ECU Voltage Fingerprinting for Ground Truth Message Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Establishing ground truth for electronic control units (ECUs) in communication networks is challenging, especially after context shifts, due to the need for retraining machine learning (ML) models without prior knowledge of all message identifications (MIDs), which is complicated by proprietary information and overlapping MID signatures.
Innovation Solution
An ECU identification device with processing circuitry and observation circuitry collapses overlapping MIDs into a single ECU label, allowing the ML model to be trained or retrained without prior knowledge of all MIDs, using voltage signatures to infer ECU labels and iteratively update the mapping between MIDs and ECU labels until the desired recall threshold is met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained to infer ECU labels from voltage signatures, then the ability to identify ECUs improves, but the complexity of establishing ground truth increases due to overlapping MID signatures and proprietary information limitations
Solution Approach 1:
The patent introduces an intermediary process that uses voltage signatures as a mediator between message identifications (MIDs) and ECU labels. Instead of directly mapping MIDs to ECU labels, the system uses voltage signatures observed during message transmission as an intermediate representation that captures physical characteristics of the transmitting ECU, thereby resolving the complexity of establishing ground truth while maintaining identification accuracy
Solution Approach 2:
The system creates a copy or representation of ECU identity through voltage signatures rather than relying on proprietary ECU identification data. By capturing and analyzing the electrical characteristics (voltage signatures) of message transmissions, the system generates a fingerprint that serves as a surrogate for the actual ECU identity, eliminating the need for direct access to proprietary ECU information
2Productivity
If the system collapses overlapping MIDs into a single ECU label, then the training process becomes simpler and faster, but the precision of message source identification may be reduced
Solution Approach 1:
The patent changes the parameter used for identification from message identification (MID) codes to voltage signature characteristics. By measuring and comparing voltage parameters during message transmission, the system can distinguish between different ECUs even when they use overlapping or identical MIDs, thereby maintaining identification precision while enabling more efficient training through collapsed ECU labels
3Reliability
If the ML model is retrained after context shifts, then the reliability of intrusion detection is improved, but the time and computational resources required increase
Solution Approach 1:
The system performs preliminary characterization of ECUs by capturing voltage signatures during normal operation and using these to establish ground truth before intrusion detection is needed. By pre-training the machine learning model with voltage signature data from legitimate ECUs, the system prepares the detection mechanism in advance, reducing the need for frequent retraining while maintaining reliability when context shifts occur
Data Source
AI summary
Systems, apparatuses, and methods to establish ground truth for an intrusion detection system using machine learning models to identify an electronic control unit transmitting a message on a communication bus, such as an in-vehicle network bus, are provided. Voltage signatures for overlapping message identification (MID) numbers are collapsed and trained on a single ECU label.