TY - GEN
T1 - A modulation classification method based on deformable convolutional neural networks for broadband satellite communication systems
AU - Li, Qian
AU - Zhang, Qi
AU - Xin, Xiangjun
AU - Tian, Qinghua
AU - Tao, Ying
AU - Shen, Yufei
AU - Cao, Guixing
AU - Liu, Naijing
AU - Qian, Jixi
N1 - Publisher Copyright:
© 2018 SPIE.
PY - 2018
Y1 - 2018
N2 - In order to solve the problem of broadband satellite modulation signal with SNR fluctuation of complex channel and various modulation signal recognizing, we propose a Deformable Convolutional Neural Networks (DCNN) classification model based on broadband satellite communication systems. In our algorithm, we propose a deformable convolution kernel, which only need to calculate the 2/3 pixel convoluting. Our algorithm not only can be used to reduce the complexity and improve the robustness, but also used to improve the accuracy. We simulate the accuracy and the complexity of the algorithm among the four neural network models of DCNN, VGG, AlexNet and ResNet. The results show that the design of the DCNN model has high recognition rate and low algorithm complexity. Then we simulate the DCNN network in variable signal-to-noise of BPSK, QPSK, 8PSK, 16APSK, 32APSK, 16QAM, 32QAM and 64QAM commonly used satellite modulation signal classification and complex channel conditions, and training the four basic modulation signal used to identify other modulation signals. The results show that the DCNN model not only can be used to maintain a high recognition rate of the modulated signal, but also used to reduce the complexity of the algorithm and improves the robustness of the algorithm.
AB - In order to solve the problem of broadband satellite modulation signal with SNR fluctuation of complex channel and various modulation signal recognizing, we propose a Deformable Convolutional Neural Networks (DCNN) classification model based on broadband satellite communication systems. In our algorithm, we propose a deformable convolution kernel, which only need to calculate the 2/3 pixel convoluting. Our algorithm not only can be used to reduce the complexity and improve the robustness, but also used to improve the accuracy. We simulate the accuracy and the complexity of the algorithm among the four neural network models of DCNN, VGG, AlexNet and ResNet. The results show that the design of the DCNN model has high recognition rate and low algorithm complexity. Then we simulate the DCNN network in variable signal-to-noise of BPSK, QPSK, 8PSK, 16APSK, 32APSK, 16QAM, 32QAM and 64QAM commonly used satellite modulation signal classification and complex channel conditions, and training the four basic modulation signal used to identify other modulation signals. The results show that the DCNN model not only can be used to maintain a high recognition rate of the modulated signal, but also used to reduce the complexity of the algorithm and improves the robustness of the algorithm.
KW - DCNN
KW - convolution kernel
KW - modulation signal classification
KW - robustness
UR - http://www.scopus.com/inward/record.url?scp=85059475893&partnerID=8YFLogxK
U2 - 10.1117/12.2506109
DO - 10.1117/12.2506109
M3 - Conference contribution
AN - SCOPUS:85059475893
T3 - Proceedings of SPIE - The International Society for Optical Engineering
BT - Tenth International Conference on Information Optics and Photonics
A2 - Huang, Yidong
PB - SPIE
T2 - 10th International Conference on Information Optics and Photonics
Y2 - 8 July 2018 through 11 July 2018
ER -