A Loss Differentiation Method Based on Heterogeneous Ensemble Learning Model for Low Earth Orbit Satellite Networks

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Satellite networks face high bit error rates, challenging traditional Transmission Control Protocol (TCP) performance.
• Existing loss differentiation methods lack sophisticated ensemble learning and feature selection strategies.

Purpose of the Study:

• To propose a novel Loss Differentiation Method based on Heterogeneous Ensemble Learning (LDM-HEL) for low-Earth-orbit (LEO) satellite networks.
• To address the limitations of current methods in distinguishing between network congestion and wireless link losses.

Main Methods:

• Utilized Relief and mutual information algorithms for effective loss differentiation feature selection.
• Employed a heterogeneous ensemble of learners including least-squares support vector machine, decision tree, logistic regression, and K-nearest neighbor.
• Implemented a stochastic gradient descent method for optimizing individual learner weights within the ensemble.

Main Results:

• The LDM-HEL method demonstrated superior accuracy, recall rate, and F1-score in simulation scenarios.
• Significant improvements in throughput performance were observed when LDM-HEL was applied to TCP.
• Outperformed the LDM-satellite integrated model by 4.37% in accuracy, 4.55% in recall, 4.87% in F1-score, and 9.28% in throughput.

Conclusions:

• The proposed LDM-HEL is an effective solution for loss differentiation in LEO satellite networks.
• This method offers a robust approach to enhance satellite communication reliability and performance.
• LDM-HEL provides a significant advancement over existing methods, particularly for TCP performance in challenging satellite environments.