Probabilistic Clustering With Deep Neural Network for Radio Frequency Fingerprinting

Abstract

Radio Frequency Fingerprint (RFF) is a hardware-based attribute of wireless devices arising from inherent imperfections. It plays a crucial role in wireless communication by enabling security and sensing applications. However, traditional methods face significant challenges in dynamic environments and with unfamiliar devices, necessitating a versatile and scalable solution. This work proposes leveraging a Deep Neural Network (DNN) with self-supervised learning for robust fingerprint extraction and soft clustering through an Infinite Gaussian Mixture Model (IGMM) with thresholding for cluster updates. IGMM facilitates adaptive clustering, allowing new signals to join existing clusters or form new ones based on their similarity to previously observed fingerprints. This probabilistic approach enables continuous learning and adaptation, providing a flexible framework for real-time signal filtering in opportunistic wireless environments.