Abstract
As concerns grow over the security of the Controller Area Network (CAN) bus used in modern cars, researchers are developing technologies to mitigate potential cyberattacks on it. Some of those technologies rely on CAN messages’ arrival timestamp (i.e., the point of time when a receiving node first observes the CAN frame) to measure slight differences between the clock references of Electronic Control Units (ECUs). Those technologies are promising, but several unknown factors still limit their efficiency and accuracy. Therefore, we investigated the properties of arrival timestamps of CAN messages and studied how network properties can impact them. First, we established a mathematical model for the case of two separate ECUs sending a message each. Then, we verified that model with a network simulator that can simulate an arbitrary number of ECUs sending arbitrary messages with different clock references. Finally, we verified those results on actual hardware and observed arrival timestamps for various scenarios. We concluded that the standard deviation of the arrival timestamp of periodic messages is a dynamic source of information about the physical network, rather than a static property of the sending ECU, as usually modeled. Those results can help researchers improve technologies that rely on CAN messages’ arrival timestamps.
