Environment-Aware Proactive Beam Prediction in mmWave V2I via Multi-Modal Prior Mask Map.

Area of Science:

• Wireless Communication
• Computer Vision
• Machine Learning

Background:

• Millimeter wave (mmWave) Vehicle-to-Infrastructure (V2I) systems require precise beam prediction for optimal performance.
• Existing single-modal methods lack comprehensive environmental data, while current multi-modal approaches underutilize stable environmental features.

Purpose of the Study:

• To develop an environment-aware proactive beam prediction method for mmWave V2I communication.
• To improve beam prediction accuracy by effectively integrating diverse data sources and stable environmental features.

Main Methods:

• Proposes a multi-modal prior mask map (MMPMM) integrating offline mapping with an online beam prediction network.
• Fuses image, point cloud, position, and MMPMM data for optimal beam index prediction.
• MMPMM extracts static V2I scene features offline, providing channel-related prior information without online overhead.

Main Results:

• Achieved a Top-3 beam prediction accuracy of up to 71.23% on real-world datasets.
• Demonstrated stable performance under dynamic and degraded environmental conditions.
• Effectively leveraged complementary information from multiple data modalities and static scene features.

Conclusions:

• The proposed MMPMM-based method significantly enhances beam prediction accuracy in mmWave V2I systems.
• Integrating offline environmental priors with online multi-modal data offers a robust solution for V2I communication.
• The approach effectively addresses limitations of existing methods by utilizing stable environmental features.