UAV Communication Signal Recognition: A New Feature Representation and Deep-Learning Method

Abstract

As the threats from unmanned aerial vehicles (UAVs) increases gradually, to recognize and classify unknown UAVs have became more and more important in both civil and military security fields. Classification of signal modulation types is one of the basic techniques for specific UAV recognition. In this paper, to represent the hidden features involved in the transmitted signals from UAVs, we propose a two-dimensional squeezing transform (TDST) to characterize the UAV communication signals in a compressed time-frequency plane. The new time-frequency representation, TDST, retains the instantaneous characteristics of the UAV signal, and is with excellent data reduction and noise suppression capabilities. The TDST plane of different modulation types are then considered as input features, and we propose a convolutional neural network (CNN) based on deep-learning to recognize the UAV signals. We design an interception system and consider 10 types of UAV signals with random initial phase, bandwidth and frequency offset. Experimental results demonstrate the effectiveness and superiority of the proposed algorithm.