Temporal QBER-Based ML Framework Detects Stealthy Eavesdropping in BB84 QKD
A new machine learning approach has been developed to spot eavesdropping attacks in BB84 Quantum Key Distribution (QKD) systems, detailed in a paper on arXiv (arXiv:2608.04047). Traditional BB84 systems use a fixed Quantum Bit Error Rate (QBER) threshold of 11% for detecting such attacks, but some covert methods can bypass this limit and still threaten security. The innovative framework focuses on temporal QBER features, identifying 63 physics-informed characteristics related to burst behavior and other dynamics. Testing involved three classifiers—Random Forest, XGBoost, and SVM-RBF—across seven attack scenarios. XGBoost excelled with an accuracy of 88.01% and a macro F1 score of 0.880. This work addresses a critical weakness in QKD, enhancing quantum communication security.
Key facts
- Traditional BB84 QKD systems use a fixed 11% QBER threshold to detect eavesdropping.
- Stealthy attacks can remain below the 11% QBER threshold while compromising security.
- The proposed framework uses temporal QBER-based features for attack detection.
- 63 physics-informed temporal features are extracted, including burst behavior, temporal instability, basis-dependent asymmetry, and QBER loss interactions.
- Three classifiers were evaluated: Random Forest, XGBoost, and SVM-RBF.
- XGBoost achieved the best performance with 88.01% accuracy and a macro F1 score of 0.880.
- The framework was tested on seven eavesdropping attacks and a normal channel scenario under noisy and lossy conditions.
- The paper is available on arXiv with identifier 2608.04047.
Entities
Institutions
- arXiv