ECU Identification Based on Voltage Signals Using Complex Value Neural Network

Nowadays, vehicles offer many personalized services and comfortable experience for users, but they also present new vulnerabilities and attack vectors. While researchers have pro-posed various defensive measures, some complex attack behaviors remain difficult to counter. To address this, we propose an intrusion detection system based on CAN signal voltage values. This method creates a fingerprint for each CAN data frame's voltage signal and, by utilizing a complex-valued neural network combined with an attention mechanism, identifies the location of compromised ECUs. Experimental results on a public dataset show that when classifying CAN signals from 50 ECUs, the accuracy of this method exceeds 92%, with a 8% improvement in classification accuracy compared to other approaches.