SWiST DeepNet: Spectral-Temporal Deep Network for Detecting Resonance-Based Attacks on MEMS Inertial Sensors

Area of Science:

• Cyber-Physical Systems (CPS)
• Sensor Security
• Deep Learning Applications

Background:

• Microelectromechanical systems (MEMS)-based inertial sensors are crucial for cyber-physical systems (CPS).
• These sensors are vulnerable to resonance-based acoustic interference, leading to inaccurate measurements and system instability.
• Existing methods lack robust detection of such subtle signal injection attacks.

Purpose of the Study:

• To develop a novel deep learning framework for detecting resonance-induced signal injections in MEMS inertial sensors.
• To enhance the resilience and trustworthiness of CPS against acoustic interference attacks.
• To evaluate the effectiveness of temporal and spectral features in attack detection.

Main Methods:

• A sliding-window-based deep learning framework (SWiST DeepNet) was proposed.
• The framework utilizes both temporal- and spectral-domain features for attack identification.
• Model performance was evaluated under static and dynamic conditions across varying attack levels (soft, medium, hard).

Main Results:

• The SWiST DeepNet reliably detected resonance-based attacks, even at low intensities (10-dB attack-to-signal ratio).
• Fusion of spectral and temporal features demonstrated superior performance in attack detection compared to individual feature domains.
• The framework proved effective across diverse operating conditions and attack intensities.

Conclusions:

• The proposed SWiST DeepNet significantly enhances the security of CPS against resonance-based acoustic attacks.
• The framework offers a reliable method for detecting low-intensity signal injection threats.
• This research contributes to building more resilient and trustworthy cyber-physical systems.