Abstract

Millimeter-wave (mmWave) communication systems rely on narrow beams to achieve sufficient receive signal power. Adjusting these beams is typically associated with large training overhead, which becomes particularly critical for highly-mobile applications. Beam selection can benefit from the knowledge of user positions to reduce the overhead in mm Wave beam training. Prior work, however, studied this problem using only synthetic data that does not accurately represent real-world measurements. In this paper, we revisit the position-aided beam prediction problem in light of real-world measurements with commercial-off-the-shelf GPS to derive insights into how much beam training overhead can be saved in practice. We also compare algorithms that perform well in synthetic data but fail to generalize with real data, and attempt to answer what factors cause inference degradation. Further, we propose a machine learning evaluation metric that better captures the end communication system objective. This work aims at closing the gap between reality and simulations in position-aided beam alignment.

Videos

image section

Bibtex

          @INPROCEEDINGS{10278998,
    author={Morais, João and Bchboodi, Arash and Pezeshki, Hamed and Alkhateeb, Ahmed},
    booktitle={ICC 2023 - IEEE International Conference on Communications}, 
    title={Position-Aided Beam Prediction in the Real World: How Useful GPS Locations Actually are?}, 
    year={2023},
    volume={},
    number={},
    pages={1824-1829},
    keywords={Training;Measurement;Neural networks;Probability;Predictive models;Prediction algorithms;Power system reliability},
    doi={10.1109/ICC45041.2023.10278998}}