5G ML Research: Beam Selection

Problem

In 5G mmWave networks, selecting the optimal beam pair between the base station and user equipment is critical but incurs high overhead if done via exhaustive search.

Approach

I investigated using Machine Learning to predict the best beam based on location and channel state information (CSI).

1. Dataset: Used the DeepMIMO dataset.
2. Models: Trained Random Forest and Neural Network models.
3. Evaluation: Compared top-k accuracy and beam search time reduction.

Tools

• Python: Data science stack.
• PyTorch: Deep learning models.
• Scikit-learn: Traditional ML models.

Output & Impact

• Achieved 90% top-3 accuracy in beam prediction.
• Reduced the beam search overhead by approximately 70% compared to exhaustive search.

What I Learned

• Data preprocessing is the most critical step in any ML pipeline.
• There is always a trade-off between model accuracy and inference latency, especially in real-time systems like 5G.