/isceic53685.2021 2021
[44]

In: 2024 16th International Conference on COMmunication Systems & NETworkS (COMSNETS), pp

Riddhi, S., Parmar, A., Captain, K., K A, D., Chouhan, A., Patel, J.: A dual-stream convolution-gru-attention network for automatic modulation clas- sification. In: 2024 16th International Conference on COMmunication Systems & NETworkS (COMSNETS), pp. 720–724 (2024). https://doi.org/10.1109/ COMSNETS59351.2024.10426812

work page arXiv 2024
[45]

In: 2023 15th Inter- national Conference on COMmunication Systems & NETworkS (COMSNETS), pp

Parmar, A., A, D.K., Chouhan, A., Captain, K.: Dual-stream cnn-bilstm model with attention layer for automatic modulation classification. In: 2023 15th Inter- national Conference on COMmunication Systems & NETworkS (COMSNETS), pp. 603–608 (2023). https://doi.org/10.1109/COMSNETS56262.2023.10041403 . Journal Abbreviation: 2023 15th International Conference...

work page doi:10.1109/comsnets56262.2023.10041403 2023
[46]

Physical Communication64, 102361 (2024) https://doi.org/10.1016/j.phycom.2024.102361 34

Parmar, A., Chouhan, A., Captain, K., Patel, J.: Deep multilevel architecture for automatic modulation classification. Physical Communication64, 102361 (2024) https://doi.org/10.1016/j.phycom.2024.102361 34

work page doi:10.1016/j.phycom.2024.102361 2024
[47]

IEEE Transactions on Cognitive Communications and Networking9(6), 1503–1518 (2023) https:// doi.org/10.1109/TCCN.2023.3309010

Chang, S., Yang, Z., He, J., Li, R., Huang, S., Feng, Z.: A fast multi-loss learning deep neural network for automatic modulation classification. IEEE Transactions on Cognitive Communications and Networking9(6), 1503–1518 (2023) https:// doi.org/10.1109/TCCN.2023.3309010

work page doi:10.1109/tccn.2023.3309010 2023
[48]

IEEE Transactions on Wireless Communications23(12), 19083–19097 (2024) https://doi.org/10.1109/ TWC.2024.3478752

Luo, Z., Xiao, W., Zhang, X., Zhu, L., Xiong, X.: Rlitnn: a multi-channel modu- lation recognition model combining multi-modal features. IEEE Transactions on Wireless Communications23(12), 19083–19097 (2024) https://doi.org/10.1109/ TWC.2024.3478752

work page arXiv 2024
[49]

EURASIP Journal on Wireless Communications and Networking2023(1), 66 (2023) https://doi.org/10.1186/ s13638-023-02275-y

Sun, S., Wang, Y.: A novel deep learning automatic modulation classifier with fusion of multichannel information using gru. EURASIP Journal on Wireless Communications and Networking2023(1), 66 (2023) https://doi.org/10.1186/ s13638-023-02275-y . Place: New York

2023
[50]

IEEE Access9, 143661–143676 (2021) https://doi.org/10.1109/access.2021.3121762

Yang, H., Zhao, L., Yue, G., Ma, B., Li, W.: Irlnet: a short-time and robust archi- tecture for automatic modulation recognition. IEEE Access9, 143661–143676 (2021) https://doi.org/10.1109/access.2021.3121762

work page doi:10.1109/access.2021.3121762 2021
[51]

IEEE Transactions on Vehicular Technology, 1–16 (2024) https://doi.org/10.1109/TVT.2024.3486079

Qu, Y., Lu, Z., Zeng, R., Wang, J., Wang, J.: Enhancing automatic modula- tion recognition through robust global feature extraction. IEEE Transactions on Vehicular Technology, 1–16 (2024) https://doi.org/10.1109/TVT.2024.3486079

work page doi:10.1109/tvt.2024.3486079 2024
[52]

Alexandria Engineering Journal104, 162–170 (2024) https://doi.org/10.1016/j.aej.2024.06.008

Cheng, R., Chen, Q., Huang, M.: Automatic modulation recognition using deep cvcnn-lstm architecture. Alexandria Engineering Journal104, 162–170 (2024) https://doi.org/10.1016/j.aej.2024.06.008

work page doi:10.1016/j.aej.2024.06.008 2024
[53]

China Communications20(5), 135–147 (2023) https://doi.org/10.23919/JCC.ja

Ying, S., Huang, S., Chang, S., Yang, Z., Feng, Z., Guo, N.: A convolutional and transformer based deep neural network for automatic modulation classification. China Communications20(5), 135–147 (2023) https://doi.org/10.23919/JCC.ja. 2022-0580

work page doi:10.23919/jcc.ja 2023
[54]

Electronics 12(2), 422 (2023) https://doi.org/10.3390/electronics12020422

Hou, S., Fan, Y., Han, B., Li, Y., Fang, S.: Signal modulation recognition algorithm based on improved spatiotemporal multi-channel network. Electronics 12(2), 422 (2023) https://doi.org/10.3390/electronics12020422 . Place: Basel

work page doi:10.3390/electronics12020422 2023
[55]

Knowledge-Based Systems300, 112191 (2024) https://doi.org/10

Liu, B., Zheng, Q., Wei, H., Zhao, J., Yu, H., Zhou, Y., Chao, F., Ji, R.: Deep hybrid transformer network for robust modulation classification in wireless com- munications. Knowledge-Based Systems300, 112191 (2024) https://doi.org/10. 1016/j.knosys.2024.112191

work page arXiv 2024
[56]

IEEE Transactions on Instrumentation and Measurement72, 1–13 (2023) https://doi.org/10.1109/TIM.2023.3244798

Ma, J., Hu, M., Wang, T., Yang, Z., Wan, L., Qiu, T.: Automatic modulation classification in impulsive noise: hyperbolic-tangent cyclic spectrum and multi- branch attention shuffle network. IEEE Transactions on Instrumentation and Measurement72, 1–13 (2023) https://doi.org/10.1109/TIM.2023.3244798

work page doi:10.1109/tim.2023.3244798 2023
[57]

IEEE Transactions on Green Communications and Networking, 1–1 (2025) https://doi

An, T.T., Argyriou, A., Puspitasari, A.A., Cotton, S.L., Lee, B.M.: Efficient 35 automatic modulation classification for next-generation wireless networks. IEEE Transactions on Green Communications and Networking, 1–1 (2025) https://doi. org/10.1109/TGCN.2025.3574278

work page doi:10.1109/tgcn.2025.3574278 2025
[58]

Digital Signal Processing129, 103650 (2022) https://doi.org/10.1016/j.dsp.2022.103650

Zhang, F., Luo, C., Xu, J., Luo, Y., Zheng, F.-C.: Deep learning based auto- matic modulation recognition: models, datasets, and challenges. Digital Signal Processing129, 103650 (2022) https://doi.org/10.1016/j.dsp.2022.103650 . ISBN: 1051-2004

work page doi:10.1016/j.dsp.2022.103650 2022
[59]

In: 2013 Fourth International Conference on Intelligent Control and Information Process- ing (ICICIP), pp

Chen, G., Xie, W., Zhao, Y.: Wavelet-based denoising: a brief review. In: 2013 Fourth International Conference on Intelligent Control and Information Process- ing (ICICIP), pp. 570–574 (2013). https://doi.org/10.1109/ICICIP.2013.6568140 . Journal Abbreviation: 2013 Fourth International Conference on Intelligent Control and Information Processing (ICICIP)

work page doi:10.1109/icicip.2013.6568140 2013
[60]

IEEE Transactions on Information Theory , volume =

Donoho, D.L.: De-noising by soft-thresholding. IEEE transactions on information theory41(3), 613–627 (1995) https://doi.org/10.1109/18.382009

work page doi:10.1109/18.382009 1995
[61]

Sensors22(2) (2022) https://doi.org/10.3390/s22020515

Salimy, A., Mitiche, I., Boreham, P., Nesbitt, A., Morison, G.: Dynamic Noise Reduction with Deep Residual Shrinkage Networks for Online Fault Classification. Sensors22(2) (2022) https://doi.org/10.3390/s22020515

work page doi:10.3390/s22020515 2022
[62]

Maddigan and T

Syed, S.N., Lazaridis, P.I., Khan, F.A., Ahmed, Q.Z., Hafeez, M., Ivanov, A., Poulkov, V., Zaharis, Z.D.: Deep neural networks for spectrum sensing: a review. IEEE Access11, 89591–89615 (2023) https://doi.org/10.1109/ACCESS.2023. 3305388

work page doi:10.1109/access.2023 2023
[63]

Uncovering the iceberg in the sea: Fundamentals of pulse shap- ing and modulation design for random ISAC sig- nals[J/OL]

Ansari, S., Alnajjar, K.A., Majzoub, S., Almajali, E., Jarndal, A., Bonny, T., Hussain, A., Mahmoud, S.: Attention-enhanced hybrid automatic modulation classification for advanced wireless communication systems: a deep learning- transformer framework. IEEE Access13, 105463–105491 https://doi.org/10. 1109/ACCESS.2025.3580574 36 Figure Titles and Legends Fi...

work page arXiv 2025