RFF Recognition Based on Adaptive Signal Transform and Weighted Feature Fusion

Radio Frequency Fingerprinting (RFF) is an effective physical layer identification technique that makes Internet of Things (IoT) systems more secure. It uses the deviation of the parameters of each electronic component in the transmitter from the nominal parameters to perform the authentication task. RFF recognition schemes based on deep learning (DL) are emerging due to their ability to automatically perform feature extraction and classification. To fully exploit the information in RF signals, researchers perform feature extraction from different signal representations and fuse these features for subsequent classification tasks. However, selecting the combination of signal representations is a time-consuming task. Therefore, in this paper, we propose an RFF recognition model based on Adaptive Signal Transform (AST) and Weighted Feature Fusion (WFF). The AST module is able to learn a signal transform form, which can greatly improve recognition accuracy and reduce the pressure of manual selection of signal representations. In addition, we use multiple convolution layers to form the WFF module. This module weighs the RFF features and highlights the feature components that favor classification. Simulation results show the ability of these modules to improve the performance of RFF recognition and the effectiveness of the proposed AST-WFF model.