Deep Learning - Powered Radio Frequency Fingerprint Identification: Methodology and Case Study

Abstract

Radio frequency fingerprint identification (RFFI) is an authentication technique that identifies wireless devices by analyzing the characteristics of the received physical layer signals. In recent years, RFFI has been significantly enhanced by deep learning. A neural network (NN) is often leveraged to predict device identity. As a data-driven approach, deep learning requires the collection of large amounts of data for NN training. In addition, the RFFI system should be evaluated on datasets collected under various conditions to assess the system's robustness. However, only a few RFFI datasets are publicly available, and there are no clear guidelines for building an RFFI testbed for data collection. This paper presents a tutorial to build both closed-set and openset RFFI systems. A LoRa- RFFI testbed is created as a case study and the implementation details are described in depth. The LoRa-RFFI testbed involves 60 commercial-off-the-shelf (COTS) LoRa development boards as devices to be identified, and a USRP N210 software-defined radio (SDR) platform for physical layer signal reception. Experiments are carried out using the implemented LoRa-RFFI testbed, and the collected datasets are made publicly available online. It is anticipated that this work can aid the research community in constructing RFFI testbeds and facilitate the development of RFFI research.