Free space optic channel monitoring using machine learning

Area of Science:

• Optical Communication
• Machine Learning Applications

Background:

• Free Space Optics (FSO) transmits signals through air, unlike fiber optics, making it susceptible to various impairments.
• Predicting these impairments is crucial for automated diagnostics and adaptive network management in FSO systems.
• While machine learning is established for fiber optics, its application to FSO impairment prediction is nascent.

Purpose of the Study:

• To predict three key FSO channel parameters: amplified spontaneous emission (ASE) noise, turbulence, and pointing errors.
• To be the first study to predict FSO channel parameters considering multiple impairments simultaneously.
• To evaluate different feature extraction methods and machine learning models for FSO impairment prediction.

Main Methods:

• Utilized asynchronous amplitude histogram (AAH) and asynchronous delay-tap sampling (ADTS) histogram features for parameter prediction.
• Compared the predictive performance of Support Vector Machine (SVM) regressor and Convolutional Neural Network (CNN) regressor.
• Investigated the CNN regressor's capability across three different transmission speeds.

Main Results:

• ADTS histogram features demonstrated superior prediction accuracy compared to AAH features.
• CNN regressor showed comparable or superior performance to SVM regressor in predicting FSO parameters.
• CNN achieved good performance for Optical-to-Signal Noise Ratio (OSNR) prediction irrespective of transmission speed.
• Prediction accuracy for turbulence and pointing errors was higher at lower transmission speeds.

Conclusions:

• ADTS features are effective for FSO channel parameter prediction.
• CNNs are a viable and often superior model for FSO impairment prediction compared to SVMs.
• FSO channel parameter prediction accuracy, particularly for turbulence and pointing errors, is influenced by transmission speed, with lower speeds yielding better results.