基于少样本学习的无线通信信号调制模式识别研究

调制模式识别是无线通信中的关键技术，其目的在于根据获取的信号识别判 断其使用的调制模式，以便对其进行正确的解调以及其他后续操作。近些年深度学 习（Deep Learning，DL）快速发展，由于其简单、便捷、快速等特点，相关的技术 被广泛用于自动调制模式识别领域。现今，一些基于深度学习的自动调制模式识别 算法已经实现了信号调制模式的高精度识别。但是，作为取得高准确率的代价，相 关算法的训练过程高度依赖于具有大量高质量样本的数据集。然而在实际场景中， 充足的训练数据往往不易获取，随着数据量的减少，这些算法的精度随之下降，在 1 ∼ 10 个样本的情境下，甚至出现模型不可用的情况。基于此，本文中我们基于与 模型无关的元学习（Model Agnostic Meta Learning，MAML）算法提出一种适用于 少样本情景的自动调制模式识别框架，我们称其为通道重要性可学习的少样本调 制识别（Channel Importance Learnable Few-shot Modulation Recognition，CILFMR） 框架。我们在 CILFMR 中设计了更加适合于少样本情境的无线 IQ 信号的特征提取 模块，提高了框架整体的识别精度。此外我们设计了单神经元分类模块和特征变换 模块，分别用于减小模型在测试阶段对数据标签排列的敏感性以及调整由于任务 分布偏移造成的特征通道的权重偏差。我们使用公开数据集 RadioML2016.10a 对 CILFMR 的性能进行测试。实验证明，在 SNR>4dB 时，CILFMR 的识别准确率高 于 90%。在少样本条件下，相较于基于深度学习的自动调制识别的基线方法，我 们提出的 CILFMR 框架性能突出。

[1] DOBRE O, ABDI A, BAR-NESS Y, et al. Survey of automatic modulation classification techniques: Classical approaches and new trends[J/OL]. IET Communications, 2007, 1(2): 137-156. DOI: 10.1049/iet-com:20050176.
[2] BHATTI F A, KHAN M J, SELIM A, et al. Shared Spectrum Monitoring Using Deep Learning[J/OL]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(4): 1171-1185. DOI: 10.1109/TCCN.2021.3071149.
[3] ZHANG W, FENG M, KRUNZ M, et al. Signal Detection and Classification in Shared Spectrum: A Deep Learning Approach[C/OL]//IEEE INFOCOM 2021 - IEEE Conference on Computer Communications. 2021: 1-10. DOI: 10.1109/INFOCOM42981.2021.9488834.
[4] WEI W, MENDEL J. Maximum-likelihood classification for digital amplitude-phase modulations[J/OL]. IEEE Transactions on Communications, 2000, 48(2): 189-193. DOI: 10.1109/26.823550.
[5] DOBRE O A, HAMEED F. Likelihood-Based Algorithms for Linear Digital Modulation Classification in Fading Channels[C/OL]//2006 Canadian Conference on Electrical and Computer Engineering. 2006: 1347-1350. DOI: 10.1109/CCECE.2006.277525.
[6] YANG Y, LIU C H. An asymptotic optimal algorithm for modulation classification[J/OL]. IEEE Communications Letters, 1998, 2(5): 117-119. DOI: 10.1109/4234.673652.
[7] O’SHEA T, CORGAN J, CLANCY T C. Convolutional Radio Modulation Recognition Networks: abs/1602.04105[A]. 2016.
[8] ZENG Y, ZHANG M, HAN F, et al. Spectrum Analysis and Convolutional Neural Network for Automatic Modulation Recognition[J/OL]. IEEE Wireless Communications Letters, 2019, 8(3): 929-932. DOI: 10.1109/LWC.2019.2900247.
[9] XU J, LUO C, PARR G P, et al. A Spatiotemporal Multi-Channel Learning Framework forAutomatic Modulation Recognition[J]. IEEE Wireless Communications Letters, 2020, 9: 1629-1632.
[10] LI L, HUANG J, CHENG Q, et al. Automatic Modulation Recognition: A Few-Shot Learning Method Based on the Capsule Network[J/OL]. IEEE Wireless Communications Letters, 2021, 10(3): 474-477. DOI: 10.1109/LWC.2020.3034913.
[11] VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching Networks for One Shot Learning [M/OL]. arXiv, 2016. DOI: 10.48550/ARXIV.1606.04080.
[12] SNELL J, SWERSKY K, ZEMEL R S. Prototypical Networks for Few-shot Learning[M/OL]. arXiv, 2017. DOI: 10.48550/ARXIV.1703.05175.
[13] SUNG F, YANG Y, ZHANG L, et al. Learning to Compare: Relation Network for Few-Shot Learning[M/OL]. arXiv, 2017. DOI: 10.48550/ARXIV.1711.06025.
[14] FINN C, ABBEEL P, LEVINE S. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks[M/OL]. arXiv, 2017. DOI: 10.48550/ARXIV.1703.03400.
[15] NICHOL A, ACHIAM J, SCHULMAN J. On First-Order Meta-Learning Algorithms[M/OL]. arXiv, 2018. DOI: 10.48550/ARXIV.1803.02999.
[16] Q, ZHANG R, MU J, et al. AMCRN: Few-Shot Learning for Automatic Modulation Classification[J/OL]. IEEE Communications Letters, 2022, 26(3): 542-546. DOI: 10.1109/LCOMM. 2021.3135688.
[17] SILLS J. Maximum-likelihood modulation classification for PSK/QAM[C/OL]//MILCOM 1999. IEEE Military Communications. Conference Proceedings (Cat. No.99CH36341): volume 1. 1999: 217-220 vol.1. DOI: 10.1109/MILCOM.1999.822675.
[18] ABDI A, DOBRE O, CHOUDHRY R, et al. Modulation classification in fading channels using antenna arrays[C/OL]//IEEE MILCOM 2004. Military Communications Conference, 2004.: volume 1. 2004: 211-217 Vol. 1. DOI: 10.1109/MILCOM.2004.1493271.
[19] LAY N, POLYDOROS A. Modulation classification of signals in unknown ISI environments [C/OL]//Proceedings of MILCOM ’95: volume 1. 1995: 170-174 vol.1. DOI: 10.1109/MILCOM.1995.483293.
[20] PANAGIOTOU P, ANASTASOPOULOS A, POLYDOROS A. Likelihood ratio tests for modulation classification[C/OL]//MILCOM 2000 Proceedings. 21st Century Military Communications. Architectures and Technologies for Information Superiority (Cat. No.00CH37155): volume 2. 2000: 670-674 vol.2. DOI: 10.1109/MILCOM.2000.904013.
[21] HONG L, HO K. BPSK and QPSK modulation classification with unknown signal level[C/OL] MILCOM 2000 Proceedings. 21st Century Military Communications. Architectures and Technologies for Information Superiority (Cat. No.00CH37155): volume 2. 2000: 976-980 vol.2. DOI: 10.1109/MILCOM.2000.904076.
[22] HONG L, HO K. Modulation classification of BPSK and QPSK signals using a two element antenna array receiver[C/OL]//2001 MILCOM Proceedings Communications for Network-Centric Operations: Creating the Information Force (Cat. No.01CH37277): volume 1. 2001: 118-122 vol.1. DOI: 10.1109/MILCOM.2001.985774.
[23] HONG L, HO K. Antenna array likelihood modulation classifier for BPSK and QPSK signals [C/OL]//MILCOM 2002. Proceedings: volume 1. 2002: 647-651 vol.1. DOI: 10.1109/MILCOM.2002.1180521.
[24] HSUE S Z, SOLIMAN S. Automatic modulation recognition of digitally modulated signals [C/OL]//IEEE Military Communications Conference, ’Bridging the Gap. Interoperability, Survivability, Security’. 1989: 645-649 vol.3. DOI: 10.1109/MILCOM.1989.104004.
[25] SWAMI A, SADLER B. Hierarchical digital modulation classification using cumulants[J/OL]. IEEE Transactions on Communications, 2000, 48(3): 416-429. DOI: 10.1109/26.837045.
[26] SAFAVIAN S R, LANDGREBE D A. A survey of decision tree classifier methodology[J].IEEE Trans. Syst. Man Cybern., 1991, 21: 660-674.
[27] SCHöLKOPF B, TSUDA K, VERT J. Advanced Application of Support Vector Machines[M]. Kernel Methods in Computational Biology, 2004.
[28] SOLTANIEH N, NOROUZI Y, YANG Y, et al. A Review of Radio Frequency Fingerprinting Techniques[J/OL]. IEEE Journal of Radio Frequency Identification, 2020, 4(3): 222-233. DOI: 10.1109/JRFID.2020.29683690.
[29] YE H, LI G Y, JUANG B H. Power of Deep Learning for Channel Estimation and SignalDetection in OFDM Systems[J/OL]. IEEE Wireless Communications Letters, 2018, 7(1): 114-117. DOI: 10.1109/LWC.2017.2757490.
[30] WANG C, WANG J, ZHANG X. Automatic radar waveform recognition based on timefrequency analysis and convolutional neural network[C/OL]//2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2017: 2437-2441. DOI:10.1109/ICASSP.2017.7952594.
[31] O’SHEA T, WEST N. Radio Machine Learning Dataset Generation with GNU Radio[J]. Proceedings of the GNU Radio Conference, 2016, 1(1).
[32] WEST N E, O’SHEA T. Deep architectures for modulation recognition[M/OL]//2017 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). 2017: 1-6. DOI: 10.1109/DySPAN.2017.7920754.
[33] ZHANG M, ZENG Y, HAN Z, et al. Automatic Modulation Recognition Using Deep Learning Architectures[C/OL]//2018 IEEE 19th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC). 2018: 1-5. DOI: 10.1109/SPAWC.2018.8446021.
[34] CHEN T, GAO S, ZHENG S, et al. EMD and VMD Empowered Deep Learning for Radio Modulation Recognition[J/OL]. IEEE Transactions on Cognitive Communications and Networking, 2023, 9(1): 43-57. DOI: 10.1109/TCCN.2022.3218694.
[35] LIN S, ZENG Y, GONG Y. Modulation Recognition Using Signal Enhancement and Multistage Attention Mechanism[J/OL]. IEEE Transactions on Wireless Communications, 2022, 21(11): 9921-9935. DOI: 10.1109/TWC.2022.3181026.
[36] CHEN Z, CUI H, XIANG J, et al. SigNet: An Advanced Deep Learning Framework for Radio Signal Classification: abs/2011.03525[A]. 2020.
[37] WANG Y, BAI J, XIAO Z, et al. MsmcNet: A Modular Few-Shot Learning Framework for Signal Modulation Classification[J/OL]. IEEE Transactions on Signal Processing, 2022, 70:3789-3801. DOI: 10.1109/TSP.2022.3191783.
[38] CHE J, WANG L, BAI X, et al. Spatial-Temporal Hybrid Feature Extraction Network for FewShot Automatic Modulation Classification[J/OL]. IEEE Transactions on Vehicular Technology, 2022, 71(12): 13387-13392. DOI: 10.1109/TVT.2022.3196103.
[39] CHEN Z, FU Y, ZHANG Y, et al. Semantic Feature Augmentation in Few-shot Learning:abs/1804.05298[A]. 2018.
[40] ZHOU F, WU B, LI Z. Deep Meta-Learning: Learning to Learn in the Concept Space:abs/1802.03596[A]. 2018.
[41] SANTORO A, BARTUNOV S, BOTVINICK M, et al. Meta-Learning with MemoryAugmented Neural Networks[C]//ICML’16: Proceedings of the 33rd International Conference on International Conference on Machine Learning - Volume 48. New York, NY, USA:JMLR.org, 2016: 1842–1850.
[42] JIA Y, DARRELL T. Latent Task Adaptation with Large-Scale Hierarchies[C/OL]//2013 IEEE International Conference on Computer Vision. 2013: 2080-2087. DOI: 10.1109/ICCV.2013. 260.
[43] MEHROTRA A, DUKKIPATI A. Generative Adversarial Residual Pairwise Networks for One Shot Learning[M]//Computer Vision and Pattern Recognition (CVPR). 2017.
[44] KIMURA A, GHAHRAMANI Z, TAKEUCHI K, et al. Few-shot learning of neural networks from scratch by pseudo example optimization[A]. 2018. arXiv: 1802.03039.
[45] CHEN Z, FU Y, WANG Y X, et al. Image Deformation Meta-Networks for One-Shot Learning [C/OL]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).2019: 8672-8681. DOI: 10.1109/CVPR.2019.00888.
[46] LAMPERT C H, NICKISCH H, HARMELING S. Learning to detect unseen object classes by between-class attribute transfer[C/OL]//2009 IEEE Conference on Computer Vision and Pattern Recognition. 2009: 951-958. DOI: 10.1109/CVPR.2009.5206594.
[47] KOCH G R. Siamese Neural Networks for One-Shot Image Recognition[C]//2015.
[48] ANTONIOU A, EDWARDS H, STORKEY A. How to train your MAML[M/OL]. arXiv, 2018. https://arxiv.org/abs/1810.09502. DOI: 10.48550/ARXIV.1810.09502.
[49] RAGHU A, RAGHU M, BENGIO S, et al. Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML[M/OL]. arXiv, 2019. https://arxiv.org/abs/1909.09157. DOI: 10.48550/ARXIV.1909.09157.
[50] YE H J, CHAO W L. How to Train Your MAML to Excel in Few-Shot Classification[M/OL]. arXiv, 2021. https://arxiv.org/abs/2106.16245. DOI: 10.48550/ARXIV.2106.16245.
[51] KAO C H, CHIU W C, CHEN P Y. MAML is a Noisy Contrastive Learner in Classification[M/OL]. arXiv, 2021. https://arxiv.org/abs/2106.15367. DOI: 10.48550/ARXIV.2106.15367.
[52] YAO H, ZHANG L, FINN C. Meta-Learning with Fewer Tasks through Task Interpolation [M/OL]. arXiv, 2021. https://arxiv.org/abs/2106.02695. DOI: 10.48550/ARXIV.2106.02695.
[53] BAIK S, CHOI J, KIM H, et al. Meta-Learning with Task-Adaptive Loss Function for Few-Shot Learning[M/OL]. arXiv, 2021. https://arxiv.org/abs/2110.03909. DOI: 10.48550/ARXIV.2110.03909.
[54] KINGMA D P, BA J. Adam: A Method for Stochastic Optimization[J]. CoRR, 2014, abs/1412.6980.
[55] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: Visual Explanations fromDeep Networks via Gradient-Based Localization[C/OL]//2017 IEEE International Conference on Computer Vision (ICCV). 2017: 618-626. DOI: 10.1109/ICCV.2017.74.
[56] LUO X, XU J, XU Z. Channel Importance Matters in Few-Shot Image Classification[M/OL]. arXiv, 2022. https://arxiv.org/abs/2206.08126. DOI: 10.48550/ARXIV.2206.08126.
[57] LOSHCHILOV I, HUTTER F. SGDR: Stochastic Gradient Descent with Warm Restarts[A].2016.