Radio Frequency Fingerprint-Based Classification Performance Analysis with ML Models in the Presence of Hardware Impairments

In recent years, physical layer security (PLS) has gained important attention to leverage physical layer signal features with the aim of providing a complementary security approach. For this purpose, radio frequency (RF) fingerprinting is considered for classification of signals. In this study, we work on an classification task for users which experience different hardware impairments. For this purpose, we extract several features like skewness, kurtosis, autocorrelation, spectral flatness, mean and variance based on the received signal. Then, by leveraging these features, several Machine Learning (ML) models are trained. The performance of user classification is analyzed in terms of accuracy, precision, recall and F1-score. Thanks to the numerical results, we show that CatBoost ML model provides better scores compared to other ones. Also, we represent the effect of Recursive Feature Elimination method on performance.