To address modulation recognition of multi-source time-frequency overlapped signals in non-cooperative wideband reception scenarios this issue, a multi-feature fusion-based modulation recognition method is proposed. Firstly, an improved DeepLabV3+ architecture is adopted, where the core innovation is embedding deformable atrous convolution into the atrous spatial pyramid pooling module, thereby enhancing dense focusing capability on signal energy core regions. Secondly, in the network multi-modal features are integrated simultaneously from both time-frequency and time-domain representations, improving recognition robustness through cross-modal feature complementarity. Finally, an end-to-end joint optimization framework is constructed to achieve efficient collaborative learning of multiple features. Experimental results demonstrate that under conditions of high aliasing and low signal-to-noise ratio, the average recognition accuracy of this method for 9 types of single signals and 21 types of aliase dsignals reaches 97.4%, which is 15.1 percentage points higher than that of the single-modal method, which validates the superiority of the multi-modal fusionstrategy.
为系统验证MFRS方法在多模态特征融合上的优越性,本文在对识别准确率开展实验的同时也对比了参数量、时延以及类内识别准确率,本文选取六类代表性方法作为基准:多模态融合TRMM[19]、ICA-WPD联合分离法[20]、SegNet[6]、DCNN[7]、支持向量机高维映射(High-Dimensional Mapping in Support Vector Machine, SVM)[21]及TF-ResNet时频融合架构[22]。
WUJ P, HANY Z, ZHANGJ M, et al. Automatic modulation recognition of digital communication signals[C]∥Proceedings of the 2010 First International Conference on Pervasive Computing, Signal Processing and Applications. Piscataway, USA: IEEE, 2010:590-593.
[2]
WANGY, WANGJ, ZHANGW, et al. Deep learning-based cooperative automatic modulation classification method for MIMO systems[J]. IEEE Transactions on Vehicular Technology, 2020,69(4):4575-4579.
CHENGQ, GONGK X, ZHANGM, et al. An efficient wide-band signal detection and extraction method[J]. MATEC Web of Conferences, 2021,336:No.04011.
[5]
李君超.基于轻量级网络的时频混叠调制信号识别[D].成都:电子科技大学,2022:1-76.
[6]
潘宁.复杂电磁环境下时频混叠信号的识别与分离[D].郑州:郑州大学,2021:1-70.
[7]
LIUZ P, LIL C, XUH Y, et al. A method for recognition and classification for hybrid signals based on deep convolutional neural network[C]∥Proceedings of the 2018 International Conference on Electronics Technology. Piscataway, USA: IEEE, 2018:325-330.
ZILBERMANE R, PACEP E. Autonomous time-frequency morphological feature extraction algorithm for LPI radar modulation classification[C]∥Proceedings of the 2006 International Conference on Image Processing. Piscataway, USA: IEEE, 2006:2321-2324.
[15]
NGIAMJ, CHENZ H, KOH P, et al. Learning deep energy models[C]∥Proceedings of the 28th International Conference on International Conference on Machine Learning. Madison, USA: Omnipress, 2011:1105-1112.
GUOC, PLEISSG, SUNY, et al. On calibration of modern neural networks[C]∥Proceedings of the 34th International Conference on Machine Learning. Worcester, USA:PMLR, 2017:1321-1330.
[18]
CIPOLLAR, GAL Y, KENDALLA. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics[C]∥Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2018:7482-7491.
[19]
ZHOUS, XUH J, ZHENGZ N, et al. A comprehensive survey on deep clustering: taxonomy, challenges, and future directions[J]. ArXiv e-Prints, 2022:arXiv:
[20]
LIUJ Y, SHAY, ZHANGW C, et al. Anomaly detection method for industrial control system operation data based on time-frequency fusion feature attention encoding[J]. Sensors, 2024,24(18):6131.