LECS
Laboratoire pour les systèmes informatiques émergents
Université Concordia · Montréal
conference

Towards Massively Parallel HyperDimensional Computing Architectures for Intelligent Satellite Links

Mohsen Asghari, Masoud Rahimi, Sébastien Le Beux, Otmane Ait Mohamed, Cos Modafferi, Ron Mankarious
CF '26: Proceedings of the 22nd ACM International Conference on Computing Frontiers · 2026
Accélérateurs d'IA embarquéePrototypage sur FPGA
CF '26: Proceedings of the 22nd ACM International Conference on Computing Frontiers 2026 Mohsen Asghari, Masoud Rahimi, Sébastien Le Beux, Otmane Ait Mohamed, Cos Modafferi, Ron Mankarious
Résumé

Hubless full-mesh VSAT modems require rapid receiver-gain adaptation under dynamic channel impairments, where conventional rule-based automatic gain control (AGC) becomes less effective at high symbol rates. This paper presents a Hyperdimensional Computing (HDC) accelerator architecture for VSAT modems. The proposed method integrates a policy-based reinforcement learning algorithm for AGC together with a configurable encoder and similarity-checking accelerator. The design enables low-latency, high-throughput online decision making for burst-mode satellite networks. The accelerator achieves a throughput of 380k inferences/s and an energy efficiency of 15.2 GIPS/W, while reducing bit errors by 89% compared to the baseline under real-world conditions.

Citation

Si vous citez ces travaux, merci d'utiliser l'entrée ci-dessous. Vous pouvez copier le BibTeX dans le presse-papier via le bouton en haut de page.

@inproceedings{mohsen2026TowardsMassivelyParallelHyperDimensionalComputingArchitecturesforIntelligentSatelliteLinks2026,
  title  = {Towards Massively Parallel HyperDimensional Computing Architectures for Intelligent Satellite Links},
  author = {Mohsen Asghari and Masoud Rahimi and Sébastien Le Beux and Otmane Ait Mohamed and Cos Modafferi and Ron Mankarious},
  booktitle = {CF '26: Proceedings of the 22nd ACM International Conference on Computing Frontiers},
  year   = {2026}
}

Remerciements

Ces travaux ont été soutenus en partie par le Conseil de recherches en sciences naturelles et en génie du Canada (CRSNG) et par le Fonds de recherche du Québec — Nature et technologies (FRQNT).