Physical Layer Discrimination of Electronic Control Units Using Wired Signal Distinct Native Attribute (WS-DNDA)

The Controller Area Network (CAN) bus is a communication system used in automobiles to interconnect the electronic components required for critical vehicle operations. These components are called Electronic Control Units (ECU) and each one exercises one or more functions within the vehicle. ECUs can provide autonomous safety features and increased comfort to drivers but these advancements may come at the expense of compromised vehicle security. Researchers have shown that the standard automobile CAN bus can be hacked by 1) compromised authorized ECUs or 2) by unauthorized devices, or ECUs, that have been physically connected. Physical layer (PHY) device fingerprinting has emerged as one accepted approach to establishing vehicle security. This work investigates the application of AFIT’s Wired Signal Distinct Native Attribute (WS-DNA) Fingerprinting using Multiple Discriminant Analysis Maximum Likelihood (MDA/ML) to achieve ECU discrimination. Demonstrations include 4-Class Cross Lot Discrimination (CLD) assessments with four Toyota Avalon ECUs with the same part number but different lot numbers as well as 9-Class Like Model Discrimination (LMD) assessments with nine Toyota Avalon ECUs of the same make and model as authorized devices. Rogue Arduino, Beagle Board, and CANable USB to CAN bus adapter are introduced and Rogue Reject Rate (RRR) estimated. Using WS-DNA features, RRR = 100% for rogue devices presenting false credentials for both the four class and nine class problem. Specific performance for compromised authorized ECU access attempts included 98% ≤ RRR ≤ 100% for the 4-class CLD assessment and 35.2% ≤ RRR ≤100% for the LMD assessment. Additionally, the Average Percent Correct Classification (%C) benchmark of %C = 90% was achieved for authorized devices at SNR∆ ≥ -8 dB for the 4-class CLD and at collected con-