An Ensemble Learning Method for RF Identification

RF transmitter identification is a big challenge due to the increasing number of wireless devices in recent years. Traditional approaches for identification are implemented by choosing specific rapid fingerprintings manually with large number of signals, which cannot identify the unknown transmitter in few shots efficiently. In order to solve this problem, an ensemble learning method with convolutional neural networks (CNNEL) is proposed by considering both the limitation of unknown transmission data and the classical fingerprintings of transmitters' signals. The proposed method includes three kinds of classifiers trained by three different RF sample datasets, which are the RF I/Q signal, the short time Fourier transform of RF I/Q signals and the wavelet transform of RF I/Q signals. The identification is evaluated by averaging and voting the training result of all types signals. Three compared deep learning methods with eight known and two unknown transmitters for RF transmitters classification and identification are investigated. Numerical experiments suggest that the higher accuracy across a range of conditions is achieved by using the ensemble learning based method.