Exploring Practical Vulnerabilities of Machine Learning-based Wireless Systems
summary
RAFA (RAdio Frequency Attack) implements a robust wireless UAP that models channel distortion and sample misalignment to craft over-the-air perturbations. A single-antenna attacker (USRP + router) transmitting 20x less power than the client and without synchronization causes up to ~7 dB SNR loss against FIRE (ML channel prediction) and increases D-LOC localization error by ~0.7 m median and nearly 3 m at the 90th percentile across 16 client locations, and adversarial training with RAFA examples partially restores performance.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:32Machine Learning-driven Wireless Systems
- 1:22ML Models are Vulnerable to Attacks
- 2:18Wireless Systems are Critical Infrastructure
- 4:07Problem Setup
- 5:38Traditional Adversarial Attacks
- 6:55Challenge 1: The attacker doesn't know the client channel H
- 8:06Challenge 2: Attacker's Noise Gets Distorted
- 10:09Modeling Transformations
- 12:14RAFA's Algorithm: Robust Wireless UAP
- 13:05Attacker Hardware Design
- 13:20Deployment
- 13:38Attacking FIRE (Communication)
- 15:12Attacking Dloc (Localization)
- 16:55Benefits of Robust Training
related talks
17mA Systematic Threat Analysis and Practical Attacks on Automated Frequency Coordination
59mWireless security testing with attack
17mOh No, My RAN! Breaking Into an O-RAN 5G Indoor Base Station
44mManna from Heaven: Improving the State of Wireless Rogue AP Attacks
44m"Attacking industrial remote controllers for fun and profit"
27mWIFI-Important Remote Attack Surface: Threat is Expanding
42mYour Wi-Fi Is the Eavesdropper’s Radar: How to Counter Privacy Threats...
17mADR-X: ANN-Assisted Wireless Link Rate Adaptation for Compute-Constrained Embedded...