Convolutional neural network for detecting fiber-bending eavesdropping attacks in optical communication systems.

Area of Science:

• Optical Communications
• Cybersecurity
• Machine Learning

Background:

• Optical fiber networks transmit over 90% of global data, making them critical infrastructure.
• Fiber-bending eavesdropping attacks pose a significant threat, leading to potential massive data leaks.
• Detecting these attacks is crucial for maintaining data security in modern communication systems.

Purpose of the Study:

• To propose and validate a novel detection scheme for fiber-bending eavesdropping attacks.
• To employ a convolutional neural network (CNN) for enhanced detection capabilities.
• To secure dual-polarization coherent optical communication systems against physical intrusion.

Main Methods:

• Established the relationship between fiber-bending radius and bending power loss coefficient.
• Constructed a 100 km, 25 GBaud dual-polarization 16 quadrature amplitude modulation (16QAM) coherent optical system.
• Collected normal and fiber-bending eavesdropping polarization data for analysis.

Main Results:

• The CNN achieved high detection accuracy rates for various bending radii: 99.8% (10.8 mm), 99.67% (12.1 mm), and 99.22% (15 mm).
• Detection accuracy for mixed bending radii reached 96.15%.
• Demonstrated the effectiveness of the CNN-based scheme in identifying malicious fiber manipulation.

Conclusions:

• The proposed CNN-based detection scheme is highly effective for identifying fiber-bending eavesdropping attacks.
• This method offers a robust solution for enhancing data security in dual-polarization coherent optical systems.
• The findings contribute to securing critical communication infrastructure against physical security threats.