The Impact of Different Kernel Functions on the Performance of Scintillation Detection Based on Support Vector...

Area of Science:

• Space Physics
• Signal Processing
• Machine Learning

Background:

• Ionospheric scintillation, caused by electron density irregularities, disrupts Global Navigation Satellite Systems (GNSS) signals.
• This disruption severely impacts GNSS receiver performance in signal acquisition, tracking, and positioning.
• Machine learning algorithms are increasingly applied for detecting and mitigating scintillation effects.

Purpose of the Study:

• To evaluate the performance of Support Vector Machines (SVM) for detecting ionospheric scintillation.
• To analyze the impact of different SVM kernel functions (linear, Gaussian, polynomial) on detection accuracy.
• To determine the optimal kernel function selection for reliable scintillation detection.

Main Methods:

• Utilized real-world GNSS signal data affected by scintillation from SANAE IV (Antarctica) and Hanoi (Vietnam).
• Implemented and compared SVM algorithms with linear, Gaussian (coarse, medium, fine), and third-order polynomial kernels.
• Statistically assessed performance using probabilities of detection and false alarm rates, considering data preparation and cross-validation.

Main Results:

• Fine Gaussian SVM demonstrated superior overall accuracy compared to linear and other Gaussian kernels.
• The third-order polynomial kernel showed improved performance over linear and coarser Gaussian kernels.
• Linear SVM exhibited the lowest complexity and fastest running time.

Conclusions:

• SVM is a viable machine learning approach for ionospheric scintillation detection using GNSS signals.
• Fine Gaussian SVM provides the best accuracy, while linear SVM offers computational efficiency.
• Kernel function selection involves a trade-off between detection accuracy and computational complexity.