一种基于深度学习的辐射源信号调制识别新算法

针对传统的辐射源信号调制识别方法需要大量特征提取的问题,提出一种基于深度学习的辐射源信号自动调制识别算法,该算法通过对辐射源信号进行幅-相域二维图像表征,基于卷积神经网络实现层次化地理解和识别电磁信号。仿真结果表明:相比基于时域的传统信号调制识别算法,所提算法在中、高信噪下识别率分别提升了2.5%和2.3%,单信号的识别时间不大于0.1 ms。

A Novel Emitter Signal Modulation Classification Method with Deep Learning

Due to the increasing usage of wireless devices, the electromagnetic environment tends to be crowded with various types of radiation signals. It becomes harder to effectively classify signals from different emitters in such environment. The traditional signal analysis method based on feature extraction has been far from unable to meet the requirements of identification. This article proposed a novel emitter signal modulation classification method using deep learning. In this method, signals of different modulation were represented by amplitude-phase images. A deep convolutional neural network was proposed to extract features from these images, and the modulation types were classified. Experiment results show that using the amplitude-phase representation for recognizing modulation could lead to performance improvements up to to 2.5% and 2.3% for medium and high SNR compared to IQ data. The prediction time is around 0.1ms for a single sample signal.