pith. sign in

arxiv: 2605.00849 · v1 · submitted 2026-04-20 · 📡 eess.SP · cs.LG· cs.SY· eess.SY

Deep Learning for Multi-Antenna Modulation Recognition of Radio Signals

Tao Chen , Shilian Zheng , Jiepeng Chen , Zhangbin Pei , Qi Xuan , Xiaoniu Yang This is my paper

Pith reviewed 2026-05-10 04:33 UTC · model grok-4.3

classification 📡 eess.SP cs.LGcs.SYeess.SY
keywords multi-antenna modulation recognitiondeep learningIQ signalsconvolutional neural networkdata augmentationfew-shot learningautomatic modulation classificationspatial diversity
0
0 comments X

The pith

Concatenating raw IQ signals from multiple antennas and feeding them into a convolutional neural network improves modulation recognition accuracy and cuts computational cost compared to voting or averaging per antenna.

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

The paper proposes an MAMR-IQ method that concatenates the raw in-phase and quadrature signals received across multiple antennas before passing the combined input to a convolutional neural network. Simulations demonstrate that this joint processing yields higher recognition accuracy and lower complexity than two prior deep-learning approaches that classify each antenna separately and then combine results via direct voting or weighted averaging. The authors further introduce a data-augmentation step that exchanges IQ sequences between any pair of antennas to create additional training samples, raising accuracy when only limited labeled data are available. A reader would care because multi-antenna receivers are already common in communications, yet automatic modulation recognition still relies on methods that under-use the available spatial diversity.

Core claim

The MAMR-IQ method concatenates the raw received in-phase and quadrature signals of multiple antennas and feeds them into a convolutional neural network; this approach outperforms two existing deep learning-based MAMR methods based on direct voting and weight average in both recognition accuracy and computational complexity. A data-augmentation technique that exchanges IQ sequences received by any two antennas further improves accuracy in few-shot scenarios.

What carries the argument

The MAMR-IQ concatenation of raw multi-antenna IQ vectors before CNN input, which directly exploits spatial diversity without separate per-antenna classification and fusion steps.

If this is right

  • Recognition accuracy rises because the CNN can learn joint spatial features across antennas rather than fusing independent decisions.
  • Computational complexity drops because a single network processes the combined input instead of running multiple classifiers followed by voting or averaging.
  • Few-shot performance improves when antenna-pair IQ exchanges generate additional training samples without requiring new channel measurements.
  • The method scales to larger antenna arrays by simply extending the input tensor width while keeping the same network architecture.

Where Pith is reading between the lines

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

  • The lower complexity could allow modulation recognition to run on edge devices attached to multi-antenna arrays without dedicated high-power processors.
  • The same concatenation principle might transfer to related tasks such as multi-antenna signal detection or direction-of-arrival estimation.
  • If the augmentation works by increasing effective sample diversity, similar pairwise exchanges could be tested on other array geometries such as circular or planar arrays.

Load-bearing premise

Simulation results obtained by concatenating raw IQ inputs under idealized channel models will translate directly to real-world multi-antenna receivers without extra impairments or hardware effects.

What would settle it

A controlled over-the-air experiment with physical multi-antenna hardware that measures recognition accuracy under the same modulation set and signal-to-noise ratios used in the simulations.

Figures

Figures reproduced from arXiv: 2605.00849 by Jiepeng Chen, Qi Xuan, Shilian Zheng, Tao Chen, Xiaoniu Yang, Zhangbin Pei.

Figure 1
Figure 1. Figure 1: The process of modulation recognition of multi antenna receiving system based on Deep Learning. where real(·) and imag(·) represent extracting the real part and imaginary part of the signal received by the i-th antenna, Ii(n) and Qi(n) are the in-phase (I) and quadrature (Q) components of the signal yi(n). After￾wards, the IQ components of the received signals are spliced to obtain Y as Y =        … view at source ↗
Figure 2
Figure 2. Figure 2: The structure of residual. [54], which improve the performance of the model through the stacking of neural networks. However, when the depth increases to a certain extent, the net￾work has a degradation problem, i.e., the performance reaches saturation and even the accuracy decreases. In order to avoid the degradation of deep network, a new network structure ResNet has been proposed. ResNet introduces the … view at source ↗
Figure 3
Figure 3. Figure 3: Structure of the constructed ResNet56, “avg￾pool” stands for average pooling layer. The repetition times of Residual block1, Residual block2 and Residual block3 are 9. We design a ResNet network structure based on the above residual blocks for MAMR in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The process of data augmentation by exchanging IQ sequences of each antenna, n = 0, 1, . . . , N −1, N represents the length of the signal. Yi,j represents the data obtained by exchanging the IQ sequence of the i-th antenna and the j-th antenna. IQi (n) represents the I-channel and Q-channel components received by the i-th antenna. The two sequences can be exchanged to obtain the augmented sample Yi,j as Y… view at source ↗
Figure 5
Figure 5. Figure 5: Recognition accuracy of different methods. formance gain of MAMR-IQ method over MAMR￾DV method and MAMR-WA method increases as the number of antennas grows. Specifically, when the number of antennas is 2, MAMR-IQ method outper￾forms the other methods by 4%. However, when the number of antennas is 16, MAMR-IQ method exhibits a surprisingly high performance gain of 13.6%. Over￾all, MAMR-IQ method we proposed… view at source ↗
Figure 6
Figure 6. Figure 6: The confusion matrices of different methods. WA methods. 4.2.3 Performance in Random Antenna Setting In real-world scenarios, we may encounter the situ￾ation of variant antenna array settings. To account for this, we use the dataset with random phase offset among the received signals of different antenna ele￾ments. The number of antennas in this experiment is 2 and the accuracy under different SNR is shown… view at source ↗
Figure 7
Figure 7. Figure 7: Recognition accuracy in random antenna setting [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Recognition accuracy under different exchange times. 4.3.1 Time Complexity Time complexity of MAMR-IQ method refers to the complexity of the ResNet56 network computation in the inference stage, whereas MAMR-DV method’s complexity includes both the neural network com￾putation and the direct voting process. Meanwhile, MAMR-WA method’s complexity comprises the neu￾ral network computation and the weight averag… view at source ↗
Figure 9
Figure 9. Figure 9: The confusion matrices of augmentation method at sample ratio = 0.01. size, Tin and Tout are the numbers of input channels and output channels of the convolution kernel. 4.3.2 Space Complexity The space complexity S is independent of the input size, but is related to the neural network model. There￾fore, the space complexity mainly includes two parts. The first part W is the total number of weight parame￾t… view at source ↗
read the original abstract

Multi-antenna receiving systems have become a prevalent technical solution in communication systems. Meanwhile, deep learning has achieved significant progress in automatic modulation recognition tasks in single-antenna systems. However, the application of deep learning in multi-antenna modulation recognition (MAMR) tasks is still limited. In this paper, we propose an MAMR method namely MAMR-IQ to fully explore the diversity gain of a multi-antenna receiving system, which concatenates the raw received in-phase and quadrature (IQ) signals of multiple antennas and feeds them into a convolutional neural network. Simulation results show that the proposed MAMR-IQ method outperforms two existing deep learning-based MAMR methods which are based on direct voting (DV) and weight average (WA) in terms of both recognition accuracy and computational complexity. To address the problem of limited training data in few-shot scenarios, we further propose a data augmentation method that involves exchanging IQ sequences received by any two antennas to generate augmented samples. Simulation results show that with the proposed augmentation method, the recognition accuracy can be further improved.

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

3 major / 2 minor

Summary. The paper proposes MAMR-IQ, a method that concatenates raw multi-antenna IQ samples and feeds them directly into a CNN for automatic modulation recognition. It claims this outperforms two prior DL-based MAMR baselines (direct voting and weighted averaging) in both classification accuracy and computational complexity. A simple IQ-exchange augmentation is introduced to improve few-shot performance, with simulation results showing further accuracy gains.

Significance. If the reported simulation gains hold under reproducible conditions, the work demonstrates that joint CNN processing of concatenated multi-antenna IQ can extract spatial diversity more effectively than post-processing ensembles, offering a low-complexity alternative for MAMR. The augmentation technique provides a lightweight way to address training-data scarcity without requiring new channel models or hardware.

major comments (3)
  1. Experimental Setup section: no details are supplied on the number of antennas, modulation formats, SNR ranges, dataset sizes, channel models (e.g., AWGN vs. fading), or training/validation splits. These parameters are load-bearing for the central claim that MAMR-IQ outperforms DV and WA, as the reported accuracy and complexity advantages cannot be assessed or reproduced without them.
  2. Results section (accuracy and complexity comparisons): the paper states superior performance but provides neither error bars, statistical significance tests, nor the exact CNN architectures and hyperparameters used for the DV and WA baselines. Without these, it is impossible to determine whether the gains arise from the concatenation approach or from differences in model capacity or training.
  3. Few-shot augmentation experiments: quantitative results on the number of augmented samples generated per original example, the specific shot levels tested, and the resulting accuracy deltas are missing. This information is required to evaluate whether the IQ-exchange method meaningfully mitigates data scarcity or merely inflates the training set size.
minor comments (2)
  1. The abstract and introduction use the acronym MAMR without an initial definition; add “multi-antenna modulation recognition (MAMR)” on first use.
  2. Figure captions for the performance curves should explicitly state the number of Monte Carlo trials and the exact SNR grid used.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects for improving the clarity and reproducibility of our work. We have prepared point-by-point responses below and will revise the manuscript to incorporate the requested details.

read point-by-point responses

  1. Referee: Experimental Setup section: no details are supplied on the number of antennas, modulation formats, SNR ranges, dataset sizes, channel models (e.g., AWGN vs. fading), or training/validation splits. These parameters are load-bearing for the central claim that MAMR-IQ outperforms DV and WA, as the reported accuracy and complexity advantages cannot be assessed or reproduced without them.

    Authors: We agree that these experimental parameters are essential for assessing and reproducing the results. The original manuscript did not provide a complete description of the simulation setup. In the revised version, we will expand the Experimental Setup section to explicitly state the number of antennas, the modulation formats considered, the SNR ranges, the dataset sizes, the channel models (AWGN and fading), and the training/validation/test splits. revision: yes

  2. Referee: Results section (accuracy and complexity comparisons): the paper states superior performance but provides neither error bars, statistical significance tests, nor the exact CNN architectures and hyperparameters used for the DV and WA baselines. Without these, it is impossible to determine whether the gains arise from the concatenation approach or from differences in model capacity or training.

    Authors: We acknowledge the need for greater transparency and statistical rigor in the comparisons. The manuscript used the same underlying CNN for all methods to enable fair evaluation, but did not report error bars, significance tests, or full hyperparameter details for the baselines. In the revision, we will add error bars to the performance plots, include statistical significance tests, and provide the exact CNN architectures and training hyperparameters employed for MAMR-IQ, DV, and WA. revision: yes

  3. Referee: Few-shot augmentation experiments: quantitative results on the number of augmented samples generated per original example, the specific shot levels tested, and the resulting accuracy deltas are missing. This information is required to evaluate whether the IQ-exchange method meaningfully mitigates data scarcity or merely inflates the training set size.

    Authors: We agree that more precise quantitative information is required to properly evaluate the augmentation technique. The original manuscript described the IQ-exchange approach at a high level but omitted the specific counts and deltas. In the revised manuscript, we will report the number of augmented samples generated per original example, the shot levels tested, and the resulting accuracy improvements to demonstrate the method's effectiveness in few-shot scenarios. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical ML study proposing MAMR-IQ (raw multi-antenna IQ concatenation into CNN) and comparing it via simulation to external baselines DV and WA. No equations, derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the argument. Performance claims rest on reported simulation accuracy and complexity metrics against independent methods, with the augmentation technique also evaluated empirically. The derivation chain is self-contained and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Paper rests on standard assumptions that CNNs can extract features from concatenated IQ tensors and that simulated channels represent real performance; no free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption Convolutional neural networks can effectively learn modulation features from raw multi-antenna IQ samples
    Implicit in the choice of CNN architecture for the concatenated input

pith-pipeline@v0.9.0 · 5504 in / 1171 out tokens · 31087 ms · 2026-05-10T04:33:33.006716+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

56 extracted references · 56 canonical work pages · 1 internal anchor

  1. [1]

    Sig- nal identification for multiple-antenna wireless sys- tems: Achievements and challenges,

    Y . A. Eldemerdash, O. A. Dobre, and M. ner, “Sig- nal identification for multiple-antenna wireless sys- tems: Achievements and challenges,”IEEE Communi- cations Surveys & Tutorials, vol. 18, no. 3, pp. 1524– 1551, 2016

    work page 2016

  2. [2]

    Auto- matic modulation classifier for military applications,

    V . Iglesias, J. Grajal, and O. Yeste-Ojeda, “Auto- matic modulation classifier for military applications,” in2011 19th European Signal Processing Conference, pp. 1814–1818, 2011

    work page 2011

  3. [3]

    Cogni- tive radio in 5G: a perspective on energy-spectral ef- ficiency trade-off,

    X. Hong, J. Wang, C.-X. Wang, and J. Shi, “Cogni- tive radio in 5G: a perspective on energy-spectral ef- ficiency trade-off,”IEEE Communications Magazine, vol. 52, no. 7, pp. 46–53, 2014

    work page 2014

  4. [4]

    Cognitive-radio-based internet of things: applica- tions, architectures, spectrum related functionalities, and future research directions,

    A. A. Khan, M. H. Rehmani, and A. Rachedi, “Cognitive-radio-based internet of things: applica- tions, architectures, spectrum related functionalities, and future research directions,”IEEE Wireless Com- munications, vol. 24, no. 3, pp. 17–25, 2017

    work page 2017

  5. [5]

    Spectrum sensing based on deep learning classifi- cation for cognitive radios,

    S. Zheng, S. Chen, P. Qi, H. Zhou, and X. Yang, “Spectrum sensing based on deep learning classifi- cation for cognitive radios,”China Communications, vol. 17, no. 2, pp. 138–148, 2020

    work page 2020

  6. [6]

    Digital modula- tion classification based on higher-order moments and characteristic function,

    L. Zhang, Z. Yang, and W. Lu, “Digital modula- tion classification based on higher-order moments and characteristic function,” in2020 IEEE 5th Interna- tional Conference on Signal and Image Processing (ICSIP), pp. 809–812, 2020

    work page 2020

  7. [7]

    Feature image- based automatic modulation classification method us- ing CNN algorithm,

    J. H. Lee, K.-Y . Kim, and Y . Shin, “Feature image- based automatic modulation classification method us- ing CNN algorithm,” in2019 International Confer- ence on Artificial Intelligence in Information and Communication (ICAIIC), pp. 1–4, 2019

    work page 2019

  8. [8]

    Au- tomatic classification of analog modulation schemes,

    H. Xiao, Y . Q. Shi, W. Su, and J. A. Kosinski, “Au- tomatic classification of analog modulation schemes,” in2012 IEEE Radio and Wireless Symposium, pp. 5–8, 2012

    work page 2012

  9. [9]

    Sur- vey of automatic modulation classification techniques: classical approaches and new trends,

    O. A. Dobre, A. Abdi, Y . Bar-Ness, and W. Su, “Sur- vey of automatic modulation classification techniques: classical approaches and new trends,”IET communi- cations, vol. 1, no. 2, pp. 137–156, 2007

    work page 2007

  10. [10]

    The modulation recognition approaches for software radio,

    N. Alyaoui, H. B. Hnia, A. Kachouri, and M. Samet, “The modulation recognition approaches for software radio,” in2008 2nd International Conference on Sig- nals, Circuits and Systems, pp. 1–5, 2008

    work page 2008

  11. [11]

    Likelihood ratio tests for modulation classification,

    P. Panagiotou, A. Anastasopoulos, and A. Polydoros, “Likelihood ratio tests for modulation classification,” inMILCOM 2000 Proceedings. 21st Century Military Communications. Architectures and Technologies for Information Superiority (Cat. No.00CH37155), vol. 2, pp. 670–674 vol.2, 2000

    work page 2000

  12. [12]

    Deep learning based automatic modulation classification for varying snr environment,

    X. Xie, Y . Ni, S. Peng, and Y .-D. Yao, “Deep learning based automatic modulation classification for varying snr environment,” in2019 28th Wireless and Optical Communications Conference (WOCC), pp. 1–5, 2019

    work page 2019

  13. [13]

    Algorithms for automatic modulation recognition of communication signals,

    A. Nandi and E. Azzouz, “Algorithms for automatic modulation recognition of communication signals,” IEEE Transactions on Communications, vol. 46, no. 4, China Communications 13 pp. 431–436, 1998

    work page 1998

  14. [14]

    Automatic modula- tion recognition of communication signals based on instantaneous statistical characteristics and SVM clas- sifier,

    X. Zhang, T. Ge, and z. Chen, “Automatic modula- tion recognition of communication signals based on instantaneous statistical characteristics and SVM clas- sifier,” in2018 IEEE Asia-Pacific Conference on An- tennas and Propagation (APCAP), pp. 344–346, 2018

    work page 2018

  15. [15]

    Parti- cle swarm optimization-based deep neural network for digital modulation recognition,

    W. Shi, D. Liu, X. Cheng, Y . Li, and Y . Zhao, “Parti- cle swarm optimization-based deep neural network for digital modulation recognition,”IEEE Access, vol. 7, pp. 104591–104600, 2019

    work page 2019

  16. [16]

    Automatic modulation classi- fication using deep learning based on sparse autoen- coders with nonnegativity constraints,

    A. Ali and F. Yangyu, “Automatic modulation classi- fication using deep learning based on sparse autoen- coders with nonnegativity constraints,”IEEE Signal Processing Letters, vol. 24, no. 11, pp. 1626–1630, 2017

    work page 2017

  17. [17]

    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

  18. [18]

    Modulation classi- fication by means of different orders statistical mo- ments,

    C. Le Martret and D. Boiteau, “Modulation classi- fication by means of different orders statistical mo- ments,” inMILCOM 97 MILCOM 97 Proceedings, vol. 3, pp. 1387–1391 vol.3, 1997

    work page 1997

  19. [19]

    Automatic modulation classification using statistical moments and a fuzzy classifier,

    J. Lopatka and M. Pedzisz, “Automatic modulation classification using statistical moments and a fuzzy classifier,” inWCC 2000 - ICSP 2000. 2000 5th In- ternational Conference on Signal Processing Proceed- ings. 16th World Computer Congress 2000, vol. 3, pp. 1500–1506 vol.3, 2000

    work page 2000

  20. [20]

    Deep learning,

    Y . Lecun, Y . Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  21. [21]

    Research on image process- ing technology of computer vision algorithm,

    X. Zhang and S. Xu, “Research on image process- ing technology of computer vision algorithm,” in2020 International Conference on Computer Vision, Image and Deep Learning (CVIDL), pp. 122–124, 2020

    work page 2020

  22. [22]

    Dynamic knowledge integration for natural language inference,

    M. Guo, Y . Chen, J. Xu, and Y . Zhang, “Dynamic knowledge integration for natural language inference,” in2022 4th International Conference on Natural Lan- guage Processing (ICNLP), pp. 360–364, 2022

    work page 2022

  23. [23]

    Complex network classification with convolutional neural network,

    R. Xin, J. Zhang, and Y . Shao, “Complex network classification with convolutional neural network,”Ts- inghua Science and Technology, vol. 25, no. 4, pp. 447–457, 2020

    work page 2020

  24. [24]

    Nas-amr: Neural ar- chitecture search-based automatic modulation recog- nition for integrated sensing and communication sys- tems,

    X. Zhang, H. Zhao, H. Zhu, B. Adebisi, G. Gui, H. Gacanin, and F. Adachi, “Nas-amr: Neural ar- chitecture search-based automatic modulation recog- nition for integrated sensing and communication sys- tems,”IEEE Transactions on Cognitive Communica- tions and Networking, vol. 8, no. 3, pp. 1374–1386, 2022

    work page 2022

  25. [25]

    Toward next-generation signal intel- ligence: A hybrid knowledge and data-driven deep learning framework for radio signal classification,

    S. Zheng, X. Zhou, L. Zhang, P. Qi, K. Qiu, J. Zhu, and X. Yang, “Toward next-generation signal intel- ligence: A hybrid knowledge and data-driven deep learning framework for radio signal classification,” IEEE Transactions on Cognitive Communications and Networking, pp. 1–1, 2023

    work page 2023

  26. [26]

    EMD and VMD empowered deep learn- ing for radio modulation recognition,

    T. Chen, S. Gao, S. Zheng, S. Yu, Q. Xuan, C. Lou, and X. Yang, “EMD and VMD empowered deep learn- ing for radio modulation recognition,”IEEE Transac- tions on Cognitive Communications and Networking, vol. 9, no. 1, pp. 43–57, 2023

    work page 2023

  27. [27]

    Automatic modula- tion recognition based on CNN and GRU,

    F. Liu, Z. Zhang, and R. Zhou, “Automatic modula- tion recognition based on CNN and GRU,”Tsinghua Science and Technology, vol. 27, no. 2, pp. 422–431, 2022

    work page 2022

  28. [28]

    Automatic modulation recognition using deep learning architec- tures,

    M. Zhang, Y . Zeng, Z. Han, and Y . Gong, “Automatic modulation recognition using deep learning architec- tures,” in2018 IEEE 19th International Workshop on Signal Processing Advances in Wireless Communica- tions (SPAWC), pp. 1–5, 2018

    work page 2018

  29. [29]

    Design of retransmission mechanism for de- centralized inference with graph neural networks,

    J. Zhang, Y . Jiang, X. Liu, M. Lee, H. Gao, and G. Yu, “Design of retransmission mechanism for de- centralized inference with graph neural networks,” in 2022 27th Asia Pacific Conference on Communica- tions (APCC), pp. 515–519, 2022

    work page 2022

  30. [30]

    Convolu- tional radio modulation recognition networks,

    T. J. O’Shea, J. Corgan, and T. C. Clancy, “Convolu- tional radio modulation recognition networks,”Inter- national Conference on Engineering Applications of Neural Networks, 2016

    work page 2016

  31. [31]

    Over-the- air deep learning based radio signal classification,

    T. J. O’Shea, T. Roy, and T. C. Clancy, “Over-the- air deep learning based radio signal classification,” IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 1, pp. 168–179, 2018

    work page 2018

  32. [32]

    Deep neu- ral network for robust modulation classification under uncertain noise conditions,

    S. Hu, Y . Pei, P. P. Liang, and Y .-C. Liang, “Deep neu- ral network for robust modulation classification under uncertain noise conditions,”IEEE Transactions on Ve- hicular Technology, vol. 69, no. 1, pp. 564–577, 2020

    work page 2020

  33. [33]

    Avgnet: Adaptive visibility graph neu- ral network and its application in modulation classi- fication,

    Q. Xuan, J. Zhou, K. Qiu, Z. Chen, D. Xu, S. Zheng, and X. Yang, “Avgnet: Adaptive visibility graph neu- ral network and its application in modulation classi- fication,”IEEE Transactions on Network Science and Engineering, vol. 9, no. 3, pp. 1516–1526, 2022

    work page 2022

  34. [34]

    Distribution test based low complexity modulation classification in MIMO sys- tems,

    Z. Gao and Z. Zhu, “Distribution test based low complexity modulation classification in MIMO sys- tems,” in2018 10th International Conference on Wire- less Communications and Signal Processing (WCSP), pp. 1–5, 2018

    work page 2018

  35. [35]

    Modulation classifi- cation in MIMO systems with distribution test ensem- ble,

    Z. Gao, Z. Zhu, and A. K. Nandi, “Modulation classifi- cation in MIMO systems with distribution test ensem- ble,”IEEE Access, vol. 8, pp. 128819–128829, 2020

    work page 2020

  36. [36]

    Automatic modulation recognition of digital signal based on auto-encoding network in MIMO system,

    M. Wei, Z. Wei, J. Yang, and L. Sang, “Automatic modulation recognition of digital signal based on auto-encoding network in MIMO system,” in2018 IEEE 18th International Conference on Communica- tion Technology (ICCT), pp. 1017–1021, 2018

    work page 2018

  37. [37]

    Automatic modu- lation classification for MIMO systems via deep learn- ing and zero-forcing equalization,

    Y . Wang, J. Gui, Y . Yin, J. Wang, J. Sun, G. Gui, H. Gacanin, H. Sari, and F. Adachi, “Automatic modu- lation classification for MIMO systems via deep learn- ing and zero-forcing equalization,”IEEE Transactions 14 China Communications on Vehicular Technology, vol. 69, no. 5, pp. 5688– 5692, 2020

    work page 2020

  38. [38]

    Automatic modulation recognition method for multiple antenna system based on convolutional neural network,

    J. Wang, Y . Wang, W. Li, G. Gui, H. Gacanin, and F. Adachi, “Automatic modulation recognition method for multiple antenna system based on convolutional neural network,” in2020 IEEE 92nd Vehicular Tech- nology Conference (VTC2020-Fall), pp. 1–5, 2020

    work page 2020

  39. [39]

    Deep learning-based automatic modulation classifi- cation over MIMO keyhole channels,

    P. Dileep, A. Singla, D. Das, and P. K. Bora, “Deep learning-based automatic modulation classifi- cation over MIMO keyhole channels,”IEEE Access, vol. 10, pp. 119566–119574, 2022

    work page 2022

  40. [40]

    A semi-supervised modulation identification in MIMO systems: A deep learning strategy,

    S. Bouchenak, R. Merzougui, F. Harrou, A. Dairi, and Y . Sun, “A semi-supervised modulation identification in MIMO systems: A deep learning strategy,”IEEE Access, vol. 10, pp. 76622–76635, 2022

    work page 2022

  41. [41]

    Deep learning-based cooperative automatic modula- tion classification method for MIMO systems,

    Y . Wang, J. Wang, W. Zhang, J. Yang, and G. Gui, “Deep learning-based cooperative automatic modula- tion classification method for MIMO systems,”IEEE Transactions on Vehicular Technology, vol. 69, no. 4, pp. 4575–4579, 2020

    work page 2020

  42. [42]

    Au- tomatic modulation recognition: A few-shot learning method based on the capsule network,

    L. Li, J. Huang, Q. Cheng, H. Meng, and Z. Han, “Au- tomatic modulation recognition: A few-shot learning method based on the capsule network,”IEEE Wireless Communications Letters, vol. 10, no. 3, pp. 474–477, 2021

    work page 2021

  43. [43]

    Amcrn: Few-shot learning for auto- matic modulation classification,

    Q. Zhou, R. Zhang, J. Mu, H. Zhang, F. Zhang, and X. Jing, “Amcrn: Few-shot learning for auto- matic modulation classification,”IEEE Communica- tions Letters, vol. 26, no. 3, pp. 542–546, 2022

    work page 2022

  44. [44]

    Data augmentation for deep learning-based radio modulation classification,

    L. Huang, W. Pan, Y . Zhang, L. Qian, N. Gao, and Y . Wu, “Data augmentation for deep learning-based radio modulation classification,”IEEE Access, vol. 8, pp. 1498–1506, 2020

    work page 2020

  45. [45]

    Deep learning based cp-ofdm signal classification with data augmentation,

    J. Combo, A. Tato, J. J. Escudero-Garz ´as, L. P. Roca, and P. Gonz´alez, “Deep learning based cp-ofdm signal classification with data augmentation,” in2022 IEEE International Black Sea Conference on Communica- tions and Networking (BlackSeaCom), pp. 352–357, 2022

    work page 2022

  46. [46]

    Data augmentation for signal modulation classifi- cation using generative adverse network,

    Z. Tang, M. Tao, J. Su, Y . Gong, Y . Fan, and T. Li, “Data augmentation for signal modulation classifi- cation using generative adverse network,” in2021 IEEE 4th International Conference on Electronic In- formation and Communication Technology (ICEICT), pp. 450–453, 2021

    work page 2021

  47. [47]

    Digital signal modulation classification with data augmentation us- ing generative adversarial nets in cognitive radio net- works,

    B. Tang, Y . Tu, Z. Zhang, and Y . Lin, “Digital signal modulation classification with data augmentation us- ing generative adversarial nets in cognitive radio net- works,”IEEE Access, vol. 6, pp. 15713–15722, 2018

    work page 2018

  48. [48]

    Joint transmit beamforming for mul- tiuser mimo communications and MIMO radar,

    X. Liu, T. Huang, N. Shlezinger, Y . Liu, J. Zhou, and Y . C. Eldar, “Joint transmit beamforming for mul- tiuser mimo communications and MIMO radar,”IEEE Transactions on Signal Processing, vol. 68, pp. 3929– 3944, 2020

    work page 2020

  49. [49]

    A new nested MIMO array with increased degrees of freedom and hole-free difference coarray,

    M. Yang, L. Sun, X. Yuan, and B. Chen, “A new nested MIMO array with increased degrees of freedom and hole-free difference coarray,”IEEE Signal Processing Letters, vol. 25, no. 1, pp. 40–44, 2018

    work page 2018

  50. [50]

    Performance analysis of signal-to-noise ratio estimators in awgn and fad- ing channels,

    L. Gopal and M. L. Sim, “Performance analysis of signal-to-noise ratio estimators in awgn and fad- ing channels,” in2008 6th National Conference on Telecommunication Technologies and 2008 2nd Malaysia Conference on Photonics, pp. 300–304, 2008

    work page 2008

  51. [51]

    Big data processing architecture for ra- dio signals empowered by deep learning: Concept, ex- periment, applications and challenges,

    S. Zheng, S. Chen, L. Yang, J. Zhu, Z. Luo, J. Hu, and X. Yang, “Big data processing architecture for ra- dio signals empowered by deep learning: Concept, ex- periment, applications and challenges,”IEEE Access, vol. 6, pp. 55907–55922, 2018

    work page 2018

  52. [52]

    Deepreceiver: A deep learning-based intelligent receiver for wireless communications in the physical layer,

    S. Zheng, S. Chen, and X. Yang, “Deepreceiver: A deep learning-based intelligent receiver for wireless communications in the physical layer,”IEEE Transac- tions on Cognitive Communications and Networking, vol. 7, no. 1, pp. 5–20, 2021

    work page 2021

  53. [53]

    Automatic modula- tion classification based on the improved alexnet,

    Z. Li, Z. Jiang, and J. Huang, “Automatic modula- tion classification based on the improved alexnet,” in 2021 International Wireless Communications and Mo- bile Computing (IWCMC), pp. 2068–2073, 2021

    work page 2021

  54. [54]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    K. Simonyan and A. Zisserman, “Very deep convo- lutional networks for large-scale image recognition,” CoRR, vol. abs/1409.1556, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  55. [55]

    A spatiotemporal multi-channel learning framework for automatic mod- ulation recognition,

    J. Xu, C. Luo, G. Parr, and Y . Luo, “A spatiotemporal multi-channel learning framework for automatic mod- ulation recognition,”IEEE Wireless Communications Letters, vol. 9, no. 10, pp. 1629–1632, 2020

    work page 2020

  56. [56]

    Generative adversarial networks,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Ben- gio, “Generative adversarial networks,” 2014. China Communications 15

    work page 2014