Voltage Based Electronic Control Unit (ECU) Identification with Convolutional Neural Networks and Walsh–Hadamard Transform

Abstract

This paper proposes an identification approach for the Electronic Control Units (ECUs) in the vehicle, which are based on the physical characteristics of the ECUs extracted from their voltage output. Then, the identification is not based on cryptographic means, but it could be used as an alternative or complementary means to strengthen cryptographic solutions for vehicle cybersecurity. While previous research has used hand-crafted features such as mean voltage, max voltage, skew or variance, this study applies Convolutional Neural Networks (CNNs) in combination with the Walsh–Hadamard Transform (WHT), which has useful properties of compactness and robustness to noise. These properties are exploited by the CNN, and in particular, the pooling layers, to reduce the size of the feature maps in the CNN. The proposed approach is applied to a recently public data set of ECU voltage fingerprints extracted from different automotive vehicles. The results show that the combination of CNN and the WHT outperforms, in terms of identification accuracy, robustness to noise and computing times, and other approaches proposed in the literature based on shallow machine learning and tailor-made features, as well as CNN with other linear transforms such as the Discrete Fourier Transform (DFT) or CNN with the original time domain representations.