|
|
Optoelectronics 2025
深度学习驱动的盲信号分离:技术进展与展望
|
Abstract:
盲信号分离(BSS)旨在无需先验信息的前提下,从混合信号中提取独立源信号。传统方法如独立成分分析(ICA)和非负矩阵分解(NMF)受限于线性假设与统计依赖性,难以应对复杂场景中的非线性混合、动态环境适应及高噪声鲁棒性问题。近年来,深度学习技术通过数据驱动策略与非线性建模能力,显著提升了BSS的性能,尤其在时频域分离(如卷积神经网络CNN)、全局特征建模(如Transformer架构)及多模态融合中取得突破。然而,模型泛化能力不足、计算复杂度高、可解释性缺失及隐私安全风险仍是主要挑战。本文系统梳理了深度学习赋能BSS的技术路径,分析了现存瓶颈,并展望了多模态深度融合、自监督学习及边缘计算部署等未来方向,为推动BSS在医疗、通信等领域的实用化提供参考。
Blind Signal Separation (BSS) aims to extract independent source signals from mixed signals without a priori information. Traditional methods such as Independent Component Analysis (ICA) and Nonnegative Matrix Factorization (NMF) are limited by linear assumptions and statistical dependencies, which make it difficult to cope with nonlinear mixing, dynamic environment adaptation, and high-noise robustness problems in complex scenes. In recent years, deep learning techniques have significantly improved the performance of BSS through data-driven strategies and nonlinear modeling capabilities, especially making breakthroughs in time-frequency domain separation (e.g., convolutional neural network CNN), global feature modeling (e.g., Transformer architecture), and multimodal fusion. However, insufficient model generalization capability, high computational complexity, lack of interpretability, and privacy and security risks are still the main challenges. In this paper, we systematically sort out the technical paths of deep learning-enabled BSS, analyze the existing bottlenecks, and look forward to the future directions of deep multimodal fusion, self-supervised learning, and edge computing deployments, which will provide references to promote the utility of BSS in the fields of healthcare, communication, and so on.
| [1] | Dian 何岭松. 《数字信号分析理论与实践》教学案例集之: 信号的盲源分离和ICA独立成分分析[EB/OL]. https://zhuanlan.zhihu.com/p/692594809, 2025-05-06. |
| [2] | 何丹丹, 刘润杰, 申金媛, 陈园园. 盲信号分离算法综述[J]. 光电子学, 2011(1): 1. |
| [3] | Kurisu, D., Fukami, R. and Koike, Y. (2022) Adaptive Deep Learning for Nonlinear Time Series Models. https://arxiv.org/abs/2207.02546 |
| [4] | Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J., Carvalhais, N., et al. (2019) Deep Learning and Process Understanding for Data-Driven Earth System Science. Nature, 566, 195-204. https://doi.org/10.1038/s41586-019-0912-1 |
| [5] | Hemanth, D.J. (2021) Automated Feature Extraction in Deep Learning Models: A Boon or a Bane? 2021 8th International Conference on Electrical Engineering, Computer Science and Informatics (EECSI), Semarang, 20-21 October 2021, 3. https://doi.org/10.23919/eecsi53397.2021.9624287 |
| [6] | Reddy, P., Wisdom, S., Greff, K., Hershey, J.R. and Kipf, T. (2023) AudioSlots: A Slot-Centric Generative Model for Audio Separation. http://arxiv.org/abs/2305.05591 |
| [7] | 王景景, 李爽, 杨星海, 吴承安, 郑轶, 鄢社锋, 乔钢, 施威, 张祥光, 郭瑛, 李海涛. 基于机器学习和FFT的盲源分离信源数目并行估计方法[P]. 中国, CN112861066B. 2022-05-17. |
| [8] | 邱媛媛, 王皓辰. 基于深度学习的领域自适应目标检测算法研究[J]. 人工智能与机器人研究, 2024, 13(3): 503-514. |
| [9] | Thompson, N.C., Greenewald, K., Lee, K. and Manso, G.F. (2022) The Computational Limits of Deep Learning. http://arxiv.org/abs/2007.05558 |
| [10] | Li, Y., Daho, M.E.H., Conze, P.H., Zeghlache, R., Boité, H.L., Tadayoni, R., et al. (2024) A Review of Deep Learning-Based Information Fusion Techniques for Multimodal Medical Image Classification. http://arxiv.org/abs/2404.15022 |
| [11] | Detection de grandeurs primitives dans un message composite par une architeture de calcul neuromimetique en apprentissage non supervise. https://www.researchgate.net/publication/27606213_Detection_de_grandeurs_primitives_dans_un_message_composite_par_une_architeture_de_calcul_neuromimetique_en_apprentissage_non_supervise |
| [12] | Jain, S. and Rai, D. (2012) Blind Source Separation and ICA Techniques: A Review. International Journal of Environmental Science and Technology, 4, 1490-1503. |
| [13] | 苗水清, 闫文耀, 张静, 吴梦蝶. 基于FFT和PCA的图像压缩方法研究[J]. 佳木斯大学学报(自然科学版), 2022, 40(3): 29-32. |
| [14] | 殷玉玲, 罗兰花. 高维数据降维算法综述[J]. 电脑知识与技术, 2025, 21(6): 12-14, 26. |
| [15] | Sadkhan, S.B. and Abbas, N.A. (2009) Higher Order Statistics and Their Roles in Blind Source Separation (BSS). MASAUM Journal of Computing, 1, 227-234. |
| [16] | Lv, S. and Zhang, C. (2014) Blind Signal Separation for Speech Signals with Noise. 2014 IEEE International Conference on Mechatronics and Automation, Tianjin, 3-6 August 2014, 1850-1855. https://doi.org/10.1109/icma.2014.6885983 |
| [17] | 陆维, 吴锡. 基于深度学习的复杂天气场景交通标志检测[J]. 软件导刊, 2025, 24(6): 175-184. |
| [18] | 深度学习: 数据特征与复杂模式的探索[EB/OL]. 百度开发者中心. https://developer.baidu.com/article/details/1847863, 2025-06-10. |
| [19] | 沈越泓, 苏巧, 袁志刚, 简伟, 黄葆华, 魏以民. 一种分离时频域混合信号的方法[P]. 中国, CN103870875A. 2014-06-18. |
| [20] | 毛琳, 任凤至, 杨大伟, 张汝波. 时频域联合全景分割卷积神经网络及应用[P]. 中国, CN113536905A. 2021-10-22. |
| [21] | Gao, X. and Gao, R. (2025) Music Signal Recognition Aids Based on Convolutional Neural Networks in Music Education. Systems and Soft Computing, 7, Article ID: 200219. https://doi.org/10.1016/j.sasc.2025.200219 |
| [22] | 孙林慧, 王春艳, 张蒙. 基于全卷积神经网络多任务学习的时域语音分离[J]. 信号处理, 2024, 40(12): 2228-2237. |
| [23] | 李端, 张洪欣, 刘知青, 黄菊香, 王田. 基于深度残差卷积神经网络的心电信号心律不齐识别[J]. 生物医学工程学杂志, 2019, 36(2): 189-198. |
| [24] | Li, J., Yan, Y., Liao, S., Yang, X. and Shao, L. (2021) Local-to-Global Self-Attention in Vision Transformers. http://arxiv.org/abs/2107.04735 |
| [25] | Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L. and Polosukhin, I. (2023) Attention Is All You Need. Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, 4-9 December 2017, 6000-6010. http://arxiv.org/abs/1706.03762 |
| [26] | 周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(6): 1229-1251. |
| [27] | 赖鸣姝. 基于Transformer的自然语言处理模型综述[J]. 人工智能与机器人研究, 2023, 12(3): 219. |
| [28] | Subakan, C., Ravanelli, M., Cornell, S., Bronzi, M. and Zhong, J. (2021) Attention Is All You Need in Speech Separation. http://arxiv.org/abs/2010.13154 |
| [29] | Fan, C., Yuan, X. and Zhang, Y. (2019) CNN-Based Signal Detection for Banded Linear Systems. IEEE Transactions on Wireless Communications, 18, 4394-4407. https://doi.org/10.1109/twc.2019.2924424 |
| [30] | Ranganath, A., Muñoz, J.O., Smith, R., Singhal, M. and Marcia, R. (2024) Image Separation Using Transformer Attention Models. 2024 IEEE International Conference on Future Machine Learning and Data Science (FMLDS), Sydney, 20-23 November 2024, 331-336. https://doi.org/10.1109/fmlds63805.2024.00066 |
| [31] | Wang, H., Wu, X., Huang, Z. and Xing, E.P. (2020) High Frequency Component Helps Explain the Generalization of Convolutional Neural Networks. http://arxiv.org/abs/1905.13545 |
| [32] | Alijani, S., Fayyad, J. and Najjaran, H. (2024) Vision Transformers in Domain Adaptation and Domain Generalization: A Study of Robustness. Neural Computing and Applications, 36, 17979-18007. https://doi.org/10.1007/s00521-024-10353-5 |
| [33] | Guo, P., Yu, M., Shen, L., Lin, Z., An, K. and Wang, J. (2024) Single-Channel Blind Source Separation in Wireless Communications: A Complex-Domain Deep Learning Approach. IEEE Wireless Communications Letters, 13, 1645-1649. https://doi.org/10.1109/lwc.2024.3384813 |
| [34] | Guo, P., Yao, F., Yu, M., Li, C., Tang, Y. and Ning, Z. (2025) Single-Channel Blind Source Separation Empowered Joint Transceiver Optimization for Wireless Communications Using Deep Learning. Digital Communications and Networks. https://doi.org/10.1016/j.dcan.2025.04.008 |
| [35] | 刘学博, 户保田, 陈科海, 张民. 大模型关键技术与未来发展方向——从ChatGPT谈起[J]. 中国科学基金, 2023, 37(5): 758-766. |
| [36] | 陈建廷, 向阳. 深度神经网络训练中梯度不稳定现象研究综述[J]. 软件学报, 2018, 29(7): 2071-2091. |
| [37] | Yue, X.B. (2015) The Influence of the Amount of Parameters in Different Layers on the Performance of Deep Learning Models. Computer Science and Application, 5, 445-453. https://doi.org/10.12677/csa.2015.512056 |
| [38] | 李宁, 陈海庭. 欠定条件下弱稀疏源信号混合矩阵盲估计[J]. 数据采集与处理, 2015, 30(4): 793-801. |
| [39] | Ruiz, H., Jarman, I.H., Martin, J.D., Ortega-Martorell, S., Vellido, A., Romero, E., et al. (2012). Towards Interpretable Classifiers with Blind Signal Separation. The 2012 International Joint Conference on Neural Networks (IJCNN), Brisbane, 10-15 June 2012, 1-7. https://doi.org/10.1109/ijcnn.2012.6252783 |
| [40] | 化盈盈, 张岱墀, 葛仕明. 深度学习模型可解释性的研究进展[J]. 信息安全学报, 2020, 5(3): 1-12. |
| [41] | 解元, 张旭, 邹涛, 马鸽, 孙为军. 面向带混响和噪声环境的心肺音混合信号盲分离[J]. 信息与控制, 2025, 54(1): 150-160. |
| [42] | 汪旭童, 尹捷, 刘潮歌, 徐辰晨, 黄昊, 王志, 张方娇. 神经网络后门攻击与防御综述[J]. 计算机学报, 2024, 47(8): 1713-1743. |
| [43] | 周纯毅, 陈大卫, 王尚, 付安民, 高艳松. 分布式深度学习隐私与安全攻击研究进展与挑战[J]. 计算机研究与发展, 2021, 58(5): 927-943. |
| [44] | 王文萱, 汪成磊, 齐慧慧, 叶梦昊, 张艳宁. 面向深度模型的对抗攻击与对抗防御技术综述[J]. 信号处理, 2025, 41(2): 198-223. |
| [45] | Webster, M.B., Lee, D. and Lee, J. (2025) Self-Supervised Autoencoder Network for Robust Heart Rate Extraction from Noisy Photo-Plethysmogram: Applying Blind Source Separation to Biosignal Analysis. http://arxiv.org/abs/2504.09132 |
| [46] | Mosqueira-Rey, E., Hernández-Pereira, E., Bobes-Bascarán, J., Alonso-Ríos, D., Pérez-Sánchez, A., Fernández-Leal, Á., et al. (2023) Addressing the Data Bottleneck in Medical Deep Learning Models Using a Human-in-the-Loop Machine Learning Approach. Neural Computing and Applications, 36, 2597-2616. https://doi.org/10.1007/s00521-023-09197-2 |
| [47] | Obradovic, D. and Deco, G. (1997) Unsupervised Learning for Blind Source Separation: An Information-Theoretic Approach. 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1, 127-130. https://doi.org/10.1109/icassp.1997.599567 |
| [48] | 杨轩, 王子颖, 张力, 赵恒, 洪弘. 基于盲源分离的多人呼吸信号检测方法[J]. 雷达学报(中英文), 2025, 14(1): 117-134. |
| [49] | 徐光辉, 侯芳芳. 一种用于安防的图像识别方法、系统、终端及存储介质[P]. 中国, CN111832478A. 2020-10-27. |
| [50] | 刘建伟, 丁熙浩, 罗雄麟. 多模态深度学习综述[J]. 计算机应用研究, 2020, 37(6): 1601-1614. |
| [51] | 杨维娜, 裴以建, 蔡光卉, 肖敏. 基于信息论的盲源信号分离[J]. 云南大学学报(自然科学版), 2008, 30(5): 460-464, 471. |
| [52] | Levin, D.N. (2008) Using State Space Differential Geometry for Nonlinear Blind Source Separation. Journal of Applied Physics, 103, Article ID: 044906. https://doi.org/10.1063/1.2826943 |
| [53] | Wang, T., Yang, F. and Yang, J. (2022) Convolutive Transfer Function-Based Multichannel Nonnegative Matrix Factorization for Overdetermined Blind Source Separation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 30, 802-815. https://doi.org/10.1109/taslp.2022.3145304 |