Convolutional Neural Network-Based Interference Recognition

Abstract

Interference signals are a huge threat to communication systems and interference recognition is an indispensable part in anti-interference communication. With the rapid development of artificial intelligence, interference recognition methods based on deep learning have been proposed. However, these methods do not allow networks to learn relationships among interference signals. In this paper, we propose new loss functions and a model to improve the recognition performance of interference signals without increasing complexity. Firstly, interference signals are converted into time-frequency images (TFIs) by short-time Fourier transform (STFT) and these TFIs are the input of convolutional neural network (CNN). Secondly, two CNNs with shared parameters are established and the samples of every batch in the training are divided into several pairs. Finally, through these pairs and proposed loss functions, which are Kullback-Leibler (KL) divergence and Euclidean distance of extracted features, the network can learn relationships of every category of interference and have better ability of generalization. The experiments prove our proposed approach significantly improves the recognition accuracy of interference signals.