A comparative study on classifying RF signals

Radio Frequency (RF) Fingerprinting refers to the hardware-specific characteristics that arise from the cumulative effects of tolerable imperfections in the electronic components used in the manufacturing of Radio Frequency (RF) circuits. These variations, while not affecting the end-user's experience, create a unique signature for each device. The process of RF fingerprint detection is a method of identifying a transmitting device by interpreting these distinct signal characteristics. Given that the RF fingerprint is unique to the hardware of RF circuits within communication devices such as Wi-Fi, Bluetooth, GSM, and two-way radios, replicating their RF transmissions is considered infeasible. Leveraging these inherent fingerprinting data, RF fingerprinting is employed as an additional physical security layer to supplement existing password-based authentication and authorization measures for network access. Its applications include preventing spoofing and identity theft, as well as detecting and verifying known and unknown RF transmissions in the field, particularly in military applications such as signals intelligence (SIGINT). In this study, transient detection from RF signals was performed, and classical machine learning methods were applied for signal classification, along with image-based classification and feature selection analyses.