Device Fingerprinting in Power Line Communications

Power Line Communication (PLC) use existing electrical infrastructure for data transmission but are susceptible to security threats such as spoofing and impersonation attacks due to their open nature. This paper proposes a novel Device Fingerprinting (DF) approach for device authentication in PLC systems. The approach leverages hardware-induced imperfections in signals transmitted over power lines to identify devices based on their physical-layer characteristics. We develop a methodology that converts raw In-Phase Quadrature (IQ) samples from PLC channels into images, enabling the use of Convolutional Neural Networks for device classification. Our approach demonstrates the feasibility of CNN-based DF in PLC environments using only physical-layer information from received signals. Our experimental validation uses 8 Software Defined Radios and 2 power line couplers in real-world PLC measurements. We evaluate multiple Convolutional Neural Network (CNN) architectures and demonstrate that the PLC device fingerprint consists of two components: radio-specific and coupler-specific characteristics. The results show classification accuracy exceeding 0.9 across different configurations, establishing the viability of DF-based authentication in PLC systems without requiring additional security layers.