Search for: All records

As the market for autonomous vehicles advances, a need for robust safety protocols also increases. Autonomous vehicles rely on sensors to understand their operating environment. Active sensors such as camera, LiDAR, ultrasonic, and radar are vulnerable to physical channel attacks. One way to counter these attacks is to pattern match the sensor data with its own unique physical distortions, commonly referred to as a fingerprint. This fingerprint exists because of how the sensor was manufactured, and it can be used to determine the transmitting sensor from the received waveform. In this paper, using an ultrasonic sensor, we establish that there exists a specific distortion profile in the transmitted waveform called physical fingerprint that can be attributed to their intrinsic characteristics. We propose a joint time-frequency analysis-based framework for ultrasonic sensor fingerprint extraction and use it as a feature to train a Naive Bayes classifier. The trained model is used for transmitter identification from the received physical waveform.