An Improved Modulation Recognition Method Based on Hybrid Kolmogorov-Arnold Convolutional Neural Network
-
摘要: 针对苛刻通信场景下调制方式识别精度低且泛化难的问题,该文提出一种改进Kolmogorov-Arnold混合卷积神经网络模型KA-CNN。首先,通过双树复小波包变换将信号分解至多维小波域,结合不同频率分量构建多尺度信号表征,促使神经网络模型学习各频率下的一致性特征;然后设计集成样条函数和非线性激活函数的深度学习结构,利用样条函数解决维度诅咒问题,增强周期性特征的持续学习能力;最后,采用Lipschitz正则化约束的多级网格训练,改善模型面对不同信号参数的适应性,增强跨通信场景的泛化能力。在公开数据集RadioML 2016.10a, RadioML 2018.01a和CSPB.ML.2023的实验表明,KA-CNN具有优异的调制识别精度,当信噪比在16 dB时能够取得90%以上的识别准确率。相较于其它深度学习方法,整体识别精度提升3%~10%,并在各种信噪比条件下具备更强的特征学习能力和泛化性。
-
关键词:
- 调制识别 /
- 深度学习 /
- Kolmogorov-Arnold模型 /
- 卷积神经网络
Abstract:Objective With the rapid growth of communication devices and increasing complexity of electromagnetic environments, spectrum efficiency has become a critical performance metric for sixth-generation communication systems. Modulation recognition is an essential function of dynamic spectrum access, aiming to automatically identify the modulation scheme of received signals to enhance spectrum utilization. In practice, wireless signals are often affected by multipath propagation, interference, and noise, which pose challenges for accurate recognition. To address these issues, this study proposes a deep learning-based approach using an end-to-end model that eliminates manual feature extraction, mitigates limitations of handcrafted features, and improves recognition accuracy. By transferring general knowledge from signal classification to modulation recognition, a well-generalized method based on a hybrid Kolmogorov-Arnold Convolutional Neural Network (KA-CNN) is developed. This approach supports reliable communication in applications such as intelligent transportation, the Internet of Things (IoT), vehicular ad hoc networks, and satellite communication. Methods The proposed modulation recognition method first decomposes the signal into a multi-dimensional wavelet domain using a dual-tree complex wavelet packet transform. Different frequency components are then combined to construct a multi-scale signal representation, enabling the neural network to learn consistent features across frequencies. A deep learning structure, KA-CNN, is designed by integrating spline functions with nonlinear activation functions to enhance nonlinear fitting and continuous learning of periodic features. Spline functions are used to address the curse of dimensionality. To improve adaptability to varying signal parameters and enhance generalization across communication scenarios, multilevel grid training with Lipschitz regularization constraints is applied. In KA-CNN, the hybrid module transfers the characteristics of the spline function into convolution operations, which improves the model’s capacity to capture complex mappings between input signals and modulation schemes while retaining the efficiency of the Kolmogorov-Arnold network. This enhances both the expressive power and adaptability of deep learning models under complex communication conditions. Results and Discussions During the experimental phase, modulation recognition performance testing, ablation study, and comparative analysis are conducted on three publicly available datasets (RadioML 2016.10a, RadioML 2018.01a, and CSPB.ML.2023) to evaluate the performance of KA-CNN. Results show that KA-CNN achieves modulation recognition accuracies of 65.14%, 65.56%, and 78.40% on RadioML 2016.10a, RadioML 2018.01a, and CSPB.ML.2023, respectively ( Figure 6 ). The main performance limitation arises in the classification of QPSK versus 8PSK, AM-DSB versus WBFM, and high-order QAM modulation types (Figure 7 ). Maximum differences in recognition accuracy of KA-CNN driven by different signal representations reach 2.04%, 3.46%, and 4.54% across the three datasets, demonstrating the effect of signal representation (Figure 8 ). The wavelet packet transform constructs a multi-scale time-frequency representation of signals that is insensitive to the maximum decomposition scale L and supports complementary learning of different modulation features. The hybrid Kolmogorov-Arnold convolutional module and the multi-dimensional perceptual cascade attention mechanism play key roles in enhancing modulation recognition accuracy, particularly under relatively high Signal-To-Noise Ratio (SNR) conditions (Figure 9 ). Additionally, finer grids and higher decomposition orders improve the model’s ability to extract discriminative signal features, thereby increasing recognition accuracy (Figure 10 ). Finally, a comparative evaluation against several deep learning models, including GGCNN, Transformer, PR-LSTM, and MobileViT, confirms the superior performance of KA-CNN (Figure 11 ).Conclusions This study proposes a hybrid KA-CNN to address the reduced modulation recognition accuracy caused by noise and parameter variation, as well as the limited generalization across communication scenarios in existing deep learning models. By integrating spline functions with nonlinear activation functions, KA-CNN mitigates the curse of dimensionality and improves its capacity for continuous learning of periodic features. A dual-tree complex wavelet packet transform is used to construct a multi-scale signal representation, enabling the model to extract consistent features across frequencies. The model is trained using multilevel grids with Lipschitz regularization constraints to enhance adaptability to varying signal parameters and improve generalization. Experimental results on three public datasets demonstrate that KA-CNN improves modulation recognition accuracy and exhibits robust generalization, particularly under low SNRs. -
表 1 信号模型参数
参数类型 RadioML 2016.10a RadioML 2018.01a CSPB.ML.2023 符号率偏移(kHz) 0.1~0.4 0.1~0.4 0.1~0.6 频率偏移(kHz) 0~0.01 0.0001 ~0.01–0.2~0.2 相位偏移 0~0.01×2π 0.0001 ~0.01×2π–0.1~0.1×2π 时钟误差 0~0.02 0~0.02 0~0.25 采样点数 2×128 2× 1024 2×128 信噪比(dB) –20:2:18 –20:2:30 2~20 噪声类型 加性高斯白噪声 加性高斯白噪声 加性高斯白噪声 信道环境 莱斯+瑞利 莱斯+瑞利 莱斯+瑞利 调制类型 11种 24种 9种 样本数量 2.2×106 2.56×107 1.2×106(非重叠) 表 2 训练过程的超参数
数据集 分解尺度L 迭代周期 批量大小$ \varOmega $ 学习率 惩罚因子$ \lambda $ 网格大小G 样条阶数D 丢弃率$ \rho $ RadioML 2016.10a 3 80 64 0.001 0.0001 5 3 0.2 RadioML 2018.01a 4 100 128 0.0001 0.0001 6 4 0.3 CSPB.ML.2023 3 100 64 0.0001 0.00001 5 3 0.2 -
[1] 尤肖虎, 张川, 谈晓思, 等. 基于AI的5G技术——研究方向与范例[J]. 中国科学: 信息科学, 2018, 48(12): 1589–1602. doi: 10.1360/N112018-00174.YOU Xiaohu, ZHANG Chuan, TAN Xiaosi, et al. AI for 5G: Research directions and paradigms[J]. Scientia Sinica: Informationis, 2018, 48(12): 1589–1602. doi: 10.1360/N112018-00174. [2] 杨伟超, 杜宇, 文伟, 等. 基于多重分形谱智能分析的卫星信号调制识别研究[J]. 电子学报, 2022, 50(6): 1336–1343. doi: 10.12263/DZXB.20210882.YANG Weichao, DU Yu, WEN Wei, et al. Modulation recognition of satellite communication signal based on intelligent analysis of multi-fractal spectrum[J]. Acta Electronica Sinica, 2022, 50(6): 1336–1343. doi: 10.12263/DZXB.20210882. [3] 王华华, 张睿哲, 黄永洪. 基于生成式对抗网络和多模态注意力机制的扩频与常规调制信号识别方法[J]. 电子与信息学报, 2024, 46(4): 1212–1221. doi: 10.11999/JEIT230518.WANG Huahua, ZHANG Ruizhe, and HUANG Yonghong. Spread spectrum and conventional modulation signal recognition method based on generative adversarial network and multi-modal attention mechanism[J]. Journal of Electronics & Information Technology, 2024, 46(4): 1212–1221. doi: 10.11999/JEIT230518. [4] 张天骐, 汪锐, 安泽亮, 等. 基于多任务学习的MIMO-OFDM信噪比估计与调制识别[J]. 北京邮电大学学报, 2022, 45(6): 95–100,121. doi: 10.13190/j.jbupt.2022-050.ZHANG Tianqi, WANG Rui, AN Zeliang, et al. MIMO-OFDM signal to noise ratio estimation and modulation recognition based on multi task learning[J]. Journal of Beijing University of Posts and Telecommunications, 2022, 45(6): 95–100,121. doi: 10.13190/j.jbupt.2022-050. [5] WANG Danyang, QI Peihan, FU Qifan, et al. Multiple high-order cumulants-based spectrum sensing in full-duplex-enabled cognitive IoT networks[J]. IEEE Internet of Things Journal, 2021, 8(11): 9330–9343. doi: 10.1109/JIOT.2021.3055782. [6] BAI Jiansheng, YAO Jinjie, GUO Yurong, et al. Automatic modulation classification in impulsive noise: Log-domain 3-D constellation diagrams and multiscale dual-convolutional 3DCNN[J]. IEEE Transactions on Cognitive Communications and Networking, 2024, 10(4): 1204–1217. doi: 10.1109/TCCN.2024.3375510. [7] LI Lin, ZHU Yaxin, and ZHU Zhigang. Automatic modulation classification using ResNext-GRU with deep feature fusion[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 2519710. doi: 10.1109/TIM.2023.3290301. [8] 李东瑾, 杨瑞娟, 李晓柏, 等. 基于核协同表示与鉴别投影的辐射源调制识别[J]. 电子学报, 2020, 48(9): 1695–1702. doi: 10.3969/j.issn.0372-2112.2020.09.005.LI Dongjin, YANG Ruijuan, LI Xiaobai, et al. Emitter signal modulation recognition based on kernel collaborative representation and discriminative projection[J]. Acta Electronica Sinica, 2020, 48(9): 1695–1702. doi: 10.3969/j.issn.0372-2112.2020.09.005. [9] YANG Faquan, YANG Ling, WANG Dong, et al. Method of modulation recognition based on combination algorithm of K-means clustering and grading training SVM[J]. China Communications, 2018, 15(12): 55–63. doi: 10.12676/j.cc.2018.12.004. [10] 占锦敏, 赵知劲, 王李军. 基于随机森林的通信信号识别算法[J]. 杭州电子科技大学学报: 自然科学版, 2020, 40(5): 1–6. doi: 10.13954/j.cnki.hdu.2020.05.001.ZHAN Jinmin, ZHAO Zhijin, and WANG Lijun. Signals identification algorithm based on random forest[J]. Journal of Hangzhou Dianzi University: Natural Sciences, 2020, 40(5): 1–6. doi: 10.13954/j.cnki.hdu.2020.05.001. [11] 左晓亚, 张俊杰, 姚如贵, 等. 残差膨胀卷积结构下的多模态特征调制方式识别[J]. 信号处理, 2023, 39(11): 2013–2021. doi: 10.16798/j.issn.1003-0530.2023.11.010.ZUO Xiaoya, ZHANG Junjie, YAO Rugui, et al. Automatic modulation classification based on residual dilated-convolution and multi-feature[J]. Journal of Signal Processing, 2023, 39(11): 2013–2021. doi: 10.16798/j.issn.1003-0530.2023.11.010. [12] KE Ziqi and VIKALO H. Real-time radio technology and modulation classification via an LSTM auto-encoder[J]. IEEE Transactions on Wireless Communications, 2022, 21(1): 370–382. doi: 10.1109/TWC.2021.3095855. [13] GHASEMZADEH P, HEMPEL M, WANG Honggang, et al. GGCNN: An efficiency-maximizing gated graph convolutional neural network architecture for automatic modulation identification[J]. IEEE Transactions on Wireless Communications, 2023, 22(9): 6033–6047. doi: 10.1109/TWC.2023.3239311. [14] ZHENG Qinghe, ZHAO Penghui, WANG Hongjun, et al. Fine-grained modulation classification using multi-scale radio transformer with dual-channel representation[J]. IEEE Communications Letters, 2022, 26(6): 1298–1302. doi: 10.1109/LCOMM.2022.3145647. [15] ZHANG Zufan, LUO Hao, WANG Chun, et al. Automatic modulation classification using CNN-LSTM based dual-stream structure[J]. IEEE Transactions on Vehicular Technology, 2020, 69(11): 13521–13531. doi: 10.1109/TVT.2020.3030018. [16] QI Peihan, ZHOU Xiaoyu, ZHENG Shilian, et al. Automatic modulation classification based on deep residual networks with multimodal information[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(1): 21–33. doi: 10.1109/TCCN.2020.3023145. [17] 肖易寒, 王亮, 郭玉霞. 基于去噪卷积神经网络的雷达信号调制类型识别[J]. 电子与信息学报, 2021, 43(8): 2300–2307. doi: 10.11999/JEIT200506.XIAO Yihan, WANG Liang, and GUO Yuxia. Radar signal modulation type recognition based on denoising convolutional neural network[J]. Journal of Electronics & Information Technology, 2021, 43(8): 2300–2307. doi: 10.11999/JEIT200506. [18] O’SHEA T J, CORGAN J, and CLANCY T C. Convolutional radio modulation recognition networks[C]. 17th International Conference on Engineering Applications of Neural Networks, Aberdeen, UK, 2016: 213–226. doi: 10.1007/978-3-319-44188-7_16. [19] O’SHEA T J, ROY T, and CLANCY T C. Over-the-air deep learning based radio signal classification[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12(1): 168–179. doi: 10.1109/JSTSP.2018.2797022. [20] SNOAP J A, POPESCU D C, and SPOONER C M. Novel nonlinear neural-network layers for high performance and generalization in modulation-recognition applications[C]. MILCOM 2023-2023 IEEE Military Communications Conference, Boston, USA, 2023: 562–567. doi: 10.1109/MILCOM58377.2023.10356331. [21] ZHENG Qinghe, TIAN Xinyu, YU Zhiguo, et al. DL-PR: Generalized automatic modulation classification method based on deep learning with priori regularization[J]. Engineering Applications of Artificial Intelligence, 2023, 122: 106082. doi: 10.1016/j.engappai.2023.106082. [22] ZHENG Qinghe, SAPONARA S, TIAN Xinyu, et al. A real-time constellation image classification method of wireless communication signals based on the lightweight network MobileViT[J]. Cognitive Neurodynamics, 2024, 18(2): 659–671. doi: 10.1007/s11571-023-10015-7. -