基于特征融合的自动调制识别算法
Automatic Modulation Recognition Algorithm Based on Feature Integration
朱敏 1陈慧贤 2王国华 1张鹏1
作者信息
- 1. 中国人民解放军陆军炮兵防空兵学院信息工程系,安徽 合肥 230031
- 2. 中国人民解放军陆军炮兵防空兵学院信息工程系,安徽 合肥 230031;偏振光成像探测技术安徽省重点实验室,安徽 合肥 230031
- 折叠
摘要
目前大多数基于神经网络的调制识别算法,只使用时域或频域的单一信息来源,忽视利用多个变换域信息特征进行优势互补.提出基于特征融合的深度学习自动调制识别算法,可有效改善只利用时域或频域单一信息来源的调制识别效果.上述算法包含时频特征提取模块,将信号在不同变换域中的特征进行融合,然后采用基于注意力机制的长短期记忆网络和全连接层进行分类,通过多个变换域信息的特征融合,实现了优势互补.仿真结果表明,相比传统的深度学习调制识别算法,基于特征融合的自动调制识别算法能够有效地提取信号特征,具有更高的识别准确度.
Abstract
Most of the present neural network-based modulation recognition algorithms use the time domain or frequency domain as the single informationsource and ignore the use of multiple transform domaininformation features for complementary advantages.This article proposes a deep learning automatic modulation recognition algorithm based on feature integration,which can effectively improve the modulation recognition effect of by using time domain or fre-quency domain as the single information source.The algorithm includes the time-frequency feature extraction module,which integrates the features of signals in different transform domains.Then,the long short-term memoryneu-ral network based on the attention mechanism and the full connection layer were used for classification.Complementa-ry advantages were achieved through the feature integration of multiple transform domain informations.The simulation experiments show that the automatic modulation recognition algorithm based on feature integration can effectively ex-tract the signal features with improved recognition accuracy,compared with that of traditional deep learning modulation recognition algorithm.
关键词
自动调制识别/特征融合/深度学习/卷积神经网络Key words
Automatic modulation recognition/Feature integration/Deeplearning/Convolutional neural network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据