Machine learning-based classification of structured light modes under turbulence and eavesdropping effects.

Area of Science:

• Optical communications
• Machine learning
• Free-space optics

Background:

• Multiplexed structured light modes offer potential for enhanced data transmission.
• Free-space optic (FSO) systems face challenges from turbulence and eavesdropping.
• Classifying these modes is crucial for reliable FSO communication.

Purpose of the Study:

• To classify multiplexed structured light modes in FSO systems.
• To evaluate the impact of turbulence and interception threats on mode classification.
• To develop and compare machine/deep learning algorithms for this classification task.

Main Methods:

• An experimental 3-m FSO system was used to transmit 16 modes (8-ary Laguerre Gaussian and 8-ary superposition LG).
• Four machine/deep learning algorithms were employed: artificial neural network, support vector machine, 1D CNN, and 2D CNN.
• A fusion approach combining outputs from these algorithms was utilized.

Main Results:

• Classification accuracy exceeded 92% in weak turbulence, 81% in moderate turbulence, and 69% in strong turbulence.
• The study is the first to concurrently address turbulence and interception threats in structured light mode classification.
• The fused model demonstrated robust performance across varying turbulence levels.

Conclusions:

• Multiplexed structured light modes show significant potential for reliable, high-capacity data transmission in FSO systems.
• Machine and deep learning algorithms are effective for classifying these modes under challenging conditions.
• The proposed classification method enhances communication reliability in turbulent FSO channels.