EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Double-Directional Wireless Channel Generation Using Statistics-Informed Machine Learning

Ferdous Pervej, Pratik Patel, Koushik Manjunatha, Prasad Shamain, Andreas F. Molisch

IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation · 2026 · 2 · 100-115

DOI: 10.1109/jsteap.2026.3701014 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation, 2026. Verfasst von Ferdous Pervej und 4 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: FP and AFM, Amazon, U.S. National Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Channel models are important for the design and testing of wireless communication systems. To represent the complete propagation channel, we need a double-directional (DD) channel model describing the directions at transmitter and receiver (Rx), as well as delays and amplitudes of the multipath components. Traditional statistical and/or semideterministic models may not perform well when the evolutions of the different MPC features are considered. While machine learning (ML) has long been used for channel prediction to match channel realizations, creating models that match channel statistics is vital for system design. Motivated by this, we propose a statistics-informed ML-based complete DD channel generation process that, as a function of Rx trajectories, produces channel realizations matching the true channel statistics. More specifically, we propose a statistical-hTransformer (Stat-HTransformer) solution, which is a combination of a hybrid transformer (hTransformer) model that uses a low-rank projected attention calculation mechanism and a customized loss function that injects statistical information of the channel, to generate the complete DD channel realizations with accurate statistics matching. We consider three customized loss functions that leverage three combinations of different (channel) statistics and investigate their performances. Our extensive simulation results validate that this statistics-aided training method effectively generates future channel realizations with statistics that closely match the ground truth, and the proposed Stat-HTransformer method mostly outperforms all other baselines.

Förderung

  • FP and AFM
  • Amazon
  • U.S. National Science Foundation

Zitieren

APA

Pervej, F., Patel, P., Manjunatha, K., Shamain, P., & Molisch, A. F. (2026). Double-Directional Wireless Channel Generation Using Statistics-Informed Machine Learning. IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation, 2, 100-115. https://doi.org/10.1109/jsteap.2026.3701014

Vancouver

Pervej F, Patel P, Manjunatha K, Shamain P, Molisch AF. Double-Directional Wireless Channel Generation Using Statistics-Informed Machine Learning. IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation. 2026;2:100-115. doi:10.1109/jsteap.2026.3701014

BibTeX
@article{Pervej2026,
  author = {Pervej, Ferdous and Patel, Pratik and Manjunatha, Koushik and Shamain, Prasad and Molisch, Andreas F.},
  title = {Double-Directional Wireless Channel Generation Using Statistics-Informed Machine Learning},
  journal = {IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation},
  year = {2026},
  volume = {2},
  pages = {100--115},
  doi = {10.1109/jsteap.2026.3701014}
}
Kostenlos starten →