基于时频融合的深度学习调制识别算法

自动调制识别(Automatic Modulation Recognition,AMR) 能够在缺少先验信息的条件下，识别出接收信号的调制类型，在非合作通信中起着至关重要的作用。为提高调制识别的准确率，提出了一种基于时频融合的深度学习调制识别算法。该算法将调制信号的时频图作为网络的输入，使用一维卷积分别提取信号的时频特征，并通过计算时频维度上的权重来突出重要的时频信息，使网络学习到更具区分度的时频特征。为了充分利用时频特征之间的互补性和相关性，使用了基于压缩和激励网络(Squeeze-and-Excitation Network，SENet)的时频特征融合策略。利用该网络对11种调制类型进行识别，实现了最高92.5%的识别准确率；在0 dB以上时，平均识别准确率达到90.87%，优于其他的深度学习算法。

A Deep Learning Modulation Recognition Algorithm Based on Time-Frequency Fusion

Automatic modulation recognition(AMR) can identify the modulation type of the received signal without a priori information,and plays a vital role in non-cooperative communication.In order to improve the accuracy of modulation recognition,a deep learning modulation recognition algorithm based on time-frequency fusion is proposed.The algorithm takes the time-frequency diagram of the modulated signal as the input of the network,uses one-dimensional convolution to extract the time-frequency characteristics of the signal respectively,and highlights the important time-frequency information by calculating the weight in the time-frequency dimension,so that the network can learn more differentiated time-frequency features.In order to make full use of the complementarity and correlation between time-frequency features,a time-frequency feature fusion strategy based on Squeeze-and-Excitation Network(SENet) is used.Using this network,11 modulation types are recognized,and the recognition accuracy is up to 92.5%.Above 0 dB,the average recognition accuracy reaches 90.87%,which is better than that of other deep learning algorithms.