Physical Layer Authentication Based on Integrated Semi-Supervised Learning in Wireless Networks for Dynamic Industrial Scenarios

Supervised machine learning algorithms are not enough to establish a low-complexity, lightweight, and high-security authentication mechanism in wireless communication systems for industrial scenarios where only the physical layer channel information of legitimate nodes is known. Therefore, this paper establishes a new physical layer authentication scheme based on an integrated semi-supervised algorithm around the positive-unlabeled (PU) learning and bootstrap aggregating (bagging) strategy, which can accurately distinguish legal nodes from illegal nodes in unmarked channel information when only the channel information label of legal nodes is known. According to the characteristics of the high dimension of industrial scene data, the characteristics of the original physical layer channel information data are extracted. Using the characteristics of amplitude, phase, carrier frequency offset (i.e. phase change), and variance, the high-dimensional complex channel information matrix is transformed into a seven-dimensional feature set, which not only reduces the dimension but also retains the original complex information. Finally, the effectiveness of feature extraction was verified by using public data set collected by the National Institute of Standards and Technology (NIST) in a real industrial scene. Simulation results show that the authentication strategy proposed in this paper could accurately distinguish legal nodes from illegal nodes with unmarked channel information in dynamic industrial scenarios.