Open AccessModulation recognition with pre‐denoising convolutional neural network
Author(s) -
Liu Yabo,
Liu Yi
Publication year - 2020
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/el.2019.3586
Subject(s) - noise reduction , convolutional neural network , computer science , artificial intelligence , pattern recognition (psychology) , residual , modulation (music) , signal to noise ratio (imaging) , noise (video) , process (computing) , artificial neural network , encoder , deep learning , reduction (mathematics) , speech recognition , algorithm , mathematics , image (mathematics) , telecommunications , philosophy , geometry , operating system , aesthetics
Modulation recognition (MR) is an important technology in modern communication systems. Traditional MR methods can be categorised as the maximum likelihood hypothesis, pattern recognition, and deep learning‐based methods. The authors can achieve high‐recognition accuracy when the signal‐to‐noise ratio (SNR) is high. However, their recognition accuracy is greatly reduced when the SNR is low, especially when the SNR is below 0 dB. In order to improve the MR accuracy at low SNRs, they propose a pre‐denoising algorithm, which is used before MR methods. The model of the pre‐denoising algorithm is a fully convolutional neural network, which is similar to an auto‐encoder. They also use residual learning to speed up the training process. Experimental results show that the proposed pre‐denoising algorithm can significantly enhance the SNRs of modulated signals and improve the accuracy of MR methods.