Sampling Point 기반 피처를 이용한 CAN 네트워크 내 ECU 식별 기법

다국어 초록 (Multilingual Abstract)

CAN(Controller Area Network) is widely used for communication protocol among ECUs(Electronic Control Units) in modern vehicles. However, since CAN protocol was not designed with security considerations, it lacks security mechanisms for sender authentication and identification. This structural vulnerability allows adversaries can inject malicious messages, threatening vehicle safety. Existing methods typically identify the transmitting ECU based on unique electrical or timing characteristics, but they are ineffective for ECUs that only receive messages without transmitting. To address this limitation, this paper proposes a novel SP(Sampling Point)-based feature for ECU identification, exploiting unique timing characteristics observed during the reception of CAN frames. The proposed approach injects precisely timed bit-flip signals at the bit-level, collects offset-dependent error responses resulting from ECU-specific SP differences, and extracts statistical features to train a machine-learning classifier. Experimental results demonstrate distinct SP-based feature patterns among identically configured boards, achieving over 90% identification accuracy with a Random Forest classifier.