A Multi-Modal Modulation Recognition Method with SNR Segmentation Based on Time Domain Signals and Constellation Diagrams
-
Published:2023-07-21
Issue:14
Volume:12
Page:3175
-
ISSN:2079-9292
-
Container-title:Electronics
-
language:en
-
Short-container-title:Electronics
Author:
Duan Ruifeng12ORCID, Li Xinze12, Zhang Haiyan12, Yang Guoting12ORCID, Li Shurui3, Cheng Peng45, Li Yonghui5
Affiliation:
1. School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China 2. Engineering Research Center for Forestry-Oriented Intelligent Information Processing of National Forestry and Grassland Administration, Beijing 100083, China 3. School of Technology, Beijing Forestry University, Beijing 100083, China 4. Department of Computer Science and Information Technology, La Trobe University, Melbourne, VIC 3086, Australia 5. School of Electrical and Information Engineering, The University of Sydney, Sydney, NSW 2006, Australia
Abstract
Deep-learning-based automatic modulation recognition (AMR) has recently attracted significant interest due to its high recognition accuracy and the lack of a need to manually set classification standards. However, it is extremely challenging to achieve a high recognition accuracy in increasingly complex channel environments and balance the complexity. To address this issue, we propose a multi-modal AMR neural network model with SNR segmentation called M-LSCANet, which integrates an SNR segmentation strategy, lightweight residual stacks, skip connections, and an attention mechanism. In the proposed model, we use time domain I/Q data and constellation diagram data only in medium and high signal-to-noise (SNR) regions to jointly extract the signal features. But for the low SNR region, only I/Q signals are used. This is because constellation diagrams are very recognizable in the medium and high SNRs, which is conducive to distinguishing high-order modulation. However, in the low SNR region, excessive similarity and the blurring of constellations caused by heavy noise will seriously interfere with modulation recognition, resulting in performance loss. Remarkably, the proposed method uses lightweight residuals stacks and rich ski connections, so that more initial information is retained to learn the constellation diagram feature information and extract the time domain features from shallow to deep, but with a moderate complexity. Additionally, after feature fusion, we adopt the convolution block attention module (CBAM) to reweigh both the channel and spatial domains, further improving the model’s ability to mine signal characteristics. As a result, the proposed approach significantly improves the overall recognition accuracy. The experimental results on the RadioML 2016.10B public dataset, with SNR ranging from −20 dB to 18 dB, show that the proposed M-LSCANet outperforms existing methods in terms of classification accuracy, achieving 93.4% and 95.8% at 0 dB and 12 dB, respectively, which are improvements of 2.7% and 2.0% compared to TMRN-GLU. Moreover, the proposed model exhibits a moderate parameter number compared to state-of-the-art methods.
Funder
Beijing Natural Science Foundation National Natural Science Foundation of China
Subject
Electrical and Electronic Engineering,Computer Networks and Communications,Hardware and Architecture,Signal Processing,Control and Systems Engineering
Reference50 articles.
1. Blind detection techniques for non-cooperative communication signals based on deep learning;Ke;IEEE Access,2019 2. Automatic modulation classification using convolutional neural network with features fusion of SPWVD and BJD;Zhang;IEEE Trans. Signal Inf. Process. Over Netw.,2019 3. Overview on intelligent wireless communication technology;Liang;J. Commun.,2020 4. Jiang, J., Wang, Z., Zhao, H., Qiu, S., and Li, J. (2021, January 20–23). Modulation recognition method of satellite communication based on CLDNN model. Proceedings of the IEEE 30th International Symposium on Industrial Electronics (ISIE), Kyoto, Japan. 5. Zhu, Z., and Nandi, A.K. (2015). Automatic Modulation Classification: Principles, Algorithms and Applications, John Wiley & Sons.
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献
|
|