Article
Chemistry, Analytical
Ke Zang, Wenqi Wu, Wei Luo
Summary: This paper introduces a neural network parameter optimization method based on sparse learning, which can reduce parameters without compromising network performance. Experimental results demonstrate that the proposed method effectively reduces model parameters and improves network generalization ability.
Article
Chemistry, Analytical
Dong Wang, Meiyan Lin, Xiaoxu Zhang, Yonghui Huang, Yan Zhu
Summary: In this study, a CNN-transformer graph neural network (CTGNet) is proposed for modulation classification, which aims to uncover complex representations in signal data by transforming them into graph structures. Extensive experiments demonstrate that our method outperforms advanced deep learning techniques and achieves the highest recognition accuracy, highlighting the significant advantage of CTGNet in capturing key features in signal data and providing an effective solution for modulation classification tasks.
Article
Computer Science, Information Systems
Biao Dong, Yuchao Liu, Guan Gui, Xue Fu, Heng Dong, Bamidele Adebisi, Haris Gacanin, Hikmet Sari
Summary: This article presents an efficient lightweight decentralized-learning-based AMC method for edge devices. The proposed method uses a spatiotemporal hybrid deep neural network based on multichannels and multifunction blocks to balance the trade-off between lightweight and classification performance. It also utilizes a cooperative approach for model updates and aggregation to enhance classification accuracy while reducing computational pressure.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Chemistry, Analytical
Fan Wang, Tao Shang, Chenhan Hu, Qing Liu
Summary: Automatic modulation classification (AMC) is crucial in intelligent wireless communication. This paper proposes a lightweight neural network-based AMC framework, which combines complex convolution with residual networks for improved classification performance. Depthwise separable convolution is used for a lightweight design, and a hybrid data augmentation scheme is introduced to compensate for any performance loss. Simulation results demonstrate that this framework reduces the number of parameters by approximately 83.34% and the FLOPs by approximately 83.77% without degrading performance.
Article
Computer Science, Information Systems
Seung-Hwan Kim, Chang-Bae Moon, Jae-Woo Kim, Dong-Seong Kim
Summary: This study designs a hybrid signal and image-based deep learning model for automatic modulation classification in cognitive radio. The experimental results show that the proposed model significantly outperforms conventional models in terms of prediction accuracy.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2022)
Article
Computer Science, Information Systems
Hany S. Hussein, Mohamed Hassan Essai Ali, Mohammed Ismeil, Mohamed N. Shaaban, Mona Lotfy Mohamed, Hany A. Atallah
Summary: This study proposes robust CNN-based AMC techniques, which eliminate the need for feature extraction and can achieve high classification accuracy. The developed techniques utilize different classification layers and can automatically learn the features from the transmitted signals during training.
Article
Computer Science, Information Systems
Seung-Hwan Kim, Jae-Woo Kim, Williams-Paul Nwadiugwu, Dong-Seong Kim
Summary: In this paper, a novel deep learning-based robust automatic modulation classification (AMC) method is proposed for cognitive radio networks. The input size is extended and then reduced to improve classification accuracy and reduce computation complexity. The simulation results demonstrate higher classification accuracy compared to conventional models across various signal-to-noise ratio (SNR) ranges.
Article
Computer Science, Information Systems
Shuo Chang, Sai Huang, Ruiyun Zhang, Zhiyong Feng, Liang Liu
Summary: Automatic modulation classification (AMC) plays a vital role in identifying the modulation type of a received signal for ensuring the physical-layer security of IoT networks. This article focuses on reproducing and evaluating popular AMC algorithms using the in-phase/quadrature (I/Q) and amplitude/phase (A/P) representations for comparison. Based on the experimental results, it is found that CNN-RNN-like algorithms using A/P as input data perform better at high signal-to-noise ratio (SNR), while the opposite is true at low SNR. Inspired by these findings, a multitask learning-based deep neural network (MLDNN) is proposed, which effectively fuses I/Q and A/P. Extensive simulations demonstrate the superior performance of the proposed MLDNN in a public benchmark.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Chemistry, Multidisciplinary
Malik Zohaib Nisar, Muhammad Sohail Ibrahim, Muhammad Usman, Jeong-A Lee
Summary: Automatic modulation classification (AMC) is a classification problem in wireless communication systems that aims to determine the modulation type of a received signal. Deep learning (DL) methods have been widely used for AMC due to their ability to automatically learn features without technical expertise. This research proposed a DL algorithm inspired by residual learning, which showed high accuracy and strong representation ability. A squeeze-and-excitation network was also employed to improve performance by exploiting interconnections between channels and recalibrating feature responses. The proposed model achieved better accuracy than existing methods and significantly reduced the number of parameters compared to convolutional neural network designs.
APPLIED SCIENCES-BASEL
(2023)
Article
Telecommunications
Satish Kumar, Rajarshi Mahapatra, Anurag Singh
Summary: In this study, a CNN-based AMC scheme was designed and implemented to address the challenges of model size and floating-point weights and activations. The use of low precision and quantized CNN, along with the proposed RU-based AMC scheme and iterative pruning-based training mechanism, improved efficiency and accuracy on hardware.
IEEE COMMUNICATIONS LETTERS
(2022)
Article
Telecommunications
Shi Yunhao, Xu Hua, Jiang Lei, Qi Zisen
Summary: Automatic modulation classification (AMC) is a key technique in wireless communication, but current methods have unsatisfactory performance. This paper proposes a novel AMC framework, using an autoencoder as the backbone and combining Convolution-AE and LSTM-AE as temporal and spatial feature extractors. Experimental results show that the proposed network achieves higher classification accuracy at low SNR with low cost, and is suitable for semi-supervised scenarios.
IEEE COMMUNICATIONS LETTERS
(2022)
Article
Remote Sensing
Min Ma, Yunhe Xu, Zhi Wang, Xue Fu, Guan Gui
Summary: Automatic modulation classification (AMC) is a promising technology for identifying modulation modes in drone communication systems. This study introduces deep neural networks (DNNs) into AMC methods, specifically proposing a decentralized learning method based on residual network (ResNet) called DecentAMC. Simulation results demonstrate that DecentAMC achieves similar classification performance to centralized AMC while improving training efficiency and protecting data privacy.
Article
Telecommunications
Xue Fu, Guan Gui, Yu Wang, Tomoaki Ohtsuki, Bamidele Adebisi, Haris Gacanin, Fumiyuki Adachi
Summary: This paper proposes a decentralized learning AMC method using model consolidation and lightweight design, which reduces the storage and computational capacity requirements, improves the training efficiency, and lowers the communication overhead.
IEEE TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING
(2022)
Article
Telecommunications
Shanchuan Ying, Sai Huang, Shuo Chang, Zheng Yang, Zhiyong Feng, Ningyan Guo
Summary: In this paper, a data-driven framework called CTDNN is proposed to enhance the classification performance of automatic modulation classification. By using modules such as convolutional neural network, transition module, and final classifier, CTDNN achieves superior classification performance compared to traditional deep models.
CHINA COMMUNICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Chenghong Xiao, Shuyuan Yang, Zhixi Feng
Summary: In this article, a novel end-to-end automatic modulation classification (AMC) model called complex-valued depthwise separable convolutional neural network (CDSCNN) is proposed, which adopts complex-valued operation units for tailored feature learning for AMC. With an overall accuracy of 62.63% on the RadioML2016.10a dataset, CDSCNN outperforms its counterparts by 1%-11%. After fine-tuning on the RadioML2016.10b dataset, the overall accuracy reaches 63.15%, demonstrating the robust recognition and generalization capability of CDSCNN. Moreover, CDSCNN exhibits lower model complexity compared to other methods.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
