Machine Learning Techniques for Blind Beam Alignment in mmWave Massive MIMO.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Millimeter wave (mmWave) massive MIMO systems require efficient beam alignment (BA) for reliable communication.
• Traditional BA methods often involve high pilot overhead, limiting system efficiency.
• Analog architectures in mmWave systems present unique challenges for beam management.

Purpose of the Study:

• To propose low-complexity Machine Learning (ML) models for efficient beam alignment (BA) in single-user, uplink, fully analog mmWave massive MIMO systems.
• To reduce pilot overhead by developing a data-driven and model-based approach for partial and blind beam sounding.
• To investigate the effectiveness of ML models in predicting un-sounded beams using Received Signal Energies (RSEs).

Main Methods:

• Utilizing a partially and blindly sounded subset of beams from large codebooks at both the User Equipment (UE) and Base Station (BS).
• Training ML models including low-rank Matrix Factorization (MF), non-negative MF (NMF), and shallow Multi-Layer Perceptron (MLP) on the sampled beam data.
• Employing Received Signal Energies (RSEs) for blind beam alignment, eliminating the need for Channel State Information (CSI).

Main Results:

• Accurate prediction of un-sounded beams was achieved by training ML models on only 10% of the total beams.
• The proposed ML-based BA methods demonstrated effectiveness across various transmitted power regimes.
• Performance remained robust even as codebook dimensions increased from 128x128 to 1024x1024 at both UE and BS.

Conclusions:

• The proposed ML-based beam alignment offers a significant reduction in pilot overhead for mmWave massive MIMO systems.
• Low-complexity ML models can effectively predict beam patterns, enabling efficient communication in analog architectures.
• This approach provides a scalable and accurate solution for beam management in next-generation wireless networks.