Machine-learning-based method for fiber-bending eavesdropping detection.

Area of Science:

• Optical Engineering
• Cybersecurity
• Machine Learning

Background:

• Fiber optic networks are vulnerable to physical eavesdropping.
• Detecting unauthorized access is crucial for secure data transmission.
• Existing methods may require specialized equipment or link modifications.

Purpose of the Study:

• To develop an effective and non-intrusive method for detecting fiber-bending eavesdropping.
• To leverage machine learning for classifying normal and eavesdropping events in optical signals.

Main Methods:

• Extraction of 5-dimensional features from time-domain optical signals.
• Application of a long short-term memory (LSTM) network for signal classification.
• Experimental validation on a 60 km single-mode fiber link with a clip-on coupler.

Main Results:

• Achieved a 95.83% detection accuracy for fiber-bending eavesdropping.
• Demonstrated successful classification of normal and eavesdropping events.
• Validated the scheme's effectiveness in a realistic transmission scenario.

Conclusions:

• The proposed feature extraction and LSTM-based scheme offers high accuracy in detecting fiber-bending eavesdropping.
• The method's reliance on time-domain signal analysis eliminates the need for additional devices or special link designs.
• This approach enhances the security of fiber optic communications without compromising network infrastructure.