Wireless Signal Identification in 230MHz Band Based on Interference Cleaning and Convolutional Neural Network

Abstract

With the development of digital wireless communication technol-ogy, the wireless signal identification has been suffering from increasingly complex electromagnetic environment and higher spectrum utilization. In this paper, we propose a wireless signal identification method based on interference cleaning and convolutional neural network (CNN) in 230MHz Band. The method firstly analyzes the received signal in time domain, building feature data sets combined with amplitudes, phases, in-phase components and orthogonal components. The method then generalizes singular value decomposition(SVD) and subspace division to preserve signal subspace, eliminate noise subspace and interference compress subspace. Finally, it utilizes the data set to train the CNN and make the wireless signals' identification through the well-trained the CNN. The experimental results with different kinds of modulation show that this method can achieve high recognition accuracy and strong anti-noise ability.