调制识别是保障软件无线电通信可靠性的关键技术，在电子对抗、军事通信侦察和链路自适应等领域具有重要应用价值。由于当前无线频谱资源匮乏以及调制样式多样化等问题，复杂信道下多类调制信号并存，使得传统调制识别算法对接收信号难以有效完成分类工作。为提高调制识别算法的有效性和实用性，本文结合深度学习方法，对软件无线电平台下的通信信号调制识别问题进行研究。主要内容如下：

（1）针对复杂电磁环境下神经网络难以挖掘信号关键特征导致识别精度低的问题，设计了一种基于多注意力机制网络的调制识别算法。该算法在信号预处理环节提取幅相信息，与I/Q序列构成双通道数据作为输入，以获取丰富的信号特征。通过引入改进卷积注意力机制的残差密集块学习信号重要空间特征，将输出特征向量融合后送入双向门控循环单元获取时间信息，并添加序列注意力机制来捕捉信号关键时间特征。利用软件无线电平台GNU Radio仿真小尺度衰落信道下的九种调制信号进行验证，实验结果表明，所提算法有效改善了高阶QAM、PSK信号的识别精度。相比于其他深度学习算法，所提算法在信噪比大于-6dB时可获得更高的识别精度。

（2）为解决有限标签数据下深度学习监督算法对通信信号调制识别准确率低的问题，设计了基于残差半监督生成对抗网络的调制识别算法。该算法直接以I/Q序列作为输入，以谱归一化残差单元替换传统半监督生成对抗网络的二维卷积层，获取信号多尺度特征并增强网络训练稳定性，通过挖掘无标签数据中的潜在信息来提高有限标签数据下的识别准确率。为验证算法实用性，利用两台软件无线电硬件外设HackRF和GNU Radio搭建数据采集平台，完成真实电磁环境下七种调制信号的收发并制备实测数据集。实验结果表明，在提供相同数量的标签数据时，所提算法在识别准确率上优于深度学习监督算法和其他半监督生成对抗网络调制识别算法。

Modulation recognition is a key technology to ensure the reliability of software defined radio communication, and has important application value in the fields of electronic countermeasures, military communication reconnaissance and link adaptation. Due to the lack of current wireless spectrum resources and the diversification of modulation patterns, multiple types of modulation signals coexist in complex channels, and it is difficult for traditional modulation recognition algorithms to effectively classify received signals. In order to improve the effectiveness and practicability of the modulation recognition algorithm, this thesis combines the deep learning method to study the modulation recognition problem of communication signals under the software defined radio platform. The main contents are as follows:

(1) Aiming at the problem that it is difficult for neural network to excavate key features of signal in complex electromagnetic environment, which leads to low recognition accuracy, a modulation recognition algorithm based on multi-attention mechanism network is designed. The algorithm extracts the amplitude and phase information in the signal preprocessing link, and forms dual-channel data with the I/Q sequence as input to obtain rich signal features. The important spatial features of the signal are learned by the residual dense block introducing improved convolutional attention mechanism, the output feature vector is fused and sent to the bidirectional gated recurrent unit to obtain temporal information, and the sequence attention mechanism is added to capture the key temporal features of the signal. The software defined radio platform GNU Radio is used to simulate nine kinds of modulation signals under small-scale fading channels for verification. The experimental results show that the proposed algorithm can effectively improve the recognition accuracy of high-order QAM and PSK signals. Compared with other deep learning algorithms, the proposed algorithm can achieve higher recognition accuracy when the signal-to-noise ratio is greater than -6dB.

(2) In order to solve the problem of low recognition accuracy of communication signal modulation recognition by deep learning supervision algorithm under limited label data, a modulation recognition algorithm based on residual semi-supervised generative adversarial network is designed. The algorithm directly takes the I/Q sequence as input, replacing two-dimensional convolutional layers of traditional semi-supervised generative adversarial networks with spectral normalized residual units, obtaining the multi-scale features of the signal and enhancing the stability of network training. The latent information in unlabeled data is excavated to improve the recognition accuracy under limited label data. In order to verify the practicability of the algorithm, two software defined radio hardware peripherals HackRF, and GNU Radio, are used to build a data collection platform to complete the transmission and reception of seven modulated signals in a real electromagnetic environment and prepare a measured data set. The experimental results show that the proposed algorithm outperforms deep learning supervised algorithms and other semi-supervised generative adversarial networks modulation recognition algorithms in recognition accuracy when the same amount of labeled data is provided.

[1] Liao J, Zhao J, Gao F, et al. A model-driven deep learning method for massive MIMO detection[J]. IEEE Communications Letters, 2020, 24(08): 1724-1728.

[2] 楼才义,徐建良,杨小牛.软件无线电原理与应用[M]. 北京: 电子工业出版社,2014.

[3] 李艳玲.非合作通信中单载波调制信号识别算法研究[D]. 西安: 西安电子科技大学,2013.

[4] Salam A O A, Sheriff R E, Al-Araji S R, et al. A unified practical approach to modulation classification in cognitive radio using likelihood-based techniques[C]//2015 IEEE 28th Canadian Conference on Electrical and Computer Engineering(CCECE). IEEE, 2015: 1024-1029.

[5] Phukan G J, Bora P K. Blind equalization for classification of digital modulations[C]// 2018 International Conference on Signal Processing and Communications(SPCOM). IEEE, 2018: 472-476.

[6] Yuan Y, Zhao P, Wang B, et al. Hybrid maximum likelihood modulation classification for continuous phase modulations[J]. IEEE Communications Letters, 2016, 20(03): 450-453.

[7] 吴斌,袁亚博,汪勃.基于记忆因子的连续相位调制信号最大似然调制识别[J]. 电子与信息学报,2016,38(10):2546-2552.

[8] Zheng J, Lv Y. Likelihood-based automatic modulation classification in OFDM with index modulation[J]. IEEE Transactions on Vehicular Technology, 2018, 67(09): 8192-8204.

[9] Tan Q, Yang A, Guo P. Blind modulation format identification using differential phase and amplitude ratio[J]. IEEE Photonics Journal, 2019, 11(01): 1-12.

[10] Liu W, Wang Q, Guo Q. Automatic radar waveform recognition based on neural network[J]. Control and Intelligent Systems, 2018, 46(02): 92-96.

[11] 刘朝阳,王安义,李蓉.一种非协作通信信号特征提取方法[J]. 西安科技大学学报,2020,40(04):706-711.

[12] Das D, Bora P K, Bhattacharjee R. Blind modulation recognition of the lower order PSK signals under the MIMO keyhole channel[J]. IEEE Communications Letters, 2018, 22(09): 1834-1837.

[13] 陈思南,龚晓峰,陈利波.共信道时频重叠的数字信号调制识别方法[J]. 计算机工程与设计,2018,39(03):612-616.

[14] Liu Y, Guo X. Modulation recognition based on wavelet transform and fractal theory[J]. International Journal of Performability Engineering, 2019, 15(03): 998-1004.

[15] 闫文康,闫毅,范雅楠,等.基于小波变换熵值及高阶累积量联合的卫星信号调制识别算法[J]. 空间科学学报,2021,41(06):968-975.

[16] Chikha W B, Dayoub I, Hamouda W, et al. Modulation recognition for MIMO relaying broadcast channels with direct link[J]. IEEE Wireless Communications Letters,2014, 3(01): 50-53.

[17] Chu P, Xie L, Dai C, et al. Automatic modulation recognition for secondary modulated signals[J]. IEEE Wireless Communication Letters, 2021, 10(05): 962-965.

[18] 任进,姬丽彬,党柳.基于深度学习的卫星信号调制识别算法[J]. 无线电工程, 2022,52(04):529-535.

[19] 弋佳东,杨洁.基于IFOA-SA-BP神经网络的雷达信号识别[J]. 系统工程与电子技术, 2020,42(12):2735-2741.

[20] Park C S, Kim D Y. A novel robust feature of modulation classification for reconfigurable software radio[J]. IEEE Transactions on Consumer Electronics, 2006, 52(04): 1193-1200.

[21] 王海滨,周正,李炳荣,等.基于数字通信信号瞬时特性的调制方式识别方法[J]. 现代电子技术,2019,42(16):22-25.

[22] Liang Y, Xiang X, Sun Y, et al. Novel modulation recognition for WFRFT-based system using 4th-order cumulants[J]. IEEE Access, 2019, 7: 86018-86025.

[23] 王安义,李立.基于高阶累积量和DNN模型的井下信号识别方法[J]. 工矿自动化, 2020,46(02):82-87.

[24] Jiang X R, Chen H, Zhao Y D, et al. Automatic modulation recognition based on mixed-type features[J]. International Journal of Electronics, 2021, 108(01): 105-114.

[25] 李正权,林媛,李梦雅,等.基于判别式受限玻尔兹曼机的数字调制识别[J]. 通信学报, 2021,42(02):81-91.

[26] Walenczykowska M, Kawalec A. Type of modulation identification using wavelet transform and neural network[J]. Bulletin of the Polish Academy of Sciences-Technical Sciences, 2016, 64(01): 257-261.

[27] Mihandoost S, Azimzadeh E. Introducing an efficient statistical model for automatic modulation classification[J]. Journal of Signal Processing Systems. 2020, 92(01): 123-134.

[28] Karami E, Dobre O A. Identification of SM-OFDM and AL-OFDM signals based on their second-order cyclostationarity[J]. IEEE Transactions on Vehicular Technology, 2015, 64(03): 942-953.

[29] 赵雄文,郭春霞,李景春.基于高阶累积量和循环谱的信号调制方式混合识别算法[J]. 电子与信息学报,2016,38(03):674-680.

[30] Dong S, Li Z, Zhao L. A modulation recognition algorithm based on cyclic spectrum and SVM classification[C]// 2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference(ITNEC). IEEE, 2020: 2123-2127.

[31] Liu M, Song T, Hu J, et al. Deep learning-inspired message passing algorithm for efficient resource allocation in cognitive radio networks[J]. IEEE Transactions on Vehicular Technology, 2018, 63(01): 641-653.

[32] Peng S, Jiang H, Wang H, et al. Modulation classification based on signal constellation diagrams and deep learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(03): 718-727.

[33] Li R, Li L, Yang S, et al. Robust automated VHF modulation recognition based on deep convolutional neural networks[J]. IEEE Communications Letters, 2018, 22(05): 946-949.

[34] 宋子豪,程伟,彭岑昕,等.基于CWD和残差收缩网络的调制方式识别方法[J]. 系统工程与电子技术,2021,43(11):3371-3379.

[35] Jeong S, Lee U, Kim S C. Spectrogram-based automatic modulation recognition using convolutional neural network[C]// 2018 Tenth International Conference on Ubiquitous and Future Networks (ICUFN), IEEE, 2018: 843-845.

[36] O’Shea T J, West N. Radio machine learning dataset generation with GNU radio[C]// Proceedings of the GNU Radio Conference, Boulder, America, GRCon, 2016: 1-6.

[37] O’Shea T J, Corgan J, Clancy T C. Convolutional radio modulation recognition networks[C]// International Conference on Engineering Applications of Neural Networks. Springer, Cham, 2016: 213-226.

[38] West N E, O’Shea T. Deep architectures for modulation recognition[C]// 2017 IEEE International Symposium on Dynamic Spectrum Access Networks(DySPAN). IEEE, 2017: 1-6.

[39] Liu X, Yang D, EI Gamal A. Deep neural network architectures for modulation classification[C]// 2017 51st Asilomar Conference on Signals, Systems, and Computers. IEEE, 2017: 915-919.

[40] Zhang Z, Luo H, Wang C, et al. Automatic modulation classification using CNN-LSTM based dual-stream structure[J]. IEEE Transactions on Vehicular Technology, 2020, 69(11): 13521-13531.

[41] Xu J, Luo C, Parr G, et al. A spatiotemporal multi-channel learning framework for automatic modulation recognition[J]. IEEE Wireless Communication Letters, 2020, 9(10): 1629-1632.

[42] 刘桥平,高兴宇,邱昕,等.基于USRP的自动调制识别[J]. 计算机应用与软件,2020,37(08):110-114+121.

[43] 任思睿,黄铭.基于改进的长短期记忆网络的调制识别算法[J]. 云南大学学报(自然科学版),2021,43(01):39-45.

[44] 史蕴豪,许华,郑万泽,等.基于集成学习与特征降维的小样本调制识别方法[J]. 系统工程与电子技术,2021,43(04):1099-1109.

[45] 苟泽中,许华,郑万泽,等.基于半监督联合神经网络的调制识别算法[J]. 信号处理,2020,36(02):168-176.

[46] Jiao J, Sun X, Zhang Y, et al. Modulation recognition of radio signals based on edge computing and convolutional neural network[J]. Journal of Communications and Information Networks, 2021, 6(03): 280-300.

[47] 樊雨沛.基于HackRF的室内无线电频谱检测仪的设计与实现[D]. 海南: 海南大学,2019.

[48] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2016: 770-778.

[49] Huang G, Liu Z, Van Der Maaten, et al. Densely connected convolutional networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2017: 2261-2269.

[50] Chung J, Gulcehre C, Cho K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[C]// Annual Conference on Neural Information Processing Systems (NIPS). Montréal: MIT Press, 2014: 1-9.

[51] Su J, Zeng J, Xiong D, et al. A hierarchy-to-sequence attentional neural machine translation model[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2018, 26(03): 623-632.

[52] Woo S, Park J, Lee J Y, et al. CBAM: Convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision (ECCV). Springer, Cham, 2018: 3-19.

[53] 程建刚,毕凤荣,张立鹏,等.基于多重注意力-卷积神经网络-双向门控循环单元的机械故障诊断方法研究[J]. 内燃机工程,2021,42(04):77-83+92.

[54] Van Der Maaten L, Hinton G. Visualizing data using t-SNE[J]. Journal of machine learning research, 2008, 9(11): 2579-2605.

[55] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial networks[J] Advances in Neural Information Processing Systems, 2014,3: 2672-2680.

[56] Tang B, Tu Y, Zhang Z, et al. Digital signal modulation classification with data augmentation using generative adversarial nets in cognitive radio networks[J]. IEEE Access, 2018, 6, 15713-15722.

[57] 于浩洋,尹良,李书芳,等.生成对抗网络小样本雷达调制信号识别算法[J]. 西安电子科技大学学报,2021,48(06):96-104.

[58] Salimans T, Goodfellow I, Zaremba W, et al. Improved techniques for trainings GANs[C]// Advances in Neural Information Processing Systems(NIPS), 2016: 2234-2242.

[59] Zhou H, Jiao L, Zheng S, et al. Generative adversarial network-based electromagnetic signal classification: a semi-supervised learning framework[J]. China Communications, 2020, 17(10): 157-169.

[60] Dai Z, Yang Z, Yang F, et al. Good Semi-supervised learning that requires a bad GAN[C]// Advances in Neural Information Processing Systems(NIPS), 2017: 6510-6520.