Beam-Selection for 5G/B5G Networks Using Machine Learning: A Comparative Study.

Area of Science:

• Wireless Communications
• Machine Learning
• Signal Processing

Background:

• Millimeter wave (mmWave) communication faces significant challenges in 5G/B5G, particularly beam selection, due to inherent signal attenuation and penetration losses.
• Exhaustive search for beam selection in vehicular mmWave links is impractical due to short contact times.
• Machine learning (ML) offers a promising solution to enhance the complexity of cellular network construction.

Purpose of the Study:

• To comparatively study various ML methods for solving the mmWave beam selection problem in vehicular scenarios.
• To improve the accuracy of existing ML-based solutions for mmWave beam selection.
• To introduce novel ensemble learning techniques for enhanced performance.

Main Methods:

• Comparative analysis of different machine learning algorithms applied to a standard mmWave vehicular dataset.
• Dataset augmentation through the generation of synthetic data.
• Development and application of a custom ensemble learning method.

Main Results:

• Achieved an approximate 30% increase in accuracy compared to baseline methods.
• Extended the dataset with synthetic data to improve model robustness.
• Attained approximately 94% accuracy using the proposed ensemble learning approach.

Conclusions:

• Ensemble learning, combined with synthetic data augmentation, significantly enhances beam selection accuracy in mmWave vehicular communications.
• The developed custom ensemble method provides a novel and effective solution for this critical 5G/B5G challenge.
• This work demonstrates the potential of ML to overcome mmWave communication limitations.